Research On Fast Inter-prediction Algorithm For HEVC Based On Visual Saliency

High Efficiency Video Coding(HEVC)is the latest video coding standard,inherited from H.264/AVC.By introducing new technologies,it can further improve coding efficiency while maintaining good video quality.Compared with the H.264 standard,HEVC adopts a new Merge mode,flexible block partitioning and a more reasonable sub-pixel interpolation algorithm for inter-frame prediction.The coding performance is also improved,but the coding calculation is complicated.Therefore,on the premise of ensuring coding efficiency,effectively reducing coding complexity is the key to expanding the application of the HEVC coding standard in the video field.This paper focuses on the HEVC inter prediction fast coding technology,optimizes the Coding Unit(CU)partition and mode selection algorithm,and reduces the inter prediction coding complexity.The innovations in this paper mainly include the following aspects:(1)Propose a fast depth termination algorithm based on residual distribution for HEVC.In the first stage,the residual value after the CU executes the Inter?2N×2N mode is obtained,then classified according to the actual CU depth and modeled with a Gaussian distribution.In the second stage,the mean value of the residual as a feature after the Inter?2N×2N mode is completed,and the probability of the residual in the Gaussian distribution is calculated,and whether the current CU spliting is determined according to the probability.(2)Propose an HEVC inter-frame depth prediction decision algorithm based on multi-reference frames.Firstly,the Structural Similarity(SSIM)value of the Co-located CTU of each Coding Tree Unit(CTU)and its two reference frames with the smallest index in the reference frame list is calculated;then the weight of the SSIM values of weighting the co-located CTU depth predict the depth range of the current CTU.At the same time,using the parent-child CU information correlation,a fast reference frame selection algorithm is proposed,which determines the best reference frame of the current CU according to the best mode and coding information of the parent CU.(3)An inter-frame prediction optimization algorithm based on saliency is proposed.In terms of CU depth partitioning,our uses the saliency detection model to obtain the spatial and global saliency of all CTUs in the current frame,and finally obtains the spatial saliency value of the CTU.The correspondence between saliency value and saliency region is established,and the depth range of CTU is finally set according to the saliency region.In terms of inter-mode prediction,the saliency of 4 × 4 blocks in the CU is calculated and is binarized,then the proportion of saliency blocks in the two parts of the prediction unit(PU)is calculated,and is determined whether to execute in the PU mode,this scheme can effectively avoid time consumption in unnecessary modes.The experimental results show that the depth decision algorithm proposed in this paper can effectively reduce the coding time while maintaining the objective quality of the video;the saliencybased interframe coding optimization scheme proposed in this paper can effectively reduce the coding time while maintaining the objective and subjective video quality.