Fast HEVC Intra Prediction Algorithm Based On Local Texture Feature

HEVC is a new generation standards of video coding,it can achieve a higher compression rate with the same compression quality,and it can meet the applications requirement of HD video.However,HEVC greatly improves the coding efficiency at the cost of the complexity,and high complexity will result in longer coding time,so it has brought great difficulties in practical applications.It has become the current main research direction that reduce the HEVC complexity and coding time while it will not influence the video coding quality.This paper does research about the key technologies in HEVC fast intra prediction,focuses on the existing prediction block division and best prediction mode selection algorithm in HEVC intra prediction.HEVC intra prediction employ more accurate prediction mode and more flexible prediction block division.The number of intra prediction modes is as high as 35 and prediction block can be divided into larger block,and they are all based on the quad-tree traversal.Therefore,HEVC intra prediction is more complicated.Aiming at the high complexity of prediction mode selection and prediction block partitioning process,this paper proposes a fast algorithm for intra prediction based on local image texture feature in these two aspects,thus realize fast mode selection and PU division.The main work of this paper includes the following aspects: Firstly,this paper proposes a 4×4 prediction block fast partitioning algorithm based on local image texture direction against the high complexity of quad-tree traversal division.The Canny operator is used to calculate the gradient,and obtain the texture direction of local image.According to the texture direction,it can achieve fast division of 4×4 PU.Secondly,this paper proposes a fast intra best prediction mode selection based on local image texture complexity and direction against the high complexity of HEVC intra best prediction mode decision traversal process.A method to measure the complexity of image texture is proposed,Hadamard transform is used to calculate the texture complexity of PU,it is combined with the texture direction of PU that calculated by Canny operator to obtain local image texture feature.And it can reduce the number of traversal times of intra prediction mode decision according to these feature,thus realize the best prediction mode decision.Finally,the improved algorithm is implemented in JCT-VC's official test software HM-16.0 by modifying code,and do a lot of experiments on the proposed fast algorithm by the official test video sequence,and then analysis the result.The experimental results show that the proposed algorithm can reduce the encoding time by 25.63%,and has no significant influence on the coding quality.