R Dugad, K  Ratakonda

Given a video sequence, identify those frames that represent changes in the parts of the scene being imaged

R. Dugad, K. Ratakonda and N. Ahuja, Robust Video Shot Change Detection, IEEE Workshop on Multimedia Signal Processing, Redondo Beach, CA, December 1998, 276-278.  Full Text

R. Dugad, K. Ratakonda and N. Ahuja, Robust Video Shot Change Detection, Indian Conference on Computer Vision, Graphics and Image Processing, New Delhi, India, December 1998, 358-363. 

R Dugad

Combine spatial and temporal filtering to suppress noise in each frame of a video sequence

R. Dugad and N. Ahuja, Video Denoising by Combining Kalman and Wiener Estimates, International Conference on Image Processing, Kobe, Japan, Oct. 1999, IV-152-156  Abstract

]]> http://vision.ai.uiuc.edu/?feed=rss2&p=1077 0 http://vision.ai.uiuc.edu/?p=1072 http://vision.ai.uiuc.edu/?p=1072#comments Fri, 08 May 2009 16:55:59 +0000 myra http://vision.ai.uiuc.edu/wordpress/?p=1072

R Dugad, K Ratakonda

We present a novel iterative approach to define any multidimensional linear transform over an arbitrary shape given that we know its definition over a hypercube

LINEAR TRANSFORMS OVER ARBITRARY SUPPORTS

Krishna Ratakonda and Narendra Ahuja
Beckman Institute for Advanced Science…

R Dugad, K Ratakonda

To develop a watermarking method that does not use the original image for watermark detection and resilient to image quality degradation

K. Ratakonda, R. Dugad and N. Ahuja, Digital Image Watermarking: Issues in Resolving Rightful Ownership, Proc. International Conference on Image Processing, Vol. 2, Chicago, IL, October 1998, 414-418.Abstract

R. Dugad, K. Ratakonda and N. Ahuja, A New Wavelet Based Scheme for Digital Image Watermarking, Proc. International Conference on Image Processing, Vol. 2, Chicago, IL, October 1998, 419-423. Abstract

R. Dugad and N. Ahuja, A Scheme for Joint Watermarking and Compression of Video, IEEE International Conference on Image Processing, Vol. 2, Vancouver, BC, Canada, September 2000, 80-84. Abstract

R Dugad, K Ratakonda

To develop fast algorithms for magnification or demagnification of a compressed image by a given amount directly in the compressed image format

K. Ratakonda and N. Ahuja, Discrete Multi-Dimensional Linear Transforms over Connected Arbitrarily Shaped Supports, IEEE International Conference on Acoustics, Speech and Signal Processing, Vol. 4, Munich, Germany, April 1997, 3041-3044. Processing, Geneva, Switzerland, September 1996, 81-84. Abstract

R. Dugad and N. Ahuja, A Fast Scheme for Downsampling and Unsampling in the DCT Domain, International Conference on Image Processing, Kobe, Japan, Oct. 1999, II-909-913. Abstract

R. Dugad and N. Ahuja, A Fast Scheme for Altering Resolution in the Compressed Domain, EEE Conference on Computer Vision and Pattern Recognition, Ft. Collins, CO, June 1999, I-213-218. Abstract

R. Dugad and N. Ahuja, A Fast Scheme for Image Size Change in the Compressed Domain, IEEE Trans. on Circuits and Systems for Video Technology, Vol. 11, No. 4, April 2001, 461-474. Abstract

S Yoon

Given a video sequence with missing frames, generate the missing frames by interpolating nearby available frames

Motion Compensated Missing Frame Interpolation

Results
We implemented the proposed algorithm using various QCIF (176×144) video sequences. Fig. 1(a) shows a typical motion vector field resulting from the block matching algorithm (BMA). The block size for the BMA is set to 16×16. We tried to make the number of triangles of a uniform mesh the same as the case of the BMA. After fixing the locations of node points, the algorithm obtains one motion vector per node in order to estimate affine parameters per triangle. Fig. 1(b) shows the improved motion vector field. Fig. 2 shows comparisons among the block-based method, the pel-recursive method, and the proposed method. The frame numbers in Fig. 2 refer to the interpolated frames. It means that we do not include the PSNRs of the transmitted frames in the graphs. The proposed algorithm shows a performance close to the pel-recursive approach in terms of PSNR. Fig. 3 shows a comparison between two interpolated images using the block-based method and the proposed method, respectively.

original image

change detection invalid triangles (visible pixels)

Fig. 1(a): block-based motion field Fig. 1(b): improved motion field

Fig. 2: Comparison of PSNR: block-based method (`–’: dashed line), pel-recursion method (`-’: solid line), and proposed method (`-.-’: dashed line with dots)

block-based interpolation improved interpolation
Fig. 3

Downloadable Paper
Seung-Chul Yoon and Narendra Ahuja, “Frame Interpolation Using Transmitted Block-Based Motion Vectors,” To appear in ICIP 2001. [Download Paper]

S.-C. Yoon and N. Ahuja, Frame Interpolation using Transmitted Block-Based Motion Vectors, Proc. International Conference on Image Processing, Thessaloniki, Greece, October 2001

Contact Information
Seung-Chul Yoon
Address:
1614 Beckman Institute
405 N. Mathews Avenue, Urbana IL 61801, USA.
Phone: (217) 333-1869 / 244-4392
Email: scyoon@vision.ai.uiuc.edu
Homepage: http://vision.ai.uiuc.edu/scyoon/

Abstract: The proposed technique is designed for interpolating video frames often dropped during compression using standards, such as MPEG-4 and H.263 at very low bit rates. The coarse motion vector field is refined at the receiving side using mesh-based motion estimation instead of using computationally intensive dense motion estimation. We propose a framework for detecting and utilizing local motion boundaries in terms of an explicit model. Motion boundaries are modeled using edge detection and Hough transform. The motion of the occluding side is represented by affine mapping. The newly appearing region is also detected. Each pixel is interpolated differently according to adaptive interpolation based on the property of the pixel: moving, static, and disoccluded.

K Ratakonda , S Yoon

Given a video sequence and a hierarchical segmentation representing the natural multiscale spatiotemporal structure, identify the interframe redundancy for efficient video coding which is adaptive to the desired level of coded detail

Multiscale Structure-based Video Compression
Seung-Chul Yoon and Krishna Ratakonda and Narendra AhujaWe developed a very low bit rate video compression algorithm using multiscale image segmentation based hierarchical motion compensation and residual coding. The proposed algorithm outperforms the H.261-like coder by 3 dB and the H.263 version 2 by 1 dB. Such gains come from the use of image segmentation and reversed motion prediction. The proposed region based reversed motion compensation strategy regulates the size and number of regions used, by pruning multiscale segmentation of video frames. Since regions used for motion compensation are obtained by segmenting the previously decoded frame, the shape of the regions need not be transmitted to the decoder. Furthermore, the hierarchical motion compensation strategy involves two stages: it refines an initial, region level, coarse motion field to obtain a dense motion field which provides pixel level motion vectors. The refinement procedure does not require any additional information to be transmitted. We also developed a residual coding technique for coding the displaced frame difference after segmentation based motion compensation. Residual coding is performed using a method which exploits the fact that the energy of the residual resulting from motion compensation is concentrated in a priori predictable positions. This residual coding technique can also be extrapolated to improve the performance of coders using a block based motion compensation strategy.

Results
We compare our coder with a generic block based coder as used in the H.261 or the H.263 standards. All performance comparison is performed on the luminance (Y) component of the video frames. In order to make an objective comparison, we used the same quantization strategies to quantize DCT coefficients for both the coders. The Huffman codes for motion vectors and DCT coefficients were the same for both the coders. The frame bit-rate was held (approximately) fixed for both the coders at 1280 bits. This bit-rate corresponds to a bit-rate of 9.6 kbps if every fourth frame is coded and a bit-rate of 38.4 kbps if all the frames are coded.

* Click one of the images to see a video sequence.

Original frame Standard Approach Our approach

Original frame Standard Approach Our approach

We also present results comparing our residual coding scheme with the usual block DCT based coding scheme. The overhead of 1 bit per coded block will be transmitted. Such a coder always performs better than the baseline block DCT scheme. The following figure shows the improvement (in dB PSNR) over the generic coder, when the quantization step size of AC coefficients is 16 and 32.

Downloadable Papers
1. Seung Chul Yoon, Krishna Ratakonda and Narendra Ahuja, “Low Bit-Rate Video Coding with Implicit Multiscale Segmentation,” IEEE Trans. on Circuits and Systems for Video Technology, Vol. 9, No. 7, pp. 1115-1129, October 1999. [Abstract ][Download Paper]

2. Seung Chul Yoon, Krishna Ratakonda and Narendra Ahuja, “Region based Video Coding using a Multiscale Image Segmentation,” Proc. IEEE Int. Conf. on Image Proc. (ICIP’97), vol. 2, pp. 510-513, Santa Barbara, 1997. [Abstract ][Download Paper]

3. Krishna Ratakonda, Seung Chul Yoon and Narendra Ahuja, “Video Compression: Coding the Displaced Frame Difference,” Proc. IEEE Int. Conf. on Image Proc. (ICIP’97), vol. 1, pp. 353-356, Santa Barbara, 1997. [Abstract ][Download Paper]

Contact Information
Seung-Chul Yoon
Address:
1614 Beckman Institute
405 N. Mathews Avenue, Urbana IL 61801, USA.
Phone: (217) 333-1869 / 244-4392
Email: scyoon@vision.ai.uiuc.edu
Homepage: http://vision.ai.uiuc.edu/scyoon/

Title: Low Bit-Rate Video Coding with Implicit Multiscale Segmentation
Abstract: Discusses a multiscale segmentation based video compression algorithm aimed at very low bit-rate applications such as video teleconferencing and video phones. We introduce novel techniques for multiscale segmentation based motion compensation and residual coding. Our region based forward motion compensation strategy (in terms of direction of motion vector, which is from the previous frame to the current frame) regulates the size and number of regions used, by pruning a multiscale segmentation of video frames. Since regions used for motion compensation are obtained by segmenting the previously decoded frame, the shape of the regions need not be transmitted to the decoder. Furthermore, our hierarchical motion compensation strategy refines an initial region level, coarse motion field to obtain a dense motion field which provides pixel level motion vectors. The refinement procedure does not require any additional information to be transmitted. This motion compensation technique effectively addresses the problem of dealing with “holes” and “overlapping regions” which are inherent to forward motion compensation. Residual coding is performed using a novel method which exploits the fact that the energy of the residual resulting from motion compensation is concentrated in a priori predictable positions. We show that this residual coding technique can also be extrapolated to improve the performance of coders using a block based motion compensation strategy. A fusion of these concepts leads to a gain of 2-3 dB in peak signal-to-noise ratio, apart from significant perceptual improvement, over a generic video coding algorithm using a block based motion compensation strategy (such as H.261 or H.263) for a variety of test sequences.

Title: Region based Video Coding using a Multiscale Image Segmentation
Abstract: This paper proposes a novel region-based video coding technique using a multiscale image segmentation method thus obtaining better quality at the same bit rate. In most of the previous region-based video coding techniques, occlusion caused degradation in terms of both PSNR and perceptual video quality.We propose a new motion estimation and compensation algorithm which solves occlusion related problems effectively. The proposed motion estimation and compensation is a two stage procedure: the first stage uses a coarse motion model while the second stage uses a dense motion model. The coarse motion model generates region level motion vectors which are then fine tuned by the dense motion model which produces pixel level motion vectors. A fusion of these concepts leads to a gain of 2~3 dB in PSNR over the block-based algorithm for a variety of test sequences using a fully functional video coder.

Title: Video Compression: Coding the Displaced Frame Difference
Abstract: Popular techniques employed to code the displaced frame difference (DFD) treat it no differently from an ordinary image for coding purposes. Since the DFD is generated by the process of motion compensation, such methods do not fully exploit the underlying redundancies. This paper proposes a DFD coding method which exploits such redundancies while incurring negligible information overhead. The key idea is to predict locations of high DFD concentration which occupy small portions of the image and use this predicted information (which is also available to the decoder without additional information transmission) to improve the quality of the decoded image. Two key features of the proposed approach are its compatibility with any transform based DFD coding scheme and negligible information overhead. Tests with a fully functional video coder show the efficacy of the proposed approach.

P Ishwar, P Moulin, K Ratakonda, M Singh

Given an image and its hierarchical segmentation representing its natural multiscale structure, use the knowledge of the image regions to smooth out the noise in the region interiors without blurring borders

Image Denoising Using Projection on Convex Sets

Krishna Ratakonda

and Narendra Ahuja
Beckman Institute for Advanced Science and Technology
Department of Electrical and Computer Engineering
University of Illinois at Urbana-Champaign

This work addresses the problem of denoising of images corrupted by AWGN. The Wiener filter is optimum in minimizing the mean-square-error under suitable assumptions of stationarity of the signal statistics. Locally, such assumptions are reasonable, as in the adaptive realization of theWiener filter whose performance is among the best known till date. Over the last few years, there has been much interest in threshold based denoising schemes. In this paper we present a novel framework for denoising signals from their compact representation in multiple domains. Each domain captures, uniquely, certain signal characteristics better than others. We define confidence sets around data in each domain and find sparse estimates that lie in the intersection of these sets, using a POCS algorithm. Simulations demonstrate the superior nature of the reconstruction (both in terms of mean-square error and perceptual quality) in comparison to the adaptive Wiener filter.

Results

The following images compare the performance of our scheme to that of Donoho and Johnstone’s.

(a)                                                                                             (b)

(c)                                                                                             (d)

(a) Lena processed with the Donoho-Johnstone scheme (best singleDaubechies filter). PSNR: 33.25 dB. (b) Lena processed with 1-3-4 wavelet filters. PSNR: 35.01 dB. Original noisy image (not shown) PSNR: 31.21 dB. (c) Goldhill processed with the Donoho-Johnstone scheme (best single Daubechies filter). PSNR: 30.59 dB. (d) Goldhill processed with 1-3-5 wavelet filters. PSNR: 33.51 dB. Original noisy image (not shown) PSNR: 29.04 dB.

PSNR as a function of the number of vanishing moments of the wavelet used.

Papers

• Segmentation based denoising using multiple compaction domainsSingh, M.; Ishwar, P.; Ratakonda, K.; Ahuja, N. Image Processing, 1999. ICIP 99. Proceedings. 1999 International Conference on , Volume: 1 , 1999 Page(s): 372 -375 vol.1
• Image denoising using multiple compaction domainsIshwar, P.; Ratakonda, K.; Moulin, P.; Ahuja, N. Acoustics, Speech and Signal Processing, 1998. Proceedings of the 1998 IEEE International Conference on , Volume: 3 , 1998 Page(s): 1889 -1892 vol.3

Publications

K. Ratakonda and N. Ahuja, Restoring Image Quality Through Structure Preserving De-Noising, 3rd Asian Conference on Computer Vision, Hong Kong, January 8-11, 1998, 33-40.  Abstract

P. Ishwar, K. Ratakonda, P. Moulin and N. Ahuja, Image De-noising Using Multiple Compaction Domains, International Conference on Acoustics, Speech, and Signal Processing, Vol. III, Seattle, WA, May 12-15, 1998, Invited, 1889-1892.  Abstract

M. Singh, P. Ishwar, K. Ratakonda and N. Ahuja, Segmentation Based Denoising Using Multiple Compaction Domains, International Conference on Image Processing, Kobe, Japan, Oct. 1999, I-372-375.  Abstract

K Ratakonda

Given an image and its hierarchical segmentation representing its natural multiscale structure, use the explicit, known geometry for image scaling, e.g., image expansion for superresolution

Details:

POCS BASED ADAPTIVE IMAGE MAGNIFICATION

Krishna Ratakonda and Narendra Ahuja
Beckman Institute for Advanced Science and Technology
Department of Electrical and Computer Engineering
University of Illinois at Urbana-Champaign

Problem Definition

Resolution enhancement involves the problem of magnifying a small image to several times its size while avoiding blurring, ringing and other artifacts.  We tackle the problem of magnifying an image without incurring the edge enhancement effects and other structural distortions characteristic of classical  image magnification techniques. We propose an iterative algorithm based on a Projections onto Convex Sets (POCS) formalization.

Applications:

1. Image Magnification.
2. Phase Retrieval.

Details:

Classical image magnification  methods include bilinear, bicubic and FIR interpolation schemes followed by a sharpening method like unsharp masking. Such interpolation schemes tend to blur the images when applied indiscriminately. Unsharp masking, which involves subtracting a properly scaled Laplacian of the image from itself, produces artifacts and increases noise. More sophisticated schemes involving wavelet- or fractal-based techniques have also been proposed. Such methods perform extrapolation of the signal in either the wavelet or fractal domain, which leads to objectionable artifacts when the assumptions behind such extrapolation are violated. It may also be noted that such extrapolatory assumptions predict and actively enhance the high-frequency content within the image, thus increasing any noise present in the unmagnified image.

The proposed method starts with an initial magnified image obtained through selective interpolation followed by an iterative procedure which aims to avoid edge-related artifacts while retaining and enhancing sharpness. The initial image is a composite image formed from a base interpolation scheme in the smooth areas of the image and from a selective interpolation mechanism in the non-smooth (or edge) areas. The proposed iterative algorithm aims to find a magnified image satisfying two constraints: one of the constraints is derived from sampling theory while the other constraint reflects the confidence that we place on the initial iterate. Both the constraints are convex sets; thus, we seek a solution which is at the intersection of these two convex sets and  can be obtained using the projection on convex sets (POCS) method.  Starting with the initial iterate, we project alternately on the two constraints. Convergence is guaranteed since we operate within the POCS formalism.

The proposed algorithm consists of three basic steps: Finding edges
Edge locations are found using a multiscale segmentation algorithm. Obtaining the initial image
This is done using a bilinear interpolation scheme in the smooth regions and a selective interpolation mechanism in the non-smooth areas POCS based iterative algorithm
The first constraint set arises from sampling theory which suggests that the unmagnified image can be viewed as being  obtained by subsampling the magnified image without aliasing. The second constraint set arises by constraining the non-edge and edge location values to vary within limits from their initial estimate.

Results:

Fig 1a – Lena, 4x magnification using bilinear interpolation                      Fig 1b -  Segmentation of  Fig 1a

Fig 1c – Proposed method  with coarse segmentation                              Fig 1d – Proposed method with fine segmentation

Fig 2a – Goldhill, 4x magnification, Bilinear interpolation              Fig 2b – Proposed scheme (note the crisper bars).

Publications:

• POCS based adaptive image magnification
  • Ratakonda, K.; Ahuja, N.

Image Processing, 1998. ICIP 98. Proceedings. 1998 International Conference on ,  1998

Page(s): 203 -207 vol.3

K. Ratakonda and N. Ahuja, POCS-Based Adaptive Image Magnification, Proc. International Conference on Image Processing, Vol. 3, Chicago, IL, October 1998, 203-207.  Abstract

K. Ratakonda and N. Ahuja, Super Resolution with Region Sensitive Interpolation, Proc. Image Understanding Workshop, New Orleans, LA, May 1997, 537-540.

K Ratakonda

Given an image and its hierarchical segmentation representing its natural multiscale structure, use the compactness of the structural description for best image compression

Structure Based Image Compression
Krishna Ratakonda and Narendra Ahuja

(a) piecewise constant model

(b) piecewise linear model

(c) linear prediction model

(d) best JPEG implementation

A. Sehgal, A. Jagmohan, N. Ahuja

This project deals with scalable coding and robust Internet streaming of predictively encoded media. We frame the problem of predictive coding as a variant of the Wyner-Ziv problem in Information theory. Subsequently, LDPC based coset code constructions are used to compress the media in a scalable, error-resilient manner.

1. A. Sehgal, A. Jagmohan, N. Ahuja “Wyner-Ziv Coding of Video: Applications to Error Resilience,” IEEE Trans. Multimedia, April 2004, pages 249 – 258. Full Text
2. A. Sehgal, A. Jagmohan, N. Ahuja “A State-free Causal Video Encoding Paradigm, ” Invited Paper, Proc. IEEE Int. Conf. Image Processing, 2003, pp. I-605-608 Full Text
3. A. Sehgal, A. Jagmohan, N. Ahuja “Scalable Predictive Coding and the Wyner-Ziv Problem, ” Proc. IEEE Int. Conf. Comm. Systems, 2002. Full Text
4. A. Sehgal, N. Ahuja, “Robust predictive coding and the Wyner-Ziv problem,” Data Compression Conference, Snowbird, Utah, 2002. pp. 103 Full Text

A. Jagmohan, A. Sehgal, N. Ahuja

The design of compression techniques for streaming of image-based rendering data to remote viewers. A compression algorithm based on the use of Wyner-Ziv codes is proposed, which satisfies the key constraints for IBR streaming, namely those of random access for interactivity, and precompression.

A. Jagmohan, A. Sehgal, N. Ahuja “Compression of Light-field Rendering Data using Coset Codes , ” Invited Paper, Proc. Asilomar Conf. on Sig., Syst., and Comp., 2003. Full Text

OBJECTIVE
Given a slanted view of a planar, homogeneously textured surface, estimate the surface slant from the image texture gradient.

APPROACH
(1) Identification of image texture elements (texels) that correspond to surface texture elements is itself a significant problem since the scale at which surface detail is captured varies continuously with the three-dimensional distance, and therefore across the image texture. The image texels may exhibit a systematic variation in a priori unknown properties, e. g., size, density or contrast. All regions are potential texels. Consequently, all regions, of all sizes and contrasts, are detected at each location and treated as candidate texels.

(2) The estimation of surface slope (slant and tilt) is integrated with the process of selecting texels from among the large number of detected regions. For any given slant and tilt, only those regions across the image are interpreted as texels whose properties, e. g., area distribution, match the spatial distribution predicted by the hypothesized slant and tilt, and which occupy the largest fraction of the image. The image area is used as a measure of  the extent of support for the particular slant-tilt pair.

(3) All possible slant-tilt values are considered as hypotheses, and a search is conducted to find the hypothesis with most support. This is the estimated surface orientation.

RESULTS
Several real life texture images are shown below. For each, estimated values of the slant-tilt angle pairs are depicted graphically by showing how a hypothetical surface containing fixed size disks would look when viewed at the estimated slant and tilt. Thus, a visual comparison of the original texture image and its graphical depiction can be made to obtain a quick assessment of the quality of the derived estimates.

Each of the following figures consists of four parts:

1. The original texture image,

2. A set of bright regions detected at multiple scales and as candidate texels,

3. The estimated surface slant and tilt values shown graphically, and

4. The subset of bright regions supporting the estimates surface orientation, i.e., the estimated image texels, shown superposed on the original image.

a. Original	b. All regions	c. Slant tilt estimates	d. Detected texels

PUBLICATIONS

1. D. Blostein and N. Ahuja, A Multiscale Region Detector, Computer Vision, Graphics and Image Processing, January 1989, 22-41. Abstract

2. S. Jackson and N. Ahuja, Elliptical Gaussian Filters, Proceedings International Conference on Pattern Recognition, Vienna, Austria, August 26-29, 1996, 775-779. Abstract

3. N. Ahuja and S. A. Jackson, Multiscale Region Detection, Proceedings Image Understanding Workshop, Palm Springs, CA, February 12-15, 1996, 961-967. Abstract

OBJECTIVE
Given a slanted view of a planar, homogeneously textured surface, estimate the surface slant from the image texture gradient.

APPROACH
(1) Identification of image texture elements (texels) that correspond to surface texture elements is itself a significant problem since the scale at which surface detail is captured varies continuously with the three-dimensional distance, and therefore across the image texture. The image texels may exhibit a systematic variation in a priori unknown properties, e. g., size, density or contrast. All regions are potential texels. Consequently, all regions, of all sizes and contrasts, are detected at each location and treated as candidate texels.

