A fast rate-distortion optimization algorithm for motion-compensated video coding

Abstract

A fast optimization algorithm that jointly selects motion vectors (MVs) and quantization levels based on a rate-distortion (R-D) criterion is developed for a motion-compensated video coding system. In contrast to an optimal R-D algorithm, which requires performing DCT and scalar quantization operations, which can be computationally excessive for practical applications, the proposed algorithm achieves a significant complexity reduction by efficiently pruning MV candidates and modeling residual bit-rate and overall distortion with simple polynomial functions of the quantization level for a given motion vector. Experimental results indicate that our fast algorithm maintains compression performance close to that of the optimal algorithm while keeping computational complexity close to that of conventional heuristic approaches.