A neural architecture for illusory contour detection

Summary form only given. A local luminance based approach to the problem of detecting illusory and real contours in a scene was developed using the properties of the Hough transform. A connectionist architecture implementing an enhanced Hough transform model was simulated using the UCLA-SFINX environment and tested on gray-level images. Results indicate that the proposed algorithm has the ability to detect illusory contours but needs additional information to monitor the filling-in process. The type of additional information needed by an illusory contour detection mechanism was considered. >