(2) The estimation of surface slope (slant and tilt) is integrated with the process of selecting texels from among the large number of detected regions. For any given slant and tilt, only those regions across the image are interpreted as texels whose properties, e. g., area distribution, match the spatial distribution predicted by the hypothesized slant and tilt, and which occupy the largest fraction of the image. The image area is used as a measure of  the extent of support for the particular slant-tilt pair.

(3) All possible slant-tilt values are considered as hypotheses, and a search is conducted to find the hypothesis with most support. This is the estimated surface orientation.

RESULTS
Several real life texture images are shown below. For each, estimated values of the slant-tilt angle pairs are depicted graphically by showing how a hypothetical surface containing fixed size disks would look when viewed at the estimated slant and tilt. Thus, a visual comparison of the original texture image and its graphical depiction can be made to obtain a quick assessment of the quality of the derived estimates.

Each of the following figures consists of four parts:

1. The original texture image,

2. A set of bright regions detected at multiple scales and as candidate texels,

3. The estimated surface slant and tilt values shown graphically, and

4. The subset of bright regions supporting the estimates surface orientation, i.e., the estimated image texels, shown superposed on the original image.

a. Original	b. All regions	c. Slant tilt estimates	d. Detected texels

PUBLICATIONS

1. N. Ahuja and T. Huang, IU at UI: An Overview and an Example on Shape from Texture, Proc. DARPA Image Understanding Workshop, Boston, April 6-8, 1988, 222-253.

2. D. Blostein and N. Ahuja, Representation and Three-dimensional Interpretation of Image Texture: An Integrated Approach, First International Conference on Computer Vision, London, England, June 8-11, 1987, 444-449. Abstract

3. D. Blostein and N. Ahuja, Shape from Texture: Integrating Texture Element Extraction and Surface Estimation, IEEE Trans. Pattern Analysis and Machine Intelligence, December 1989, 1233-1251. Abstract

1. Tianli Yu, Ning Xu and Narendra Ahuja, Recovering Shape and Reflectance Model of Non-Lambertian Objects from Multiple Views, 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’04) Volume 2, pp. 226-233, 06 27 – 07 02, 2004, Washington, D.C., USA Full Text
2. Tianli Yu, Ning Xu and Narendra Ahuja, Shape and View Independent Reflectance Map from Multiple Views, ECCV 2004, LNCS 3024, pp. 602-616, May 11-14, Prague. Full Text

1. D. Hougen and N. Ahuja, Integration of Stereo and Shape from Shading using Color, Proc. Second International Conf. on Automation, Robotics and Computer Vision, Vol 1, Singapore, September 15-18 1992, pp. CV-6.6.1 – CV-6.6.5.

2. D. Hougen and N. Ahuja, Estimation of the Light Source Distribution and its Use in Shape Recovery from Stereo and Shading, 4th Int. Conf. on Computer Vision, Berlin, Germany, May 11-14, 1993, 148-155.

3. Darrell R. Hougen and Narendra Ahuja, Adaptive Polynomial Modelling of the Reflectance Map for Shape Estimation from Stereo and Shading, Proceedings IEEE Conference on Computer Vision and Pattern Recognition, Seattle, WA, June 1994, 991-994.

4. D. Hougen and N. Ahuja, Shape from Appearance: A Statistical Approach to Surface Shape Estimation, Proceedings European Conference on Computer Vision, Cambridge, England, April 1996, 421-429.

5. M. Singh and N. Ahuja, Estimating Light Sources, Indian Conference on Computer Vision, Graphics and Image Processing, New Delhi, India, December 1998, 76-81.Abstract and Full Text

1. Tianli Yu, Ning Xu and Narendra Ahuja, Reconstructing a Dynamic Surface from Video Sequences Using Graph Cuts in 4D Space-Time, Pattern Recognition, 17th International Conference on (ICPR’04) Volume 2, pp. 245-248 08, 2004 Cambridge UK Full Text
2. Ning Xu, Tianli Yu and Narendra Ahuja. Shape from color consistency using node cut. In Proceedings of Asian Conference on Computer Vision, Jeju Island, Korea. January 2004. Abstract and Full Text
3. Ning Xu and Narendra Ahuja. A Three-view Matching Algorithm Considering Foreshortening Effects. In Proceedings of International Conference on Computer Vision, Pattern Recognition and Image Processing, pp. 635-638, Cary, NC. September 2003. Abstract and Full Text

1. A. Jagmohan, M. Singh, and N. Ahuja, Dense Two View Stereo Matching Using Kernel Maximum Likelihood Estimation, Pattern Recognition, 17th International Conference on (ICPR’04) Volume 3, pp. 28-31, 08, 2004 Cambridge UK IEEE Reference
2. M. Singh, H. Arora and N. Ahuja, Robust Registration and Tracking Using Kernel Density Correlation, 2004 Conference on Computer Vision and Pattern Recognition Workshop on Image and Video Registration, CVPRW’04 Volume 11, p. 174, 06 27 – 07 02, 2004, Washington, D.C., USA Full Text

Given multiple images of a scene, taken from multiple cameras and different viewpoints, find the 3D depth map and surfaces

W. Hoff and N. Ahuja, Surfaces from Stereo, Proc. DARPA Image Understanding Workshop, Miami, December 9-10, 1985, 98-106. Full Text

W. Hoff and N. Ahuja, Surfaces from Stereo, 8th International Conference on Pattern Recognition, Paris, France, October 28-31, 1986, 516-518.

J. Weng, N. Ahuja and T. Huang, Two-View Matching, Second International Conference on Computer Vision, Tarpon Springs, December 5-8, 1988, 64-73. Abstract and Full Text

W. Hoff and N. Ahuja, Extracting Surfaces from Stereo Images: An Integrated Approach, First International Conference on Computer Vision, London, England, June 8-11, 1987, 284-294.

W. Hoff and N. Ahuja, Depth from Stereo, IEEE Conf. on Computer Vision and Pattern Recognition, San Francisco, June 9-13, 1985.

W. Hoff and N. Ahuja, Generating Range Map from Stereo Images, 4th Scandinavian Conf. on Image Analysis, Trondheim, Norway, June 18-20, 1985, 761-768.

W. Hoff and N. Ahuja, Surfaces from Stereo: Integrating Feature Matching, Disparity Estimation and Contour Detection, IEEE Trans. Pattern Analysis and Machine Intelligence, February 1989, 121-136. Abstract and Full Text

E. Altman and N. Ahuja, A Dynamical Systems Approach to Integration in Stereo, Proc. DARPA Image Understanding Workshop, Pittsburgh, Sep 11-13, 1990, 423-427.

A regression-based model which admits a realistic framework for automatically choosing bandwidth parameters which minimizes a global error criterion. This is used for automatic segmentation of images at any input resolution scale (for e.g., the wavelet decomposition scale).

1. M. Singh and N. Ahuja, “Regression based Bandwidth Selection for Segmentation using Parzen Windows”, in Ninth IEEE International Conference in Computer Vision, Proceedings, vol. 1, pp. 2-9, Oct. 2003, Nice, France. Full Text
2. Singh and N. Ahuja, Mean-Shift Segmentation with Wavelet-based Bandwidth Selection, IEEE Workshop on Applications in Computer Vision, pp. 43-50, Dec. 3-4, 2002, Florida. Full Text

Models of spatial variation in images are central to a large number of low-level computer vision problems including  egmentation, registration, and 3D structure detection. Often, images are represented using parametric models to characterize (noise-free) image variation, and, additive noise. However, the noise model may be unknown and parametric models may only be valid on individual segments of the image. Consequently, we model noise using a nonparametric kernel density estimation framework and use a locally or globally linear parametric model to represent the noise-free image pattern. This results in a  ovel, robust, redescending, M- parameter estimator for the above image model which we call the Kernel Maximum Likelihood estimator (KML). We also provide a provably convergent, iterative algorithm for the resultant optimization problem. The estimation framework is empirically validated on synthetic data and applied to the task of range image segmentation.

1. M. Singh, H. Arora and N. Ahuja, “A Robust Probabilistic Estimation Framework for Parametric Image Models”, European Conference on Computer Vision, LNCS 3021, pp. 508-522, 2004. Full Text

]]> http://vision.ai.uiuc.edu/?feed=rss2&p=924 0 http://vision.ai.uiuc.edu/?p=883 http://vision.ai.uiuc.edu/?p=883#comments Thu, 07 May 2009 21:44:21 +0000 xianbiao http://vision.ai.uiuc.edu/wordpress/?p=883 To demonstrate the featured capabilities of the HMPD technology, and explore its application for distance collaboration in interactive augmented environments

| Demo | Details | Publications | Back to Theme |

To demonstrate the featured capabilities of the HMPD technology, and explore its application for distance collaboration in interactive augmented environments

| Demo | Details | Publications | Back to Theme |

]]>

| Demo | Details | Publications | Back to Theme |

An automatic object detection, localization and segmentation system is proposed for object categories. Object categories are modelled as templates of patches around interest points, encoding both location and appearance information. The automatic segmentation algorithm integrates the localization information with the edge information in the image.

Publications

1. S. Todorovic and N. Ahuja, Extracting subimages of an unknown category from a set of images,  in Proc. IEEE Comp. Soc. Conf. Computer Vision and Pattern Recognition (CVPR 2006), vol. 1, pp. 927-934, New York, NY, 2006 Full Text

A

1. Tianli Yu, Hongcheng Wang, Narendra Ahuja, Wei-Chao Chen, Sparse Lumigraph Relight by Illumination and Reflectance Estimation from Multi-View Images, Eurographics Symposium on Rendering (EGSR), 2006 Full Text
2. Tianli Yu, Hongcheng Wang, Narendra Ahuja, Wei-Chao Chen, Sparse Lumigraph Relight by Illumination and Reflectance Estimation from Multi-View Images, Technical Sketch, SIGGRAPH, 2006 Full Text

J Cocatre-Zilgien, F Delcomyn, Z Ding, J Hart, G Kremesec, L Lu, M Nelson, J Reichler, K Tan, Design and implementation of a pneumatic, six-legged robot with the geometry, number of leg joints, and joint functionality modeled after the American Cockroach

|Demo|Details|Presentation|Poster|Publications|

1.Hongcheng Wang, Ning Xu, Ramesh Raskar and Narendra Ahuja, Videoshop: A New Framework for Video Editing in Spatio-Temporal Gradient Domain, IEEE, Video Proceedings, International Conference on Computer Vision and Pattern Recognition, 2005  Full Text
2. Hongcheng Wang, Ramesh Raskar and Narendra Ahuja, Seamless Video Editing, Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on, Aug. 2004, pp. III-858-861  Full Text

J Chuang, Y Hwang, R Ruff

Given a mobile object required to move from a source location/orientation to a destination location/orientation, compute a path that it can follow and the orientation and velocity values it must assume along the path to efficiently and smoothly move from the source to the destination.

| Demo | Details | Publications|

A sketching interface is an appropriate choice for image editing because the pen and pencil are ubiquitous tools for creating graphics, and sketch-based interaction would be very natural for use in handheld personal digital assistants and the emerging class of �tablet� computers. A typical example of the kind of sketch allowed by the selection tool we will describe is shown in figure 1(a). With the sketch shown, our selection tool was able to extract the head of the crane, as shown in figure 1(b). Despite the fairly convoluted profile of the crane�s head, which consists of both sharp boundaries along the beak and complicated boundaries around the crane�s crown, our selection tool needed only the simple sketch shown, which is natural and can be drawn quickly, without much skill and dexterity. We believe that this tool will enable sophisticated and high-fidelity graphical selection and manipulation operations in situations where the user may not have much time, such as during a live sports broadcast, or may not have much dextrous control, such as on a mobile handheld device, or may want to avoid the monotony and tedium of selection, without compromising the quality of the selection.

Algorithm

Step 1: Draw Sketch User draws sketch over the image. Users can make three types of sketch strokes: Points Lines Regions (closed loops) Step 2: Image Segmentation Compute a global segmentation of the image. We can use any existing segmentation method. The segment map illustrated is computed by recursive binary-split vector quantization in the (x,y,rgb) space. Step 3: Sketch Processing Use sketch to select a number of segments from the image that form the foreground object. Step 4: Simplicial Decomposition We want to decompose the image into small triangles so they can be processed individually. We can use: Delaunay Triangulation of the segment centroids, [1] or Use the segmentation to compute a triangulation [2]. Step 5: Object Extraction Each triangle is processed separately as follows: Find a linear approximation to the local object boundary. Re-segment the image within the boundary using the same segmentation algorithm as in step 2. Classify the new segments into foreground/background using the linear boundary. Pick a window of pixels around each of the new segment centroids and use them as samples of foreground/background colors Estimate alpha channel Step 6: Complete Selection Composite the result from each triangle back into a complete selection.

Figure 2: Steps in selection algorithm.

Results
Sketch	Resulting Selection

Figure 3: Results computed with the Delaunay Triangulation decomposition method.

Publications

1. Kar-Han Tan and Narendra Ahuja. Selecting Objects with Freehand Sketches, Proc. International Conference on Computer Vision, Vancouver, Canada, July 2001, 337-344.  Full Text
2. Kar-Han Tan and Narendra Ahuja. A Representation for Image Structure and its Application in Object Selection with Freehand Sketches. In Proceedings IEEE Computer Society Conference on Computer Vision and Pattern Recognition 2001. Full Text

A new framework for seamless video editing in gradient domain with the objective of replacing video segments in one video sequence from those in another, composing video sequences by juxtaposing multiple other video sequences, etc.

1. Hongcheng Wang, Ning Xu, Ramesh Raskar and Narendra Ahuja, Videoshop: A New Framework for Video Editing in Spatio-Temporal Gradient Domain, IEEE, Video Proceedings, International Conference on Computer Vision and Pattern Recognition, 2005

2. Hongcheng Wang, Ramesh Raskar and Narendra Ahuja, Seamless Video Editing, Pattern Recognition, 2004. ICPR 2004. Proceedings of the 17th International Conference on, Aug. 2004, pp. III-858-861

Related projects:

Videoshop: A New Framework for Video Editing in Gradient Domain

Narendra Ahuja,  Donald Biggar Willet Professor of Engineering

Leading Expert in Computer Vision

Q: What is your area of expertise?
A: My area is computer vision, a field related to image processing, pattern recognition, sensing, and robotics. The main goal is to automatically understand the contents of images and video–like human vision does routinely–which is why the field is also called “image understanding.” Specific tasks range from very simple and general ones like estimating whether an orange on the tree is ripe to eat or is it within reach, to very complex ones like assigning semantic and possibly subjective labels such as “an attractive outfit” or “that serene view of the skies.”

Q: Give me a brief synopsis of your education and career.
A: My bachelor’s is in electronics engineering, my master’s is in electrical communication engineering, and my PhD is in computer science. My bachelor’s is from the Birla Institute of Technology and Science at Pilani, India, my master’s is from the Indian Institute of Science in Bangalore, India, and my PhD is from the University of Maryland, College Park. I grew up not far from the Taj Mahal, about 30 or 40 miles away. After my PhD, I came to Illinois.

Q: You have been at Illinois since 1979. What do you enjoy most about being here?
A: The longest I have been at any one place is here at Illinois! I never thought I would be here this long. But it’s been one pleasant journey. I like the supportive environment UIUC provides. Urbana-Champaign offers a lifestyle conducive to research. You don’t have to spend an hour and half each way to work. Collegial department, great colleagues, and wonderful students.

Q: Why did you become an engineer?
A: When I was growing up, engineering was one of the most desired professions. If you were doing well in school, you basically went into engineering or medicine. I knew definitely that I didn’t want to become a doctor. I didn’t have any interest in drugging a frog unconscious and dissecting it, which the biology students did regularly. And I don’t like the sight of misery and suffering, although the thought of being able to do something about it was very enticing. Engineering appealed to me on the other hand. I always liked physics when I was in school. Chemistry never really appealed to me as much, nor were my friends passionate about it! I basically made a choice between the wet sciences and the dry sciences, in favor of the dry.

Q: How did you become interested in researching this particular field?
A: When I was thinking about the PhD, we had a professor from Maryland, Azriel Rosenfeld, visit my institute in Bangalore. He was a founder of the field of computer vision. He had come to give a talk, and I thought, this is as good a field as anything I have considered so far. I was also doing my master’s thesis in speech, a reasonably related area. I wrote to him and came to Maryland, and from then on it has been just a tunnel.

Q: What keeps you interested in the field?
A: The challenge posed by the fact that vision is indeed possible. Without a living proof of it in biology, the feasibility of vision would probably have been a controversial notion. It is a formidable challenge to engineering. It will take several generations of many, many people to get even close to the stage where we may duplicate it. If you don’t pay close attention to our ability to see, it seems like an easy task. However, it’s not only difficult by itself, and by that I mean the sensing of light and converting it into something meaningful. But it’s also difficult because it involves lots of bits of knowledge that come from other modalities of sensing and knowledge acquired in the past. So there is this top-down element that regulates how we interpret the sensory signals that are coming in. It keeps me interested because it keeps me busy trying to solve the next problem.

Q: Tell me about a research accomplishment you’re proud of?
A: I have had different periods of emphases on different groups of problems. I have worked on the integration of diverse information sources. We have different cues coming in from the three dimensional world, from texture to stereo to motion to color. So the engineering design question that has run through my work is how to analyze them individually and fuse them together into a holistic construct. In the early ’90s, I had sort of an intellectual accident, and we wound up developing a camera called the omnifocus camera. The whole field of new camera designs has since been an active research area for many researchers across the world. The camera we developed was able to produce a 3D panoramic image with every object in focus along with estimates of depth. We have since developed about eight more cameras. One of these cameras acquires a 360 × 170 degree image, which records a whole hemisphere of an image. You can come back to the recorded video and watch any specific part of it at your leisure. You can break it down by sections of the hemisphere and replay an event that may have occurred. One of these cameras is installed in the main branch of the Busey Bank in Urbana and another one is at work in Missouri. More recently, we have found a way of discovering visual themes present in large sets of images that help unify diverse appearances in terms of a simple explanation.

Q: What do you enjoy most about teaching?
A: Teaching keeps giving you new ideas, in addition to the opportunities to enjoy the moments when you see that the students are learning and understanding, and then using and extending that knowledge. I’ve had roughly 45 PhD students. Often they start with very little, and they falter and run into blind alleys and then suddenly take off. And then they go and do their own things. It’s fun to see them spread their wings and fly. They teach you, too, when you are teaching them. Research comes with teaching, the two are interdependent. Research has a major impact on the contents of the courses and classroom teaching.

Q: What role do students play in your research?
A: They are the central pillars of my research. I enjoy my meetings with them, discussing issues, raising questions, planning investigations, and of course their coming back and reporting the results. It’s a weekly cycle. We look at what happened last week, what this week’s results showed us, and then plan how to move forward.

Q: What award have you won that is the most meaningful?
A: I was probably most influenced by the very first national level award I won in the final year of my high school. I actually received two different awards that year, both from the president of India. For one, I got a scholarship for subsequent six years of my education. The other one was for being a President’s Scout–I think it is called Eagle Scout in the US. I got to shake hands with the president, which is etched in my mind because he was an eminent philosopher besides being the president. The other awards that mean a lot to me are those that come from peers and students because it means that those who are affected by it the most feel that I am doing something worthwhile.

Q: What does the future hold?
A: Computer vision has come an incredibly long way. The theory and the applications are both flourishing. A lot has been accomplished. And of course, cameras are everywhere now. However, compared to what still needs to be done, we’ve only scratched the surface so far. It’s just an extremely large problem. So we can expect a lot of excitement to come our way in this field for a very long time to come.

Q: What else do you hope to accomplish with your research?
A: There are so many problems within apparent reach that I hardly know where to begin the list. One such problem is to improve the content of the raw data acquired by the cameras I mentioned before. I would certainly like to push that up, to a level where they don’t miss anything left or right, near or far, dark or light, coarse or detailed. Another major problem I wish to help solve is extraction and precise characterization of natural structure in images and video. Overall, I want to enhance our understanding of the vision process, and apply it to everyday life in ways that themselves need research, and many of the most creative of which, I am sure, are yet to be discovered.

Resource: http://www.ece.illinois.edu/news/experts/fe-ahuja.html

Publications

1. E. Altman and N. Ahuja and F. Kishino, Hand Trajectory Recognition Using Dynamical Systems, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 321-324.
2. M.-H. Yang and N. Ahuja, Extracting Gestural Motion Trajectory, 1998 IEEE International Conference on Automatic Face and Gesture Recognition, Nara, Japan, April 1998, 10-15.  Abstract
3. M.-H. Yang and N. Ahuja, Extraction and Classification of Motion Patterns for Hand Gesture Recognition, IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Santa Barbara, CA, June 1998, 892-897.  Abstract
4. M.H. Yang and N. Ahuja, Gaussian Mixture Modeling of Human Skin Color and Its Applications in Image and Video Databases, Proc. of the SPIE: Storage and Retrieval for Image and Video Databases VI, Vol. 3656, San Jose, CA, Jan. 1999, 458-466.  Abstract

Publications

1. M.-H. Yang, D. Kriegman and N. Ahuja, Detecting Faces in Images: A Survey, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2001.  Full Text
2. M.-H. Yang, D. Kriegman and N. Ahuja, Face Detection Using Multimodal and Density Modes, Computer Vision and Image Understanding, 2001.  Full Text
3. M.H. Yang and N. Ahuja, Detecting Human Faces in Color Images, IEEE Conference on Computer Vision and Pattern Recognition, Ft. Collins, CO, June 1999, I-466-472.  Full Text
4. M.-H. Yang, N. Ahuja, D. Kriegman, Face Detection using a Mixture of Factor Analyzers, International Conference on Image Processing, Kobe, Japan, Oct. 1999, III-612-616.  Abstract
5. M.-H. Yang, N. Ahuja and D. Kriegman, Face Detection Using Mixtures of Linear Subspaces, Fourth IEEE Int. Conference on Automatic Face and Gesture Recognition (FG 2000), Grenoble, France, March 2000, 70-76.  Full Text
6. D. Roth, M.-H. Yang and N. Ahuja, A SNoW-Based Face Detector, Proc. Advances in Neural Information Processing Systems (NIPS’99), Denver, CO, December 1999, 862-868.  Full Text
7. M.-H. Yang and N. Ahuja, Detecting Human Faces in Color Images, IEEE International Conference on Image Processing, Vol. 1, Chicago, IL, October 1998, 127-130.
8. M.-H. Yang, D. Roth and N. Ahuja, Face Detection Using Large Margin Classifiers, Proc. International Conference on Image Processing, Thessaloniki, Greece, October 2001.  Full Text

N Bridwell, J Chuang, F Kishino, Y Kitamura, H Takemura, R Yen, R Chien

Given a set of objects moving in a known fashion and a set of still obstacles, detect or predict collisions between specific pairs of objects.

