Research And Improvement On H.264/AVC-based Video Coding

The traditional video coding techniques mainly utilize reducing temporal and spatialredundant information to raise the compression ratio, these researches are verysophisticated and have achieved high compression at present, so it’s difficult to improveany more. Therefore, how to save bitrate without influencing the quality, and how toachieve better subjective visual quality at low bitrate, these two topics have been hot inrecent years. In this dissertation, on the basis of hybrid coding scheme,we address theabove-mentioned problems from the angle of the HVS and Rate Control scheme.In the respect of HVS, we intensively research the human visual model in Chapter2:JND model and ROI model.We describe the implementation of the pixel domain JNDand the DCT domain JND. Moreover, we also realize the transformation of JND from theDCT domain to the Integer Transform domain to match the hybrid coding scheme. Basedon Chapter2,we propose two new video coding models based on ROI and JND in Chapter3: Residual Filter Model and Graded Quantization Model. The Residual Filter Modelbased on ROI and JND, combines the ROI, consider of the Integer Transform domainJND, utilize the pixel domain and transform domain JND at the same time to adjust theresidue adaptively. Compared with other literatures, our Residual Filter Model caneffectively retain the important parts, can save5%-28%bitrate while keeping thesubjective quality about the same. The Graded Quantization Model based on ROI andJND, this model divides frame into two grades: ROI and non-ROI, after that we separatethe ROI area into two grades, then we assign different QP to every grade. Compared withother schemes, our Graded Quantization Model can accord with the human visual systembetter and get an impressively subjective quality.In the respect of the Rate Control algorithm, a new CBR Rate Control Scheme Basedon a circular Control Strategy is proposed to solve the unsatisfactory Quality problem inthe first several frames of X264.We compute the CPLX of the k-th frame of the video asthe input of the first frame,this can effectively solve the shortage of CPLX lackexperience.Results from experiments show that our scheme can greatly improve thecoding quality of the whole video, and is also competitive in objective quality such asPSNR and SSIM.