Research On Power Rate Distortion Model For Video Coding And Its Low Complexity Algorithm

The rate distortion (R-D) theory proposed by Shannon in 1948 has played an important role in advanced communication especially in wireless multimedia communication. However, the R-D theory mainly research on the R & D under the rate constraint. With the rapid development of wireless communication, the enhancement of the wireless channel bandwidth, and the popularity of mobile devices with multimedia data processing ability, the constraints on the rate become weaker and weaker. At the meantime, the processing capabilities insufficiency of mobile device and the microprocessor's power-constraint problem caused by battery become the major restriction to the development of wireless multimedia applications. Sponsored by the National Natural Science Foundation of China (Grant No.60572063, No.60372067) and Doctoral Foundation of Ministry of Education in China (Grant No.20040487009), this thesis focuses on analyzing the power consumption behaviors of the typical video encoding system, extending power consumption as another dimension to the traditional rate-distortion analysis of video encoding system and developing a parametric power-rate-distortion analysis model. The work aims to provide a theoretic guideline for system design and performance optimization in mobile video communication under energy constraints, and achieve the best-effort visual quality when the power level is fluctuant or limited.In general, the research of this thesis can be summarized as follows:Firstly, based on the dynamic voltage scaling, an energy consumption management technology recently developed in CMOS circuits design. We clarify that the energy consumption scalability is equivalent to the complexity scalability of the video encoder. The time complexity of an algorithm is defined to represent the actual power consumption. After analyzing the main encoding modules of a typical video encoding system, we model the power consumption behaviors of these encoding modules and develop a parametric power consumption model on a typical video encoding system.Secondly, based on the fact that the intra rate also has important effect on the bit rate and distortion, we extend the parameter set of the classical R-D model which is the function of quantization parameter. and then revise it. After that we set up the R-D model where the bit rate and the distortion are the functions of the quantization parameter and the intra rate, respectively. It can be shown that the classical R-D model is one of the special case of the proposed model when given the intra rate.Thirdly, after analyzing the traditional rate-distortion characteristics of video encoding system, the proposed power consumption model is extended to the R-D model and then a power-rate-distortion (P-R-D) analysis framework is established, which chooses the bit rate and INTRA rate chosen as general control parameters. The P-R-D model reveals the intrinsic relationgship among the distortion, tbit rate and power consumption of the video encoding and provides a theoretic guideline for system design and performance optimization in mobile video communication under energy constraints adaptive to the different requirements in practice.Fourthly, variable block sizes motion estimation results in extremely high computation complexity, which prevents the H.264/AVC from practical use. Based on the Lagrangian R-D optimization, an efficient mode selection algorithm is proposed to reduce the complexity of H.264 encoder. The main idea is to classify the video into different types of regions with the motion vectors information and the coded modes information. After that, we process the different regions distinctly based on the Lagrangian R-D optimization cost. Experimental results show that the algorithm can save the encoding time up to 51% on average compared to the conventional method in the JVT JM8.6 reference encoder at the cost of negligible performance degradation and the bit rate augment.Fifthly, channel transmitted distortion is one of the most important components of the end-to-end distortion. The general method researching on the channel transmitted distortion is to fix error bit rate (BER) of the channel. However, such method can't characterize the time-vary channel well. In this part, the performance of Turbo codes under the influence of channel codec parameters and channel statistic is investigated firstly, and then a BER model over Rayleigh channel is deduced using mathematical tools based on the simulation results. By using the BER model, the channel BER can be predicted accurately in the encoder when given the coding parameters and the wireless channel characteristics. Therefore, the proposed BER model can be used in joint source-channel coding, end-to-end R-D optimization and distortion prediction.