Research On Fast Coding Optimization Technology Of HEVC And 3D-HEVC

With the rapidly development of advanced video acquisition and display technology,the video resolution has also increased,and the video has gradually developed from the standard video to the current high-definition(HD)and ultra-high-definition(Ultra HD)video.Massive video data introduces greater challenges for data storage and network transmission.Video coding can efficiently compress video data,which makes the research of video coding technology a hot issue in industry and academia.In order to meet the user's demand for HD,UHD video,a new generation of high efficiency video coding standard(HEVC)and its 3D extension standard(3D-HEVC)have been developed.Although more and more advanced video compression technologies can greatly improve the compression performance,advanced video compression standard also brings huge computational complexity while achieving high compression performance,which seriously affect Real-time application of compression standard.Therefore,under the premise of keeping the video compression quality basically unchanged,effectively reducing the computational complexity of coding is a necessary premise for real-time application of video compression standard,and it is also an urgent need for large-scale application of HD and Ultra HD video.The video compression standard improves compression performance by adopting more flexible and finer algorithms,but it also increases the coding computational complexity.Therefore,in order to reduce the computational complexity of coding,this paper mainly studies the optimization of intra frame coding and inter frame coding.The main works are summarized as follows:Firstly,an early SKIP mode decision algorithm based on unimodal stopping model(USM)is proposed to reduce the complexity of inter frame mode decision for High Efficiency Video Coding(HEVC).First,according to the depth temporal correlation of Coding Unit(CU),each CU is divided into rare-used or frequent-used.For the rare-used CU,the SKIP mode is directly determined as the optimal mode of the current CU,and the remaining mode decision process is early terminated.For frequent-used CU,by exploring the hierarchical mode correlation and RD cost property,an USM is established to early terminate the SKIP mode decision.The experimental results show that the proposed algorithm saves on average 58.5% and 54.8% of coding time under the conditions of random access(RA)and low delay B(LD B),BDBR increased by 0.8% and 0.8%.Secondly,a fast inter-frame CU depth decision algorithm based on optimal selection model and coding parameters is proposed to reduce the computational complexity of HEVC inter-frame coding.According to the CU depth temporal correlation,an optimal selection model is established for predicting the depth range of the current coding CU.To reduce the prediction error of the current CU depth,the maximum CU depth predicted by the optimal selection model and the Coded Block Flag(CBF)of the current coding CU are jointly used to determine the maximum depth of the current CU.Meanwhile,the prediction unit(PU)and CBF of the current coding CU are also used to determine the maximum depth of the current coding CU.The experimental results show that the proposed fast CU depth decision algorithm saves the average coding time of 56.3% and 51.5% under the test conditions of RA and LD B,while the average BDBR increase is only 1.3% and 1.1%.Thirdly,a fast CU and PU decision algorithm based on hybrid stopping model is proposed to reduce the computational complexity of dependent texture views in 3D-HEVC inter-frame coding.By using the inter-view correlation,the optimal CU depth partition and PU prediction mode of the current coding CU are roughly predicted,and then the rate distortion cost(RD cost)and the CBF are used to determine whether the current predicted CU depth partition and the PU prediction mode are optimal or not.Furthermore,an early Merge mode decision algorithm based on probability model is proposed to reduce the computational complexity of 3D-HEVC inter-frame dependent texture views and dependent depth maps coding.First,using the hierarchical and the inter-view correlations,a prior probability model is established to roughly predict whether the optimal mode of the current coding block is the Merge mode.Second,using the CBF of the current coding block,a posterior probability model is established to further determine whether the optimal mode of the current coding block is the Merge mode.The experimental results show that the proposed early Merge mode decision algorithm saves 45.2% and 30.6% coding time for dependent texture views and dependent depth maps coding,respectively,while keeping coding efficiency almost lossless.Fourthly,an early Merge mode decision algorithm based on self-learning residual model is proposed to reduce the computational complexity of both texture views and depth maps in 3D-HEVC.Firstly,a self-learning model is established by using the residual signals of those already encoded CUs with the Merge mode as their optimal modes.Secondly,the residual-based self-learning model is exploited to predict the residual signal of the current coding CU,which can early determine whether the Merge mode is the optimal mode for it or not.Experimental results show that the proposed approach achieves 41.9%,24.3% and34.4% coding time savings on average for texture views,depth maps and total encoding with only negligible coding performance degradation,respectively.Fifthly,a self-learning residual model-based fast coding unit(CU)size decision approach is proposed for the intra coding of both texture views and depth maps in 3D-HEVC.Residual signal,which is defined as the difference between the original luminance pixel and the optimal prediction luminance pixel,is firstly extracted from each CU to express the feature of each CU.It has a strong relationship with optimal CU partitions.Then,a binary output classifier with self-learning residual model is developed to early terminate CU size decision for both texture views and depth maps.The self-learning residual model,which is established by intra feature learning,iteratively learns the features of previously encoded coding tree unit(CTU)that is generated by itself.Experimental results show that the proposed fast intra CU size decision method achieves 33.3% and 49.3% encoding time reduction on average for texture views and depth maps with negligible loss of overall video quality,respectively.