Advances on Video Coding Algorithms for next Generation Mobile Applications

Visual communication via mobile devices is becoming more and more popular. However, current video coding standards are not initially designed for this type of application area. Traditionally, high compression efficiency is usually achieved at the expense of increasing computations. The newest standard, H.264/AVC significantly outperforms the others in terms of coding efficiency, however, the complexity is greatly increased. This sets challenges on implementation, especially on mobile devices, which typically possess constrained computational resources. Thus, there is great significance and interest in reducing the computations for video coding. Currently, the emerging standard, i.e., High Efficiency Video Coding (HEVC), claims an aggressive goal to realize high efficiency video coding with low complexity. This thesis deals with the problems of advancing video coding algorithms for real-time high performance, focusing on applications for the next generation mobile devices. The goal of the research is to develop new techniques in order to reduce the coding complexity and improve the real-time performance while maintaining competitive video quality. The research work presented in this thesis can be categorized into two areas. First, different techniques are provided for low complexity video encoding and/or decoding for standardized video codec, e.g., H.264/AVC and MPEG-4 Visual. The efforts particularly focus on the prediction of zeroquantized DCT coefficients which can be skipped for complexity reduction. Second, new algorithms and methods are applied to 3-D DCT video coding for improvements of both the coding efficiency and complexity reduction. Compared to existing techniques, the proposed algorithms in this thesis extend the prediction from 2-D DCT to 1-D DCT on the row and column directions. Thus, they are particularly efficient for partially zero-quantized DCT blocks. In addition, the proposed algorithms lend themselves the butter-fly implementation structure which is commonly used in video coding. The experiments show that the proposed method outperforms the earlier methods in terms of complexity reduction. As good supplement to the effort on inter transformation, complexity reduction for intra 2-D DCT computations has also been intensively researched in this thesis. In addition, variable size 3-D DCT video coding methods are proposed in the thesis. Compared to competing techniques, the proposed approaches improve the 3-D DCT video coding scheme in both the complexity reduction and coding efficiency. In addition, the evaluation criterion on this topic is further improved by lending the algorithms to the implementation structure. For instance, since the DCT coefficients are not individually computed, the discussion has been focused on the complexity reduction in terms of 1-D transforms and arithmetic operation numbers instead of individual coefficients. Furthermore, PSNR and bitrate have been separately compared in the related references and thus “negligible video quality degradation” is drawn as conclusion. However, rate-distortion comparisons in the thesis show a comparable overall video quality to the original codec, which means the proposed models do not necessarily result in any additional visual distortion.