Area-Based Estimation of Stereo Disparity Using Hierarchical Windows

We propose a new solution to stereo matching which is combined the area-based stereo method using hierarchical windows with the pixel-based stereo methods using global optimization. The areabased method has a problem that the best size of area is unknown. Since the proposed method changes the sizes of windows hierarchically, the local minima problem and a-priori knowledge of the size can be removed. The proposed method brings better robustness in the analysis of stereo images having periodic texture and textureless areas of images, and a scene having abrupt disparity changes, compared to the ordinary stereo method using hierarchical multi-resolution approach.