Stereo matching using generalized symmetry transform

We propose a way to find corresponding points of a stereoscopic image by applying an algorithm of a generalized symmetry transform. A stereoscopic image entails the acquisition of a left image and right image simultaneously. Firstly, by using corner detection for the edge of both left and right images, feature points are analyzed, a square domain on the center of each feature point is found, and symmetry magnitude is such that edges within this domain have cumulated to positions of a feature point. Points are selected below the threshold value to correspond by executing a comparison of symmetry magnitude between edges of the right image and results of the left image. A more subdivided comparison of these corresponds in reserve through symmetry magnitude comparison of the radius unit within the domain, and if point correspondence could not be obtained in this way, corresponding points can be analyzed by measuring color correspondence rate of corresponding points in reserve. As for calculating correspond points, the proposed algorithm is more accurate than execution with feature points.