Mark Tabb and Narendra Ahuja

This paper is concerned with the problem of detecting low-level structure in images. It describes an algorithm for image segmentation at multiple scales.The detected regions are homogeneous and surrounded by closed edge contours.The segmentation is achieved by utilizing a new concept of scale which is integrated into a nonlinear transform. The use of the transform helps in collecting spatially distributed evidence for edges and regions and in making it available at contour locations, thereby facilitating integrated detection of edges and regions without restrictive geometric models. In a sense, Gestalt analysis is used to do local structure identification. The resulting multiscale segmentation extracts structure of complex geometry without smoothing of the boundaries.

Keywords: computer vision; multiscale image segmentation; recent transform; low-level structure detection; algorithm; multiple scales; multiple scale method; closed edge contour; image region; restrictive geometric model; Gestalt analysis; local structure identification.