Research On Fast Motion Estimation Algorithm For Inter Prediction In VVC

As people's demand for high-quality video becomes more urgent,video transmission and compression face more challenges and opportunities.In order to adapt to higher resolutions and more video application scenarios,in July 2020,ITU-T(International Telecommunication Union for Telecommunication Standardization Sector)launched a new generation coding standard H.266/VVC(Versatile Video Coding),compared with the previous generation coding standard H.265/HEVC(High Efficiency Video Coding),VVC not only saves the average bit rate,but also brings high computational costs.It is found that Affine Motion Mode(AMM)brings huge computational complexity in the new coding framework.This topic mainly studies the fast algorithm of affine motion estimation in VVC inter-frame prediction,mainly from two aspects: the advance decision of affine motion mode and the fast decision of affine four and six parameters,so as to reduce the coding time.Aiming at the problem of high computational complexity in affine motion estimation decision-making,this paper proposes the FARD(Fast Affine Region Determination)algorithm.The algorithm first makes an advance decision on the affine motion area based on the inter-frame difference method and the Prewitt texture gradient operator;then,according to the Multiple Tree Type(MTT)structure in VVC,when the parent block belongs to the affine motion area,it is judged that it belongs to the affine motion area.The optimal prediction mode of the sub-block,when there are multiple sub-blocks with the affine motion mode as the optimal prediction mode,the parent block is used to encode the affine motion mode;at the same time,the threshold value of the inter-frame difference method is adjusted to reduce Part of the ME situation with smooth motion is eliminated,thereby further reducing the computational complexity of the ME.In the process of affine four-and six-parameter model matching,a large number of RDO calculations are required,resulting in high complexity.In the research of this paper,a four-and six-parameter fast decision-making algorithm based on SATD(Sum of Absolute Transformed Difference)cost is proposed for merging candidates of time-domain co-located blocks.The algorithm first obtains the co-located block information through the forward/backward reference frame,expands the current candidate list,and merges the CPMV of the co-located block with the current block candidate;The merged candidate list is sorted from low to high cost;finally,the fourth and sixth parameters are quickly decided according to the cost order.At the same time,the precision of MVD is refined according to the residual values of fourand six-parameter motion compensation.The experimental results show that,compared with the original test framework of VTM11.0,on the premise that the subjective video image quality is basically unchanged,in RA(Random,random access)mode and LDP(low delay P frame)In the mode,the encoding time of the FARD algorithm is saved by an average of 17.52%and 14.64%,and the BD-RATE is only increased by 0.43% and 0.71%.In the joint test of the four-and six-parameter fast decision-making algorithms based on SATD and the FARD algorithm,the two modes of RA and LDP were also tested,and the coding time was saved by 21.44% and 24.49%,respectively,and the BD-RATE increased by 0.64% and 0.74%.