Extraction of Illusory Contours by Perceptual Grouping

The detection of arbitrary objects in images is still a major challenge in image processing. Many approaches rely on the robust extraction of contours. Object contours, however, are frequently occluded by other objects. How can one deal with this problem? One important source for inspirations in such diicult situations is the human visual system. There are several systems that describe the early stages of visual contour processing up to the level of complex cell responses; only few models describe even end-stopped cells. However, the processes of grouping these local cell responses together are still hardly understood. This paper builds on a previously described model of early vision and investigates the logic of grouping. As a result, we propose a method to treat the induction of contours by corners and line-ends in a uniied fashion. The properties of these grouping methods are demonstrated for illusory contour stimuli.