General layered neural network for stereo disparity detection

The general layered network for stereo (GLNS), a method for creating the general layered neural network to get stereo disparity maps, is presented. By using GLNS, disparity maps can be obtained without iterative calculations because it is a three-layered neural network. GLNS accepts gray levels in stereo images and output stereo disparity maps of the input images. A multistage learning method for training a large-scale network without falling into worthless local minimum points is proposed. The method consists of four learning stages that progress from simple tasks in a small network to complicated ones in a large network. The parameters from one task are used as initial parameters in the succeeding stages. Connection parameters of GLNS can be decided by the learning. Feature extraction and any heuristics or knowledge about images are not required. The matching property and sensitivity of GLNS can be easily designed. Random binary patterns with various disparities between left and right are used as training sets. GLNS can be changed by a small amount of additive learning into specific neural networks for stereo which are tuned to specific kinds of stereo images