Research On Key Technologies Of 3D Video Coding

In the application of multimedia technology, digital video has attracted much more attention due to its huge amount of information, flexible and diversified expression, and rich content. Nowadays, applications such as digital TV broadcasting, Internet video, digital video surveillance, and digital mobile video are ubiquitous in people's daily life. However, people pursue technical progress all the time,and how to provide higher quality video services with limited storage space and transmission bandwidth becomes a new challenge of the digital video technology. Therefore, research and application of novel digital video techniques have also become a hot topic in both industries and research institutes.Among the numerous new research directions of digital video, 3D video which can bring people the fresh sensation of stereo andthe real perception has caused wide concern. Meanwhile, 3D video has been preliminarily applied in some circumstance. However, 3D videocoding still facesproblems such as slow standardization process, low coding efficiency, high coding complexity, and lack of effectivemetric for coded video quality, which influencethe application, the extension and the industrialization process of 3D video. This paper starts with the basic principles of 3D video, makes explorationon problems mentioned above, and proposes some effective coding techniques and useful algorithm models.The detailed contents of this dissertation are as follows.In 3D video coding, main problems are the slow standardization process and the low coding effeciency. So, this paper proposes a intra-frame/inter-view prediction mode, which has been adopted by standard workgroup. Liti profile for 3D video presented byAudio Video coding Standard Workgroup of China (AVS) comprises binocular combined coding scheme and stereo packing mode, in which stereo packing mode has advantages such as high compatibility and low complexity. However, in the traditional packing mode, part of information is lost because images are down sampled when merging, and the correlation of images from different views in one frame is ignored, so that its coding efficiency is lower compared with the binocular combined coding scheme. In order to improve the performance of the stereo packing mode, this paper proposes a intra-frame/inter-view prdiction mode. In this mode,utilizing the correlation of different views, the inter-view redundancy is removed, and thecoding efficiency is increased. In detail, for the packing frame, after reconstructing the image from the first view, the image from the second view may refer the reconstructed image from the first view in the same frame, besides referring to the temporal neighboring frames when encoding. After that, the optimal mode is selected by rate-distortion optimization.During the acquisition of 3D video, because of the difference of parameters, angles and positions of the two neighboring cameras, illumination change exists between images from different views at the same time instance, which will influence the accuracy of inter-view prediction, and furthermore decrease the coding performance. To solve the problem, this paper presents a template based illumination compensation algorithm, which predicts the illumination change of the current block and the reference block from the change of L templates of these two blocks. Moreover, several prediction models can beconstructed to improve the accuracy of the prediction and to remove the influence on disparity estimation caused by the illumination change. For example, different prediction model can be built according to different texture directions. Then, optimal model is selected by prediction model selection strategy to improve coding efficiency.3D video has attracted great attention from industries and research institutes, but compared with triditional 2D video,the heavy computational complexity limits itspractical applications. So this paper presents a fast disparity andmotion estimation algorithm for 3D video coding based on the correlations between the neighboring camerasand between the motion and the disparity. In the proposedapproach, first, a search region estimation is proposed toreduce the disparity estimation complexity according to thatthe camera set is usually fixed and the disparitybetween the two neighboring views can be limited to anestimable range. Second, a motion vector derivation is givenbased on the geometric relationship between the motion andthe disparity in this paper. Finally, this paper provides an early termination strategy to reduce the number of reference frames.The experimental results show that comparedwith the anchor in multiview video coding test modelJSVM, the proposed algorithm can save roughly 50% coding time with negligible coding efficiency loss. Finally, in order to define the least distortion people can notice for 3D video, this paper exploresthe just noticeable distortion model (JND) for 3D image/video. Using human's binocular suppression and depth perception ability,this paperconstructs jointJND for 3D image/video. First, an image is decomposed into the occlusion region and non-overlapped region by disparity estimation. Then, different JND thresholds are computed according to the depths of different regions, which are derived from the disparity field obtained from the previous disparity estimation. Experimental results verified the validity of the joint just noticeable distortion model for 3D image/video.In conclusion, this paper makes deep exploration on issues of standardization, coding efficiency, complexity, and quality assessment system of 3D video coding, proposes correspondingtechnique schemes, and ahcievesgoodperformance.