Publications

N. Ahuja, R. T. Chien, R. Yen and N. Bridwell, Interference Detection and Collision Avoidance Among Three Dimensional Objects, Proc. 1st National Conf. on Artificial Intelligence, Stanford University, August 19-21, 1980, 44-48. Abstract

Y. Kitamura, H. Takemura, N. Ahuja and F. Kishino, Efficient Interference Detection Among Objects Using Octree and Polyhedral Shape Representations, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 775-779. Abstract

J.-H. Chuang and N. Ahuja, An Analyticaly Tractable Potential Field Model of Free Space and Its Application in Obstacle Avoidance, IEEE Trans. on Systems, Man, and Cybernetics, Vol. 28, No. 5, October 1998, 729-736. Abstract

Y. Kitamura, H. Takemura, N. Ahuja and F. Kishino, Colliding Face Detection among 3-D Objects using Octree and Polyhedral Shape Representation, Journal of the Robotics Society of Japan, Vol. 14, No. 5, 1996, 733-742. Abstract

Y. Kitamura, H. Takemura, N. Ahuja and F. Kishino, A Study of Interference Detection in Virtual Cooperative Workspace for an Operator Assistant, Workshop on Human Interface, Tokyo, May 24, 1993, vol. 8, no. 2, 247-254. (Part in English, part in Japanese.) Also published as a Technical Report of Technical meeting on Human Communication Engineering of IEICE, HC93-17 (in Japanese). Abstract

Y. Kitamura, H. Takemura, N. Ahuja and F. Kishino, Interference Detection Among Objects for Operator Assistance in Virtual Cooperative Workspace, International Workshop on Robot and Human Communication, Tokyo, Japan, pp. 442-447. IEEE, November 1993.. Abstract

A Choudhary, S Das, C Debrunner, J Patel, M Sharma, S Swamy

To develop computationally efficient, e.g., divide-and-conquer or DSP chip oriented, algorithms for different classes of computer vision algorithms, and to define special purpose, e.g., parallel multiprocessor, architectures that efficiently execute the algorithms

Publications

N. Ahuja and S. Swamy, Multiprocessor Pyramids for Bottom-up Image Analysis, Proc. IEEE Conf. on Pattern Recognition and Image Processing, Las Vegas, June 13-17, 1982, 380-385. Abstract

N. Ahuja and S. Swamy, Interleaved Pyramid Architectures for Bottom-Up Image Analysis, Proc. 6th Int. Conf. on Pattern Recognition, Munich, Germany, October 19-22, 1982, 388-391.  Abstract

N. Ahuja, Efficient Planar Embedding of Trees for VLSI Layouts, IEEE Int. Conf. on Systems, Man and Cybernetics, Bombay/New Delhi, India, December 30, 1983 – January 7, 1984, Proc. 28-32. Abstract

N. Ahuja and S. Swamy, Multiprocessor Pyramid Architectures for Bottom-Up Analysis, IEEE Trans. Pattern Analysis and Machine Intelligence, July 1984, 463-475. Abstract

N. Ahuja, Efficient Planar Embedding of Trees for VLSI layouts, Proc. 7th Int. Conf. on Pattern Recognition, Montreal, Canada, July 30 – Aug 2, 1984, 460-464. Abstract

N. Ahuja and S. Swamy, Multiprocessor Pyramid Architectures for Bottom-up Image Analysis, Test and Measurement World, November 1985, 66-76. Abstract

M. Sharma, J. H. Patel and N. Ahuja, NETRA: A Multiprocessor Computer Architecture for Image Understanding, IEEE Workshop on Computer Architecture for Pattern Analysis and Image Database Management, Miami, November 18-20, 1985, 92-98. Abstract

N. Ahuja, Efficient Planar Embedding of Trees for VLSI Layouts, Computer Vision, Graphics and Image Processing, May 1986, 189-203.Abstract

A. N. Choudhary, S. Das, N. Ahuja, and J. H. Patel, A Reconfigurable and Hierarchical Parallel Processing Architecture: Performance Results for Stereo Vision, 10th International Conference on Pattern Recognition – Computer Vision, Vol. 2, Atlantic City, June 1990, 389-393. Abstract

A. N. Choudhary, J. H. Patel, and N. Ahuja, NETRA: A Hierarchical and Partitionable Architecture for Computer Vision Systems, IEEE Trans. Parallel and Distributed Systems, Vol. 4, No. 10, October 1993, 1092-1104. Abstract

C. H. Debrunner and N. Ahuja, A Bottom-up Minimum Spanning Tree Algorithm for Multiprocessor Pyramid Architectures, Proc. Workshop on Algorithm-Guided Parallel Architectures for Automatic Target Recognition, Leesburg, VA, July 16-18, 1984, 51-78.Abstract

A. N. Choudhary, S. Das, N. Ahuja and J. Patel, Surface Reconstruction from Stereo Images: An Implementation on a Hypercube Multiprocessor, Fourth Conference on Hypercube Concurrent Computers and Applications, Monterey, CA, March 1989, 1045-1052.Abstract

M Aggarwal

To improve the Throughput of Flexible-Precision DSPs via Algorithm Transformation

Publications

M. Aggarwal, N. Shanbhag and N. Ahuja, Improving the Throughput of Flexible-Precision DSPs via Algorithm Transformation, International Conference on Acoustics, Speech, and Signal Processing, Vol. V, Seattle, WA, May 12-15, 1998, 3069-3072. Abstract

1. Approach

Statistical learning theory has had an influence on many applications ranging from classification and object recognition, grouping and segmentation, illumination modeling, scene reconstruction and others. The rising role of learning methods, made possible by significant improvements in computing power and storage, is largely motivated by the realization that explicit modeling of complex phenomena in a messy world cannot be done without a significant role of learning. Learning is required for model and knowledge acquisition as well  as to support generalization and avoid brittleness. However, many statistical and probabilistic learning models require making explicit assumptions, e.g., on the distribution that governs the occurrences of instances in the world. For visual inference problems such as recognition, categorization and detection, making these assumptions seems unrealistic.This work develops a distribution free learning theory account to an archetypical visual recognition problem: object recognition. The problem is viewed as that of learning a representation of an object that, given a new image, is used to identify the target object in it. The learning account is developed within the PAC (Probably Approximately Correct) model of learnability. This framework allows us to (1) quantify success relative to the experience with previously observed objects, without making assumptions on the distribution, and (2) study the theoretical limits of what can be learned from images in terms of the expressivity of the intermediate representation used by the learning process. That is, learnability guarantees that objects sampled from the same distribution as the one that governs the experience of the learner are very likely to be recognized correctly. In addition, the framework gives guideline to developing practical algorithmic solutions to the problem.

Earlier works have discussed the possibility of identifying the theoretical limits of what can be learned from images and found that learning in terms of the raw representation of the images is computationally intractable. Attempts to explain this focused the dependence of learnability on the representation of the object but failed to provide a satisfactory explanation for it, or a practical solution.

The approach developed here builds on suggestions made in and relies heavily on the development of a feature efficient learning approach. This is a learning process capable of learning quickly and efficiently in domains in which the number of potential features is a very large but any concept of interest actually depends on a fairly small number of them. At the heart of the learning approach are two assumptions that we abstract as follows:

Representational:

There exists a (possibly infinite) collection M of “explanations” such that an object o can be represented as a simple function of elements in M.

Procedural:

There exists a process that, given an image in which the target object o occurs, generates efficiently “explanations” in M that are present in the image and  such that, with high probability, at least one of them is in the representation of o.
Under these assumptions we prove that there exists an efficient algorithm that, given a collection of images labeled as positive or negative examples of the target object, can learn a good representation of the object. That is, it can learn a representation that, with high probability, would make correct predictions on future images that contain (or do not contain) the object. Furthermore, we show that under these conditions, the learned representations are robust under realistic noise conditions.  A significant non-assumption of our approach is that it has no prior knowledge on the distribution of images nor it is trying to estimate it. The learned representation is guaranteed to perform well when tested on images sampled from the distribution that governed the data observed in its learning experience.

The framework developed here is very general.  The explanations alluded to above can represent a variety of computational processes and information sources that  operate on the image. They can depend on local properties of the image, the relative positions of primitives in the image, and even external information sources or context variables. Thus, the theoretical support given here applies also to an intermediate learning stage in a hierarchical process. In order to generate the explanations efficiently, this work assumes that they are syntactically simple in terms of the raw image. However, the explanation might as well be syntactically simple in terms of previously learned or inferred predicates, giving rise to hierarchical representation.

2. Results

We use the Columbia Object Image Library (COIL-100) database in all the experiments below (COIL is available at http://www.cs.columbia.edu/CAVE). The COIL-100 dataset consists of color images of 100 objects where the images of the objects that were taken at pose intervals of 5 degrees, i.e., 72 poses per object. The images were also normalized such that the larger of the two object dimensions (height and width) fits the image size off 128 x 128 pixels.  Figure 1 shows the images of the 100 objects taken in frontal view, i.e., zero pose angle. The 32 highlighted objects in 1 are considered more difficult to recognize by other researchers; we use all 100 objects including these in our experiments. Each color image is converted to a gray-scale image of 32 x 32 pixels for 128 x 128 pixels.  Figure 1 shows the images of the 100 objects taken in frontal view, i.e., zero pose angle. The 32 highlighted objects in Figure 1 are considered more difficult to recognize by other researchers; we use all 100 objects including these in our experiments. Each color image is converted to a gray-scale image of 32 x 32 pixels for our experiments.



Figure 1. Columbia Object Image Library (COIL-100) consists of 100 objects of varying poses (5 degrees apart). The objects are shown in row order where the highlighted ones are those considered more difficult to recognize by other researchers.

2.1 Ground Truth of the COIL-100 Dataset

At first glance, it seems difficult to recognize the objects in the COIL dataset because it consists of a large number of objects with varying pose, texture, shape and  size. Since each object has 72 images of different poses (5 degrees apart), many view-based recognition methods use 36 (10 degrees apart) of them for training and the remaining images for testing. However, it turns out that under these dense sampling conditions the recognition problem is not difficult (even when only grey-level images are used). Namely, in this case, instances that belong to the same object are very close to each other in the image space (where each data point represents an image of an object in a certain pose). We verified this by experimenting with a simple nearest neighbor classifier (using the Euclidean distance), resulting in an average recognition rate of 98.50% (54 errors out of 3,600 tests).  Figure 2 shows some of the objects misclassified by nearest neighbor method.In principle, one may want to avoid using the nearest neighbor method since it requires a lot of memory for storing templates and its recognition time complexity is high. The goal here was simply to show that this simple method is comparable to the complex SVM approaches for the case of dense sampling. Therefore, the abovementioned recognition problem is not appropriat for comparison among different methods.



Figure 2. Mismatched objects using the nearest neighbor method. (x:a,y:b) means that object x with view angle a is recognized as object y with view angle b. It shows some of the 54 errors (out of 3,600 test samples) made by the nearest neighbor classifier when there are 36 views per object in the training set. (See paper for datils).

Table 1. Recognition rates of nearest neighbor classifier

	30 objects randomly selected from COIL dataset	32 objects shown seletected by Pointil and Verri	The whole 100 objects in COIL dataset
Errors/Tests	14/1080	46/1152	54/3600
Recognition Rate	98.70%	96.00%	98.50%

It is interesting to see that the pairs of the objects on which the nearest neighbor method misclassified have similar geometric configurations and similar poses.  A close inspection shows that most of the recognition errors are made between the three packs of chewing gums, bottles and cars. Other dense sampling cases are easier for this method. Consequently, the set of selected objects in an experiment has direct effects on the recognition rate.  This needs to be taken into account when evaluating results that use only a subset of the 100 objects (typically 20 to 30) from the COIL dataset for experiments. Table 1 shows the recognition rates of nearest neighbor classifiers in several experiments in which 36 poses of each object are used for templates and the remaining 36 poses are used for tests.

Given this baseline experiment we have decided to perform our experimental comparisons in cases in which the number of views of objects available in training is limited. Some of our preliminary results were presented in companion papers.

2.2 Experiment Setups

Applying SNoW to 3D object recognition requires specifying the architecture used and the representation chosen for the input images. As described above, to perform object recognition we associate a target unit with each target object. This target learns a definition of the object in terms of the input features extracted from the image. We could either define a single SNoW unit which contains target subnetworks for all the 100 different target objects, or we may define different units, each with  several  (e.g., two) competing target objects. Statistically, this approach is advantageous although, clearly, it requires a  lot more computation. The architecture selected affects both the training time, where learning a definition for object a makes use of negative examples of other objects that are part of the same unit but, more importantly, it makes a difference in testing; rather that two competing objects for a decision, there may be a hundred. The chances for a spurious mistake caused by an incidental view point are clearly much higher. On the other hand, it has significant advantages in terms of space complexity and the appeal of the evaluation mode.

SVMs are two-class classifiers which, for an c-class pattern recognition problem, need to train c*(c-1)/2 binary classifiers. Since we compare the performance of the proposed SNoW-based method with SVMs, in order to maintain a fair comparison we have to perform it in the {\em one-against-one} scheme. That is, we use SNoW units of size two. To classify a test instance, tournament-like pair-wise competition between all the machines is performed and the winner determines
the label of the test instance.  The recognition rates of the SVM and SNoW based methods shown in Table 2 were performed using the one-against-one scheme. That is, we trained 4,950 classifiers for each method and evaluated 99 (=50+25+12+6+3+2+1) classifiers on each test instance.

2.3 Results Using Pixel-Based Representation

Table 2. shows the recognition rates of the SNoW-based method, the SVM-based method (using linear dot product for the kernel function), and the nearest neighbor classifier using the COIL-100 dataset. The important parameter here is that we vary the number of views of an object (v) during training and use the rest of the views (72-v) of an object for testing.Table 2. Experimental results of three classifiers using the 100 objects in the COIL-100 dataset

	# of View/Test
	36/3600 tests	18/5400 tests	8/6400 tests	4/6800 tests
SNoW	95.81%	92.31%	85/13%	81.46%
Linear SVM	96.03%	91.30%	84.80%	78.50%
Nearest Neighbor	98.50%	87.54%	79.52%	74.63%

The experimental results show that the SNoW-based method performs as well as the SVM-based method when many views of the objects are present during training and outperforms SVM-based method when the numbers of views is limited. Although it is not surprising to see that the recognition rate decreases as the number of views available during training decreases, it is worth noticing that both SNoW and SVM are capable of recognizing 3D objects in the COIL-100 dataset with satisfactory performance if enough views (e.g., >18) are provided. Also they seems to be fairly robust even if only a limited number of views (e.g., 8 and
4) are used for training; the performance of both methods degrades gracefully.

To provide some more insight into these methods, we note that in the SVM-based methods, only 27.78% (20 out of 72) of the input vectors serves as support vectors. For SNoW, out of 262,144 potential features in the pixel-based representation, only 13,805 were active in the dense case (i.e., 36 views). This shows the advantage gained from using the sparse  rchitecture. However, only a small number of those may be relevant to the representation of each target, as a more careful look as the SNoW output hypothesis reveals.

An additional potential advantage of the SNoW architecture is that it does not learn discriminators, but rather can learn a representation for each object, which can then be used for prediction in the one-against-all scheme or to build hierarchical representations. However, as is shown in Table~\ref{chap6-table4}, this implies a significant degradation is the performance.  Finding a way to make better predictions in the one-against-all scheme is one of the important issues for future investigation, to better exploit the advantages of this approach.

Table 3. Recognition rates of SNoW using two learning paradigms

	# of View
SNoW	36	18	8	4
One-against-one	95.81%	92.31%	85.13%	81.46%
Linear SVM	90.52%	85.50%	81.85%	76.00%

2.4  Results Using Edge-Based Representation

For each 32 x 32 edge map, we extract horizontal and vertical edges (of length at least 3 pixels) and then encode as our features conjunctions of two of these edges. The number of potential features of this sort is 2,096,128. However, only an average of 1,822 of these is active for objects in the COIL-100 dataset.  To reduce the computational cost the feature vectors were further pruned and only the 512 most frequently occurring features were retained in each image.Table 3 shows the performance of the SNoW-based method when conjunctions of edges are used to represent objects.  As before, we vary the number of views of an object n during training and use the rest of the views 72-n of an object for testing.  The results indicate that conjunctions of edges provide useful information for object recognition and that SNoW is able to learn very good object representations using these features.  The experimental results also
exhibit the relative advantage of this representation increases when the number of views per object is limited.

Table 4. Experimental results of four classifiers using the 100 objects in the COIL-100 dataset

	# of View/Test
	36/3600 tests	18/5400 tests	8/6400 tests	4/6800 tests
SNoW w/ conjunction of edges	96.25%	94.13%	89.23%	88.28%
SNoW w/ intensity values	95.81%	92.31%	85.13%	81.46%
Linear Support Vector Machine	96.03%	91.30%	84.80%	78.50%
Nearest Neighbor	98.50%	87.54%	79.52%	74.63%

3. Conclusion

The main contribution of this work we view in proposing a learning framework for visual learning and exhibiting its feasibility. In this approach learnability can be rigorously studied without making assumptions on the distribution of the observed objects but, via the PAC model, the learned hypothesis’ performance naturally depends on its prior experience. An important feature of the approach is that learning is not studied directly in terms of the raw data but rather with respect to intermediate representations extracted from it and can thus be quantified in terms of the ability to generate expressive intermediate representations. In particular, it makes explicit the requirements from these representations to allow learnability. We believe that research in vision should concentrate on the study of these intermediate representations.We evaluated the approach and demonstrated its feasibility in a large scale experiment in the context of learning for object recognition. Our experiments allowed us also to perform a fair comparison between two successful and related  learning methods and study them in the context of object recognition. We have illustrated our approach in a large scale experimental study in which we use the SNoW learning architecture to learn representations for the 100 objects in COIL-100.  Although it is
clear that object recognition in isolation is not the ultimate goal, this study shows the potential of this computational approach as a basis for studying and supporting more realistic visual inferences.

We note that for a fair comparison among different methods, we have used pixel-based presentation in the experiments. It is clear, however, that the edge-based representation used is even more effective and robust and should be the starting point for future research. There is no question that the RGFs used in this
work are not general enough to support more challenging recognition problems; the intention was merely to exhibit the general approach. We believe that pursing the direction of using complex intermediate representations will benefit future work on recognition and, in particular, robust recognition under various types of noise.

Publications

1. Dan Roth, Ming-Hsuan Yang and Narendra Ahuja, Learning to Recognize Objects, to appear in Neural Computation, 2001.  Abstract
2. Ming-Hsuan Yang, Dan Roth and Narendra Ahuja, Learning to Recognize 3D Objects With SNoW, podium presentation, in Proceedings of the Sixth European Conference on Computer Vision (ECCV 2000) , pp. 439-454, vol. 1, Dublin, June, 2000.  Full Text
3. Dan Roth, Ming-Hsuan Yang and Narendra Ahuja, Learning to Recognize Objects, in Proceedings of the 2000 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2000), pp. 724-731, vol. 1, Hilton Head, June, 2000.  Abstract

1. Hongcheng Wang, Qing Wu, Lin Shi, Yizhou Yu and Narendra Ahuja, Out-of-Core Tensor Approximation of Multi-Dimensional Matrices of Visual Data, in ACM SIGGRAPH 2005.

Publications

1. Y. Kitamura, J. Ohya, N. Ahuja and F. Kishino, Computational Taxonomy and Recognition of Facial Expressions, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 434-437.
2. J. Ma, N. Ahuja, C. Neti and A. Senior, Recovering Frontal-Pose Image from a Single Profile Image, IEEE International Conference on Image Processing, Vol. 2, Vancouver, BC, Canada, September 2000, 243-247.  Abstract
3. M.-H. Yang, N. Ahuja and D. Kriegman, Face Recognition Using Kernel Eigenfaces, IEEE Int. Conference on Image Processing, Vol. 1, Vancouver, BC, Canada, Sept. 2000, 37-41.  Full Text
4. M.H. Yang and N. Ahuja, A Geometric Approach to Train Support Vector Machines, IEEE International Conference on Computer Vision and Pattern Recognition, Vol. I, Hilton Head, SC, June 2000, 430-437.  Abstract
5. M.H. Yang and N. Ahuja, Face Detection and Hand Gesture Recognition for Vision-Based Human Computer Interaction, Kluwer Academic Publishers, 2001.  Abstract

Two-channel predictive multiple description coding is posed as a variant of the Wyner-Ziv coding problem. Practical code constructions are proposed within this framework, and the performance of the proposed codes is compared with conventional approaches, for communication of a first-order Gauss-Markov source over erasure channels with independent failure probabilities.

1. A. Jagmohan, A. Sehgal, N. Ahuja, “WYZE-PMD based Multiple Description Video Codec,” Proc. IEEE Int. Conf. Multimedia Expo, 2003, pp. I-569-572 Full Text
2. A. Jagmohan, N. Ahuja, “Wyner-Ziv Encoded Predictive Multiple Descriptions,” Proc. Data Compression Conference, p. 213 2003. Full Text
3. A. Jagmohan, A. Sehgal, N. Ahuja, “Predictive Encoding using Coset Codes, ” Invited Paper, Proc. IEEE Int. Conf. Image Processing, pp. I-29-32, 2002. Full Text

Publications

1. M.H. Yang, N. Ahuja and D. Roth, View-Based 3D Object Recognition Using SNoW, Proc. of the Fourth Asian Conference on Computer Vision, Vol. 2, Taipei, January 2000, 830-835.  Abstract
2. D. Roth, M.-H. Yang and N. Ahuja, Learning to Recognize Objects, IEEE International Conference on Computer Vision and Pattern Recognition, Vol. I, Hilton Head, SC, June 2000, 724-731.
3. J. Weng, N. Ahuja and T. Huang, Learning, Recognition and Segmentation of 3-D Objects from 2-D Images, 4th Int. Conf. on Computer Vision, Berlin, Germany, May 11-14, 1993, 121-128.

Publications

1. J. Weng, N. Ahuja and T. S. Huang, Learning Recognition and Segmentation Using the Cresceptron, International Journal of Computer Vision, 25(2), 1997, 109-143.  Abstract
2. N. Srinivasa and N. Ahuja, A Topological and Temporal Correlator Network for Spatiotemporal Pattern Learning, Recognition and Recall, IEEE Transactions on Neural Networks, Vol. 10, No. 2, March 1999, 356-371.  Full Text
3. J. Weng, N. Ahuja and T. S. Huang, Cresceptron: A Self-Organizing Neural Network Which Grows Adaptively, Proc. Int. Joint Conf. on Neural Networks, Vol. 1, Baltimore, Maryland, June 1992, 576-581.  Abstract
4. B. Perrin, N. Ahuja and N. Srinivasa, Learning Multiscale Image Models of 2D Object Classes, 3rd Asian Conference on Computer Vision, Hong Kong, January 8-11, 1998, 323-331.  Full Text

Publications

1. Hongcheng Wang and Narendra Ahuja, Rank-R Approximation of Tensors Using Image-as-Matrix Representation, IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), 2005.  Full Text
2. Hongcheng Wang and Narendra Ahuja, Compact Representation of Multidimensional Data Using Tensor Rank-One Decomposition, Pattern Recognition, 17th International Conference on (ICPR’04) Volume 1, pp. 44-47 08 23 – 08, 2004, Cambridge UK. Full Text

Publications

1. Hongcheng Wang, Narendra Ahuja, Facial Expression Decomposition, Ninth IEEE International Conference on Computer Vision Volume 2, p. 958 10 13 – 10, 2003, Nice, France. Full Text

1. C. Gao, H. Hua and N. Ahuja, A Hemispherical Imaging Camera, Computer Vision and Image Understanding, 2009, in press. Abstract Full Text
2. S. Todorovic and N. Ahuja, Texel-based Texture Segmentation, International Conference on Computer Vision (ICCV), Kyoto, Japan, September-October 2009.
3. E. Akbas and N. Ahuja, From ramp discontinuities to segmentation tree, 9th Asian Conference on Computer Vision (ACCV), Xi’an, China, September 2009. Full text
4. Q.Kong, A.Kumar, N.Ahuja and Y.Liu, Robust Segmentation of Freight Containers in Train Monitoring Videos, Workshop on Applications of  Computer Vision(WACV), Snowbird, Utah, December 2009.

