SELECTION AND USE OF IMAGE FEATURES FOR SEGMENTATION OF BOUNDARY IMAGES.

An algorithm is developed to segment arbitrary boundary images into sets of boundaries which represent a single object, and to group together lines which correspond to a single object or object part. The algorithm is based on features which were found to be used by humans in the early stages of visual processing, and which have a high correlation with perceptually significant aspects of images. In addition, the data structure used is based on the image representation used in the primate visual cortex. By using perceptually valid features, the algorithm is able to enhance the perceptually significant edges in an image using simple, local, parallel computations. The algorithms demonstrates that selective processing can occur in the parallel stages of early visual processing, without domain specific knowledge, iterative processing, or top-down control of some mechanism to shift attention.