Stereo disparity perception for monochromatic surface by self-organization neural network

The stereo correspondence of two retinal images is one of the most difficult problems in stereo vision because the reconstruction of a 3-D scene is a typical visual ill-posed problem. So far there still have been many unsolved problems, one of which is to reconstruct a 3-D scene for a monochromatic surface. We consider this problem with a two layered self-organization neural network to simulate the competitive and cooperative interaction of binocular neurons. A refined pretreatment approach of a similarity map is proposed in order to carry out computation efficiently. We extend our previous neural network model by expanding the cooperation effect from the neighboring region. We are successful in obtaining stereo disparity perception for a monochromatic surface enclosed by random dot region and two vertical stripes. The experimental results with real scenes show that the monochromatic surface between two black vertical stripes is recognized efficiently through our neural network model.