Study Of Stereo Video Coding Based On View Mapping

Stereo video technique is regarded as an important development direction of video technique in the future. In many applications, two-dimensional color image has a very good visual effect, as living standard improved, however, people are not satisfied with the flat two-dimensional visual effects. Images and video with the stereoscopic are being welcomed by the people because of its "realism".In addition, in many practical applications, it is necessary to add more authenticity. For example, if the user could roam in a virtual three-dimensional space, then the virtual visit, the remote operation in hazardous environments and telemedicine would be more real and effective. The principle of binocular parallax of human eyes is utilized in stereo video technique. Binocular independently receives camera images of the same scene. Left eye views the left image and right eye views the right image. So the binocular parallax was achieved. The brain can gain the depth information of the image. In the end, people can enjoy a three-dimensional visual image with a strong "sense of depth" and "realism".However, stereo video image will give rise to dramatic increase in video data. It is more difficult to store and transmit these video data. In the future, for real-time transmission of stereo video, the higher bandwidth of network must be required. At the same time, as improving of image quality that people eventually required, communication of high-resolution and multi-view images would be a "killer" of future network. At present, in theory and practice, compression of two-dimensional image has been made great progress. For three-dimensional effects and multi-view transmission, however, there are so many things to be done. Therefore, in the future, one of the keys for utility of stereo video is how to encode for compressing stereo video data efficiently and reduce the amount of data. To sum up, although interactive stereo video technique has exciting prospects, it is a very challenging task.In this paper, a new stereo video encoding method based on view mapping is studied and implemented. After summing up a large number of domestic and foreign literatures, analyzing plentiful simulation results, considering the complexity of algorithm and experimental results comprehensively, the stereo video encoding based on view mapping has been put forward.The coding algorithm in this paper takes full advantage of the information of color image. The procedure is as follows: First, according to the color stereo video image, three-dimensional color space model is built. Second, the pixel point is acted as a matching primitive. Third, under the conditions of similar constraint in color, the add function with the smallest minus the average absolute difference is selected as matching criterion. The experimental results show that the algorithm can achieve the dense disparity map and reconstruct another view in the certain bit-rate.The main work of this thesis is summarized as follows:(1) In this paper, a new stereo video coding algorithm based on view mapping is put forward, designed, and implemented on the platform of VC++6.0. The algorithm is proposed for color images. At the encoding end, first of all, three-dimensional color space model is constructed according to the color image of left and right channels, and then the appropriate matching criteria is selected using the color similarity constraints. After completing the joint stereo matching of the left and the right channel views, a more dense disparity map can be achieved. At the decoding end, after combining the view of one channel and the disparity map, the other channel view can be restored.(2) Accurate and dense disparity map can be obtained using the algorithm of this paper. First of all, the algorithm chooses a pixel as the matching unit. It can exactly match to a level of pixel. In this way, it provides a basic condition for better restoring the details of information. Secondly, the algorithm makes full use of the color image information. After building three-dimensional color space model of the left and right images, and using the constraints of color similarity, the more accurate stereo matching can be realized. Thirdly, the matching criterion called the add function with smallest minus the average absolute difference, can weaken the impact of noise, such as the light which has an uneven distribution, reinforce the relevance of intra-pixel, and improve the accuracy of stereo matching. Finally, for different sequences, the different matching window can be chosen, the different search scopes can be set, and the more dense and accurate disparity map can be obtained.(3) The algorithm in this paper has achieved the purpose that the view of the other channel can be reconstructed in a certain bit-rate. From this perspective, namely the subjective evaluation of experimental results, the output view of H.264 decoder (all coded as I frames) is fuzzy, especially in the detail region, and also has the "blocky effect". In contrast, the algorithm in this paper can reconstruct the details better, and obtain the better subjective visual effect. The shortcoming of the algorithm is that there is significant noise in the reconstructed views, when the scenes are complex and the object moves quickly. From this perspective, namely the objective evaluation of experimental results, comparing this algorithm with the right-channel videos encoded independently based on H.264 encoding platform which codes all frames as I frames, the PSNR of reconstructed images has some improvement under the similar compression ratio.To sum up, the algorithm has achieved the purpose that a more dense disparity map has been obtained, and the view of the other channel has been reconstructed in a certain bit-rate. But when the scenes are complex and the object moves quickly, there is significant noise in the reconstructed images. Therefore, in order to make the algorithm have a better versatility and get more dense disparity map, the next research work will focus on looking for the appropriate pre-processing methods, choosing for the more reasonable matching criterion, changing adaptively the scope of the search and making the necessary post-processing of disparity map.