Optimization Of AVS Video Encoder And Research On Algorithms

AVS is China's second generation source coding technology standard, which is developed with independent intellectual property rights. It has advantages of high-performance, low computational complexity, and low licensing cost, etc. Its implementation not only indicates that China is presently at the leading position in the fields of multimedia research in the world, but also will create considerable economic and social benefits. However, the AVS codec chip and software are not yet mature, which restrict the promotion of the industrialization of AVS. Therefore, research on the AVS codec is of vital significance.This dissertation aims at AVS-P2 video encoder optimization and fast algorithm development for the core module, in order to improve encoding speed. It mainly provides a solid foundation for futher applications which demand real-time encoding, such as DSP transplanting for surveillance, etc.The main work of this dissertation includes:It investages characteristics of some commonly used video compession technologies, such as entroy coding, predictive coding and transform coding. Historical development and characteristics of the prevalent international video compression standards, including the MPEG series and H.264 series, are introduced. Then, we carry out comparison analysis between AVS and other standerds to highlight its superiority over H.264, MPEG-2 and MPEG-4. After in-deep study of the AVS system architecture, the AVS are reviewed, including the integer transform and quantization, intra prediction, inter prediction, 1/4 pixel interpolation, entropy coding and the loop-filtering, which lay the foundation for AVS encoder optimization.We optimize the AVS video encoder. First, the overall encoder structure is adjusted to remove redundancy, so that the each module is distinct and convenient for futher independent study. Then, we carry out various optimization strategies to do code optimization for computing power enhancement. Module optimization is mainly for interpolation, loop-filtering, intra prediction and entrpy coding modules. Interpolation is optimized through interpolation function calling, separate processing of the interpolated image borders and code optimization. Loop-filtering optimization is implemented by acquiring boundary strength. Experimental results show that the optimized AVS encoder reduces the encoding time by about 27%-34%. Based on study of the AVS encoder algorithms, two fast motion estimation algorithems are proposed. In the first algorithem, the even subsampling method is adopted for computation of SAD to reduce the computational complexity. The second algorithem uses all-zero block to skip much computing and to exit in advance to save coding time. The ultimate aim is to combine these two methods. Experimental results show that better results are obtained by combining of these two methods than any single method. Then, based on substantial research on motion search algorithms, a new method is put forward in which double sampling and diamond-rectangle search are combined together. Experimental results prove that these algorithms can reduce encoding time by 5% -22% under guaranteed picture quality. After research on AVS loop filter, a fast AVS loop-filtering algorithm at pixel level which takes advantage of the correlation between pictures is proposed. Experimental results show that, the proposed method can efficiently reduce the deblocking time up to 40%-60% and raise the coding speed while maintaining almost the same subjective and objective quality.