Improve accuracy of disparity map for stereo images using SIFT and weighted color model

Usually, stereo images are used to visualize to human eyes in 3D. Many research projects used stereo images to reconstruct a 3D model from the stereo images. Basically, information on x-y plane can be extracted from one of stereo images and depths of fields are computed from disparity distances between the same objects on different images. Matching the same object in both images requires high computation. Therefore, accuracy and speed of depth estimation depends on those of disparity map analysis. In this paper, an algorithm to compute disparity map of stereo images is proposed. Search area to match objects is bounded by SIFT algorithm in order to speed up. Weighted color mode is used to improve the accuracy. Dataset of Middlebury is used in the experiment. SAD algorithm is used as a baseline. The experimental results show that accuracy and speed of the proposed method are 25% and 22% as much as of the baseline, respectively.