2008

2007

2006

1. C. Gao and N. Ahuja, A Refractive Camera for Acquiring Stereo Depth and Super-Resolution Images, Proceedings IEEE Conference on Computer Vision and Pattern Recognition, New York, NY Vol. 2, June 2006, 2316-2323. Full Text
2. S. Todorovic and N. Ahuja, Extracting subimages of an unknown category from a set of images,  in Proc. IEEE Comp. Soc. Conf. Computer Vision and Pattern Recognition (CVPR 2006), vol. 1, pp. 927-934, New York, NY, 2006. Full Text
3. S. Todorovic and N. Ahuja, 3D texture classification using the belief net of a segmentation tree, in Proc. 18th Int. Conf. Pattern Recognition (ICPR 2006), Hong Kong, China, 2006. Full Text
4. Alexia Briassouli, Narendra Ahuja, Spatial and Fourier Error Minimization for Motion Estimation and Segmentation, ICPR 2006, Hong Kong Full Text
5. Alexia Briassouli, Narendra Ahuja, Estimation of Multiple Periodic Motions from Video, ECCV 2006, Graz, Austria
6. Hongcheng Wang, Yunqiang Chen, Tong Fang, Jason Tyan and Narendra Ahuja, Gradient Adaptive Image Restoration and Enhancement, in Proc. IEEE International Conference on Image Processing (ICIP), Atlanta, VA, Aug. 2006 Full Text
7. Tianli Yu, Hongcheng Wang, Narendra Ahuja, Wei-Chao Chen, Sparse Lumigraph Relight by Illumination and Reflectance Estimation from Multi-View Images, Eurographics Symposium on Rendering (EGSR), 2006 Full Text
8. Tianli Yu, Hongcheng Wang, Narendra Ahuja, Wei-Chao Chen, Sparse Lumigraph Relight by Illumination and Reflectance Estimation from Multi-View Images, Technical Sketch, SIGGRAPH, 2006 Full Text
9. Himanshu Arora, Narendra Ahuja, Analysis of ramp discontinuity model for multiscale image segmentation. ICPR(1), 2006.
10. T. Yu, N. Ahuja and W-C. Chen, SDG Cut: 3D Reconstruction of Non-Lambertian Objects Using Graph Cuts on Surface Distance Grid, Proceedings IEEE Conference on Computer Vision and Pattern Recognition, New York, NY, Vol. 2, June 2006, 2269-2276. Full Text
11. C-B. Liu, R-S. Lin, M-H. Yang, N. Ahuja and S. Levinson, Object Tracking Using Globally Coordinated Nonlinear Manifolds, International Conference on Pattern Recognition, Vol. 1, Hong Kong, August 2006, 844-847. Full Text
12. S.Yi and Narendra Ahuja, An Omnidirectional Stereo Vision System Using a Single Camera, International Conference on Pattern Recognition, Vol. 4, Hong Kong, August 2006, 861-847. Full Text
13. S. Yi and N. Ahuja, A Novel Omnidirectional Stereo Vision System with a Single Camera, International Conference on Image Analysis and Recognition, Portugal, September 2006.  Full Text

2005

1. Hongcheng Wang and Narendra Ahuja, Rank-R Approximation of Tensors Using Image-as-Matrix Representation, IEEE International Conference on Computer Vision and Pattern Recognition (CVPR), 2005 Full Text
2. Hongcheng Wang, Qing Wu, Lin Shi, Yizhou Yu and Narendra Ahuja, Out-of-Core Tensor Approximation of Multi-Dimensional Matrices of Visual Data,  in ACM SIGGRAPH 2005. Full Text
3. Hongcheng Wang, Ning Xu, Ramesh Raskar and Narendra Ahuja, Videoshop: A New Framework for Video Editing in Spatio-Temporal Gradient Domain, IEEE, Video Proceedings, International Conference on Computer Vision and Pattern Recognition, 2005  Full Text
4. Tianli Yu, Jiebo Luo and Narendra Ahuja Shape Regularized Active Contour using Iterative Global Search and Local Optimization, accepted by CVPR 2005, June 20-26 2005, San Diego, CA, USA Full Text
   
5. Tianli Yu, Jiebo Luo, Amit Singhal, and Narendra Ahuja Shape regularized active contour based on dynamic programming for anatomical structure segmentation, SPIE Medical Imaging 2005, February 12-17 2005, San Diego, CA, USA Full Text
   
6. Alexia Briassouli, Narendra Ahuja, Integrated Spatial and Frequency Domain 2D Motion Segmentation and Estimation, ICCV 2005, Beijing, China, [Full Text]

2004

1. M. Aggarwal and N. Ahuja, Split Aperture Imaging for High Dynamic Range, International Journal on Computer Vision, Vol. 58, No. 1, June 2004, 7-17. Abstract and Full Text
2. Che-Bin Liu and Narendra Ahuja. Vision Based Fire Detection. Pattern Recognition, 17th International Conference on (ICPR’04) Volume 4, pp. 134-137, 08 23 – 08, 2004, Cambridge UK Full Text
   
3. Chunyu Gao, Narendra Ahuja, Single camera stereo using planar parallel plate , Pattern Recognition, 17th International Conference on (ICPR’04) Volume 4, pp. 108-111 08, 2004, Cambridge UK Full Text
   
4. Alexia Briassouli, Narendra Ahuja, Fusion of frequency and spatial domain information for motion analysis , Pattern Recognition, 17th International Conference on (ICPR’04) Volume 2, pp. 175-178, 2004 Cambridge UK Full Text
   
5. Hongcheng Wang and Narendra Ahuja, Compact Representation of Multidimensional Data Using Tensor Rank-One Decomposition, Pattern Recognition, 17th International Conference on (ICPR’04) Volume 1, pp. 44-47 08 23 – 08, 2004, Cambridge UK Full Text
6. Ning Xu and Narendra Ahuja. A Three-view Matching Algorithm Considering Foreshortening Effects. In Proceedings of International Conference on Computer Vision, Pattern Recognition and Image Processing, pp. 635-638, Cary, NC. September 2003. Abstract and Full Text
7. Ning Xu and Narendra Ahuja. Generating Omnifocus Images Using Graph Cuts and a New Focus Measure. Pattern Recognition, 17th International Conference on (ICPR’04) Volume 4. pp 697-700, 08, 2004 Cambridge UK. Full Text
8. A. Jagmohan, M. Singh, and N. Ahuja, Dense Two View Stereo Matching Using Kernel Maximum Likelihood Estimation, Pattern Recognition, 17th International Conference on (ICPR’04) Volume 3, pp. 28-31, 08, 2004 Cambridge UK Abstract
9. M. Singh, H. Arora and N. Ahuja, Robust Registration and Tracking Using Kernel Density Correlation, 2004 Conference on Computer Vision and Pattern Recognition Workshop on Image and Video Registration, CVPRW’04 Volume 11, p. 174, 06 27 – 07 02, 2004, Washington, D.C., USA Full Text
10. Che-bin Liu, Narendra Ahuja, A Model for Dynamic Shape and Its Applications , 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’04) Volume 2, pp. 129-134 06 27 – 07 02, 2004, Washington, D.C., USA Full Text
    
11. Tianli Yu, Ning Xu and Narendra Ahuja, Recovering Shape and Reflectance Model of Non-Lambertian Objects from Multiple Views, 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’04) Volume 2, pp. 226-233, 06 27 – 07 02, 2004, Washington, D.C., USA Full Text
12. Tianli Yu, Ning Xu and Narendra Ahuja, Shape and View Independent Reflectance Map from Multiple Views, ECCV 2004, LNCS 3024, pp. 602-616, May 11-14, Prague. Full Text
13. M. Singh, H. Arora and N. Ahuja, A Robust Probabilistic Estimation Framework for Parametric Image Models, European Conference on Computer Vision, LNCS 3021, pp. 508-522, 2004. Full Text
14. A. Sehgal, A. Jagmohan, N. Ahuja Wyner-Ziv Coding of Video: Applications to Error Resilience, IEEE Trans. Multimedia, April 2004, pages 249 – 258. Full Text
15. Ning Xu, Tianli Yu and Narendra Ahuja. Shape from color consistency using node cut. In Proceedings of Asian Conference on Computer Vision, Jeju Island, Korea. January 2004. Abstract and Full Text

N. Ahuja and A. Rosenfeld, Mosaic Models for Textures, 18th IEEE Conf. Decision and Control, Fort Lauderdale, December 1979, Proc. 60-70. Abstract

N. Ahuja, R. T. Chien, R. Yen and N. Bridwell, Interference Detection and Collision Avoidance Among Three Dimensional Objects, Proc. 1st National Conf. on Artificial Intelligence, Stanford University, August 19-21, 1980, 44-48. Abstract

N. Ahuja, Dot Pattern Processing Using Voronoi Polygons as Neighborhoods, Proc. 5th Int. Joint Conf. on Pattern Recognition, Miami Beach, December 1-4, 1980, 1122-1127. Abstract

N. Ahuja, Dot Pattern Processing Using Voronoi Polygons as Neighborhoods, presented at IEEE Image Segmentation Workshop, VPI, Blacksburg, May 15-16, 1980. Abstract

N. Ahuja, Approaches to Recursive Image Decomposition, Proc. IEEE Conf. on Pattern Recognition and Image Processing, Dallas, August 3-5, 1981, 75-80.

N. Ahuja, N. Bridwell, C. Nash and T. S. Huang, Three-Dimensional Robot Vision, Proc. IEEE International Workshop on Industrial Applications of Machine Vision, Raleigh-Durham, May 3-5, 1982, 206-213. Abstract

N. Ahuja and S. Swamy, Multiprocessor Pyramids for Bottom-up Image Analysis, Proc. IEEE Conf. on Pattern Recognition and Image Processing, Las Vegas, June 13-17, 1982, 380-385. Abstract

N. Ahuja and S. Swamy, Interleaved Pyramid Architectures for Bottom-Up Image Analysis, Proc. 6th Int. Conf. on Pattern Recognition, Munich, Germany, October 19-22, 1982, 388-391. Abstract

M. Tuceryan and N. Ahuja, Segmentation of Dot Patterns Containing Homogeneous Clusters, Proc. 6th Int. Conf. on Pattern Recognition, Munich, Germany, October 19-22, 1982, 392-394. Abstract

B. An, N. Ahuja and B. Schachter, Image Representation Using Voronoi Tessellation, Proc. IEEE Conf. on Computer Vision and Pattern Recognition, Washington, D. C., June 19-23, 1983, 188-189. Abstract

C. Nash and N. Ahuja, Updating Octrees of Translating Objects, Proc. IEEE Conf. on Computer Vision and Pattern Recognition, Washington, D. C., June 19-23, 1983, 380-381. Abstract

M. Tuceryan and N. Ahuja, Perceptual Segmentation of Nonhomogeneous Dot Patterns, Proc. IEEE Conf. on Computer Vision and Pattern Recognition, Washington D. C., June 19-23, 1983, 47-52. Abstract

H. C. Chen, N. Ahuja and T. S. Huang, Septree Representation of Moving Objects Using Hexagonal Cylinderical Decomposition, Proc. Intelligent Robots: 3rd Int. Conf. on Robot Vision and Sensory Controls, Cambridge, November 6-10, 1983. Abstract

N. Ahuja, Dot Pattern Perception, Proc. National Symposium on Image Signal Processing, Indian Institute of Technology, Madras, India, January 9-12, 1984.

N. Ahuja, Efficient Planar Embedding of Trees for VLSI Layouts, IEEE Int. Conf. on Systems, Man and Cybernetics, Bombay/New Delhi, India, December 30, 1983 – January 7, 1984, Proc. 28-32. Abstract

N. Ahuja, Efficient Planar Embedding of Trees for VLSI layouts, Proc. 7th Int. Conf. on Pattern Recognition, Montreal, Canada, July 30 – Aug 2, 1984, 460-464. Abstract

N. Ahuja and W. Hoff, Augmented Medial Axis Transform, Proc. IEEE Workshop on Computer Vision, Annapolis, Maryland, April 30 – May 2, 1984, 251-256.

C. H. Debrunner and N. Ahuja, A Bottom-up Minimum Spanning Tree Algorithm for Multiprocessor Pyramid Architectures, Proc. Workshop on Algorithm-Guided Parallel Architectures for Automatic Target Recognition, Leesburg, Virginia, July 16-18, 1984, 51-78. Abstract

W. Osse and N. Ahuja, Efficient Octree Representation of Moving Objects, Proc. 7th Int. Conf. on Pattern Recognition, Montreal, Canada, July 30 – Aug 2, 1984, 821-823. Abstract

R. Ruff and N. Ahuja, Robot Path Planning in a Three-Dimensional Environment, Proc. 7th Int. Conf. on Pattern Recognition, Montreal, Canada, July 30 – Aug 2, 1984, 188-191. Abstract

N. Ahuja and W. Hoff, Augmented Medial Axis Transform, Proc. 7th Int. Conf. on Pattern Recognition, Montreal, Canada, July 30 – Aug 2, 1984, 336-338. Abstract

N. Ahuja and M. Tuceryan, Representing Geometric Structure in Dot Patterns, Geobild ©85, Georgenthal, DDR, January 14-18, 1985.

J. Veenstra and N. Ahuja, Octree Generation from Silhouette Views of an Object, IEEE Int. Conf. on Robotics and Automation, St. Louis, March 25-28, 1985, 843-848. Abstract

N. Ahuja and M. Tuceryan, Extracting Perceptual Structure in Dot Patterns, IEEE Workshop on Languages for Automation: Cognitive Aspects in Information Processing, Palma de Mallorca, Spain, June 28-29, 1985, 86-91.

W. Hoff and N. Ahuja, Generating Range Map from Stereo Images, 4th Scandinavian Conf. on Image Analysis, Trondheim, Norway, June 18-20, 1985, 761-768.

J. Weng and N. Ahuja, Octree Representation of Objects in Arbitrary Motion, IEEE Conf. on Computer Vision and Pattern Recognition, San Francisco, June 9-13, 1985, 524-529. Abstract

W. Hoff and N. Ahuja, Depth from Stereo, IEEE Conf. on Computer Vision and Pattern Recognition, San Francisco, June 9-13, 1985.

M. Sharma, J. H. Patel and N. Ahuja, NETRA: A Multiprocessor Computer Architecture for Image Understanding, IEEE Workshop on Computer Architecture for Pattern Analysis and Image Database Management, Miami, November 18-20, 1985, 92-98. Abstract

W. Hoff and N. Ahuja, Surfaces from Stereo, Proc. DARPA Image Understanding Workshop, Miami, December 9-10, 1985, 98-106.

J. Veenstra and N. Ahuja, Efficient Octree Generation from Silhouettes, IEEE Conf. on Computer Vision and Pattern Recognition, Miami, June 22-26, 1986, 537-542. Abstract

T. S. Huang, J. Weng and N. Ahuja, 3-D Motion from Image Sequences: Modeling, Understanding and Prediction, IEEE Workshop on Motion: Representation and Analysis, Kiawah Island, May 6-9, 1986, 516-518.

W. Hoff and N. Ahuja, Surfaces from Stereo, 8th International Conference on Pattern Recognition, Paris, France, October 28-31, 1986, 516-518.

J. Weng, N. Ahuja and T. Huang, 3-D Motion from Image Sequences: Modeling, Estimation and Prediction, 8th International Conference on Pattern Recognition, Paris, France, October 28-31, 1986, 1107-1109.

W. Hoff and N. Ahuja, Extracting Surfaces from Stereo Images: An Integrated Approach, First International Conference on Computer Vision, London, England, June 8-11, 1987, 284-294.

D. Blostein and N. Ahuja, Representation and Three-dimensional Interpretation of Image Texture: An Integrated Approach, First International Conference on Computer Vision, London, England, June 8-11, 1987, 444-449.

J. Weng, T. Huang and N. Ahuja, Error Analysis of Motion Parameter Estimation from Image Sequences, First International Conference on Computer Vision, London, England, June 8-11, 1987, 703-707.

S. Srivastava and N. Ahuja, An Algorithm for Generating Octrees from Object Silhouettes in Perspective Views, Proc. IEEE Workshop on Computer Vision, Miami Beach, November 30 -December 2, 1987, 363-365. Abstract

J. Weng, T. Huang and N. Ahuja, A Two-Step Approach to Optimal Motion and Structure Estimation from Feature Correspondences, Proc. IEEE Workshop on Computer Vision, Miami Beach, November 30 -December 2, 1987, 355-357.

Y. Hwang and N. Ahuja, Path Planning Using a Potential Field Representation, Proc. IEEE International Conference on Robotics and Automation, Philadelphia, April 25-29, 1988, 648-649. Abstract

J. Weng, N. Ahuja and T. Huang, Closed-Form Solution+Maximum Likelihood: A Robust Approach to Motion and Structure Estimation, IEEE Conf. on Computer Vision and Pattern Recognition, Ann Arbor, June 1988, 381-386. Abstract and Full Text

N. Ahuja and T. Huang, IU at UI: An Overview and an Example on Shape from Texture, Proc. DARPA Image Understanding Workshop, Boston, April 6-8, 1988, 222-253.

J. Weng, Y. Liu, T. Huang and N. Ahuja, Determining Motion/Structure from Line Correspondences: A Robust Linear Algorithm and Uniqueness Theorems, IEEE Conf. on Computer Vision and Pattern Recognition, Ann Arbor, June 1988, 387-392. Abstract and Full Text

J. Weng, N. Ahuja and T. Huang, Motion and Structure from Point Correspondences: A Robust Algorithm for Planar Case with Error Estimation, 9th Int. Conference on Pattern Recognition, Rome, Italy, November, 1988, 247-251.

J. Weng, N. Ahuja and T. Huang, Two-View Matching, Second International Conference on Computer Vision, Tarpon Springs, December 5-8, 1988, 64-73. Abstract and Full Text

L. Abbott and N. Ahuja, Surface Reconstruction by Dynamic Integration of Focus, Camera Vergence and Stereo, Second International Conference on Computer Vision, Tarpon Springs, December 5-8, 1988, 532-543. Abstract and Full Text

A. N. Choudhary, S. Das, N. Ahuja and J. Patel, Surface Reconstruction from Stereo Images: An Implementation on a Hypercube Multiprocessor, Fourth Conference on Hypercube Concurrent Computers and Applications, Monterey, CA, March 1989, 1045-1052. Abstract

J. Weng, T. Huang and N. Ahuja, Motion from Images: Image Matching, Parameter Estimation and Intrinsic Stability, IEEE Workshop on Visual Motion, Irvine, March 20-22, 1989, 359-366. Abstract and Full Text

Y. Hwang and N. Ahuja, Path Planning using a Potential Field Representation, IEEE Conference on Computer Vision and Pattern Recognition, San Diego, June 4-8, 1989, 569-575. Abstract

J. Weng, N. Ahuja and T. Huang, Optimal Motion and Structure Estimation, IEEE Conference on Computer Vision and Pattern Recognition, San Diego, June 4-8, 1989, 144-152. Abstract & PDF Full Text

S. Das, A. L. Abbott and N. Ahuja, Surface Reconstruction from Focus and Stereo, Proc. 5th International Conference on Image Analysis and Processing, Positano, Italy, September 1989. Abstract

S. Das and N. Ahuja, Integrating Mutilresolution Image Acquisition and Coarse-to-fine Surface Reconstruction from Stereo, IEEE Workshop on Interpretation of 3D Scenes, Austin, November 26-29, 1989, 9-15.

C. Debrunner and N. Ahuja, A Hankel Matrix Based Motion Estimation Algorithm, 10th International Conference on Pattern Recognition – Computer Vision, Vol. 1, Atlantic City, June 1990, 384-389.

A. N. Choudhary, S. Das, N. Ahuja, and J. H. Patel, A Reconfigurable and Hierarchical Parallel Processing Architecture: Performance Results for Stereo Vision, 10th International Conference on Pattern Recognition – Computer Vision, Vol. 2, Atlantic City, June 1990, 389-393.Abstract

J. Weng, T. S. Huang, and N. Ahuja, Estimating Motion and Structure from Line Matches, 10th International Conference on Pattern Recognition – Computer Vision, Vol. 1, Atlantic City, June 1990, 168-172. Abstract and Full Text

S. Das and N. Ahuja, Active Stereo Based Image Acquisition and Surface Reconstruction, Proc. 5th IEEE Int. Symposium on Intelligent Control, Philadelphia, Sep 5-7, 1990, 233-238. Abstract

N. Ahuja, IU at UI: An Overview of Research during 1988-90, Proc. DARPA Image Understanding Workshop, Pittsburgh, Sep 11-13, 1990, 134-140.

S. Das and N. Ahuja, Integrated Multiresolution Image Acquisition and Surface Reconstruction from Active Stereo, Proc. DARPA Image Understanding Workshop, Pittsburgh, Sep 11-13, 1990, 418-422. Abstract

E. Altman and N. Ahuja, A Dynamical Systems Approach to Integration in Stereo, Proc. DARPA Image Understanding Workshop, Pittsburgh, Sep 11-13, 1990, 423-427.

X. Hu and N. Ahuja, A Robust Algorithm for Plane Motion Estimation, First International Conf. on Automation, Robotics and Computer Vision, Singapore, Sep 18-21, 1990.

S. Das and N. Ahuja, Multiresolution Image Acquisition and Surface Reconstruction, Proc. 3rd Int. Conf. on Computer Vision, Osaka, Japan, Dec 4-7, 1990, 485-488. Abstract

A. L. Abbott and N. Ahuja, Active Surface Reconstruction by Integrating Focus, Vergence, Stereo and Camera Calibration, Proc. 3rd Int. Conf. on Computer Vision, Osaka, Japan, Dec 4-7, 1990, 489-492. Abstract

S. Das and N. Ahuja, Dynamic Integration of Visual Cues for Position Estimation, Proc. SPIE Conference on Intelligent Robots and Computer Vision IX: Neural, Biological and 3-D Methods, Vol. 1382, Boston, November 1990, 341-352. Abstract

J. Chuang and N. Ahuja, Path Planning using Newtonian Potential, Proc. IEEE Conference on Robotics and Automation, Sacramento, April 8-10, 1991, 558-563. Abstract

N. Ahuja, Issues in the Design of Integrated Vision Systems, Int. Joint Conference on Artificial Intelligence Workshop W-7, Sydney, Australia, August 24, 1991, Part III. Abstract

X. Hu and N. Ahuja, Motion and Structure from Orthographic Images, Proc. IEEE Conference on Acoustics, Speech and Signal Processing, Toronto, Canada, May 14-17, 1991, 2445-2448.

S. Sull and N. Ahuja, Estimation of Motion and Structure of Planar Surfaces from a Sequence of Monocular Images, Proc. IEEE Conference on Computer Vision and Pattern Recognition, Maui, Hawaii, June 1991, 732-733.

S. Sull and N. Ahuja, Segmentation and Estimation of Structure and Motion of Textured Piecewise Planar Surfaces, Proc. IEEE Workshop on Visual Motion, Princeton, New Jersey, October 7-9, 1991, 274-279.

J. Chuang and N. Ahuja, Skeletonization Using a Generalized Potential Field Model, Proc. 8th Israeli Symposium on Artificial Intelligence and Computer Vision, December 30-31, 1991. Abstract

N. Ahuja and T. S. Huang, IU at UI: An Overview of Research during 1990-91, Proc. DARPA Image Understanding Workshop, San Diego, January 27-29, 1992, 127-135.

S. Sull and N. Ahuja, Integrated 3D Recovery and Visualization of Flight Image Sequences, Proc. DARPA Image Understanding Workshop, San Diego, January 27-29, 1992, 473-477.

C. Debrunner and N. Ahuja, Motion and Structure Factorization and Segmentation of Long Multiple Motion Image Sequences, Proc. DARPA Image Understanding Workshop, San Diego, January 27-29, 1992, 543-547.

