Realizing visual functions with the reaction-diffusion mechanism

The present paper proposes a computational model for the realization of visual functions of edge and/or feature detection and segmentation. The model utilizes a reaction-diffusion model which is an extended version of the diffusion-based Difference of Gaussians (DOG) filter previously proposed by Marr and Hildreth as an edge detection model. The proposed model self-organizes spatial patterns having edges and/or features and segments. These patterns are sustained by the intrinsic mechanism of the proposed model under specific conditions. In addition, the model also helps to solve the stereo matching problem in random dot stereograms and the aperture problem in optical flow computation. These Visual functions of the proposed model are demonstrated with both artificial and real images.