Research On Fast Intra Prediction Algorithms For H.264/AVC

Abstract:Video has been widely utilized in many fields, such as broadcasting, online video, video conferencing, etc. As uncompressed video occupies such a huge space, video coding is of vital importance for its storage and transmission. And H.264Recommendation is the most popular industry standard now. As optimal coding mode is determined by rate distortion optimization model through exhaustive search over possible modes, fast intra prediction algorithms are useful and necessary for practical encoder implementation.In this work, we propose the RR (Reliability Refinement) to improve the overall coding efficiency. It consists of reliability detection and essential modes refinement, and has been implemented on top of the two typical fast algorithms, i.e., Pan’s algorithm and HIMD (Hierarchical Intra Mode Decision). Extensive simulation results demonstrate that, with slight sacrifice of encoding speed, the proposed method could greatly improve the video quality with about50%decrease of BD-Rate on average.A two-stage block size decision scheme for intra prediction is proposed in this paper. The SAAPE (Sum of Absolute Adjacent Pixel Error) is proposed to classify the coding block into texture or homogeneous groups. Extensive experimental results have shown that proposed algorithm leads to over45%reduction in encoding time for H.264, with negligible degradation.