The Research On Acceleration Algorithms Of HEVC/H.265 Inter Coding

With the popularization of video applications and the gradual increase of video resolution and frame rate,the existing H.264/AVC can not meet the storage and transmission requirements of massive video data.JCT-VC formulated and officially announced HEVC in 2013.Compared to H.264,HEVC can reduce the video stream by 50%while guaranteeing the same or similar quality of reconstructed videos.Inter coding with high compression ratiois is one of the important components of HEVC.To find the best coding mode of CTU,inter coding of HEVC will utilize the quadtree technique to traverse all possible block combinations which imposes a huge computational burden on video coding.Therefore,inter coding acceleration has very important research significance.This dissertation designs two types of inter coding acceleration algorithms for CU/PU partition process:the algorithm based on statistical analysis and the SVM-based algorithm.There are three kinds of the algorithm based on statistical analysis:the algorithm based on residual analysis,the algorithm based on image texture and the algorithm based on depth information.The details of the algorithms are as follows:(1)The algorithm based on residual analysis:Residuals can reflect the matching effect of the current block in the reference frame,and the matching effect will directly affect the decision of CU partition.The algorithm uses dispersion to analyze the residuals,and the dispersion can be measured by the optimal RQT depth.According to statistical analysis,the prediction of CU partition can be completed by thresholds comparing.TH_up and TH bottom can be predicted from the encoded images.(2)The algorithm based on image texture:There is a high probability that the sub-regions of the homogenous region have similar motion trajectories.Therefore the CU with full homogenous region should not be divided into smaller CUs.The algorithm makes use of edge gradient to determine whether the current CU is a full homogenous region.(3)The algorithm based on depth information:Statistical data shows that the decision of current CU partition is very likely to be consistent with the reference CUs when decisions of them are consistent.Therefore,the reference CUs can be used to predict the decision of CU partition.The SVM-based algorithm takes residuals after dimension reduction as an input feature,and uses the online SVM model to complete the pre-judgment of partition of CU and PU.In order to reduce RD loss,an error control module is set for the SVM model,which can achieve adaptive calculation of the safety distance.Experimental results show that the designed algorithms can greatly reduce the execution time of video encoding on the premise of guaranteeing image quality and compression rate,among which the algorithm based on residual analysis achieves the best acceleration effect.