Research On Optimization For HEVC Intra Prediction

The High Efficiency Video Coding(HEVC)is the next-generation video coding standard.Compared with the previous-generation standard H.264/AVC,efficiency of the HEVC is approximately doubled under the same situation.But the application and promotion of the HEVC have been obstructed for the reason of the sharp increasement in computational complexity.Based on the deep research for HEVC intra coding,this dissertation focuses on two processes that significantly affects the complexity of intra coding: prediction mode selection and coding block size selection.A fast algorithm for intra prediction mode selection based on mode grouping is proposed.At the same time,another fast algorithm for coding block size selection based on multiple textures is slao proposed.The specific work is as follows:1.The rough mode decision(RMD)is used to select modes from 35 prediction modes based on Hadamard transform-based costs.By such pre-selection,the number of modes tried in the rate-distortion optimization process are reduced,but the complexity is still high.In this paper,a fast algorithm for intra prediction mode selection based on mode grouping is proposed to reduce the number of modes entering RMD.By introducing the gradient information of prediction unit(PU)in four directions,horizon,verticality,45° and 135° respectively,the intra prediction mode for PU can be selected efficienty.2.The coding unit(CU)size is decided by searching quad-tree recursive partitioning of largest CU in rate-distortion optimization process,resulting in very high computational complexity.Based on the statistical analysis of the relationship between CU texture features and CU size,this paper extracts the teatures of content complexity and direction complexity of the CU,the difference in content complexity and the difference in direction complexity of the sub-CUs.A CU size decision mechanism using two support vector machines(SVM)is proposed,which use above mentioned features as the input parameters.The decision mechanism can acknowledge the split decision of CUs with complex texture and the split termination of CUs with simple texture in advance,which fasts the intra CU size selection obviously.In this paper,the proposed algorithms are tested on the standard test set published by JCT-VC.The test results show that by employing the two fast algorithms in intra encoding of open source encoder x265-1.7,it can save 42.80% encoding time,with 0.98% increasement in bitrate and 0.018 d B loss of peak signal to noise ration of luminance.While ensuring the encoding quality,the computational complexity is effectively reduced.