C. Debrunner and N. Ahuja, Motion and Strucuture Factorization and Segmentation of Long Multiple Motion Image Sequences, Second European Conference on Computer Vision, Santa Margherita Ligure, Italy, May 18-23, 1992, 217-221.

J. Weng, N. Ahuja and T. S. Huang, Cresceptron: A Self-Organizing Neural Network Which Grows Adaptively, Proc. Int. Joint Conf. on Neural Networks, Vol. 1, Baltimore, Maryland, June 1992, 576-581.

M. Tuceryan, A. K. Jain and N. Ahuja, Supervised Classification of Early Perceptual Structure in Dot Patterns, 11th Int. Conf. on Pattern Recognition – Computer Vision, The Hague, August 30 – September 2, 1992, B88-91.  Abstract

X. Hu and N. Ahuja, Estimating Motion of Constant Acceleration from Image Sequences, 11th Int. Conf. on Pattern Recognition – Computer Vision, The Hague, August 30 – September 2, 1992, A655-659. Abstract and Full Text

J. Chuang and N. Ahuja, Robot Arm Path Planning Using the Newtonian Potentiial, Int. Conf. on Control and Robotics, Vancouver, Canada, August 14-17, 1992, 14-17. Abstract

D. Hougen and N. Ahuja, Integration of Stereo and Shape from Shading using Color, Proc. Second International Conf. on Automation, Robotics and Computer Vision, Vol 1, Singapore, September 15-18 1992, pp. CV-6.6.1 – CV-6.6.5.

A. L. Abbott and N. Ahuja, The University of Illinois Active Vision System, Proc. SPIE Conf. on Applications of Artificial Intelligence XI: Algorithms, Techniques and Active Vision, Boston, November 1992, 757-768. Abstract

X. Hu and N. Ahuja, Long Image Sequence Motion Analysis Using Polynomial Motion Models, IAPR Workshop on Machine Vision Applications, Dec 7-9, Tokyo, 1992, 109-113.

Z. Hong and N. Ahuja, Target Tracking from Binocular Image Sequence Using the Autoregressive Moving Average Model IAPR Workshop on Machine Vision Applications, Dec 7-9, Tokyo, 1992, 317-320.

Z. Hong and N. Ahuja, Target Tracking and Cumulative Depth Map Generation from Binocular Image Sequences, Proc. 3rd Int. Conf. on Intelligent Autonomous Systems, Pittsburgh, PA, February 1993. Abstract

X. Hu and N. Ahuja, Feature Extraction and Matching as Signal Detection, Proc. SPIE Conf. on Applications of Artificial Intelligence XI: Machine Vision and Robotics, 1993, 51-65.

N. Ahuja and T. S. Huang, IU at UI: An Overview of Research during 1991-92, Proc. DARPA Image Understanding Workshop, Washington, April 19-21, 1993, 117-125.

S. Thirumalai and N. Ahuja, Detection and Segmentation of Feature Trajectories in Multiple, Discontinuous Motion Image Sequences, Proc. DARPA Image Understanding Workshop, Washington, April 19-21, 1993, 621-628.

N. Ahuja, A Transform for Detection of Multiscale Image Structure, Proc. DARPA Image Understanding Workshop, Washington, April 19-21, 1993, 893-903.

A. Krishnan and N. Ahuja, Range Estimation from Focus Using a Non-frontal Imaging Camera, Proc. DARPA Image Understanding Workshop, Washington, April 19-21, 1993, 959-965.  Full Text

J. Weng, N. Ahuja and T. Huang, Learning, Recognition and Segmentation of 3-D Objects from 2-D Images, 4th Int. Conf. on Computer Vision, Berlin, Germany, May 11-14, 1993, 121-128. Abstract and full-text

D. Hougen and N. Ahuja, Estimation of the Light Source Distribution and its Use in Shape Recovery from Stereo and Shading, 4th Int. Conf. on Computer Vision, Berlin, Germany, May 11-14, 1993, 148-155.

Y. Kitamura, H. Takemura, N. Ahuja and F. Kishino, A Study of Interference Detection in Virtual Cooperative Workspace for an Operator Assistant, Workshop on Human Interface, Tokyo, May 24, 1993, vol. 8, no. 2, 247-254. (Part in English, part in Japanese). Also published as a Technical Report of Technical Meeting on Human Communication Engineering of IEICE, HC93-17. (in Japanese). Abstract

N. Ahuja, A Transform for Detection of Multiscale Image Structure, Proc. IEEE Conference on Computer Vision and Pattern Recognition, New York, June 1993, 780-781.

S. Das and N. Ahuja, Performance Analysis of Focus, Vergence and Stereo as Depth Cues for Active Vision, Proc. IEEE Conference on Computer Vision and Pattern Recognition, New York, June 1993, 194-199. Abstract and Full Text

A. Krishnan and N. Ahuja, Range Estimation from Focus Using a Non-frontal Imaging Camera, Proc. American Association for Artificial Intelligence Conference, Washington, July 1993, 830-835.  Full Text

A. Krishnan and N. Ahuja, Use of a Non-frontal Camera for Extended Depth of Field in Wide Scenes, Proc. SPIE Conf. on Intelligent Robots and Computer Vision XII: Algorithms and Techniques, September 1993, 62-72.  Full Text

X. Hu and N. Ahuja, Necessary and Sufficient Conditions for a Unique Solution of Plane Motion and Structure, Proc. 5th Int. Conference on Computer Analysis of Images and Patterns, Budapest, Hungary, September 13-15, 1993.

T. Srikanth and N. Ahuja, Parallel Distributed Detection of Feature Trajectories in Multiple Discontinuous Motion Image Sequences, Int. Joint Conf. on Neural Networks, Vol. 3, Nagoya, Japan, October 25-29, 1993, 2069-2072.

Y. Kitamura, H. Takemura, N. Ahuja and F. Kishino, Interference Detection Among Objects for Operator Assistance in Virtual Cooperative Workspace, 2nd IEEE Int. Workshop on Robot and Human Communication (RO-MAN©93), November 3-5, 1993, Tokyo, Japan, 442-446. Abstract

A. Krishnan and N. Ahuja, Stereo Display of Large Scenes from Monocular Images Using a Novel Non-frontal Camera, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 133-136.  Full Text

M. Tabb and N. Ahuja, Detection and Representation of Multiscale Low-level Image Structure Using a New Transform, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 155-158.

S. Sull and N. Ahuja, Integrated Matching and Segmentation of Multiple Features in Two Views, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 213-216.

J. Chuang, W. Lai and N. Ahuja, Skeletal Representation of 3D Objects Based on Generalized Potential Field, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 236-239. Abstract

K. Iwasaki, N. Ahuja and F. Kishino, Coarse-to-fine Pose Estimation Using Coarse-to-fine Filtering, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 664-667.

X. Hu and N. Ahuja, Mirror Uncertainity and Uniqueness Conditions for Determining Shape and Motion from Orthographic Projection, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 272-275.

E. Altman and N. Ahuja and F. Kishino, Hand Trajectory Recognition Using Dynamical Systems, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 321-324.

Y. Kitamura, H. Takemura, N. Ahuja and F. Kishino, Efficient Interference Detection Among Objects Using Octree and Polyhedral Shape Representations, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 775-779. Abstract

Y. Kitamura, J. Ohya, N. Ahuja and F. Kishino, Computational Taxonomy and Recognition of Facial Expressions, First Asian Conference on Computer Vision, November 23-25, 1993, Osaka, Japan, 434-437.

S. Sull and N. Ahuja, Integrated 3D Analysis of Flight Image Sequences, European Conference on Computer Vision, Vol. 1, Sweden, May 1994, 211-216.

S. Das and N. Ahuja, Performance Evaluation of Depth Cues for Active Vision, NSF/DARPA Workshop on Performance vs. Methodology in Computer Vision, Seattle, WA, June 1994, 227-236. Abstract

Darrell R. Hougen and Narendra Ahuja, Adaptive Polynomial Modelling of the Reflectance Map for Shape Estimation from Stereo and Shading, Proceedings IEEE Conference on Computer Vision and Pattern Recognition, Seattle, WA, June 1994, 991-994.

J. Weng, Y. Cui, N. Ahuja and A. Singh, Integration of Transitory Image Sequences, Proceedings IEEE Conference on Computer Vision and Pattern Recognition, Seattle, WA, June 20-23, 1994, 966-969. Abstract and Full Text

M. Tabb and N. Ahuja, A Multiscale Region-Based Approach to Image Matching, Proceedings International Conference on Pattern Recognition, Jerusalem, Israel, October 10-13, 1994, 415-419. Abstract and Full Text

Y. Kitamura, H. Takemura, N. Ahuja and F. Kishino, Efficient Collision Detection Among Objects in Arbitrary Motion Using Multiple Shape Representations, Proceedings International Conference on Pattern Recognition, Jerusalem, Israel, October 10-13, 1994, 390-396.

T. Joshi, N. Ahuja and J. Ponce, Towards Structure and Motion Estimation From Dynamic Silhouettes, Proc. IEEE Workshop on Motion of Nonrigid and Articulated Objects, Austin, TX November 11-12, 1994, 166-171.

S. Sull and N. Ahuja, Estimation and Segmentation of Displacement Field Using Multiple Features, International Conference on Image Processing, Austin, TX, November 14-17, 1994, 53-57.

N. Ahuja and T. S. Huang, IU at UI: An Overview of Research During 1993-94, Proceedings DARPA Image Understanding Workshop, November 1994, 133-142.

S. Das and N. Ahuja, Performance Evaluation of Depth Cues in Active Stereo Vision, Proceedings DARPA Image Understanding Workshop, November 1994, 1041-1047. Abstract

M. Tabb and N. Ahuja, Multiscale Image Segmentation Using a Recent Transform, Proceedings DARPA Image Understanding Workshop, November 1994, 1523-1530.

T. Joshi, N. Ahuja and J. Ponce, Silhouette-based Structure and Motion Estimation of a Smooth Object, Proceedings DARPA Image Understanding Workshop, November 1994, 1237-1243.

A. Krishnan and N. Ahuja, Obtaining Focused Images Using a Non-frontal Imaging Camera, Proceedings DARPA Image Understanding Workshop, November 1994, 617-620.  Full Text

T. Joshi, N. Ahuja and J. Ponce, Structure and Motion Estimation from Dynamic Silhouettes Under Perspective Projection, 5th Int. Conf. on Computer Vision, Cambridge, Massachusetts, June 20-23, 1995, 290-295.

R. Charan and N. Ahuja, Feature Guided Pixel Matching and Segmentation in Motion Image Sequences, Int. Symposium on Computer Vision, November 19-21, 1995, Coral Gables, Florida, I277-I281.

R. Charan and N. Ahuja, Curve Detection in 3D Dot Patterns, 2nd Asian Conf. on Computer Vision, Singapore, December 5-8, 1995, 199-203.Abstract

N. Ahuja and R. Charan, Pixel Matching and Motion Segmentation in Image Sequences, 2nd Asian Conf. on Computer Vision, Singapore, December 5-8, 1995, I310-I314.

N. Ahuja, A Nonfrontal Imaging Camera, Int. Computer Science Conference, Hong Kong, December 11-13, 1995, p. 397.

N. Ahuja and T. Huang, IU at UI: An Overview of Research During 1994-95, Proceedings Image Understanding Workshop, Palm Springs, CA, February 12-15, 1996, 159-164.

P. Bajcsy and N. Ahuja, Segmentation of Multidimensional Images, Proceedings Image Understanding Workshop, Palm Springs, CA, February 12-15, 1996, 937-942.

N. Ahuja and S. A. Jackson, Multiscale Region Detection, Proceedings Image Understanding Workshop, Palm Springs, CA, February 12-15, 1996, 961-967.

D. Hougen and N. Ahuja, Shape from Appearance: A Statistical Approach to Surface Shape Estimation, Proceedings Image Understanding Workshop, Palm Springs, CA, February 12-15, 1996, 1095-1101.

D. Hougen and N. Ahuja, Resolution and Accuracy of Stereo, Focus, and Shading Methods, Proceedings Image Understanding Workshop, Palm Springs, CA, February 12-15, 1996, 1133-1139.

D. Hougen and N. Ahuja, Shape from Appearance: A Statistical Approach to Surface Shape Estimation, Proceedings European Conference on Computer Vision, Cambridge, England, April 1996, 421-429.

A. Krishnan and N. Ahuja, Panoramic Image Acquisition, Proceedings IEEE Conference on Computer Vision and Pattern Recognition, San Francisco, CA, June 18-20, 1996, 379-384.  Full Text

P. Bajcsy and N. Ahuja, Uniformity and Homogeneity-based Hierarchical Clustering, Proceedings International Conference on Pattern Recognition, Vienna, Austria, August 26-29, 1996, 96-100. Abstract

S. Jackson and N. Ahuja, Elliptical Gaussian Filters, Proceedings International Conference on Pattern Recognition, Vienna, Austria, August 26-29, 1996, 775-779.

T. Courtney and N. Ahuja, Segmentation of Volume Images using a Multiscale Transform, Proceedings International Conference on Pattern Recognition, Vienna, Austria, August 26-29, 1996, 432-436.

K. Ratakonda and N. Ahuja, Segmentation Based Reversible Image Compression, International Conference on Image Processing, Geneva, Switzerland, September 1996, 81-84.Abstract

K. Ratakonda and N. Ahuja, Discrete Multi-Dimensional Linear Transforms over Connected Arbitrarily Shaped Supports, IEEE International Conference on Acoustics, Speech and Signal Processing, Vol. Abstract

4, Munich, Germany, April 1997, 3041-3044.

P. Bajcsy and N. Ahuja, A New Framework for Hierarchical Segmentation using Homogeneity Analysis, Proc. First International Conference on Scale-Space in Computer Vision, Vol. 1252, Springer LNCS, Utrecht, Netherlands, July 1997, 319-322.  Abstract

N. Ahuja and T. S. Huang, IU at UI: An Overview of Research During 1996-97. Proc. Image Understanding Workshop, New Orleans, LA, May 1997, 465-472.

P. Bajcsy and N. Ahuja, Hierarchical Image Segmentation Using Similarity Analysis, Proc. Image Understanding Workshop, New Orleans, LA, May 1997, 541-545. Abstract

K. Ratakonda and N. Ahuja, Super Resolution with Region Sensitive Interpolation, Proc. Image Understanding Workshop, New Orleans, LA, May 1997, 537-540. Abstract

N. Srinivasa and N. Ahuja, Learning to Fixate on 3D Targets with Uncalibrated Active Cameras, Proc. Image Understanding Workshop, New Orleans, LA, May 1997, 129-134. Abstract

S. C. Yoon, K. Ratakonda and N. Ahuja, Region Based Video Coding Using a Multiscale Image Segmentation, International Conference on Image Processing, Vol. II, Santa Barbara, CA, October 26-29, 1997, 510-513. Abstract and Full Text

K. Ratakonda, S. C. Yoon and N. Ahuja, Video Compression: Coding the Displaced Frame Difference, International Conference on Image Processing, Vol. I, Santa Barbara, CA, October 26-29, 1997, 353-356. Abstract and Full Text

A. Castano and N. Ahuja, Omnifocused 3D Display Using the Nonfrontal Imaging Camera, Proc. IEEE and ATR Workshop on Computer Vision for Virtual Reality Based Human Communications, CVVRHC©98, Bombay, India, January 3, 1998, 28-34.

K. Ratakonda and N. Ahuja, Restoring Image Quality Through Structure Preserving De-Noising, 3rd Asian Conference on Computer Vision, Hong Kong, January 8-11, 1998, 33-40. Abstract

B. Perrin, N. Ahuja and N. Srinivasa, Learning Multiscale Image Models of 2D Object Classes, 3rd Asian Conference on Computer Vision, Hong Kong, January 8-11, 1998, 323-331.  Full Text

N. Srinivasa and N. Ahuja, A Learning Approach to Fixate on 3D Targets with Active Cameras, 3rd Asian Conference on Computer Vision, Hong Kong, January 8-11, 1998, 623-631. Abstract

P. Bajcsy and N. Ahuja, Hierarchical Texture Segmentation, 3rd Asian Conference on Computer Vision, Hong Kong, January 8-11, 1998, 291-298.  Abstract

M.-H. Yang and N. Ahuja, Extracting Gestural Motion Trajectory, 1998 IEEE International Conference on Automatic Face and Gesture Recognition, Nara, Japan, April 1998, 10-15. Abstract and full-text

R. Dugad and N. Ahuja, Unsupervised Multidimensional Hierarchical Clustering, International Conference on Acoustics, Speech, and Signal Processing, Seattle, WA, Vol. V, May 12-15, 1998, 2761-2764. Abstract

M. Aggarwal, N. Shanbhag and N. Ahuja, Improving the Throughput of Flexible-Precision DSPs via Algorithm Transformation, International Conference on Acoustics, Speech, and Signal Processing, Vol. V, Seattle, WA, May 12-15, 1998, 3069-3072. Abstract Full Text

P. Ishwar, K. Ratakonda, P. Moulin and N. Ahuja, Image De-noising Using Multiple Compaction Domains, International Conference on Acoustics, Speech, and Signal Processing, Vol. III, Seattle, WA, May 12-15, 1998, Invited, 1889-1892. Abstract

M.-H. Yang and N. Ahuja, Extraction and Classification of Motion Patterns for Hand Gesture Recognition, IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Santa Barbara, CA, June 1998, 892-897. Abstract and full-text

J. Ma and N. Ahuja, Dense Shape and Motion from Region Correspondences, IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Santa Barbara, CA, June 1998, 219-224. Abstract and Full Text

K. Ratakonda and N. Ahuja, POCS-Based Adaptive Image Magnification, Proc. International Conference on Image Processing, Vol. 3, Chicago, IL, October 1998, 203-207. Abstract

K. Ratakonda, R. Dugad and N. Ahuja, Digital Image Watermarking: Issues in Resolving Rightful Ownership, Proc. International Conference on Image Processing, Vol. 2, Chicago, IL, October 1998, 414-418. Abstract

R. Dugad, K. Ratakonda and N. Ahuja, A New Wavelet Based Scheme for Digital Image Watermarking, Proc. International Conference on Image Processing, Vol. 2, Chicago, IL, October 1998, 419-423. Abstract

M.-H. Yang and N. Ahuja, Detecting Human Faces in Color Images, IEEE International Conference on Image Processing, Vol. 1, Chicago, IL, October 1998, 127-130. Abstract and full-text

R. Dugad, K. Ratakonda and N. Ahuja, Robust Video Shot Change Detection, IEEE Workshop on Multimedia Signal Processing, Redondo Beach, CA, December 1998, 276-278.Abstract

R. Dugad, K. Ratakonda and N. Ahuja, Robust Video Shot Change Detection, Indian Conference on Computer Vision, Graphics and Image Processing, New Delhi, India, December 1998, 358-363. Abstract

M. Singh and N. Ahuja, Estimating Light Sources, Indian Conference on Computer Vision, Graphics and Image Processing, New Delhi, India, December 1998, 76-81. Abstract and Full Text

M.-H. Yang and N. Ahuja, Gaussian Mixture Modeling of Human Skin Color and Its Applications in Image and Video Databases, Proc. of the SPIE: Storage and Retrieval for Image and Video Databases VI, Vol. 3656, San Jose, CA, Jan. 1999. Abstract

M-H. Yang and N. Ahuja, Detecting Human Faces in Color Images, IEEE Conference on Computer Vision and Pattern Recognition, Ft. Collins, CO, June 1999, I-466-472. Abstract and full-text

R. Dugad and N. Ahuja, A Fast Scheme for Altering Resolution in the compressed Domain, IEEE Conference on Computer Vision and Pattern Recognition, Ft. Collins, CO, June 1999, I-213-218. Abstract

M.-H. Yang and N. Ahuja, A Data Partition Method for Parallel Self-Organizing Map, IEEE International Joint Conference on Neural Networks, Washington, DC, July 1999.

M.-H. Yang, N. Ahuja, D. Kriegman, Face Detection using a Mixture of Factor Analyzers, International Conference on Image Processing, Kobe, Japan, Oct. 1999, III-612-616. Abstract

R. Dugad and N. Ahuja, A Fast Scheme for Downsampling and Unsampling in the DCT Domain, International Conference on Image Processing, Kobe, Japan, Oct. 1999, II-909-913. Abstract

R. Dugad and N. Ahuja, Video Denoising by Combining Kalman and Wiener Estimates, International Conference on Image Processing, Kobe, Japan, Oct. 1999, IV-152-156. Abstract

M. Singh, P. Ishwar, K. Ratakonda and N. Ahuja, Segmentation Based Denoising Using Multiple Compaction Domains, International Conference on Image Processing, Kobe, Japan, Oct. 1999, I-372-375. Abstract and Full Text

D. Roth, M.-H. Yang and N. Ahuja, A SnoW-Based Face Detector, Proc. Advances in Neural Information Procesing Systems (NIPS’99), Denver, CO, Dec. 1999, 862-868.

M.-H. Yang, N. Ahuja and D. Roth, View-Based 3D Object Recognition Using SnoW, Proc. of the Fourth Asian Conference on Computer Vision, Vol. 2, Taipei, Jan. 2000, 830-835.  Abstract

M.-H. Yang, N. Ahuja and D. Kriegman, Face Detection Using Mixtures of Linear Subspaces, Fourth IEEE Int. Conference on Automatic Face and Gesture Recognition (FG 2000), Grenoble, France, March 2000, 70-76.  Full Text

M.-H. Yang and N. Ahuja, A Geometric Approach to Train Support Vector Machines, IEEE Int. Conference on Computer Vision and Pattern Recognition, Vol. I, Hilton Head, SC, June 2000, 430-437. Abstract and full-text

D. Roth, N.-H. Yang and N. Ahuja, Learning to Recognize Objects, IEEE Int. Conference on Computer Vision and Pattern Recognition, Vol. I, Hilton Head, SC, June 2000, 724-731.  Abstract and full-text

J. Ma and N. Ahuja, Region Correspondence by Global Configuration Matching and Progressive Delauney Triangulation, IEEE Int. Conference on Computer Vision and Pattern Recognition, Vol. II, Hilton Head, SC, June 2000, 637-642.

