Research On Extending Technology For Video Coding Standard HEVC

With the fast development of multimedia technology and computer network,videos have been widely applied in various fields.Meanwhile the large amount of video data brings great challenges to network transmission and storage devices,therefore needs to be compressed.After several decades of development,a series of video coding standards have been established.HEVC,the latest standard,has excellent coding performance and network adaptability.Compared to previous standards,it achieves about 50% compression efficiency improvement.In recent years,the development of various video applications has brought technologies such as Virtual Reality,Augmented Reality and High-Dynamic Range videos into our lives.As a result,videos with larger resolution and higher frame rate become a tendency of new applications,which leads to the growing amount of video data.However,the existing coding technologies hardly meet the demand of high compression performance in practical application.Hence,there is an urgent need for new video coding technology with higher compression ratio than HEVC's.Concerning the problem,this paper researches on extending technology for HEVC standard.We take a deep look into different modules to exploit the potential for further data compression,and implement coding methods with higher compression efficiency.In the thesis,we mainly focus on modules of coding block partition structure and intra prediction,and make a deep research on extension of coding technology.For the coding block partition structure,we propose extended coding structure and optimization algorithm.Firstly,Coding Tree Unit size is enlarged to 256×256 pixels.Meanwhile,we enlarge the size of Coding Unit,Prediction Unit and Transform Unit accordingly,and support 64×64 transform.Although extending block sizes will reduce coding overhead of block partition information and greatly improve compression efficiency,it will increase coding partition levels and lead to massive increment in computational complexity.Therefore,we need to optimize and speed up the coding process.On the basis of extended coding structure,we propose the fast depth decision algorithm based on spatial correlation and gradient information.The optimization method uses coding information of adjacent blocks and texture features of current area to determine the depth range of coding unit,which can reduce the recursive levels of block partition.Experimental results show that the extended coding structure and optimization algorithm can obtain about 0.88%,1.68% and 1.81% gains for different configurations,which improves coding performance significantly.The increment of encoding time is limited within 20% at the same time.For intra prediction,we propose two novel algorithms.The first one is hybrid four-tap interpolation filter algorithm.This method extends two-tap to four-tap interpolation,and employs the combination of two interpolation filters.Besides,we set different threshold values for prediction unit and select the optimal one through experiments.The other method is extended boundary prediction filter algorithm.For certain angular modes,the method can further modify boundary samples in the prediction unit,and use the weight factor to indicate the proportion of filtering.In the intra prediction process,we combine two algorithms above and optimize the prediction signal generation by doing filtering process twice,which improves the prediction precision of intra coding.Experimental results show that the combined algorithm can achieve about 0.85% BD-Rate saving on average while the increment of coding time is less than 5%.