Research On Rate Control And Scalable Coding For Video Communication

With the development of video coding and network technology, the demand for video communication is increasing. The target of video coding has been changed from traditional storage application to transmission. More and more attention has been paid to the technology of video coding for network transmission by both academia and industry. This dissertation introduces a comprehensive research on several key technologies for the video coding in video communication, including rate control, scalable video coding and optimization for DSP-based implementation.1) Rate control algorithmRate control is an important part in video coding and transmission. For constant bit rate (CBR) rate control, a robust and adaptive rate control algorithm is proposed for real-time video communication. In this algorithm, a joint PID-based buffer controller is utilized to maintain the buffer stability and adjust the target bits based on the frame complexity, and then an improved quadratic R-D model and update scheme are applied to achieve a stable buffer and to decrease the PSNR fluctuation. A Cauchy-distribution-based rate-distortion model is drived by the rate distortion analysis for the DCT coefficients in H.264. Based on this new model, two rate control algorithms are proposed to achieve an optimum bit allocation and accurate rate control, through Lagrange optimization and twice RDO coding method, respectively. Comparative experiments and discussion are also presented.For variable bit rate (VBR) rate control, an efficient joint source-channel rate control algorithm is proposed for real-time wireless video transmission with automatic repeat request (ARQ) scheme. In this algorithm, the Markov model is used to simulate the wireless channel and the Cauchy-distribution-based rate and distortion model is used as the source model. Then, joint source-channel rate distortion optimization is applied. Experimental results show that this VBR rate control algorithm can decrease the number o f skipped frames and improve the average PSNR.2) Scalable video codingScalable video coding (SVC) is an important way to solve the heterogeneity of network and diversity of user requirements. This paper investigates the key technology of H.264 SVC model and proposes an optimization scheme to implement the encoder based on scalable video coding of motional region of interest (MROI). In this scheme, the information of motion vector and coding mode in base layer are used to pick up the MROI, then this region is coded as an independent slice in the enhance layer to achieve spatial, temporal, and quality scalability. This scheme can decrease the complexity and improve the image quality.Moreover, for context-based SVC, a fast and efficient moving object segmentation algorithm in the H.264 compressed domain is proposed. The raw motion information in the H.264 compressed domaion is processed by normalization, spatio-temporal processing and global motion compensation to obtain the reliable and salient motion information of the moving object. Then, the mean shift-based clustering segmentation algorithm is applied to abstract the moving object. Based on the mask of the moving object, we use prioritized block coding to achieve moving object-based SNR scalable video coding. Experimental results show that the proposed segmentation algorithm can abstract the moving object accurately at real-time for the video with still or moving background. The SNR SVC based on the moving object mask can improve the PSNR performance of the object and subject quality of the whole image.3) Optimization and implementation of video coding for DSPIt is well know that the volume of video data is large and the video coding algorithm is of high complexity and requires real-time processing. For the video encoder implementation based on DSP, two kinds of optimization strategies are proposed, including algorithm level optimization and architecture level optimization. For the implementation of MPEG-4 encoder based on C64x DSP, an algorithm optimization mothed based on adaptive stop technology is proposed to decrease the computation complexity in motion estimaton and texture coding. Meanwhile, several architecture optimization methods are proposed to enhance the memory access and software executive efficiency. The implemented encoder is fast enough to encode four channel CIF video at real-time. At last, a wireless video communication system is designed and implemented based on DM642.