R. Dugad and N. Ahuja, A Scheme for Joint Watermarking and Compression of Video, IEEE Int. Conference on Image Processing, Vol. 2, Vancouver, BC, Canada, Sept. 2000, 80-84. Abstract

M.-H. Yang, N. Ahuja and D. Kriegman, Face Recognition Using Kernel Eigenfaces, IEEE Int. Conference on Image Processing, Vol. 1, Vancouver, BC, Canada, Sept. 2000, 37-41. Abstract and full-text

J. Ma, N. Ahuja, C. Neti and A. Senior, Recovering Frontal-Pose Image from a Single Profile Image, IEEE Int. Conference on Image Processing, Vol. 2, Vancouver, BC, Canada, Sept. 2000, 243-247. Abstract and full-text

M. Aggarwal and N. Ahuja, Camera Center Estimation, Int. conference on Pattern Recognition, Vol. 1, Barcelona, Spain, Sept. 2000, 876-880.  Full Text

M. Aggarwal and N. Ahuja, Estimating Sensor Orientation in Cameras, Int. Conference on Pattern Recognition, Vol. 1, Barcelona, Spain, Sept. 2000, 896-899.  Abstract Full Text

M. Aggarwal and N. Ahuja, On Generating Seamless Mosaics with Large Depth of Field, Int. Conference on Pattern Recognition, Vol. 1, Barcelona, Spain, Sept. 2000, 588-591.  Full Text

L. Lu and N. Ahuja, Identification and control of a Pneumatic Actuator for Use With the Hexapod Project, Proc. ICARCV 2000, Singapore. Dec. 2000, (CD ROM). Abstract

M. Aggarwal and N. Ahuja, High Dynamic Range Panoramic Imaging, Proc. International Conference on Computer Vision, Vancouver, Canada, July 2001, 2-9.  Full Text

M. Aggarwal and N. Ahuja, Split Aperture Imaging for High Dynamic Range, Proc. International Conference on Computer Vision, Vancouver, Canada, July 2001, 10-17.  Full Text

M. Aggarwal and N. Ahuja, A New Imaging Model, Proc. International Conference on Computer Vision, Vancouver, Canada, July 2001, 82-89. Full Text

M. Aggarwal, H. Hua and N. Ahuja, On Cosine-Fourth and Vignetting Effects in Lenses, Proc. International Conference on Computer Vision, Vancouver, Canada, July 2001, 472-479.  Full Text

K.-H. Tan and N. Ahuja, Selecting Objects with Freehand Sketches, Proc. International Conference on Computer Vision, Vancouver, Canada, July 2001, 337-344. Abstract and Full Text

K.-H. Tan and N. Ahuja. A Representation for Image Structure and its Application in Object Selection with Freehand Sketches. In Proceedings IEEE Computer Society Conference on Computer Vision and Pattern Recognition 2001. Abstract and Full Text

S.-C. Yoon and N. Ahuja, Frame Interpolation using Transmitted Block-Based Motion Vectors, Proc. International Conference on Image Processing, Thessaloniki, Greece, October 2001, to appear. Abstract and Full Text

H. Hua, L.D. Brown, C. Gao, N. Ahuja, J.P. Rolland, F. Biocca, A Head-Mounted Projective Display and Its Applications in Interactive Augmented Environments, Conference Abstracts and Applications – Sketches and Applications, SIGGRAPH, Los Angeles, CA, August 2001.  Full Text

H. Hua, C. Gao, L.D. Brown, N. Ahuja and J.P. Rolland, Using a Head-Mounted Projective Display in Interactive Augmented Environments, Proceedings of IEEE and ACM International Symposium on Augmented Reality, New York, NY, October 2001, 217-223. Full Text

M.-H. Yang, D. Roth and N. Ahuja, Face Detection Using Large Margin Classifiers, Proc. International Conference on Image Processing, Thessaloniki, Greece, October 2001, to appear.   Abstract and full-text

A. Sehgal , A. Jagmohan , N. Ahuja, High Capacity Data Embedding in the Wavelet Domain, IEEE International Conference on Image Processing (ICIP), 2001, Thessaloniki, Greece, pages 979 – 982. Abstract and Full Text

A. Sehgal , A. Jagmohan , N. Ahuja, `Wireless Video Conferencing using Multiple Description Coding,’ IEEE International Symposium on Circuits and Systems (ISCAS), 2001, Sydney, Austrailia, pages 303 – 306. Abstract and Full Text

H. Hua and N. Ahuja, A High Resolution Panoramic Camera, IEEE Conf. On Computer Vision and Pattern Recognition, December 11-13, 2001, Hawaii, I-960-967.  Full Text

Ning Xu, Ravi Bansal and Narendra Ahuja. Object boundary segmentation using graph cuts based active contours. CVPR2001 Technical Sketches, pp. 87-90, Dec. 2001. Abstract and Full Text

Ning Xu, Narendra Ahuja and Ravi Bansal. Automated lung nodule segmentation using dynamic programming and EM based classification. SPIE Medical Imaging 4684-70, pp. 666-676, San Diego, CA. February, 2002. Abstract and Full Text

H. Hua, C. Gao, L. Brown, N. Ahuja and J. P. Rolland, A Testbed for Precise Registration, Natural Occlusion and Interaction in an Augmented Environment using a Head-Mounted Projective Display (HMPD), IEEE VR 2002 Proceedings, March 2002, Orlando, FL, 81-89.  Full Text

A. Jagmohan , A. Sehgal , N. Ahuja, `Isotropic Error-Diffusion Halftoning,’ IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2002, Orlando, Florida, pages 3329-3332. Abstract and Full Text

A. Sehgal , A. Jagmohan , N. Ahuja, `Scalable Predictive Coding and the Wyner-Ziv Problem,’ IEEE International Conference on Communication Systems (ICCS), 2002, Singapore, pages 101-106. Abstract and Full Text

A. Jagmohan, A. Sehgal, N. Ahuja, `Multiple Description Coding Using Coset Codes,’ IEEE International Conference on Communication Systems (ICCS), 2002, Singapore, pages 732- 737. Abstract and Full Text

A. Jagmohan, A. Sehgal, N. Ahuja, `Predictive Coding Using Coset Codes,’ IEEE International Conference on Image Processing (ICIP), 2002, Rochester, NY, vol. 2, pages 29- 32. Abstract and Full Text

K.-H. Tan, H. Hua and N. Ahuja, Multiview Mirror Pyramid-based Panoramic Cameras, Proceedings of the IEEE Workshop on Omnidirectional Vision (Omnivis), June 2002, Copenhagen, Denmark, 87-93.  Full Text

Ning Xu and Narendra Ahuja. Iterative 3D surface modeling from a sparse set of matched points. In Proceedings of IEEE International Conference on Multimedia and Expo, vol. 1, pp. 893 –896, Lausanne, Switzerland. August 2002. Abstract and Full Text

Ning Xu and Narendra Ahuja. Object contour tracking using graph cuts based active contours. In Proceedings of IEEE International Conference on Image Processing, vol. 3, pp. 277-280, Rochester, NY. September 2002. Abstract and Full Text

H. Hua, C. Gao, and N. Ahuja, Calibration of a head-mounted projective display for augmented reality systems, in Proceedings of IEEE International Symposium on Mixed and Augmented Reality 2002, Darmstadt, Germany, Sep. 30-Oct. 1st , 2002 Abstract Full Text

K.-H. Tan, D. Kriegman and N. Ahuja, Appearance-based Eye Gaze Estimation, Proceedings IEEE Workshop on Applications of Computer Vision, December 2002, Orlando, FL, 191-195.  Full Text

M. Singh and N. Ahuja, Mean-Shift Segmentation with Wavelet-based Bandwidth Selection, IEEE Workshop on Applications in Computer Vision, pp. 43-50, Dec. 2002, Florida. Abstract and Full Text

Chunyu Gao, Hong Hua, and N. Ahuja, Easy calibration of a head-mounted projective display for augmented reality systems, in Proceedings of IEEE VR 2003, p. 53-60, March 24-26, 2003, Los Angeles, CA. Abstract Full Text

H. Hua, L. Brown, C. Gao and N. Ahuja, A New Collaborative Infrastructure: SCAPE, Proc. IEEE Virtual Reality 2003, Los Angeles, CA, March 2003, 171.  Full Text

Ning Xu, Ravi Bansal and Narendra Ahuja. Object Segmentation Using Graph Cuts Based Active Contours. In Proceedings of IEEE International Conference on Computer Vision and  Pattern Recognition, vol. 2, pp. 46-53, Madison, WI. June 2003. .Abstract and Full Text

Ning Xu and Narendra Ahuja. A Three-view Matching Algorithm Considering Foreshortening Effects. In Proceedings of International Conference on Computer Vision, Pattern Recognition and Image Processing, pp. 635-638, Cary, NC. September 2003. Abstract and Full Text

Hongcheng Wang and Narendra Ahuja. Facial Expression Decomposition. In Proceedings of International Conference on Computer Vision, vol 2, pp. 958-965, Nice, France. October 2003. Abstract and Full Text

M. Singh and N. Ahuja, “Regression based Bandwidth Selection for Segmentation using Parzen Windows”,  Ninth International Conference in Computer Vision, Proceedings, vol. 1, pp. 2-9,  Oct. 2003, Nice, France. Abstract and Full Text

A. Sehgal, N. Ahuja, `Robust Predictive Coding of Video Using Coset Codes,’ IEEE Data Compression Conference (DCC), 2003, Snowbird, Utah, pages 103-112.Abstract and Full Text

Jagmohan, A.; Sehgal, A.; Ahuja, N. WYZE-PMD based multiple description video codec, Multimedia and Expo, 2003. Proceedings. 2003 International Conference on , Volume: 1 , 6-9 July 2003 Page(s): 569 -572 Abstract and Full Text

Jagmohan, A.; Ahuja, N. Wyner-Ziv encoded predictive multiple descriptions,  Data Compression Conference, 2003. Proceedings. DCC 2003 , March 25-27, 2003 Page(s): 213 -222 Abstract and Full Text

Ning Xu, Tianli Yu and Narendra Ahuja. Shape from color consistency using node cut. In Proceedings of 6th Asian Conference on Computer Vision, Jeju Island, Korea. January 2004. Abstract and Full Text

Ning Xu and Narendra Ahuja. On the use of depth-from-focus in 3D object modeling from multiple views. In Proceedings of 6th Asian Conference on Computer Vision, Jeju Island, Korea. January 2004. Abstract and Full Text

Ning Xu and Narendra Ahuja. Interactive Object Selection Using S-t Minimum Cut. In Proceedings of 6th Asian Conference on Computer Vision, Jeju Island, Korea. January 2004. Abstract and Full Text

T. Yu, N. Xu and N. Ahuja, Shape and View Independent Reflectance Map from Multiple Views, Proc. European Conference on Computer Vision,  Prague, May 2004.  Full Text

M. Singh, H. Arora and N. Ahuja, Robust Probabilistic Estimation for Parametric Image Models, Proc. European Conference on Computer Vision,  Prague, May 2004, to appear.  Abstract

M. Singh, H. Arora and N. Ahuja, Robust Registration and Tracking Using Kernel Density Correlation, 2nd IEEE Conference on Image and Video Registration, Washington, DC. June 2004, to appear.

C.-B. Liu and N. Ahuja, A Model for Dynamic Shape and Its Applications, IEEE Conference on Computer Vision and Pattern Recognition, Washington, DC, June 2004, Vol. 2, 129-134.  Full Text

T. Yu, N. Xu and N. Ahuja, Recovering Shape and Reflectance Model of Non-Lambertian Objects from Multiple Views, IEEE Conference on Computer Vision and Pattern Recognition, Washington, DC, June 2004, 226-233.  Full Text

A. Jagmohan, M. Singh and N. Ahuja, Dense Two View Stereo Matching Using Kernel Maximum Likelihood Estimation, IEEE International Conference on Pattern Recognition, Cambridge, UK, August, 2004.  Abstract

C.-B. Liu and N. Ahuja, Vision Based Fire Detection, IEEE International Conference on Pattern Recognition, Cambridge, UK, August 2004, Vol. 4, 134-137.  Full Text

N. Xu, K. Tan, H. Arora and N. Ahuja, Generating Omnifocus Images Using Graph Cuts and a New focus Measure, IEEE International Conference on Pattern Recognition, Cambridge, UK, August 2004, Vol. 4, 697-700.  Abstract

T. Yu, N. Xu and N. Ahuja, Reconstructing a Dynamic Surface from Video Sequences Using Graph Cuts in 4D Space-Time, IEEE International Conference on Pattern Recognition, Cambridge, UK, August 2004, 245-248.  Abstract

A. Briassouli and N. Ahuja, Fusion of Frequency and Spatial Domain Information for Motion Analysis, IEEE International Conference on Pattern Recognition, Cambridge, UK, August 2004, 175-178.  Abstract

C. Gao and N. Ahuja, Single Camera Stereo Using Planar Parallel Plate, IEEE International Conference on Pattern Recognition, Cambridge, UK, August 2004, Vol. 4, 108-111.  Full Text

H. Wang, R. Raskar and N. Ahuja, Seamless Video Editing. IEEE International Conference on Pattern Recognition, Cambridge, UK, August 2004, Vol. 3, 858-861.  Full Text

H. Wang and N. Ahuja, Compact Representation of Multidimensional Data Using Tensor Rank-One Decomposition, IEEE International Conference on Pattern Recognition, Cambridge, UK, August 2004, Vol. 1, 44-47.  Full Text

C.-B. Liu and N. Ahuja, Motion Based Retrieval of Dynamic Objects in Videos, ACM Multimedia Conference, October 2004, 288-291.  Abstract

H. Wang, Q. Wu, L. Shi, Y. Yu and N. Ahuja, Out-of-Core Tensor Approximation of Multi-Dimensional Matrices of Visual Data, ACM SIGGRAPH, 2005.  Abstract

T. Yu, J. Luo, A. Singhal and N. Ahuja, Shape Regularized Active Contour Based on Dynamic Programming for Anatomical Structure Segmentation, SPIE Medical Imaging 2005, San Diego, CA, February 2005.  Full Text

H. Wang and N. Ahuja, Rank-R Approximation of Tensors Using Image-as-Matrix Representation, IEEE International Conference on Computer Vision and Pattern Recognition, San Diego, CA, June 2005, to appear.  Abstract

H. Wang, N. Xu, R. Raskar and N. Ahuja, Videoshop: A New Framework for Video Editing in Spatio-Temporal Gradient domain, Video Proceedings, IEEE International Conference on Computer Vision and Pattern Recognition, San Diego, CA, June 2005, to appear.

T. Yu, J. Luo and N. Ahuja, Shape Regularized Active Contour using Iterative Global Search and Local Optimization, IEEE International Conference on Computer Vision and Pattern Recognition, San Diego, CA, June 2005, to appear.  Full Text

C. Gao and N. Ahuja, A Refractive Camera for Acquiring Stereo Depth and Super-Resolution Images, Proceedings IEEE Conference on Computer Vision and Pattern Recognition, New York, NY Vol. 2, June 2006, 2316-2323. Full Text

S. Todorovic and N. Ahuja, Extracting subimages of an unknown category from a set of images,  in Proc. IEEE Comp. Soc. Conf. Computer Vision and Pattern Recognition (CVPR 2006), vol. 1, pp. 927-934, New York, NY, 2006. Full Text

S. Todorovic and N. Ahuja, 3D texture classification using the belief net of a segmentation tree, in Proc. 18th Int. Conf. Pattern Recognition (ICPR 2006), Hong Kong, China, 2006. Full Text

Alexia Briassouli, Narendra Ahuja, Spatial and Fourier Error Minimization for Motion Estimation and Segmentation, ICPR 2006, Hong Kong Full Text

Alexia Briassouli, Narendra Ahuja, Estimation of Multiple Periodic Motions from Video, ECCV 2006, Graz, Austria Full Text

Hongcheng Wang, Yunqiang Chen, Tong Fang, Jason Tyan and Narendra Ahuja, Gradient Adaptive Image Restoration and Enhancement, in Proc. IEEE International Conference on Image Processing (ICIP), Atlanta, VA, Aug. 2006 Full Text

Tianli Yu, Hongcheng Wang, Narendra Ahuja, Wei-Chao Chen, Sparse Lumigraph Relight by Illumination and Reflectance Estimation from Multi-View Images, Eurographics Symposium on Rendering (EGSR), 2006 Full Text

Tianli Yu, Hongcheng Wang, Narendra Ahuja, Wei-Chao Chen, Sparse Lumigraph Relight by Illumination and Reflectance Estimation from Multi-View Images, Technical Sketch, SIGGRAPH, 2006 Full Text

Himanshu Arora, Narendra Ahuja, Analysis of ramp discontinuity model for multiscale image segmentation. ICPR(1), 2006.  Abstract

T. Yu, N. Ahuja and W-C. Chen, SDG Cut: 3D Reconstruction of Non-Lambertian Objects Using Graph Cuts on Surface Distance Grid, Proceedings IEEE Conference on Computer Vision and Pattern Recognition, New York, NY, Vol. 2, June 2006, 2269-2276. Full Text

C-B. Liu, R-S. Lin, M-H. Yang, N. Ahuja and S. Levinson, Object Tracking Using Globally Coordinated Nonlinear Manifolds, International Conference on Pattern Recognition, Vol. 1, Hong Kong, August 2006, 844-847. Full Text

S.Yi and Narendra Ahuja, An Omnidirectional Stereo Vision System Using a Single Camera, International Conference on Pattern Recognition, Vol. 4, Hong Kong, August 2006, 861-847. Full Text

S. Yi and N. Ahuja, A Novel Omnidirectional Stereo Vision System with a Single Camera, International Conference on Image Analysis and Recognition, Portugal, September 2006.  Abstract

A. Kumar, N. Ahuja, J. Hart, U.K. Visesh, P.J. Narayanan, C.V. Jawahar, A Vision System for Monitoring Intermodal Freight Trains, Workshop on Applications of Computer Vision (WACV), Austin, TX, February 2007. Full Text

B. Freid, C. Barkan, N. Ahuja, J. Hart, S. Todorvic, N. Kocher, Multispectral Machine Vision for Improved undercarriage Inspection of Railroad Rolling Stock, Proc. International Heavy Haul Conference, Specialist Technical Session, Kiruna, Sweden, June 2007, 737-744. Full Text

H. Arora, N. Loeff, D. Forsyth and N. Ahuja, Unsupervised Segmentation of Objects using Efficient Learning, Proc. IEEE Conference on Computer Vision and Pattern Recognition,  Minneapolis, MN, June 2007, 1-7 Full Text

H. Arora and N. Ahuja, Modeling Objects using Distribution and Topology of Multiscale Region Pairs, IEEE Conference on Computer Vision and Pattern Recognition, Beyond Patches Workshop, Minneapolis, MN, June 2007, 1-8. Full Text

Bernard Ghanem and Narendra Ahuja, Phase PCA for Dynamic Texture Video Compression, Proc.of the IEEE International Conference on Image Processing 2007, Sept. 16-19, San Antonio, TX Full Text

N. Ahuja and S. Todorovic, Learning the Taxonomy and Models of Categories Present in Arbitrary Images,  Proc. of the IEEE International Conference on Computer Vision 2007, Oct. 14-20, Rio De Janeiro, Brazil Full Text

N. Ahuja and S. Todorovic, Extracting Texels in 2.1D Natural Textures,  Proc. of the IEEE International Conference on Computer Vision 2007, Oct. 14-20, Rio De Janeiro, Brazil Full Text

Bernard Ghanem and Narendra Ahuja, Phase Based Modelling of Dynamic Textures, Proc. of the IEEE International Conference on Computer Vision 2007, Oct. 14-20, Rio De Janeiro, Brazil Full Text

Varsha Hedau, H. Arora and N. Ahuja, “Matching Images under Unstable Segmentation”,  in Proc. IEEE Comp. Soc. Conf. Computer Vision and Pattern Recognition (CVPR 2008), Anchorage, AL, 2008. Full Text

N. Ahuja and S. Todorovic, “Connected segmentation tree – a joint representation of region layout and hierarchy,” in Proc. IEEE Comp. Soc. Conf. Computer Vision and Pattern Recognition (CVPR 2008), Anchorage, AL, 2008. Full Text

S. Todorovic and N. Ahuja, Learning subcategory relevances to category recognition,” in Proc. IEEE Comp. Soc. Conf. Computer Vision and Pattern Recognition (CVPR 2008), Anchorage, AL, 2008. Full Text

Bernard Ghanem and N. Ahuja, “Extracting a Fluid Dynamic Texture and the Background from Video”, in Proc. IEEE Comp. Soc. Conf. Computer Vision and Pattern Recognition (CVPR 2008), Anchorage, AL, 2008. Full Text

Q. Yang, K-H. Tan and N. Ahuja, Real-time 0(1) Bilateral Filtering, IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2009), Miami, FL, June 2009.  Full Text

N. Ahuja and A. Rosenfeld, A Note on the Use of Second-Order Gray-Level Statistics for Threshold Selection, IEEE Trans. Syst., Man, Cyber., SMC-8, December 1978, 895-898. Abstract

B. Schachter and N. Ahuja, Random Pattern Generation Processes, Computer Graphics and Image Processing, 10, June 1979, 95-114. Abstract

B. Schachter and N. Ahuja, A History of Visual Flight Simulation, Computer Graphics World, 3, May 1980, 16-31.  Abstract

N. Ahuja, R. M. Haralick and A. Rosenfeld, Neighbor Gray Levels as Features in Pixel Classification, Pattern Recognition, 12, August 1980, 251-260. Abstract

N. Ahuja, T. Dubitzki and A. Rosenfeld, Some Experiments with Mosaic Models for Images, IEEE Trans. Syst., Man, Cyber., 10, November 1980, 744-749. Abstract

N. Ahuja and A. Rosenfeld, Mosaic Models for Textures, IEEE Trans. Pattern Analysis and Machine Intelligence, January 1981, 1-11. Abstract

N. Ahuja, Mosaic Models for Images, 1: Geometric Properties of Components in Cell Structure Mosaics, Information Sciences, 23, March 1981, 69-104. Abstract

N. Ahuja, Mosaic Models for Images, 2: Geometric Properties of Components in Coverage Mosaics, Information Sciences, 23, April 1981, 159-200. Abstract

R. Chellappa, R. L. Kashyap and N. Ahuja, Decision Rules for Choice of Neighbors in Random Field Models of Images, Computer Graphics and Image Processing, l5, April 1981, 301-318. Abstract

N. Ahuja, Mosaic Models for Images, 3: Spatial Correlation in Mosaics, Information Sciences, 24, June 1981, 43-70. Abstract

N. Ahuja and B. Schachter, Image Models, ACM Computing Surveys, December 1981, 373-398. Abstract

N. Ahuja, Dot Pattern Processing Using Voronoi Neighborhoods, IEEE Trans. Pattern Analysis and Machine Intelligence, May 1982, 336-343.Abstract

N. Ahuja, On Approaches to Polygonal Decomposition for Image Representation, Computer Vision, Graphics and Image Processing, November 1983, 200-214. Abstract

N. Ahuja and C. Nash, Octree Representations of Moving Objects Computer Vision, Graphics and Image Processing, May 1984, 207-216.Abstract

N. Ahuja and S. Swamy, Multiprocessor Pyramid Architectures for Bottom-Up Analysis, IEEE Trans. Pattern Analysis and Machine Intelligence, July 1984, 463-475. Abstract

H. C. Chen, N. Ahuja and T. S. Huang, Septree Representations of Moving Objects Using Hexagonal Cylinderical Decomposition, Journal of Optical Engineering, September 1984, 531-535. Abstract

N. Ahuja, B. An and B. Schachter, Image Representation Using Voronoi Tessellation, Computer Vision, Graphics and Image Processing, March 1985, 286-295. Abstract

N. Ahuja and S. Swamy, Multiprocessor Pyramid Architectures for Bottom-up Image Analysis, Test and Measurement World, November 1985, 66-76. Abstract

N. Ahuja, Efficient Planar Embedding of Trees for VLSI Layouts, Computer Vision, Graphics and Image Processing, May 1986, 189-203.Abstract

J. Weng, T. Huang and N. Ahuja, 3-D Motion Estimation, Understanding and Prediction from Noisy Image Sequences, IEEE Trans. Pattern Analysis and Machine Intelligence, May 1987, 370-389.

J. Weng and N. Ahuja, Octrees of Objects in Arbitrary Motion: Representation and Efficiency, Computer Vision, Graphics and Image Processing, August 1987, 167-185. Abstract

J. Veenstra and N. Ahuja, Line Drawings of Octree-Represented Objects, ACM Transactions on Graphics, January 1988, 61-75. Abstract

X. Zhuang, T. S. Huang, N. Ahuja and R. M. Haralick, A Simplified Optic Flow-Motion Algorithm, Computer Vision, Graphics and Image Processing, June 1988, 334-344.

D. Blostein and N. Ahuja, A Multiscale Region Detector, Computer Vision, Graphics and Image Processing, January 1989, 22-41.

