Stereo Matching Using Synchronous Hopfield Neural Network

Deriving depth information has been an important issue in computer vision. In this area, stereo vision is an important technique for 3D information acquisition. This paper presents a scnaline-based stereo matching technique using synchronous Hopfield neural networks (SHNN). Feature points are extracted and selected using the Sobel operator and a user-defined threshold for a pair of scanned images. Then, the scanline-based stereo matching problem is formulated as an optimization task where an energy function, including dissimilarity, continuity, disparity and uniqueness mapping properties, is minimized. Finally, the incorrect matches are eliminated by applying a false target removing rule. The proposed method is verified with an experiment using several commonly used stereo images. The experimental results show that the proposed method solves effectively the stereo matching problem and is applicable to various areas.