Research On Inter Prediction And Transform Coding In Next Generation Video Coding Standard

In recent years,with the rapid development of acquisition and display technology,the quality of video is improving,and the amount of video data is increasing explosively.The problem of video storage and online transmission become more and more serious,so it puts forward higher requirements for the compression of video data.At present,to deal with compression of ultra-resolution video of 8K and above in the future,developed work of H.266,which is the next generation video coding standard has been put on the agenda.The exploration of H.266 has began in 2015 and has achieved periodical results.H.266 still follows the traditional hybrid video coding framework.Compared with the latest video coding standard HEVC,H.266 has introduced many new efficient coding techniques in the key links to improve the coding perfermance,such as block partition,intra prediction,inter prediction,transform coding and so on.However,there are still some places need to be optimized and improved for these efficient coding techniques.In order to make some contributions to the development of H.266,this thesis will make some researches and related optimization works on the coding techniques,which are introduced by H.266 in transform coding and inter prediction.Firstly,in order to reduce the coding complexity and bit overhead of Explicit Multiple Core Transform(EMT)technology,intensive study and optimization about EMT is done in this thesis.And next,by analyzing the characteristics of the different transform cores and the distribution characteristics of the residual signals generated by different prediction mode,an inter-frame mode dependent transform method is proposed to further optimize EMT.Compared with transform scheme in HEVC,the experimental results show that the proposed method achieves average BD-Rate savings of 1.03% and 1.02% under the low delay configuration respectively.Compared with EMT,the experimental results show that average BD-Rate savings of 0.12% and 0.16% are achieved for the proposed method respectively,at the same time,the coding complexity of the whole encoder is reduced by 4% and 6% respectively.The coding performance and the coding efficiency are improved together.Furthermore,this method can reduce the coding complexity of the whole encoder by about 10% with little loss of coding performance.On this basis,the energy concentration effect of DST7 and DCT2 on the residual information generated by template matching technology is further analyzed,and a new hybrid transform,named DST7-DCT2 is constructed to replace the original DST7 transform by combining the advantages of these two transforms cores.Under the random access configuration,the experimental result demonstrates that average BD-Rate saving of 0.031% is achieved using new constructed transform.Then,in this thesis,the template matching technology in Frame Rate Up Conversation(FRUC)mode introduced by H.266 is analyzed in detail,and finds that the prediction direction of this technology is determined without using pixel information of coding block after deriving the motion information,which leads to the problem of inaccurate prediction direction.This problem seriously decreases the coding performance of this mode.To make up this deficiency,three different schemes on prediction direction decision are proposed by considering three factors of prediction direction accuracy,bit overhead and coding complexity in this thesis.The experimental results show that three schemes achieve significant BD-Rate savings under the random access configuration,which are 0.23%,0.47% and 0.51% respectively.When only FRUC technique is turned on,the average BD-Rate savings are up to 0.48%,0.68% and 0.77% for three proposed schemes.