Research On Fast Algorithm For Intra Prediction Of HEVC Screen Content Coding

The screen content video is widely used in multimedia applications,e.g.,remote desktop interfacing,shared screen collaboration,wireless display,and so on.Screen Content Coding(SCC)is the extension of High Efficiency Video Coding(HEVC).However,the intrinsic HEVC recursive partitioning scheme and the additional SCC tools impose significant computational burden on the encoder,primarily during the seeking of optimal combinations of CU partitions and CU modes.The high computational complexity brings an obstacle for the application of HEVC-SCC.Thus,it is very important to optimize the coding algorithm to reduce the coding complexity for HEVC Screen content coding.This thesis introduces the basic knowledge of video coding and HEVC screen content coding firstly,and then summarizes the history of video coding standard briefly.The thesis also analysis the structure of HEVC prediction coding and introduce the new coding tools which import high computational complexity for video coding.After that this thesis enumerate some important fast video coding method.To reduce the high computational complexity of HEVC screen content coding,two fast coding methods were proposed.The first,this thesis presents a fast Intra prediction method based on screen content.First,CUs are classified into Completely smooth content CUs(PSCU),natural contend CUs(NCCU)and screen content CUs(SCCU)by based on the statistic characteristics of the content.For PSCUs,the newly adopted prediction modes,including the DC mode,Planar mode,horizontal angle mode and vertical angle mode,the IBC and PLT modes are skipped.For NCCUs,the traditional 35 kinds of intra mode are selected as candidates,and the IBC and PLT modes are skipped.For SCCUs,the IBC model is used to predict and then the spatial correlation is used to determine whether to skip the traditional intra prediction mode.Under all-intra configurations,experimental results show that the algorithm we proposed can save about 30.82% on average coding time compared with SCM-3.0,with only negligible bitrate increase.The Intra Block Copy mode is one of very efficient coding tools for the screen content coding,which greatly improves the coding efficiency for the screen content coding and greatly brings along high computational complexity.To further reduce the complexity,The thesis proposes to speed up the local search by checking the hash values of the current block and block candidates.For the current CU block,the hash value of the current block is generated according to the hash function firstly.Then,the blocks in the reconstruction area sharing the same hash values with the current CU are selected as prediction candidates.Last,the candidate with the lowest RD cost will be selected as the best prediction block.Experimental results show that the proposed method can save more than 10% on average coding time compared with SCM-3.0,with only negligible bitrate increase.