| Summary: | 博士 === 國立成功大學 === 電機工程學系碩博士班 === 96 === In this dissertation, we propose several coding algorithms to improve the coding efficiency for intra-prediction coding scheme defined in H.264/AVC. First, a fast mode decision method for intra-prediction is proposed to reduce the computational complexity of H.264/AVC encoders. With edge information, we propose a novel fast estimation algorithm for mode selection, where the edge direction of each coding block is detected from only part of the transformed coefficients. Hence, the computation complexity is greatly reduced. Experimental results show that the proposed fast mode decision method can efficiently reduce the encoding time for all test sequences with acceptable degradation of PSNR and bitrates.
Secondly, instead of traditional zig-zag scan when performing CAVLC in H.264, an adaptive scan method is proposed, which uses the best fitted scan order for each intra-prediction mode. In H.264/AVC, the residuals after various intra-prediction modes have different statistical characteristics. After analyses of the transformed residuals, we derive an adaptive scan order scheme based on Markov model, which could match up with the intra-prediction mode, to further improve the efficiency of intra coding. Simulation results show that the proposed adaptive scan scheme can aid the context-adaptive variable length coding (CAVLC) to achieve better rate-distortion performances in H.264/AVC video coders. When the zig-zag scan order is replaced with the derived adaptive scan order, the coding performance can be improved with PSNR increased and no extra bits required.
Thirdly, for lossless intra coding included in H.264/AVC FRExt, the so-called fidelity range extension, we propose an adaptive scan order method to improve the coding efficiency. In image/video coding standards, the zig-zag scan, which transfers the coefficients of a two-dimensional transformed matrix from low-frequency to high-frequency components, successfully help to achieve an effective entropy encoder. Generally, the optimal scan of two-dimensional residual samples should be scanned in the descending order of their statistically-averaged power levels such that we can achieve more efficient variable length coding. For H.264/AVC with sample-wise DPCM lossless intra coding, which is included in the FRExt, the so-called fidelity range extension, the residuals after different directional prediction modes should have different statistical characteristics. After analyzing the power of the residuals, an adaptive scan order method to optimally match up with each intra-prediction mode for improving coding efficiency is proposed.
Finally, an efficient shape coding scheme, called block-conditioned context-based arithmetic-encoding (BCAE) is proposed to improve the coding efficiency of video object shape coding. By performing simple block detection, we first setup an efficient conditional source model in a block-by-block fashion and then derive the probability modes based on context-based arithmetic-encoding (CAE), by training various test video sequences. CAE is an efficiency coding scheme defined in MPEG-4 shape coding, and could be exploited by H.264/AVC for the applications of video object coding. With re-trained probability tables derived by our proposal, the coding efficiency of the proposed BCAE algorithm is better than that of the standardized CAE.
|
|---|
