Research On Efficient Intra Prediction Of H.266/VVC

H.266/Versatile Video Coding(VVC)is the next-generation video coding standard that responds to the need of the present time,which is expected to be released in 2020.Its goal is to reduce the BD-rate by 50% while keeping the video quality consistent with H.265/ High Efficiency Video Coding(HEVC)standard.So far,H.266/VVC has achieved large performance boost.On the one hand,it gets benefit from the adoption of many new technologies,such as multi-tree partitioning,independent coding of luma and chroma component,Cross Component Linear Prediction(CCLM)prediction mode,etc.On the other hand,it is derived from the optimization of the original coding technology,including the 33 angular modes are extended to 65,edge filtering and boundary filtering are replaced with PDPC technology and so on.Intra prediction coding is a key part of video coding technology,because intra frame is used as the reference image of subsequent frames and its coding performance will affect subsequent frames.In order to reduce the BD-rate of H.266/VVC and take into account the complexity of the algorithm,the specific research content includes three parts:1.The CCLM mode is a newly introduced chroma prediction mode of H.266/VVC,which uses neighboring reconstructed luma and chroma samples to establish a linear model and then implements the prediction of chroma component based on the corresponding luma prediction values by the linear model.The CCLM mode can improve coding efficiency greatly,but the reference samples of adjacent regions are used in the calculation of linear model parameters.Due to the large difference between adjacent regions and coding blocks,the coding performance of CCLM mode is not ideal in this case.In H.266/VVC,luma component is encoded prior to chroma component,so in the CCLM stage the luma value of the coding block has been reconstructed.Based on this fact,a linear model parameter derivation algorithm is proposed that considers the luma characteristics of the coding block.Specifically,the luma values and corresponding distribution of the coding block participate in linear model parameter derivation.With this method,the linear model parameters are closer to the optimal linear model parameters,which can lead to more accurate prediction signal.Compared to VTM2.0 anchor,the Y,U and V components have a bit reduction of 0.14%,0.57% and 0.63% respectively in All Intra configuration.However,the codec complexity increase obviously,especially the decoding complexity.Therefore,this paper further proposes hybrid CCLM,which determine whether to use the improved CCLM mode or the original CCLM mode according to the coding block size.In All Intra configuration,the Y,U and V components of the hybrid CCLM have a bit reduction of 0.13%,0.55% and 0.65% respectively,with no complexity increase.2.Chroma mode candidate list of H.266/VVC includes two parts.The first part inherits the five prediction modes in H.265/HEVC,namely the prediction mode of the luma component,DC mode,PLANAR mode,HOR mode and VER mode.The second part is three CCLM modes,including LM,LM_L and LM_T mode.Since LM_L and LM_T are newly added modes,the reference samples used among HOR mode,VER mode,LM_T mode and LM_L mode are overlapped somehow.Since the independent partition of luma and chroma,one chroma block may correspond to multiple luma blocks.Based on the above two reasons,a chroma mode candidate list optimization based on multi-luma mode is proposed.Based on the coding performance of VTM3.0,the Y,U and V components of All Intra configuration have 0.04%,0.58% and 0.62% bit reductions respectively,with no complexity increase.3.For the luma mode coding,H.266/VVC adopts the coding method of the 6 MPM list.The modes belonging to the MPM list use a truncated unary code.The non-MPM modes use a truncated binary code.Due to the introduction of multi-reference line intra prediction techniques,MPM list construction of the zero reference line and the non-zero reference line are not uniform,which increases the codec complexity and requires multiple hardware processing units.Besides,the non-MPM modes are not prioritized,resulting in an increase in code bits.Based on this,this paper proposes the unification of MPM list construction and the priority encoding of non-MPM modes.Based on the coding performance of VTM3.0,the Y,U and V components of All Intra configuration have 0.02%,0.03% and 0.01% bit reductions respectively.the algorithm saves little BD-rate,and the most ipmotant is that it realizes the unification of MPM list.