| With the development of the Internet technology and mobile devices,multimedia applications based on image/video are emerging,such as digital television,Internet video,virtual reality,live video,etc.These multimedia applications have gradually become parts of people's daily life.The huge number of videos poses a serious challenge to video storage and video transmission.In order to store and transmit videos efficiently,the video coding technology plays an important role,which also makes the research on video coding a hot topic in academic and industry community.The latest international video coding standard HEVC(High Efficiency Video Coding)was released in 2013 in order to compress the increasing high and ultra-high resolution videos efficiently.Compared to the previous video coding standard H.264/AVC,HEVC can achieve about 50% bitrate reduction under the same perceptual video quality,which improves the encoding efficiency significantly.In 2014,the AVS working group released China's second generation video coding standard AVS2,which is mainly used for ultra-high resolution videos.AVS2 is competitive with HEVC in terms of performance.However,since the latest video coding standards adopted many novel techniques,such as more flexible coding unit partition structure and finer intra prediction modes,the coding efficiency achieves significant improvement while the encoding complexity also increases tremendously.Therefore,how to optimize the encoder to reduce the encoding complexity is crucial for practical applications.In the meantime,with the increase of high resolution videos,future video coding standards need to handle more video data.Therefore,how to further improve the coding efficiency based on current video coding standards is still of great importance.Intra coding is one of the most core modules in video coding.How to optimize intra coding is critical to the whole process of video coding.This dissertation mainly focuses on the research on reducing the complexity of intra coding and improving intra coding efficiency.The content of the dissertation can be divided into four sections that are detailed as follows.First,a fast intra mode and coding unit(CU)size decision method is proposed to reduce the complexity of intra coding.Intra coding of the latest standards employs many prediction modes(HEVC adopted 35 modes,AVS2 adopted 33 modes),and a quadtree structure for CU partition.Since the best intra prediction mode and CU size are decided by rate distortion optimization(RDO)process,the encoding complexity is very high.In order to reduce the intra coding complexity for HEVC without decreasing the coding efficiency obviously,this dissertation proposes a fast intra mode decision method and a fast CU size decision method.In the proposed fast intra mode decision,the proposed average gradients are used to reduce the number of modes for rough mode decision(RMD)and RDO so that the complexity of intra mode decision can be reduced efficiently.In the proposed fast CU size decision,firstly,the homogenous blocks are earl y detected and encoded with large CU sizes directly.Then,for the inhomogeneous blocks,a weighted linear SVM is applied in early CU split decision.At last,for the blocks which are not early split,another weighted linear SVM is employed in early CU termination decision.The experimental results suggest the proposed fast intra coding method can reduce the encoding time by over 50% while the RD loss can be neglected when compared to original HEVC reference software HM-14.0 under all intra configuration.Second,a hybrid intra prediction method based on local and non-local correlations is proposed to improve intra prediction efficiency.In current video coding standards,angular intra modes can deal well with blocks with strong directional patterns,while planar mode(or plane mode)and DC mode are used for predicting smooth blocks.However,for the more complicated blocks,current intra prediction will be inefficient.In order to improve the intra prediction efficiency of current video coding standards and also provide some potential techniques for future video coding standards,based on the analyses of the complicated coding blocks,this dissertation proposes a hybrid intra prediction based on local and non-local correlations.The proposed method consists of three parts: 1)adaptive template matching prediction used for predicting blocks with complex textures by utilizing non-local correlations,2)combined local and non-local prediction used for blocks with both complex textures and strong directional textures,and 3)combined neighboring modes prediction used for blocks with blurring edges by utilizing more local correlations.The experimental results suggest the proposed method can achieve about 3% bitrate reduction when compared to original HEVC reference software HM-14.0 under all intra configuration.Third,an adaptive chroma intra prediction is proposed to improve the efficiency of chroma prediction.YCb Cr420 color space is widely used in current video coding standards.In order to reduce the inter-channel redundancy among Y,Cb and Cr,LM(Linear Model)mode was proposed,in which the reconstructed Y component is used to predict the chroma components Cb and Cr.Considering the drawbacks in the estimation of parameters in LM mode,an adaptive template selection method is proposed in this dissertation to get more accurate parameters so that the prediction efficiency of Cb and Cr components can be improved.Besides,considering that the correlation between chroma components is not fully exploited,an adaptive Cr prediction is proposed in this dissertation to further improve the coding efficiency for Cr component.Specifically,in the prediction of Cr component,besides the correlation between Cr and the reconstructed Y,the correlation between Cr and the reconstructed Cb is also exploited.The experimental results suggest that the proposed method can improve the performance of LM mode,and the encoding and decoding complexity introduced by this method can be neglected.Fourth,a signal dependent transform for intra coding is proposed to improve the transform efficiency in intra coding.The existing video coding standards adopted signal independent transforms.For example,DCT or DST is used as transforms for intra prediction residuals in HEVC,and AVS2 mainly uses DCT.Although fixed transforms DCT and DST have been demonstrated to be close to KLT in some situations,they are signal independent and cannot be adaptive to the content of residuals.To handle the limitation of fixed transforms,this dissertation proposes a signal dependent transform based on SVD for intra coding.The proposed method utilizes the structural similarity between the residual block and the synthetic block generated from the predictions to obtain a transform pair by applying SVD on the synthetic block.Then,the derived transforms are used for transforming the corresponding residual block.Thus,the proposed transform can be adaptive to the residual block in some degree.Besides,the proposed adaptive transform is further extended to template matching prediction in order to improve the transform efficiency for residuals by TMP.The experimental results suggest the proposed method can improve the efficiency of the transform s in current standards.Thus,the intra coding efficiency is improved. |
PDF Full Text Request