A Novel Rate-Quantization Source Modeling Framework for H.264/AVC

Abstract

How to perform rate control on emerging H.264/AVC, which includes the non-normative but widely accepted rate-distortion optimization (RDO) process, i.e., the rate-constrained motion estimation and mode decision, has become a challenging and difficult research issue. Usually, rate controller is useful at given residual signals to determine quantization parameter (QP). However, to perform RDO, a QP should first be predetermined. This leads to a classical chicken and egg dilemma. Therefore, how to understand and predict the coding behavior after RDO with different QPs is the core problem of the H.264/AVC rate controller. In this work, we attempt to adopt coding behaviors of customized residues that are extracted from part of RDO to forward estimate the coding behavior of RDO with different QPs. Experiments show that the proposed rate-quantization (R-Q) model is able to provide a more accurate estimation of the actual bit rate than existing MAD-based model in JM 10.2.