New Algorithms for Object Segmentation and Stereo Correspondence in Stereo Video Coding
立体视频编码中的目标分割与立体匹配算法

Video object segmentation and stereo correspondence is a key technology in object-based stereo video coding. In this paper, based on an algorithm for extracting video object in one view-channel, a new contour tracking and correspondence algorithm is first proposed for extracting stereo video object pairs. Then, a disparity estimation algorithm based on video objects is proposed. It takes a characteristic pixel as a correspondence primitive and the disparity estimation is implemented with disparity constraints. The disadvantages of discontinuous and imprecise disparities with the traditional block-based algorithms can be improved and a precise and smooth disparity field can be finally acquired.