IV. IMAGE STRUCTURE DETECTION AND REPRESENTATION

     The work in this area is aimed at the extraction of spatial structure in multidimensional images and dot patterns although most work has focused on two-dimensional and three-dimensional cases.

IV.1. Statistical models of pixel value variations have been developed and analyzed. Some of the work focuses on kernel density estimators to develop such models. Consequently, statistical theory of density estimators can be used for various tasks including segmentation of locally/globally parametric image signals; scale estimation and object registration.

IV.2. Robust medical image segmentation based on non-linear shape prior knowledge.

IV.3. Approaches have been developed to detect the multiscale structure present in natural images without restricting their geometric, photometric and topological complexity. Structure is detected at all prevailing, a priori unknown, geometric and photometric scales, and represented as a tree. This is accomplished through an image transformation for hierarchical image segmentation that maps scalar (univariate, e.g., grayscale) or vector (multivariate, e.g., color) images onto a tree of segmentations, to capture the perceived regions and their recursive embedding. Data in both two and three dimensions have been used.

IV.4. Another effort has been aimed at the detection and representation of multiscale perceptual structure in two types of nonimage signals: 2D and 3D dot patternst, to model perceptual grouping processes; and 1D biological, sensory signals, to explore neural mechanisms and computational algorithms to extract signals embedded in an ensemble of similar signals.

IV.5. Work has been done on modeling, analysis and synthesis of repetitive (statistically) image structure - the image texture. Focus has been on texture models that capture interregional, instead of interpixel, relationships. Generative, random mosaic models have been proposed for texture analysis as well as synthesis. Model identification involves analytical and numerical solutions of problems in random geometry concerning relationships between the types and parameters of the underlying random geometric processes and the properties of the resulting textures.

IV.6. Algorithms have been developed for the extraction of 2D shape representations, such as medial axis, of image regions.

IV.7. Methods for Efficient generation and maintenance of a hierarchical (octree) representation of the occupancy of 3D space by translating and rotating objects have been proposed. Algorithms for deriving octrees of objects directly from their silhouettes have been developed. Octree updating for objects under translation is realized by a simple binary arithmetic.