基於CABAC熵編碼之H.264/AVC改良式位元率量化模型

Abstract

H.264/AVC標準中，位元率-失真度最佳化(RDO)是個能提升編碼效能的技術，但使用此程序也會讓編碼器的複雜度大幅上升。為了降低計算量的成本，我們需要ㄧ個快速的位元率估計方法以避免模式決策(mode decision)程序裡熵編碼的大量計算。本篇論文中，我們嘗試利用從部分RDO程序擷取之特製殘值(residual)的編碼行為，進而估計在不同量化參數(QP)下RDO的輸出位元率。在所需的特徵值被專屬位元率量化模型抽取後，我們再藉由分析基於上下文的自適應二進制算術編碼(CABAC)之語法元素特性，以建構此位元率估計系統。實驗結果顯示，此篇論文提出的架構能夠比現存的其他方法提供更正確且有效率的實際輸出位元率估計。

Rate-distortion optimization (RDO) is an efficient technique to improve coding performance in H.264/AVC, but the process also incurs a considerable complexity increase of the encoder. To reduce the computational cost, a fast rate estimation method is needed to avoid heavy computations of entropy coding during mode decision process. In this work, we attempt to adopt the coding behaviors of customized residuals that are extracted from part of RDO, to further estimate the output bit-rate of RDO with different quantization parameters (QPs). After the features are extracted by our dedicated rate-quantization (R-Q) model, we then analyze the characteristics of the syntax elements of context-adaptive binary arithmetic coding (CABAC) to construct our rate estimation scheme. Experiments show that the proposed method is able to efficiently provide a more accurate estimation of the actual output bit-rate than existing works.