| H.264/AVC video coding standard is provided formally by JVT (Joint Video Team) in March 2003, and it will be the second successful video compression standard after the MPEG-2/H.262 in future. The standard is designed to give a new technical solution for a broad range of applications, such as very low bit rate conversational services and entertainment quality broadcast, interactive video-on-demand services etc. For the enhancement of the coding efficiencies, the new standard adopts new tools as followed: multiple reference picture, variable block-size with seven block sizes in motion prediction, quarter-pixel accuracy for motion vector, short word-length integer transform, context-adaptive entropy coding and loop filter deblocking.At the same time, the complexity of the H.264/AVC codec is several times higher than that of existing standards (MPEG-2 and H.263). Accordingly, the software-based real-time video codec demands higher-powered processor and faster-speed algorithms. So the main endeavor of this paper is to achieve maximal processing speed and better video compression quality through optimization for motion estimation, integer transformation in the H.264/AVC encoding process, and then a software-based real-time H.264/AVC video decoder is realized on the personal computer.Firstly, as the most important module in the H.264/AVC encoder, motion estimation is accelerated in four perspectives. (1) With the initialized searching point prediction, adaptive search window changing and half-stop principle, two fast algorithms, multiple level successive elimination algorithm and enhanced diamond search are provided to reduce computation in block-matching process. (2) Through the analysis of mathematical model of the Minimum Mean Square Error, a parabola is constructed to predict the search mode for 1/2 pixel accuracy motion vector. A further optimization for the 1/4 pixel accuracy is obtained by the related location of critical positions from 1/2-pixel results. (3) Based on analytic results of final selected mode, SAD (Sum of Absolute Difference) and motion vector were used as thresholds to judge whether other coding block modes should be chosen. SAD and motion vector were the middle results in motion estimation and didn't need more computation further. However, the redundant computation is deleted in the unnecessary searching process. (4) Furthermore, optimal multiple frames predictive scheme is given through analyzing the relation between the original sum of absolute differences (OSAD) and minimum sum of absolute differences (MS AD), both of them are the middle results in the searching process too.Secondly, for most sequences at middle or low bit rates, the values of motion prediction residuals usually are very little, and most of them will be reduced to zero after transformation and quantization. In order to reduce the complexity of the encoder, a novel early detection method for all-zero coefficients block for H.264/AVC is proposed in this paper. The SAD is used as the criterion, which is the middle result from motion estimation, so no additional computation is needed. The all-zero block detection threshold is very critical to image quality and encoder complexity, so a feasible range is firstly analyzed theoretically, and then based on extensive experiments on classical video sequences, the best threshold is statistically presented in this paper with a great balance between mis-justice and leak-out ratio. So all-zero blocks can be detected correctly in most of sequences, and the computational complexity is reduced greatly in the encoder. In addition, the degradation of decoding image quality is almost imperceptible.Lastly, a software-based real-time H.264/AVC is realized on personal computer by optimization for algorithms, program code together with processor proceeding. Context-adaptive variable length coding (CAVLC) is chosen as entropy coding method in the baseline profile and main profile. According to the different coefficients, CAVLC defines multiple code tables and chooses the different tables dynamically for co...
|
PDF Full Text Request