Research on Computational Complexity Optimization for Intra and Inter Prediction in Video Compression

Due to the popularization of digital image capturing devices, video content is growing in volume every year. To compress the huge amount of data to much smaller capacity so that it can be manipulated with low cost for video conference, broadcasting, online video-on-demand and surveillance system, video coding technology has been developed since 1989. H.264/AVC is the latest international standard which provides high compression efficiency while introducing heavy computational load and high power consumption. As video content is also growing in resolution and 4k2k content becomes the mainstream, since 2010, a successor to H.264/AVC called High Efficiency Video Coding (HEVC) for next generation video coding is under standardization. HEVC can provide doubled compression efficiency compared to H.264/AVC, while tremendous computational cost is required which leads to longer encoding time and higher power consumption. In both standards, the most effective while computationally intensive tool is Rate Distortion Optimization (RDO) based hybrid encoding, which reduces spatial and temporal redundancy by brutally trying a wide range of block types and prediction modes. In hybrid encoding, intra prediction reduces data redundancy in neighboring blocks, it is not that complexity consuming but affects throughput due to high data dependency; inter prediction estimates motions by finding a matching block in other several frames, consequently, large memory as well as high power consumption will be consumed. The problem becomes even critical in HEVC in that quadtree-structured coding block candidates for hybrid encoding have been increased by a factor of 2. Since II video-related entertainment applications require real-time transmission and portable devices suffer from short battery life and limited hardware resources, it is desirable and urgent to develop optimization techniques for video encoders. The target of this dissertation is to reduce the computational complexity in video encoder by half while keeping the compression performance in terms of both video quality and encoded bits, so that video data can be compressed efficiently within halt time or under lower power consumption. Firstly, a mode decision scheme is proposed and implemented in to H.264/AVC, in which both prediction modes in intra coding and block types in hybrid prediction is reduced largely by means of texture analysis. Proposed texture analysis achieves better accuracy than previous works and the proposed mode decision scheme is proved to be capable of reducing computational complexity of H.264/AVC encoder by 52.5% averagely, which outperforms Liu's method [IEEE Trans. of CSVT 2009] Secondly, a content adaptive hierarchical decision …