Early Recurrence Improves Edge Detection

A biologically motivated computational model of early recurrence is proposed for edge detection. Studies of the primate vision suggested that visual features are transmitted in the two visual pathways with different speeds (with the dorsal pathway processing faster than that of the ventral pathway) and the presences of extensive recurrent connections across the two pathways. It is thus likely that the dorsal perception facilitates the ventral perception via early recurrent mechanism. Following these neural principles, we hypothesize that early recurrence enables responses to high-spatial frequency features (fine edges) to be suppressed by low-spatial frequency features (coarse edges) in a multiplicative manner. Using real images, we quantitatively compared contours calculated by our work with another well-known biologically motivated model. To further explore early recurrence in solving machine vision problems, the representation is used to boost different popular edge algorithms. Results from both experiments lead to the conclusion that early recurrence has a positive and consistent influence on edge detection.