Research On Object-based Stereo Video Coding

With the dramatic development of technology and the improvement of living standard, the ordinary video has been unable to satisfy people’s visual requirements. By adding the depth information of scene, stereo video can realize stereo feeling and sense of reality, and is extremely popular compared with ordinary video. But at the same time, the data volume of stereoscopic video is enormous and far greater than ordinary video, which makes it difficult to store and transmit video data. So it is of great theoretical and practical significance to improve efficiency of stereo video coding method. Compared with the traditional waveform-based coding method, object-based stereo video coding method can not only effectively reduce the coding errors, but also can provide content-based interactive function and has the strong ability of content description and extensive application prospect. This article, starting from the issues of practical process of object-based stereo video coding, mainly analyses the aspects of image segmentation and stereo matching.Image segmentation is a hot research issue and a classical difficulty in the field of image processing and computer vision. To solve problems of large time consumption and low segmentation precision at color similarity area of classical Grabcut algorithm, a new algorithm was proposed by using super-pixel technique and improving the Grabcut algorithm aiming to acquire high efficiency and accurate segmentation results. For a given image, it is firstly transformed into super-pixel-based representation with the edges of foreground objects well preserved based on an improved Slic-based algorithm. Then, with super-pixel as processing unit, segmentation is performed based on the framework of Grabcut, where the energy function is improved and rebuilt by incorporating color and gradient information. The proposed algorithm can improve the segmentation accuracy and is more efficient compared with the original algorithm.Stereo image segmentation is a crucial and difficult issue in the field of object-based stereo image processing. Based on graph cuts and inter-view correlation, a novel and efficient stereo image segmentation scheme was proposed to solve the problems of low segmentation precision and re-segmentation in different channel images. The left image is firstly segmented and foreground object is extracted by using improved Grabcut algorithm. Then, based on visual correlation of left and right images, the right image foreground object is obtained by using contour correspondence that fusing color and texture features of images. The experimental results show that the proposed method not only can accurately extract the stereo image objects, but also avoid the phenomenon that the segmentation result is sensitive to the accuracy of disparity image in the disparity-based segmentation method.Stereo matching is one of the key technologies in stereo video coding and can effectively remove the spatial redundancy of video data. In view of the low matching accuracy of existing matching algorithms, especially at occlusion regions, textureless regions and discontinuity regions, a new matching method based on image segmentation and adaptive weight was proposed. The proposed algorithm can acquire accurate dense depth image, which outperforms other existing common stereo matching algorithms.