Research And Applications Of Bidirectional Multi-reference Frame Motion Information Scalability

With the development of wireless communication technology, wireless video communication has been the main aspect of wireless mobile communication and their applications have been more various, such as wireless video broadcast, wireless video talk, wireless video-on-demand (VOD), wireless video monitor and so on. In fact, since there are different networks and different user terminal, the requirements to the quality of video are different. Therefore, in order to reduce the redundancy of video coding devices, the scalable video coding(SVC) is proposed to obtain a single bit stream and the quality of video can be adjusted by transmitting parts of the bit stream according to different network width, different display devices and different decoding ability of play back devices.Generally, the motion information of SVC is nonscalable. However, the motion information is a core constituent of video coding, its nonscalability in SVC is not reasonable and therefore the research of motion information scalability is very significant. The aim of motion information scalability is that it provides the best video communication quality by assigning the bit rate of motion information and the bit rate of texture information properly for a fixed bit rate. Especially, it improves the coding efficiency and ability of video coding in the medium to low range of decoding bit rate and is more suitable for wireless video communication. The technology of motion information scalability has been the main aspect of new generation video coding.To resolve the limitation problem of motion information scalability in SVC system based on discrete wavelet transform (DWT), this dissertation researches the algorithms and rate-distortion feature of motion information and its scalability deeply. Finally, a fully and completely scalable motion model is developed. The contributions of this dissertation are as follows.1. Implement the bidirectional multi-reference frame motion information scalability. The scalable coding techniques on motion information which have been proposed are mainly based on unidirectional motion with single reference frame and their applications are limited. This dissertation proposes the scheme of bidirectional multi-reference frame scalable motion model and corresponding motion estimation algorithm. The simulation results show that they can implement the bidirectional multi-reference frame motion information scalability effectively.Some simulations show that when the movement intensity is high, because of enormous bits of bidirectional multi-reference frame motion information, video bit stream can't support lower bit rate. To well resolve this problem, some penalty multipliers for cost functions are proposed by analyzing the data features of scalable motion information. In motion estimation, the accuracy of motion and the complexity of algorithm are opposite. Especially for bidirectional multi-reference frame motion estimation, its motion estimation is accurate, however, the computational complexity is very high. By applying the improved sub process of bidirectional multi-reference frame motion estimation and inverse motion prediction based on the assumption of uniform motion in a straight line, the complexity of motion estimation is reduced without sacrifice of motion accuracy.2. Build the rate-distortion (R-D) function of video coding based on motion information scalability. There is corresponding relationship between the bit rate and the distortion of video coding system. With the deep analysis of texture information R-D function and motion information R-D function, a fully rate-distortion (R-D) function of video coding based on bidirectional bi-reference frame motion information scalability is set up for the global analysis of video coding.3. Propose the fast search algorithms for the best level of motion information scalability. When motion information scalability is taken into consideration, the balance between motion information bit rate and texture information bit rate is required to obtain the best video communication quality and the best level of motion information scalability should be decided in the bit stream extractor. Based on the R-D function of video coding which is based on bidirectional motion information scalability, this dissertation proves some features of the best level of motion information scalability. Based on these features, some fast search algorithms for the best level of motion information scalability are proposed. The simulations show that these algorithms can reduce the complexity effectively.4. Propose the algorithm for obtaining the bit rate boundary of the best level of motion information scalability which makes the best levels of motion information scalability have more accurate application range of bit rate. It provides the evidence for video data extract in video decoder of video coding system with motion information scalability.5. The motion information scalabilities which have been proposed are almost based on the SVC systems which are based on wavelet transform. This dissertation first applies the motion information scalability into H.264/SVC video coding system. The simulations verify the superiorities of proposed scalable motion over non scalable ones.