Fast Intra Coding Algorithm For HEVC

In the past few decades,International Telegraph Union(ITU-T)and International Standard Organization(ISO)have formulated a series of video coding standard.The MPEG series and H.26x series were widely applied and researched.However,existing H.264/AVC or MPEG-4 standard has been unable to meet the demand for high video compression rate with the development and popularization of high-definition or super high-definition video.In recent years,therefore,the Moving Picture Experts Group(MPEG)of ISO and Video Coding Experts Group(VCEG)of ITU-T jointly founded the Joint Collaborative Team on Video Coding(JVT-VC)for next generation video coding standard,which was the High Efficient Video Coding(HEVC).The compression performance of HEVC is twice of H.264/AVC.With the improving of compression performance,computation complexity of HEVC has been increased enormously.The variety of structure of block partition was increased in HEVC,including symmetric and asymmetric structure.Arborescence partition model was adopted as well as,and the block size was expanded to 64×64 to improve the flexibility.There was more prediction modes than H.264/AVC in HEVC,made it more accurate for intra or inter prediction.So the whole process will be extremely complex,made the implementation difficulty far higher.This paper focus on the optimization for intra coding process.We analyzed the complexity of CU partition and PU prediction to make improvements.The main contents of this paper are as follows:1)This paper uses the decision tree models combined the rate and distortion component of R-D cost to decide the CU partition model.The original HEVC CU partition for intra coding is based on quadtree,Most of the traditional methods of improvement methods are based on some features and use thresholding method to decide CU partition mode.Results show that CU partition based on decision tree can reduce about 35%of the time,the output bitrate increase by about 1.3%on average,Peak Signal to Noise Rate(PSNR)was down no more than 0.1 dB averagely compared to original HEVC model.2)Then we use the structure correlation among current PU and other PUs to reduce the times of R-D optimization at some depth in this paper.The original HEVC needs to do R-D Optimization many times in PU prediction process.Most of the traditional methods use the texture feature to analysis possible prediction mode of current PU for narrowing the scope of PU prediction modes selection to reduce the complexity of the R-D Optimization.Results show that it can reduce about 27 of the time with the bitrate increased by less than 1%and PSNR down no more than 0.05dB compared to original HEVC model.This paper combined the CU partition algorithm based on Decision Tree and PU prediction algorithm based on structure correlation,and selected sereval HEVC standard test sequences to test.It show that we can obtain about 50%promotion of the coding speed at the same configuration,with the bitrate increased about 2%and PSNR down less than 0.1 dB compared to original HEVC model.