Stereo matching using 2-D Hopfield network with multiple constraints

Stereo matching is one of the classic difficult problems in the computer vision, and its complexity and precision hedge the capability of vision system to reconstruct the 3-D scene. This paper presents a new matching method based on neural network. On the condition that stereo rectification has been performed, the energy function is built on the basis of uniqueness, compatibility and similarity constraints, which reflects the constraint relations of all matching units of the same lines. It is then mapped onto a 2-D neural network for minimization, whose final stable state indicates the possible correspondence of the matching units. The depth map can be acquired through performing the above operation on the all epipolar lines. The algorithm has two traits relative to the traditional approach. Firstly, individual pixel point but not scene point or edge line is adopted as matching unit and dense depth map could be obtained directly. Secondly, the external input of the nodes is not constant again and is the function of gray similarity of correspondent points. The experiments on the synthetic and real images demonstrate the feasibility of the proposed approach.