Optimization For AVS Video Encoder And Research On Mode Decision

Video codec has become one of the hot issues of the research and industrial applications for saving the network bandwidth and storage space. AVS is a video/audio encoding and decoding standard developed by China independently, and is one of the most advanced video compression standards available. Compared with other video standards, it has the advantages of good performance, low computational complexity, and low patent licensing fees.etc. Although the reference software for enterprise has been currently provided by AVS standard group, its encoder suffers from poor efficiency, thus is not suitable for practical applications. Therefore, from the perspective of the practical application, this paper firstly carries out AVS video encoder optimization in aspects of structural adjustment, source code optimization and optimized implementation of core modules, then improves the algorithms of mode decision, which is extremely time-consuming, according to the characteristics of AVS video coding. These lead to obvious raise in encoding efficiency.Firstly, a brief introduction of the prevalent international video coding standards is given, together with the comparison between AVS and H.264/AVC. Then we review the AVS coding framework, main encoding process and its coding technology, which includes intra/inter prediction, sub-pixel interpolation, transform/quantization, de-blocking and entropy coding,etc.Secondly, we optimize the reference software rm52j, provided by AVS standard group, with structural adjustment, source code optimization and optimization implementation of core modules. Structural adjustment makes the main encoding process more legible while code and modules optimizations improve the performance of program and raise the coding efficiency. The simulation results show that coding time reduction of the encoder is up to 26%～35% for various videoes in different bit rates with these optimizations.Thirdly, to solve the problem that rate distortion optimization mode decision in AVS is severely time-consuming, two novel fast algorithms for inter-mode decision of AVS is proposed by referencing some classical fast methods. The first fast algorithm which is based on early detection of Skip macroblock and texture characteristic of macroblock can drastically improve the AVS coding efficiency, that speed improvement can achieve above 60% for video with fixed background. Consequently, this method applies well to video applications which have fixed background and request real-time coding such as video conference and so on. The second fast algorithm utilizes the correlation among macroblocks and adopts optional 8×8 and intra-mode prediction. As all of the judgement criterions are based on the characteristic of the video, this algorithm can save the coding time obviously while maintaining the coding performance for videoes with different motion and texture characteristics. Algorithm selection can be conducted according to the specific requirements for practical applications.Finally, this paper proposed an efficient fast algorithm for intra-mode decision selection based on analysis of existing intra prediction technologies and intra-mode decision of AVS. In this algorithm, correlation between the best modes of adjacent blocks and the directivity of luminance variations within a block are adopted to reduce the candidate modes to be tested. Experimental results show that the coding time of the proposed algorithm is decreased by 17%-29%, with negligible loss of PSNR and increment of bit rate.