N. Ahuja and J. Veenstra, Generating Octrees from Object Silhouettes in Orthographic Views, IEEE Trans. Pattern Analysis and Machine Intelligence, February 1989, 137-149. Abstract

W. Hoff and N. Ahuja, Surfaces from Stereo: Integrating Feature Matching, Disparity Estimation and Contour Detection, IEEE Trans. Pattern Analysis and Machine Intelligence, February 1989, 121-136. Abstract and Full Text

J. Weng, T. Huang and N. Ahuja, Motion and Structure from Two Perspective Views: Algorithm, Error Analysis, and Error Estimation IEEE Trans. Pattern Analysis and Machine Intelligence, May 1989, 451-476.

D. Blostein and N. Ahuja, Shape from Texture: Integrating Texture Element Extraction and Surface Estimation, IEEE Trans. Pattern Analysis and Machine Intelligence, December 1989, 1233-1251.

N. Ahuja and M. Tuceryan , Extraction of Basic Perceptual Structure in Dot Patterns: Integrating Region, Boundary and Component Gestalt, Computer Vision, Graphics and Image Processing, December 1989, 304-356. Abstract

S. Srivastava and N. Ahuja, Octree Generation from Object Silhouettes in Perspective Views, Computer Vision, Graphics and Image Processing, January 1990, 68-84. Abstract

X. Hu and N. Ahuja, Motion Estimation Under Orthographic Projection, IEEE Trans. on Robotics and Automation, December 1991, 848-853.Abstract and Full Text

J. Weng, N. Ahuja and T. S. Huang, Motion and Structure from Point Correspondences: Planar Surfaces, IEEE Trans. Signal Processing, vol. 39, No. 12, December 1991, 2691-2717. Abstract and Full Text

Y. Hwang and N. Ahuja, A Potential Field Approach to Path Planning, IEEE Transactions on Robotics and Automation, Vol. 8, February 1992, 23-32. Abstract

O. Faugeras, J. Mundy, N. Ahuja, C. Dyer, A. Pentland, R. Jain, K. Ikeuchi and K. Bowyer, Why Aspect Graphs Are Not (Yet) Practical for Computer Vision, Computer Vision, Graphics and Image Processing: Image Understanding, March 1992, 212-218. Abstract

J. Weng, T. S. Huang and N. Ahuja, Motion and Structure from Line Correspondences: Closed-form Solution, Uniqueness and Optimization, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 14, March 1992, 318-336. Abstract and Full Text

J. Weng, N. Ahuja and T. S. Huang, Matching Two Perspective Views, IEEE Trans. Pattern Analysis and Machine Intelligence, August 1992, 806-825. Abstract and Full Text

Y. Hwang and N. Ahuja, Gross Motion Plannig – A Survey, ACM Computing Surveys, Vol 24, No. 3, 1992, 219-292. Abstract

S. Das and N. Ahuja, Integrating Coarse-to-fine Image Acquisition and Estimation from Multiple Cues, Sadhana, Vol. 18, Part 2, June 1993, 223-238.

X. Hu and N. Ahuja, Sufficient Conditions for Double or Unique Solution of Motion and Structure, Computer Vision, Graphics and Image Processing: Image Understanding, September 1993, 161-176.  Abstract and Full Text

J. Weng, N. Ahuja and T. S. Huang, Optimal Motion and Structure Estimation from Two Views, IEEE Trans. Pattern Analysis and Machine Intelligence, September 1993, 864-884. Abstract and Full Text

X. Hu and N. Ahuja, Motion and Structure Estimation using Long Sequence Motion Models, Journal of Image and Vision Computing, 11, November 1993, 549-569.

N. Ahuja and A. L. Abbott, Active Stereo: Integrating Disparity, Vergence, Focus, Aperture and Calibration for Surface Estimation, IEEE Trans. Pattern Analysis and Machine Intelligence, October 1993, 1007-1029. Abstract and Full Text

A. N. Choudhary, J. H. Patel, and N. Ahuja, NETRA: A Hierarchical and Partitionable Architecture for Computer Vision Systems, IEEE Trans. Parallel and Distributed Systems, Vol. 4, No. 10, October 1993, 1092-1104. Abstract

S. Sull and N. Ahuja, Integrated 3D Analysis and Analysis Guided Synthesis of Flight Image Sequences, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 16, No. 4, April 1994, 357-372.

X. Hu and N. Ahuja, Matching Point Features with Ordered Geometric, Rigidity and Disparity Constraints, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 16, No. 10, October 1994, 1041-1049. Abstract and Full Text

X. Hu and N. Ahuja, Mirror Uncertainty and Uniqueness Conditions for Determining Shape and Motion from Orthographic Projection, Int. Journal of Computer Vision, 13, 3, November 1994, 295-309.

X. Hu and N. Ahuja, Feature Extraction and Matching as Signal Detection, Int. Journal of Pattern Recognition and Artificial Intelligence, 8, 6, December 1994, 1343-1379.

X. Hu and N. Ahuja, Necessary and Sufficient Conditions for a Unique Solution of Plane Motion and Structure, IEEE Transactions on Robotics and Automation, Vol. 11, No. 2, April 1995, 304-308.  Abstract and Full Text

N. Ahuja, On Detection and Representation of Multiscale Low-Level Image Structure, ACM Computing Surveys, Vol. 27, No. 3, September 1995, 304-306.

S. Sull and N. Ahuja, Integrated Matching and Segmentation of Multiple Features in Two Views, Computer Vision, Graphics and Image Processing: Image Understanding, Vol. 62, No. 3, November 1995, 279-297.

S. Das and N. Ahuja, Performance Analysis of Stereo, Vergence and Focus as Depth Cues for Active Vision, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 17, No. 12, December 1995, 1213-1219. Abstract and Full Text

T. Srikanth and N. Ahuja, Parallel Distributed Detection of Feature Trajectories in Multiple Discontinuous Motion Image Sequences, IEEE Trans. on Neural Networks, Vol. 7, No. 3, May 1996, 594-603.

S. Das and N. Ahuja, Active Surface Estimation: Integrating Coarse-to-fine Image Acquisition and Estimation from Multiple Cues, Artificial Intelligence, Vol. 83, May 1996, 241-266. Abstract

A. Krishnan and N. Ahuja, Range Estimation from Focus Using a Nonfrontal Imaging Camera, Int. Journal of Computer Vision, Vol. 20, No. 3, 1996, 169-185. Full Text

Y. Kitamura, H. Takemura, N. Ahuja and F. Kishino, Colliding Face Detection among 3-D Objects using Octree and Polyhedral Shape Representation, Journal of the Robotics Society of Japan, Vol. 14, No. 5, 1996, 733-742. Abstract

N. Ahuja, A Transform for Multiscale Image Segmentation by Integrated Edge and Region Detection, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 18, No. 12, December 1996, 1211-1235.

N. Ahuja and Jen-Hui Chuang, Shape Representation Using a Generalized Potential Field Model, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 19, No. 2, February 1997, 169-176. Abstract

M. Tabb and N. Ahuja, Unsupervised Multiscale Image Segmentation by Integrated Edge and Region Detection, IEEE Transactions on Image Processing, Vol. 6, No. 5, May 1997, 642-655.

J. Weng, Y. Cui and N. Ahuja, Transitory Image Sequences, Asymptotic Properties, and Estimation of Motion and Structure, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 19, No. 5, May 1997, 451-464. Abstract and Full Text

J. Weng, N. Ahuja and T. S. Huang, Learning Recognition and Segmentation Using the Cresceptron, International Journal of Computer Vision, 25(2), 1997, 109-143.

C. Debrunner and N. Ahuja, Segmentation and Factorization-Based Motion and Structure Estimation for Long Image Sequences, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. 20, No. 2, February 1998, 206-211.

P. Bajcsy and N. Ahuja, Location and Density-Based Hierarchical Clustering, IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol. 20, No. 9, September 1998, 1011-1015. Abstract

J.-H. Chuang and N. Ahuja, An Analytically Tractable Potential Field Model of Free Space and Its Application in Obstacle Avoidance, IEEE Trans. on Systems, Man, and Cybernetics, Vol. 28, No. 5, October 1998, 729-736. Abstract

N. Srinivasa and N. Ahuja, A Topological and Temporal Correlator Network for Spatiotemporal Pattern Learning, Recognition and Recall, IEEE Transactions on Neural Networks, Vol. 10, No. 2, March 1999, 356-371. Abstract and full-text

T. Joshi, N. Ahuja and J. Ponce, Structure and Motion Estimation from Dynamic Silhouettes under Perspective Projection, International Journal of Computer Vision, 31(1), 1999, 31-50.

S. C. Yoon, K. Ratakonda and N. Ahuja, Low Bit-Rate Video Coding with Implicit Multiscale Image Segmentation, IEEE Transactions on Circuits and Systems for Video Technology, Vol. 11, No. 4, April 2001, 461-474. Abstract and Full Text

R. Dugad and N. Ahuja, A Fast Scheme for Image Size Change in the Compressed Domain, IEEE Trans. on Circuits and Sstems for Video Technology, Vol. 11, No. 4, April 2001, 461-474. Abstract Full Text

Z.-P. Liang, H. Pan, R. L. Magin, N. Ahuja and T. S. Huang, Automated Image Registration by Maximization of a Region Similarity Metric, International Journal of Imaging Systems and Technology, special issue on MR Signal Processing, to appear.  Abstract

M.-H. Yang, D. Kriegman and N. Ahuja, Face Detection Using Multimodal and Density Modes, Computer Vision and Image Understanding, Vol. 84, October 2001, 1-21.  Full Text

M.-H. Yang, D. Kriegman and N. Ahuja, Detecting Faces in Images: A Survey, IEEE Transactions on Pattern Analysis and Machine Intelligence, January 2002, 34-58.  Full Text

M. Aggarwal and N. Ahuja, A Pupil-Centric Model of Image Formation, International Journal of Computer Vision, 48(3), 2002, 195-214.  Full Text

M.-H. Yang, N. Ahuja and M. Tabb, Extraction of Motion Trajectories and Its Application to Hand Gesture Recognition, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol., 24, No. 8, 2002, 1061-1074.  Abstract

D. Roth, M.-H. Yang and N. Ahuja, Learning to Recognize Objects, Neural Computation, Vol. 14, No. 5, 2002, 1071-1104.  Abstract

R. Dugad and N. Ahuja, A Scheme for Spatial Scalability Using Nonscalable Encoders, IEEE Transactions CSVT,  Vol. 13, No. 10, October 2003, 993-999.  Full Text

K.-H. Tan, H. Hua and N. Ahuja, Multiview Panoramic Cameras Using Mirror Pyramids, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 26, No. 6, June 2004.  Full Text

M. Aggarwal and N. Ahuja, Split Aperture Imaging for High Dynamic Range, International Journal on Computer Vision, Vol. 58, No. 1, June 2004, 195-214.  Full Text

A. Sehgal, A. Jagmohan and N. Ahuja, Robust  Wyner-Ziv Coding of Video, IEEE Transactions on Multimedia, Vol. 6, No. 2, April 2004, 249-258.

M. Tabb and N. Ahuja, Two-Dimensional Motion Estimation by Matching a Multiscale Set of Region Primitives, IEEE Transactions on Pattern Analysis and Machine Intelligence, to appear.

J.-H. Chuang, N. Ahuja, C.-C. Lin, C.-H. Tsai and C.-H. Chen, A Potential-Based Generalized Cylinder Representation, Computers and Graphics, Vol. 28, 2004, 907-918.

M. Aggarwal and N. Ahuja, Split Aperture Imaging for High Dynamic Range, International Journal on Computer Vision, Vol. 58, No. 1, June 2004, 7-17. Abstract and Full Text

Hongcheng Wang, Ning Xu, Ramesh Raskar and Narendra Ahuja, Videoshop: A New Framework for Video Editing in Gradient Domain, Graphical Models (GM), Volume 69, Issue 1, Jan. 2007, pp 57-70 Full Text

S. Yi, B. Choi and N. Ahuja, Real-time Omni-directional Distance Measurement with Active Panoramic Vision, International Journal of Control, Automation, and Systems, Vol. 5, No. 2, April 2007, 184-191.Full Text

Hong Hua, C. Gao, and N. Ahuja, Calibration of an augmented reality system using head-mounted projective displays, IEEE Transactions on Systems, Man, Cybernetics (Part A: Systems), 37(3), 416-30, May 2007. Full Text

T. Yu, N. Xu and N. Ahuja, Shape and View Independent Reflectance Map from Multiple Views, International Journal of Computer Vision, Vol. 73, No. 2, June 2007, 123-138. Full Text

A. Briassouli and N. Ahuja, Extraction and Analysis of Multiple Periodic Motions in Video Sequences, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 29, No. 7, July 2007, 1244-1261. Full Text

H. Hua, N. Ahuja and C. Gao, Design Analysis of a High-Resolution Panoramic Camera Using Conventional Imagers and a Mirror Pyramid, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 29, No. 2, 2007, 356-361 Full Text

N. Xu, N. Ahuja, R. Bansal, Object segmentation using graph cut based active contours, Computer Vision and Image Understanding,Volume 107, Issue 3, September 2007, Pages 210-224 Full Text

Lai, Y-C., C.P.L. Barkan, J. Drapa,  N. Ahuja, J.M. Hart, P.J. Narayanan, C.V. Jawahar, A. Kumar, L. Milhon and M.P. Stehly 2007.  Machine-vision analysis of the energy efficiency of intermodal freight trains. Journal of Rail and Rapid Transit 221: 353-364. Full Text

S. Todorovic and N. Ahuja, Region Based Hierarchical Image Matching, Int. Journal of Computer Vision, to appear. Full Text

A. Briassouli and N. Ahuja, Integration of Frequency and Space for Multiple Motion Estimation and Shape-Independent Object Segmentation, IEEE Transactions on Circuits, Systems and Video Technology, to appear

H. Wang and N. Ahuja, A Tensor Approximation Approach to Dimensionality Reduction, International Journal of Computer Vision, 76:3, March 2008, 217-229. Full Text

S. Todorovic and N. Ahuja, Unsupervised Category Modeling, Recognition, and Segmentation in Images, IEEE Trans. Pattern Analysis and Machine Intelligence, 2008, to appear. Full Text

]]> http://vision.ai.uiuc.edu/?feed=rss2&p=382 0 http://vision.ai.uiuc.edu/?p=380 http://vision.ai.uiuc.edu/?p=380#comments Thu, 07 May 2009 06:50:40 +0000 sanketh http://vision.ai.uiuc.edu/wordpress/?p=380 N. Ahuja, Mosaic Models for Textures. A. Rosenfeld (ed.), Image Modeling, Academic Press, 1981, 1-8. (Also appears in Digital Image Processing and Analysis: Vol. 1, R. Chellappa and A. Sawchuk (Eds.), IEEE Computer Society Press, Washington D. C., 1986.)

N. Ahuja…

J. Weng, T. S. Huang and N. Ahuja, Motion and Structure from Image Sequences, Springer-Verlag, 1992.

M. H. Yang and N. Ahuja, Face Detection and Hand Gesture Recognition for Vision-Based Human Computer Interaction, Kluwer Academic Publishers, 2001.

N. Ahuja and M. Tabb, Multiscale Image Edge and Region Detection Method and Apparatus, U. S. Patent, Issued September 1998.

R. Dugad and N. Ahuja, Transform Domain Significant Coefficient Digital Image Watermarking Method, U.S. patent Application filed, June 2000.

H. Hua and N. Ahuja, Method and Apparatus for a High-Resolution and Real-Time Panoramic Camera, Patent Application Filed, November 2001.

R. Dugad and N. Ahuja, Transformation of Image Parts in Different domains to Obtain Resultant Image Size Different From Initial Image Size, U.S. Patent, Issued October 2004.

1. Sooyeong Yi, Narendra Ahuja, An Omnidirectional Stereo System Using a Single Camera, 18th International Conference on (ICPR’06), 2006, Hong Kong, China Full Text
   
2. Sooyeong Yi, Narendra Ahuja, A Novel Omnidirectional Stereo System Using a Single Camera, International Conference on Image Analysis and Recognition (ICIAR), 2006 Full Text

In developing the new opto-geometric configurations, we have found that certain classical models and approaches cease to be adequate. For example, the long-established Gaussian model of image formation fails to adequately predict the acquired images, and the optical and geometric phenomena ignored in the traditional characterization of the most focused scene point make the traditional methods of focus analysis unacceptable. We have the old models with new, more rigorous, and satisfactory models. These new models are also useful in contexts other than next generation camera designs – they are useful in improving the performance of currently “acceptable” systems, and in extending the applicability of computer vision methods to many scenarios and applications which were out of reach otherwise.

M. Aggarwal and N. Ahuja, A New Imaging Model, Proc. International Conference on Computer Vision, Vancouver, Canada, July 2001, 82-89. Full Paper(PDF)

M. Aggarwal, H. Hua and N. Ahuja, On Cosine-Fourth and Vignetting Effects in Lenses, Proc. International Conference on Computer Vision, Vancouver, Canada, July 2001, 472-479. Full Paper(PDF)

M. Aggarwal and N. Ahuja, A Pupil-Centric Model of Image Formation, International Journal on Computer Vision, 48(3), 2002, 195-214.  Secured Access to Text

We have developed a camera which is capable of acquiring very large field of view (FOV) images at high and uniform resolution, from a single viewpoint, at video rates. The FOV can range from being nearly hemispherical, to being nearly omni-directional, barring some small scene parts being obstructed by image sensors themselves. The camera consists of multiple imaging sensors and a hexagonal prism made of planar mirror faces. Each sensor is paired with a planar face of the prism. The sensors are positioned in such a way that they image different parts of the scene from a single virtual viewpoint, either directly or after reflections off the prism. A panoramic image is constructed by concatenating the images taken by different sensors. The resolution of the panoramic image is proportional to the number of sensors used and therefore a multiple of that of an individual sensor. Further, the resolution is substantially uniform across the entire panoramic image.

1. Introduction

A panoramic camera is an imaging device capable of capturing a very large field of view (FOV). Like any other camera, it is desirable that such cameras acquire the entire FOV from a single viewpoint, in real time, at high resolution which is uniform across the FOV, with large dynamic range, and over a large depth of field. Such devices find applications in many areas including tele-conferencing, surveillance and robot navigation. Many efforts have been made to achieve various subsets of these properties (i.e. wide FOV, high and uniform resolution, large depth of field, high dynamic range, a single viewpoint, and real-time acquisition. These methods of capturing panoramic or omni-directional images fall into two categories: dioptric methods, where only refractive elements (lenses) are employed, and catadioptric methods, where a combination of reflective and refractive components is used.

Typical dioptric systems include camera clusters, panning cameras, and fisheye lenses. Catadioptric methods include curved mirror systems where a conventional camera captures the scene reflected off a single non-planar mirror (e.g. parabolic or hyperbolic mirror), and planar mirror systems such as mirror pyramid systems where multiple conventional cameras image the scene reflected off the faces of a mirror-pyramid. The cameras that use a parabolic- or a hyperbolic-mirror to map an omni-directional view onto a single sensor are able to capture a large FOV from a single viewpoint at video rate. However, the FOV shape is hemispherical minus a central cone which is blocked by self-occlusion. The overall resolution of the acquired images is limited to that of the sensor used, and further it varies with the viewing direction across the ring-like FOV, e.g., from a high just outside the central blind spot to a low in the periphery. The cameras using a spherical or conical mirror have similar properties as those using parabolic or hyperbolic mirrors except that they do not possess a single viewpoint.

Many of the aforementioned systems provide a cylindrical shape FOV which is 360 degrees wide in azimuth, but has limited height in elevation (Fig. 1a). In certain applications such as robot navigation and surveillance, however, a hemispherical shape FOV is highly desirable (Fig. 1b). We have developed a system which is capable of acquiring hemispherical panoramic images in real time, with high and substantially uniform resolution, and from a single viewpoint. By substantially uniform resolution we mean the same level of uniformity as delivered by a conventional, non-panoramic camera.

2. Prototype

Figure 2. Prototype in a ceiling mount

Fig. 3 shows a sample image acquired by the prototype. The effective FOV from the top boundary to the bottom boundary is exactly 140 degrees.

Stereo is one of the most widely explored sources of scene depth. Stereo usually refers to spatial stereo, wherein two cameras, separated by a baseline, simultaneously capture stereo image pairs. The spatial disparity in the images of the same scene feature then captures the feature’s depth. More than two cameras can also be used to capture the disparity information across multiple views.

An alternative to such spatial stereo is temporal stereo wherein a single camera is relocated to the same set of viewpoints to capture the two or more images sequentially. This loses the parallel imaging capability and therefore the ability to handle fast moving objects, but it reduces the number of cameras used to one as well as eliminates the need for photometric calibration of the camera if needed for feature/intensity matching of stereo images.

We propose a single-camera depth estimation system that captures a large number of images, after the incoming light from a scene point has been deflected in a manner that depends on the object depth. The deflection mechanism is the passage of light through a thick glass plate placed in front of an ordinary camera at a certain orientation to the optical axis. In order to estimate depth, at least two images captured under two different plate poses are necessary. However, a larger number of images, containing redundant depth information, are acquired by changing the plate orientation sequentially, e.g. by rotating and/or reorienting the plate. Rotating the plate at a fixed orientation with respect to the optical axis is a mechanically convenient way of obtaining a large number of depth-coded images, followed, if desired, by more plate orientations and rotations, to acquire more images, as illustrated in Fig 1. An analysis of the correspondences among the set of images yields depth estimates. High quality, dense depth estimation distinguishes this new camera from other single camera stereo systems.

Depth Dependent Pixel Displacement

It is well known from optics that light ray passing through a planar plate will encounter a lateral displacement. For a camera-plate system, this phenomenon is shown as the pixel shifts in the image. For a given object point, we assume that the point is shifted by the plate in a plane parallel with image plane.

The displacements of a pixel depend differently on changes in the plate tilt angle and rotation angle. Using both changes leads to robust estimation because the lack of sensitivity to one type of change is complemented by higher sensitivity to the other. Depth estimation is done by minimizing a cost function which is overdetermined due to the large number of images available. The cost function includes the error due to the fit of the result to the multiple estimates available at a single pixel, and the local roughness of the surface at the pixel.

The dimensions of the plate and its tilt angle, among other parameters, have to be selected properly to achieve good depth estimates. The plate parameters affect the amount of depth-sensitive displacement in the image. Larger refractive index and thicker plate yield a larger displacement, which corresponds to a higher depth resolution. However, larger refractive index and thicker plate also introduce larger chromatic aberration, which may degrade the image quality. The tilt angle of the plate also affects the displacement – larger the tilt angle, larger the pixel displacement for the same depth. However, with the increase of the tilt angle, the size of the plate increases dramatically.

Prototype

We have developed a prototype with the rotating plate oriented at a fixed tilt angle. It requires calibration only once, in the beginning, after which the system acquires images continuously without the need for any further calibration. This is achieved by rotating the plate continuously about the x, y and/or z axes, acquiring images at video rate. The images of a scene point, acquired during rotation, lie along a 4-dimensional manifold in the space defined by the object depth in the viewing direction, and the three rotation angles. To estimate object depth in a given direction, we simply find the best estimate of the intersection of the line in that direction with the manifold samples obtained from the images acquired during rotation. Each pixel in each image thus contributes to the number of manifold samples. Since the locations of pixels in different images define a denser array of directions than possible with a single (orientation) image grid, the system yields depth estimates along a direction array denser than the original images.

We used a Sony DFX900 color camera equipped with a 16mm lens, and an ordinary glass plate with 13mm thickness and refractive index of around 1.5. The plate was mounted on a rotary stage and tilted approximately 45 degrees with respect to the image plane. A total 36 plate poses , evenly distributed within the range of 360 degrees, were calibrated and used to recover depth and synthesize the super-resolution images.

