Discontinuity-preserving Stereo Matching With Occlusions And Illumination Variations

In the digital age, the traditional 2D video can not meet the need of people’s daily life. 3D technology has become an important area of scientific and technological development. Among them, 2D cover to 3D technology can transform the traditional flat video into stereoscopic video. This method has a low cost and a high applicability. However, due to the fact that the depth information of the video is limited, the e ffect can not meet the actual needs.Currently, a lot of method can obtain the depth information. Among them, the stereo matching is one of the most potential topics. Through analysis the disadvantages of the tradition stereo matching method, o ur thesis has launched discussions on disparity estimation and has mainly undertaken our tasks over following areas:(1) A stereo matching with certain region detection and adaptive disparity refinement method is put forward. Considering the relationship between the initial disparity maps of different viewpoint, we obtain the certain and uncertain regions. Based on the certain region information, the photometric and geometric properties of the stereo images, we refine the disparity in uncertain regions. The experiment results demonstrate that the proposed method successfully solve the errors information in uncertain region and achieve high accurate disparity map.(2) A novel reliability- guided discontinuity-preserving stereo matching method is proposed, through the matching cost in disparity estimation, we detecting the reliability of every pixel in the initial disparity map. Then based on the obtained reliability factor, we calibrate the initial disparity map. The experiment results demonstrate that the proposed method successfully calibrates the errors in initial disparity map and improves the accuracy.(3) A novel photometric invariant stereo matching based on the histogram cumulative distribution is put forward. Based on the cumulative distribution function, the stereo matching is utilized to remove the adverse effect from the illumination problem and increase the robust of the stereo matching. In computing the matching cost, we consider the segmentation information and a Gaussian weighting function to adjust the weighting factor, and increase the accuracy of the stereo matching. The experiment results demonstrate that the proposed method successfully estimate the high accurate disparity map and robust to the different illumination conditions.