Research On High Efficiency Algorithms Of Reducing Complexity Of HEVC

Nowadays, videos with high definition and ultra-high definition are becomingmore and more popular, which means the amount of video data to be dealed with arebecoming huger. Traditional video coding standards like H.264/AVC do not performvery well when compress these videos. To ease the heaven burden on networksimposed by growing data rates, HEVC is standardized by JCT-VC (JointCollaborative Team on Video Coding) working group and it is the most efficient andlatest video codec standard.HEVC adopted many new technologies in order to improve coding efficiency.These new technologies include flexible data representation and numerous intra andinter prediction modes, which improved coding efficiency greatly. But, these newadded technologies increase the computational complexity of encoder vastly. Thiscould hinder its practical application. For this reason, research on fast and efficientand low complexity coding algorithms of HEVC is crucial.This paper is focus on fast algorithms of getting optimal partition mode of CU(Coding Unit), PU (Prediction Unit), TU (Transform Unit) of HEVC. Three fastalgorithms were made to obtain optimal partition modes of these units by taking fulluse of spatio-temporal correlation and feature of coding tools, after analyzing theRDO process of getting the best mode of CU, PU and TU. These algorithms increasethe coding efficiency greatly, while guarantee the coding quality.1) RDO process of CTU was analyzed firstly. Rate-distortion cost of all thedifferent depths which from depth0to depth3are compared transversally to get theoptimal CU. Thus, a rapid classification decision on CU depth based ontemporal-spatial correlation is proposed in this paper to improve this problem. Firstof all, correlation weights of temporal-spatial adjacent CTUs/CUs are analyzed.Then, the possible depth ranges and depth values of current CTU/CU can bepredicted in advance via depths of these coded CTUs/CUs. As a result, unnecessaryCU size decisions can be skipped and terminated. According to the simulation results, conclusions are obtained that encoding complexity on average can bereduced by42%under condition of different encoding configurations, whilemaintaining almost the same RD performance as the original HEVC encoder.2) HEVC adopted many prediction modes in order to obtain highercompression efficiency, while increase the computational complexity of the encoder.Thus, a fast decision algorithm of PU based on classification of texture and motioncomplexity is proposed. There is strong correlation among prediction modeinformation of temporal-spatial adjacent area. The information can be used todetermine degree of comprehensive complexity of current CU. Prediction modeswere classified firstly. Then, temporal-spatial adjacent prediction modes werecorrelated by adaptive weights, and get comprehensive prediction complexity value,which can determine degree of comprehensive complexity of current CU andprediction mode need to be calculated. Finally, temporal correlation of skip mode isused to make further reduction. Simulation shows that encoding complexity onaverage can be reduced by45%under condition of different encoding configurations,while maintaining almost the same RD performance as the original HEVC encoder.3) HEVC uses RQT (Residual Quad-tree Transform) in transform coding. Largenumber of calculation is conducted to get optimal TU mode in each CU. Thus, earlytermination of TU based on all-zero blocks was proposed. According to theory, aftertransformation and quantization of residual matrix, if all the absolute values ofcoefficient are less than1, current coefficient matrix is considered as all-zero block.After derivation, the threshold of early termination can be obtained. Simulationshows that encoding complexity on average can be reduced by13%under conditionof different encoding configurations, while maintaining almost the same RDperformance as the original HEVC encoder.