We tested our system with two objects: house and monster head. The experimental results of the house object are shown in Fig 2. Fig. 2a is one of the input images taken through the plate, Fig. 2b shows the recovered depth map, and Fig. 2c shows a new view generated from the reconstructed house model.

Fig 3 shows the experimental results for the monster head object. Figs 3a~3c show one of input images, the recovered depth map, and a new view of the reconstructed head model, respectively. This result indicates that the camera performs well for objects having a limited amount of texture.

1. Chunyu Gao, Narendra Ahuja, Single camera stereo using planar parallel plate , Pattern Recognition, 17th International Conference on (ICPR’04) Volume 4, pp. 108-111 08, 2004, Cambridge UK Full Text
   
2. Chunyu Gao and Narendra Ahuja, “A Refractive Camera for Acquiring Stereo and Super-resolution  Images”, Computer Vision and Pattern Recognition, IEEE Computer Society Conference on, Volume 2,  17-22 June 2006 Page(s):2316 – 2323 Full Text

In this work, we have extended our work on the Double Mirror Pyramid Panoramic Camera, that acquires panoramic images from a single viewpoint, to multiple viewpoints.

Kar-Han Tan, Hong Hua, and Narendra Ahuja. Multiview Mirror Pyramid Cameras.
IEEE Transactions in Pattern Analysis and Machine Intelligence. To appear.
Download Manuscript: [PDF]
PowerPoint: [ZIP]

High-resolution panoramic capture is highly desirable in many applications such as immersive virtual environments, tele-conferencing, surveillance, and robot navigation. In addition, a single viewpoint for all viewing directions, a large depth-of-field (omni-focus), and real-time acquisition are desired in some imaging applications (e.g. 3D reconstruction and rendering). The FOV of a conventional camera is limited by the size of its sensor and the focal length of its lens. For example, a typical 16mm lens with 2/3″ CCD sensor has a FOV. The number of pixels on the sensor (640 x 480 for NTSC camera) determines the resolution. The depth-of-field is limited and is determined by various imaging parameters such as aperture, focal length, and the scene location of the object.

Many approaches have been presented to achieve various subsets of these properties: wide FOV, high resolution, large depth-of-field, a single viewpoint, and real-time acquisition. Among these, mirror-pyramid (MP)-based camera systems offer a promising approach to capturing high-resolution, wide-FOV panoramas as they provide single-viewpoint images at video rate. Such systems use planar mirrors assembled in pyramid or prism shapes, and as many cameras as the number of mirror faces, each located and oriented to capture the part of the scene reflected off one of the flat mirror faces. Images from the individual cameras are concatenated to yield a 360-degree wide panoramic image. Compared to designs using parabolic or hyperbolic mirrors, flat mirrors are easier to design and produce, and they introduce minimal optical aberrations.

We have developed a double-mirror-pyramid design that doubles the size of the visual field of the single-pyramid based systems. With this prototype, we have developed methods for optimally choosing the parameters of MP-based camera systems, e.g., camera placement, pyramid geometry, sensor usage, and uniformity of image resolution, and how the resultant image quality can be evaluated.

2. Overview of panoramic imaging

The existing methods of capturing panoramas fall into one of the two categories: dioptric methods, where only refractive elements (lenses) are employed, and catadioptric methods, where a combination of reflective and refractive components is used. Typical dioptric systems include: the camera cluster method where multiple cameras point in different directions to cover a wide FOV; the fisheye method where a single camera acquires a wide FOV image through a fisheye lens; and the rotating camera method where a conventional camera pans to generate mosaics, or a camera with a non-frontal, tilted sensor pans around its viewpoint to acquire panoramic omni-focused images. The catadioptric methods include: sensors in which a single camera captures the scene as reflected off a single curved mirror, or sensors in which multiple cameras image the scene as reflected off the planar mirror surfaces.

The dioptric camera clusters are capable of capturing high-resolution panoramas at video rate. However, the cameras in these clustersdue to physical constraints, which makes it difficult or even impossible to mosaic individual images to form a true panoramic view, while apparent continuity across images may be achieved by ad hoc image blending. The sensors with fisheye lens are able to deliver large FOV images at video rate, but suffer from low resolution, irreversible distortion for close-by objects, and non-unique viewpoints for different portions of the FOV.� The rotating cameras deliver high-resolution wide FOV via panning, as well as omni-focus when used in conjunction with non-frontal imaging, but they have limited vertical FOV. Furthermore, because they sequentially capture different parts of the FOV, moving objects may be imaged incorrectly. typically do not share a unique viewpoint

The catadioptric sensors that use a parabolic- or a hyperbolic-mirror to map an omni-directional view onto a single sensor are able to achieve a single viewpoint at video rate, but the resolution of the acquired image is limited to that of the sensor used and varies significantly with the viewing direction across the visual fields. Analogous to the dioptric case, this resolution problem can be alleviated partially by replacing the simultaneous imaging of the entire FOV with panning and sequential imaging of its parts, followed by mosaicing the images, at the expense of video rate. Another category of the catadioptric sensors employs a number of planar mirrors assembled in the shape of right mirror-pyramids, together with as many cameras as the number of pyramid faces. Each of these cameras, capturing the part of the scene reflected off one of the faces, is located and oriented strategically such that the mirror images of their viewpoints are co-located at a single point inside the pyramid. Effectively, this creates a virtual camera that captures wide-FOV, high-resolution panorama at video rate.

Proposed Double-Mirror-Pyramid Camera

The main challenge in constructing a panoramic camera from multiple sensors is to co-locate the entrance pupils of the multiple cameras so that adjacent cameras cover contiguous FOV without obstructing the view of other cameras or their own. Nalwa first used a right mirror pyramid (MP) formed from planar mirrors for this purpose. He reported an implementation using a 4-sided right pyramid and 4 cameras. The pyramid stands on its horizontal base. Each triangular face forms a 45-degree angle with the base. The cameras are positioned in the horizontal plane that contains the pyramid’s vertex such that the entrance pupil of each camera is equidistant from the vertex and the mirror images of the entrance pupils coincide at a common point, C, on the axis of the pyramid. The cameras are pointed vertically downward at the pyramid faces such that the virtual optical axes of the cameras are all contained in a plane parallel to the pyramid base, effectively viewing the world horizontally outward from the common virtual viewpoint .

The vertical dimension of the panoramic FOV in each of the aforementioned cases is the same as that of each of the cameras used � only their horizontal FOVs are concatenated to obtain a wider, panoramic view. We have developed a panoramic design that uses a dual mirror-pyramid (DMP), formed by joining two mirror-pyramids such that their bases coincide (Fig. 2), together with two layers of camera clusters.� Such a DMP-based design thus doubles the vertical FOV while preserving the ability to acquire panoramic high-resolution images from an apparent single viewpoint at video rate.

]]> http://vision.ai.uiuc.edu/?feed=rss2&p=303 0 http://vision.ai.uiuc.edu/?p=299 http://vision.ai.uiuc.edu/?p=299#comments Thu, 07 May 2009 05:37:51 +0000 sanketh http://vision.ai.uiuc.edu/wordpress/?p=299

Standard imaging sensors have limited dynamic range and hence are sensitive to only a part of the illumination range present in a natural scene. The dynamic range can be improved by acquiring multiple images of the same scene under different…

We discuss how to generate omnifocus images from a sequence of different focal setting images. We first show that the existing focus measures would encounter difficulty when detecting which frame is most focused for pixels in the regions between intensity edges and uniform areas. Then we propose a new focus measure that could be used to handle this problem. In addition, after computing focus measures for every pixel in all images, we construct a three dimensional (3D) node-capacitated graph and apply a graph cut based optimization method to estimate a spatio-focus surface that minimizes the summation of the new focus measure values on this surface. An omnifocus image can be directly generated from this minimal spatio-focus surface. Experimental results with simulated and real scenes are provided.

Ning Xu and Narendra Ahuja. Generating Omnifocus Images Using Graph Cuts and a New Focus Measure. Pattern Recognition, 17th International Conference on (ICPR’04) Volume 4. pp 697-700, 08, 2004 Cambridge UK. Full Text

Most imaging sensors have a limited dynamic range and hence can satisfactorily respond to only a part of illumination levels present in a scene. This is particularly disadvantageous for omnidirectional and panoramic cameras since larger fields of view have larger brightness ranges. We propose a simple modification to existing high resolution omnidirectional/panoramic cameras in which the process of increasing the dynamic range is coupled with the process of increasing the field of view. This is achieved by placing a graded transparency(mask) in front of the sensor which allows every scene point to be imaged under multiple exposure settings as the camera pans, a process anyway required to capture large fields of view at high resolution. The sequence of images are then mosaiced to construct a high resolution,high dynamic range panoramic/omnidirectional image.Our method is robust to alignment errors between the mask and the sensor grid and does not require the mask to be placed on the sensing surface. We have designed a panoramic camera with the proposed modifications and have discussed various theoretical and practical issues encountered in obtaining a robust design. We show with an example of high resolution, high dynamic range panoramic image obtained from the camera we designed.

M. Aggarwal and N. Ahuja, High Dynamic Range Panoramic Imaging, Proc. International Conference on Computer Vision, Vancouver, Canada, July 2001, 2-9. Full Text

Recall that the mosaicing the set of images taken by Nicam as it pans requires correspondence of a scene point across the set. The presence of moving objects upsets the correspondence between images in the sequence, resulting in a distorted appearance of the moving objects in the final mosaic (see part (a) of the figure). We avoid these artifacts and create large depth of field mosaics of scenes with moving objects. The basic idea is to combine the sequence of images in a manner such that each entire moving object is selected from a single best focused frame and is pasted, as it is, in the final mosaic, and other regions are chosen and pasted similarly from those images where they are best focused and not occluded by the moving objects. This approach ensures that all regions, except possibly those related to moving objects, are best focused and no artifacts due to moving objects are present.

The necessary algorithms require segmenting the moving objects in each image, selecting where is it overall in best focus, and for each still scene point determining in which frame it is best focused. The algorithms use a set of tests to perform the segmentation of moving objects. The tests that we use are point intensity difference test, point correlation test, and focus trajectory test. The point difference and point correlation tests measure how image intensity and local texture vary over the sequence of images (the correspondence between images is known for still scene points). If the variation is small between adjacent frames the point is more likely to be static. The focus trajectory test checks for variation in the focus measure across the image sequence against expected behavior. For instance, for a still scene point variation in focus measure is expected to be unimodal, while for a moving scene point it will be random, since the correspondences between images for such points are incorrect. One result thus obtained is shown in the figure below.

                                         (a)                                        (b)

An omnifocus image acquired in the presence of moving objects: (a) Nicam approach. (b) New intermediate approach.

The concept of omnifocus nonfrontal imaging camera, OMNICAM or NICAM, initiated a new chapter in imaging and digital cameras. NICAM has introduced hitherto nonexistent imaging capabilities, in addition to overcoming some problems with previous methods. NICAM is capable of acquiring seamless panoramic images and range estimates of wide scenes with all objects in focus, regardless of their locations. To understand the impact of NICAM, first consider imaging with conventional cameras.The camera’s field of view is generally much smaller than the entire visual field of interest. Consequently, the camera must pan across the scene of interest, focus on a part at a time, and acquire an image of each part. All the resulting images together then capture the complete scene. As byproduct of focusing, the range of the objects in the scene can also be estimated. Usual methods for focusing as well as range estimation from focusing mechanically relocate the sensor plane, thereby varying the focus distance setting in the camera. When a scene point appears in sharp focus, the corresponding depth and focus distance values satisfy the lens law. The depth for the scene point can then be calculated length and the focus distance.

The conventional methods therefore involve two mechanical (and hence slow) actions, those of panning and, for each chosen pan angle, focusing i.e. finding the best focus distance setting. The purpose of the first action is to accumulate data for the entire visual field from the camera’s narrower fields of view. This action is therefore essential. An innovation of nonfrontal imaging is in the elimination of the second action. The nonfrontal imaging camera has a sensor plane which is not perpendicular to the optical axis as is standard. This imaging geometry eliminates the time consuming mechanical translation of the sensor plane. Camera panning, required for panoramic viewing anyway, in addition enables focusing. Further, a range-from-focus estimate for each visible scene point is also computed as a by-product of identifying the sharpest image. Thus, from pan motion alone, nonfrontal imaging obtains a composite focused image of all objects/points in a wide scene regardless of their depths, which is in complete registration with a range map obtained in parallel. While it is well known that focus distance control yields both range and focused images, nonfrontal imaging has made it possible for the first time to realize this dual functionality simultaneously for all visible scene points. Further, this functionality is achieved passively, i.e, without any active illumination of the scene, e.g., using laser.

Thus, nonfrontal imaging has the following novel capabilities: (i) It provides panoramic (up to 360⁰) images of a scene without any visible seams. (ii) Each object in an image is in complete focus regardless of its location, i.e., that there is no need to explicitly perform the standard focusing action (accomplished mechanically in “manual” cameras and automatically in “automatic” cameras, but requiring mechanical movement in each case). (iii) Along with the sharp visual image, the camera also delivers the location (coordinates) of each focusable, visible scene point. One consequence of these capabilities is that a single nonfrontal imaging camera can provide stereo pairs of images for three-dimensional, omnifocused, viewing of the entire scene in natural lighting. In fact, this visual 3D experience is even more informative in some ways than natural, human viewing of real world, since humans have finite depth of field while the NICAM driven display shows all parts in focus.

UNIQUENESS

Nonfrontal imaging represents qualitative leaps in what is feasible with the current technology. It
makes it possible to achieve hitherto infeasible functionalities and performance levels in imaging. None of the available techniques can deliver seamless focused panoramic images. Most methods choose a scene point/object and bring it into focus by (manually or automatically) controlling sensor location. Thus, they can focus at objects one by one. Similarly, they can estimate range from focus one object/point at a time. The following paragraphs describe the differences between NICAM and the various related existing technologies.

Panoramic Images: In conventional photography, generating panoramic images has been more of an art, pursued by artists who take independent photographs and create a mosaic from them. There are some fundamental problems with this, however. Whenever an image is taken by the camera, a choice of focus distance must be made. Usually, this is done by imaging in focus that object which appears at the image center. This means that any objects at other distances in the camera’s field will not be in focus. In particular, the image borders may have different amounts of blur along them. Since in general different focus settings are used to obtain photographs of contiguous scene parts, when they are mosaiced to form a panoramic image the discontinuities in the image sharpness of the scene parts straddling the borders give rise to seams. Thus in the panoramic image neither all objects are imaged in focus nor is the mosaic seamless. One may attempt to alleviate this problem by reducing the size of the camera’s visual field, but this does not eliminate the problem because the objects in the scene are not of the same shape as the camera’s visual field (e.g., rectangular) so they always straddle across image borders. Of course, the smaller the visual field size the larger is the number of images in the mosaic, which increases the seam density. In fact, a major issue in the construction of panoramic images has been how to process the mosaic to camouflage seams in order to avoid perceptual detection.

Panoramic Cameras: A number of panoramic cameras have been designed over the years for photographic applications. The scene scan is performed by moving the camera mechanically, or pointing it at a special reflector surface such as a conical mirror. To image certain scene points in sharp focus, either the points have to be at a specified depth from the camera, or the depth of field of the camera must be made sufficiently large by a combination of reducing the aperture and increasing the focal length. These solutions are not acceptable since they require that the ambient light intensity levels be high or that the scene objects of interest all lie in a narrow depth range.

Panoramic Range Acquisition: Analogous to panoramic image acquisition, panoramic range acquisition methods are not very common. Typically a narrow field of view device is pointed in different directions to obtain a panoramic range image. Or, multiple devices are arranged in a circle such as in sonar sensor rings. Almost all methods are invasive, i.e., they involve scene illumination, e.g., using a laser beam or structured lighting.

Patented Technologies: Most of the relevant U.S. patents in the last 5 years have been filed by Japanese corporations. However, none of these come close to the NICAM methodology. The closest idea to NICAM was in a patent filed for the Asahi Kogaku Kogyo Kabushiki Kaisha of Tokyo, Japan. In this patent, a camera was described that could select one of three tilt angles between the CCD plane and the optical axis. The application suggested was in photographing one frame of a scene with two subjects. The range to the subjects was determined by an unspecified “range-determining means”. EMI Limited, England has a patent on using a tilted plane sensor to determine the focus motor drive signal. Some of the other companies that hold patents in slightly related ideas are Olympus Optical Corporation, Japan, Canon Kabushiki Kaisha and Hitachi Ltd, Tokyo, Japan. A camera with a tilted sensor plane and a scanning mirror was also used by a researcher at Jet Propulsion Laboratory to determine the range of scene points.

SAMPLE IMAGES TO ILLUSTRATE PERFORMANCE

Images 1-5 illustrate the performance of NICAM. Typically, a panoramic image is divided into multiple rows, each showing the view over smaller than the entire angle covered. If the entire length of the panoramic image were printed in a single row, the height will be reduced significantly. To avoid such excessive compression of detail, and to maximize legibility by using all available space, a complete panoramic view is divided into smaller contiguous subangles, and the corresponding subimages are shown in successive rows of the image. Thus, the right end of the row connects to the left end of the following row.

Images 1, 2 and 3 are examples of omnifocused panoramic images acquired using NICAM. Image 1 shows a 100⁰ view of an outdoor scene, shown split into two rows of 50⁰ each. The objects in the scene are at a range of distances (flowers 1.5′, tree 4.5′, bench 8′, chair 30′ and building 50′ and larger) but all are imaged in focus and no seams exist across the entire panoramic view. Image 2 shows a 60⁰ view of a room inside the Beckman Institute where the distances range from a few feet to about 20 feet. Finally, Image 3 contains a 360⁰ panoramic view of the Computer Vision and Robotics Laboratory in the Beckman Institute where object distances of 2 to 30 feet from the camera are indicated.

Image 4 demonstrates the omnifocusing performance of NICAM compared to the limited depth of field of a regular camera. The upper row in Image 4 (i) shows a 40⁰ (angle chosen by user) omnifocused image acquired by NICAM and the lower row shows a 20⁰ (angle dictated by the camera) view of the same scene acquired using a regular camera focused at 4′ (a choice must be made as to which object to focus on). The progressive loss of focus for objects located closer or farther than the focused depth of 4′ can be seen. Image 4 (ii) shows a “panoramic” view constructed from multiple images taken by a regular camera, by concatenating images of contiguous parts of the scene. Since the different images are taken when the camera is focused at different objects, the borders between images give rise to seams. Further, since the camera is focused on a specific object as each image is acquired, not all parts within even a single image are in focus. This should be contrasted with the seamless, panoramic imaging capability of NICAM shown in Images 1-3.

Image 5 (i) demonstrates the impact of the range estimation capability of NICAM A pair of contiguos planar patches formed by wrapping newspaper on a step-like structure is placed in front of NICAM. The resulting omnifocused panoramic image is shown in the first row. The second row shows the range estimation capability wherein the step structure recovered by NICAM is depicted. The range estimate available for each pixel in the omnifocused image determines the position and irradiance of the corresponding scene point. The omnifocused image and the recovered shape are combined to produce the 3D omnifocused step structure shown through a perspective view in the bottom row. If we assume that the irradiance due to this scene point is invariant for small perturbations of the viewpoint, then the intensity and range information can also be combined to produce pseudostereo images as would be acquired by a pair of hypothetical cameras placed around NICAM. Such stereo images when viewed through a stereo mechanism, e.g. stereo glasses, depict the scene in full 3D, using data obtained by a single NICAM! Image 5 (ii) shows such stereo pairs for three scenes. For ease of viewing, the left and right “eye” images have been color coded red and green and overlapped; when viewed through the enclosed red and green glasses on the different eyes, the scenes can be seen in 3D and omnifocused. The top left scene consists of two planes perpendicular to the line of sight, the nearer one at a distance of 2 ft. from the camera (right plane) and the farther one at a distance of 3 ft. (left plane). The top right scene contains a single chess piece at the distance 1 in. The bottom row shows multiple chess pieces placed at three different distances from the camera: 17 in., 20 in. and 25 in. All parts of all scenes are in focus and the 3D structure is visible through stereo viewing. It appears that such 3D depiction using a single visual

camera has never been accomplished before!

Image 1: 100⁰ Outdoor View

Image 2: 60⁰ Indoor View

Image 3: 360⁰ Indoor View

Image 4(i): 40⁰ Omnifocused View

Image 4(i): 20⁰ Standard Camera View of the Same Scene as in 4(i), Focused at 4 ft.

Image 4(ii): Regular Camera Mosaic

Image 5(i): On the following page:

Image 5(ii)

APPLICATIONS OF OMNICAM

Following are some examples of applications that depend on NICAM’s unique imaging capabilities.

Photography: Suppose a photographer wishes to capture a scene around the Washington monument. She must make two decisions before she can push the camera button. First, she must decide which part of the scene she wants to capture in the photograph, and then direct the camera to point in the desired direction using the appropriate zoom lens. Second, within the visible scene, she must determine whether the photograph should show the Washington monument in sharp focus, or the trees in front, or the buildings behind, and set the focus control accordingly. The result will be a picture showing, say the monument, in sharp focus, and the lawn to the left and right as well as the trees and the buildings blurred. If she wants to show a wider scene than the camera’s field of view, she must take multiple pictures by moving the camera across the scene and then manually “pasting” the individual photographs to make a mosaic showing the large scene. Each picture in the mosaic will show a pre-selected object in sharp focus, with consecutive photographs taken, in general, with different focus settings. Consequently, the degree of blur across a picture boundary would change visibly, causing perceptible sharpness transitions across objects that happen to straddle inter-picture borders. Of course, even within each pasted image the objects other than those focused on appear blurred proportional to their relative distances from the focused object. Further, the entire process is very time consuming because of the required mechanical redirecting of cameras, mechanical focusing in each direction, and the subsequent cutting and pasting. Because of these difficulties, such mosaicing of images has been practiced as an art by photographers wherein they attempt to smooth out the interimage transitions. Using the NICAM technology, the photographer could capture the Washington monument, the buildings behind, the trees in front, the lawn and objects as far to the left and right as she wants, up to a complete 360⁰ view, all of which would be completely focused in a single photograph. The significance of NICAM to nature photography is mentioned in a letter (Appendix A) written by a former staff photographer of the National Geographic Magazine.

Security and Surveillance: Another major application area is that of security systems and surveillance. For example, consider multiple cameras located at one or more posts outside a building to achieve visibility in all directions. Typically, a guard inside the building would monitor the images delivered by the cameras on separate monitors placed in front of him. The cameras of course show certain objects in focus while others, outside of the depth of field, appear blurred. The guard could control the cameras to focus at different objects in different directions, but that would only result in switching among the areas monitored best. If a fast panning NICAM is used, a single camera will replace the entire set of cameras and still obtain a focused image of all parts of the scene visible from the watchpost, with no loss of detail due to blurring. This image could be displayed over a surrounding screen, as well as on the usual separate monitors. Further, the guard could also be shown a 3D stereo display of the 360⁰ surround, using a single NICAM. His display could be viewed in 3D using a headmounted display. Alternatively, the guard could see it on a surrounding 360⁰ screen, e.g., using stereo glasses, while being able to turn his head in any direction as if he were on the watchpost itself.

Analogously, inside a building, a small number of NICAM’s can cover the entire premises to be monitored, instead of a much larger number of normal cameras. For example, a 360⁰ view of a large building lobby may be covered by a few NICAM’s instead of, say 15, ordinary cameras which may still not cover as much depth as NICAM. When located inside a nuclear plant or another hazardous area, clear views of the entire scene can facilitate much faster teleoperation with no adjustment of the camera.

Surgery: In endoscopic or laproscopic surgery, it is a common problem that the surgeon cannot clearly see the interior body structure in the vicinity of the area under operation which leads to imprecision in surgery. A miniature NICAM could imag