Research On Video Codng Based On Dual Frame Motion Compensation

Currently, the video coding has increasingly become important and popular in industry and research. The popular video coding standards include H.264/AVC, MPEG2/4 and domestic audio&video coding standard AVS. With the fast development of channel transmission technique and source coding technique, high compression ratio is still the pursuit. In recent years, dual frame motion compensation (DFMC) based video coding is becoming a hotspot. In DFMC, for each frame, two reference frame are existed, the first reference frame is the most recently decoded frame, which is named as short-term reference frame (STR), the second reference frame is a HQF which is previous decoded frame, and is periodly updated, the frame is named as long-term reference frame (LTR). The kind of coding method can bring higher compression ratio, and meanwhile, it can satisfy the development of channel, such as cognitive channel and multichannel. DFMC based video coding still has some problem to be resolved, such as the jump period and bit allocation of LTR; some application based on dual frame motion compensation still has some problem to be solved, such as error resilience and rate control scheme. For the two unsolved problem, some analysis has been done in the paper, and the optimal LTR interval and bit allocation scheme. Furthermore, for packet loss channel, an error resilient jump udpate dual frame motion compensation scheme is given. And for the video coding application, the rate control scheme is given. Finally, jump update DFMC based multiview scheme is proposed. The main innovation of the paper include:1) Two problems in DFMC need to be resolved: the jump period and bit allocation of the LTR. In the traditional video coding scheme, every frame is allocated nearly the same bit rate, but in DFMC, the LTR is usually allocated more bits. In the view of rate distortion, the coding efficiency of the LTR will loss performance, but the LTR can be utilized as reference frame for following several frames, then the coding efficiency of the following frames will improve. One frame's performance loss brings performan gain in following several frames, then the overall performance will improve. However, the LTR interval and bit allocation has not been resolved. Based on the motion compensation analysis of information theory, the paper give a analysis of the LTR interval and bit allocation, and then LTR jump period and bit allocation method are obtained, the research results are very significant to the study of DFMC.2) When transmitted in wireless channel or network, video program usually loss packet. In response to this situation, an error resilient DFMC is proposed. In recent study of DFMC, some existing error resilient methods were introduced into DFMC. But for the DFMC coding structure itself, how to enhance error resilence has not been resolved. The paper firstly gives a robust DFMC scheme, the LTR interval and bit allocation can be adaptively adjusted to enhance error resilience. Furthermore, an error resilient DFMC scheme is given, which includes three aspects: firstly, error resilient prediction structure is proposed, the reference frame is adaptively selected in different packet loss rate to enhance error resilence. Secondly, the macroblock data is separated into header information and texture information, and then a new end-to-end distortion model is proposed. Thirdly, as the reference frame of many frame, the LTR is more important than the other frames, in the case, a frame level rate distortion scheme is proposed to give more protection to LTR. The proposed error resilient DFMC scheme has solved the error resilience problem , and makes the DFMC video program more robust to error-prone channel.3) Rate control in DFMC. Rate control can control the bit stream and improve video quality in video coding. The existing rate control schemes are based on that every frame has similar quality, and the current fame has the same distance to reference frame. Because of the coding structure of DFMC (different distance from each frame to reference frame, and difference quality in different frame), the existing rate control scheme is not adapt to DFMC. At present, there is no rate control scheme which can be utilized for DFMC. In the paper, rate control for DFMC is resolved in three aspects: firstly, bit allocation scheme, the HQF is allocated more bits, the LQF is allocated bits with linear relationship to obtain similar quality. Secnodly, because of different quality in reference frame and different distance to reference frame, different rate-Qstep model are proposed for different frames. Finally, adaptive MAD and overhead prediction method are proposed for different frames. The proposed rate control scheme can accurately control bit rate, and make the DFMC go from theory deduction to practical application.4) In multiview video coding, different video sequences of the same object are generated simutaneously, so high coding efficiency is very important. In temporal view, the relevance is important, the coding efficiency can be further improved. In inter-view, many camera capture the same object, lots of related information is existed. In the paper, JU-DFMC are introduced into multiview coding, the coding efficience are improved from three aspects: firstly, in temporal view, enhanced dual frame motion compensation scheme is proposed; secondly, enhanced inter-view prediction is proposed, the HQF is utilized as reference frame for the frames in the other views; finally, bit allocation scheme in multiview is proposed. The performance of the proposed JU-DFMC based multiview scheme is better than the existing multiview scheme, which make the application of dual frame motion compensation go from single view to multiview.