Efficient rate-distortion optimization techniques for H.264

Abstract

This thesis addresses the rate-distortion optimization (RDO) techniques for H.264. The research of the current project aims at improving the coding efficiency of video sequences. It proposes several efficient RDO techniques for H.264. These techniques include rate control, perceptual bit allocation and fast mode decision. To solve the rate control problem in H.264, we propose a motion complexity measure and a minimum distortion based mode decision process. The motion complexity measure can represent the complexity of a video frame’s motion contents effectively and is used to estimate the frame’s target bit in frame level bit allocation. Different from other existing H.264 rate control methods, quantization parameter (QP) of a macroblock is accurately estimated based on the minimum distortion after full-search of all coding modes. According to the motion complexity measure and minimum distortion mode decision process, an efficient rate control scheme is proposed. To solve the perceptual bit allocation problem in H.264, we propose a Lagrangian multiplier update algorithm in mode decision. The Lagrangian multiplier update algorithm decides the macroblock coding mode according to the perceptual characteristics of video contents, so that more bits are allocated to pixels which are perceptually more important and fewer bits are allocated to pixels which are perceptually less important.