Research On Fast Discrimination Algorithm Of HEVC Inter-Frame Mode

HEVC is a new generation of high-performance video codec standard,compared with H.264,a lot of advanced video coding technology is added,such as flexible coding unit division,asymmetric prediction unit division and multi-mode prediction.This makes HEVC increase the compression ratio by about 50% while maintaining the same compression quality,and the cost is nearly doubled.With the development of mobile networks,people tend to use mobile phones and other wireless terminal equipment to see video information,Compared with the PC,mobile terminal equipment computing power is weak,which makes the advantages of HEVC can not be a very good expression.Therefore,it is necessary to reduce the computational complexity by optimization.Inter-frame prediction in HEVC accounts for about 70% of the total calculated.Among inter modes,SKIP/MERGE mode detection attracts lots of researcher's interest,Because SKIP/MERGE mode is selected more than 88%.The MERGE mode is like a direct mode.By evaluating the MERGE mode,it is possible to find the motion information(motion vector,reference picture index,and reference picture list usage flag)of the neighboring PU for estimating the current PU.The encoder then selects the best merge candidate from the candidate list formed by the spatial and temporal prediction unit PU and sends the corresponding index of the selected candidateSome of the existing MERGE pattern discrimination algorithms have some limitations.Some existing MERGE pattern discriminant algorithms have some limitations.On the basis of studying some existing algorithms,this paper realizes the algorithm of calculating only the single mode to judge whether it is the best mode.And then the MERGE mode can be used to quickly determine and obtain the current coding unit CU information by using the merge candidate,thus realizing the fast identification algorithm of the MERGE mode,which can effectively reduce the computational complexity of the coding mode between the coding units in HEVC.The specific process of this algorithm is,First,the motion smoothing region and the single motion region are identified by the merging candidate of the previous frame and the current frame.Second,we propose a fast Merge pattern of statistical information based on the rate distortion cost,since about 98% of the same rate of distortion is Merge mode.Finally,it is proposed to use 2N × 2N motion estimation to obtain the SAD value of the prediction unit PU and compare these SAD values for fast mode discrimination.The algorithm is implemented in HEVC's standard test system HM and strictly follows the general test requirements of HEVC.In order to compare with other algorithms,the same quantization parameter Qp({24,28,32,36})is used in the algorithm,and the standard test sequences with different resolutions of different types are used to test.The algorithm performance was evaluated using the parameters BDBR(?R),BDPSNR(?P)and time saving(?T).Experimental results show that the BDBR only 1.8% increase in the case of coding time can save about 35% compared with the HM standard codec,Especially for scenarios such as the scene of a single video resources,coding time can save about 45%.Compared with the MERGE pattern discrimination algorithm in the literature,BDBR increased by about 1.5%,but the coding time savings of about 15%.