Perception of occluding contours: Neural mechanisms and a computational model

Recent studies of neuronal responses to illusory contour figures were reviewed and the evidence for a simple bottom-up processing model was discussed. The model is based on the idea that the system infers occluding contours by combining the results of multiple parallel paths of computation, each of which exploits a specific cue. Its basic structure and many details were derived from single cell studies of areas V1 and V2 of the monkey. A computer implementation of the model was shown to predict the appearance and exact shape of various illusory contour figures, as well as the figure-ground effect. It also generated contour maps on complex natural images that were significantly better than conventional edge maps. New experiments showed that the contour cells in monkey V2 encode also the implied direction of figure and ground, as predicted from the model. Other models that have been proposed use recurrent processing to explain illusory contours and figure-ground segregation. Our results suggest that the mechanism might actually be simpler, involving convolutions and nonlinear operations, but no feedback.