An Adaptive Rate-distortion Optimization Method In HEVC

High Efficiency Video Coding(HEVC)is a new generation of video coding standard for diversified video application and high-resolution video encoding.HEVC adopts many new coding techniques,which include a quadtree-based coding unit division,more flexible intra-prediction modes.Compared with previous video coding standard H.264/AVC,HEVC improves the compression ratio by about 50% at the same visual quality.However,it is at the sacrifice of significant increase of computational complexity.HEVC adopts rate-distortion optimization(RDO)technique to compromise between the bitrate and distortion,which determines the division of coding unit and the optimal prediction mode.Therefore,RDO is the key factor of HEVC,which has influence on both visual quality and compression efficiency.In this thesis,we research on the RDO techniques for HEVC video coding so as to decrease the computation complexity and improve the bitrate-distortion performance.The main research works are summarized as follows:Firstly,to address the huge computational complexity of intra prediction coding in the RDO process of HEVC,a method for adaptive Lagrange multiplier determination for rate-distortion optimization has been proposed.It can make RDO predict the best value of Lagrange multiplier so as to reduce the computational complexity.We set a factor in the Lagrange multiplier and give it a series of different values.The training video sequences will be encoded by each different Lagrange multiplier factor and get the rate-distortion curve.According to the Bjontegaard delta measurements,we finally get a exclusive best Lagrange multiplier factor for each training video sequences.Secondly,inspired by the spatial and temporal similarity in dynamic texture,we consider that it has a huge space for improving the encoding.Therefore,we extract some of the features from dynamic texture.After that,the features and the best Lagrange multiplier factor will be used for linear regression so that we only need to extract the corresponding features and we can predict the best Lagrange multiplier factor.Experimental results show that the proposed method can achieve better rate-distortion performance and provide better visual quality in the case of promising the computational complexity Finally,inspired by the temporal regularity in the video sequences which included motion objects.We believe that we can extract the features of the motion objects and get the best Lagrange multiplier factor according to the Bjontegaard metric.After that,we have the linear regression function so that we can predict the best Lagrange multiplier factor from the relevant features value.Experimental results show that the proposed method can promote the coding efficiency compared to the original codec.