IMPROVING STEREO MATCHING QUALITY WITH SCANLINE-BASED ASYNCHRONOUS HOPFIELD NEURAL NETWORKS

Stereo matching problem has been a critical task in a stereovision system. This paper presents an asynchronous Hopfield neural network for solving a scanline-based stereo matching problem. The primitive of matching is the edge point extracted using Sobel operator in stereo images. Then, the matching problem is formulated as a '0-1' integer programming with a dissimilarity objective function and disparity and uniqueness constraints, and subsequently is transformed into the form of an energy function scanline by scanline. Finally, the asynchronous Hopfield neural network is used to obtain the minimum energy value that corresponds to the approximated optimal solution of the problem. A set of commonly used stereo images is used to verify the proposed method. Experimental results show that the proposed method can be suitably implemented on the PC platform, and gain efficiency by using scanline-based matching criterion.