Researches Of Rate Control Algorithm For High Efficiency Video Coding

With the development of the technology,communication and multimedia technology have been developed rapidly.Videos are the mostly used communication medium which contain the largest information.Normally,videos need to be coded when they are stored in the limited storage space and transmitted in the limited transmission bandwidth.As a new generation coding standard,high efficiency video coding(HEVC)has improved the coding efficiency significantly comparing with H.264/AVC.However,it still an important and difficult issue to make full use of the bandwidth resources.Rate control is the key technology to transmit the encoded video in the limited bandwidth,which aims at reconstructing the quality of the encoded video as well as possible at a target bit rate.In HEVC,a ? domain rate control algorithm is adopt in which the Lagrange multiplier ? is regarded as the most important factor for rate control,and the bit rate is allocated adaptively with the updating parameters when selecting the Lagrange multiplier ?.However,the selected ? is always not optimal.Moreover,with some special situations,the bit allocation method and ldecision strategy need to further research.This dissertation mainly concerns on the human visual characteristics,surveillance videos and the optimal ? decision.The main contents and innovations are:(1)The bit rate is allocated according to the spatial-temporal visual sensitivity.Firstly,as human vision has different sensitivity for different texture in the spatial domain,an employed just noticeable difference(JND)model is used to obtain the maximum visual distortion of a frame.Then,the maximum visual distortion of every coding tree unit(CTU)can be obtained and the spatial sensitive weights of all the CTUs can be calculated.Secondly,in the temporal domain,the frame level bit rate will be allocated according to the difference of the frames,where the frames with large variation will be coded well to improve the subjective performance.Meanwhile,the bottom-up attention region can be obtained with the difference of the frames during the continuous frames.Then,the temporal visual sensitivity of every CTU can be obtained.Finally,combining with the spatial and temporal visual sensitivity,the spatial-temporal weights can be obtained for allocating bit rate to all the CTUs more reasonable.The experimental results show that the proposed algorithm improves the subjective performance efficiently and keeps the rate control accuracy at a high level.(2)For the surveillance videos,a separated ? decision algorithm is proposed.Firstly,a low complexity pixel based segmentation method is used to obtain the foreground and the background.Based on the characteristics of the surveillance videos,in which the background always keeps still for a long time and theforeground is caused by the variation of the adjacent frames,the foreground and background pixels can be obtained by the difference of the four temporal continues frames.As the foreground pixels are always assembled in a region,the foreground CTU can be obtained from the foreground pixels.Secondly,according to the experimental results,the rate distortion characteristics of the foreground and the background obey the hyperbolic model.Then,with the rate distortion optimization(RDO),the average CTU ? value of the foreground or the background should be equal to the frame ?.By the clipping method for the foreground CTU ? and the background CTU ?,the average ? value of the foreground or the background will gradually tend to the frame ?.Finally,a separated updating process is used to obtain the more suitable parameters for the foreground and the background.The experimental results show that the proposed algorithm improves the coding quality of the foreground significantly with keeping good rate control accuracy and coding quality of the whole frame.(3)A multilevel optimal ? decision algorithm for the rate control is proposed at Low Delay configuration.Firstly,an approximate linear bit allocation relationship,which is obtained through the experiment,is used between the motion compensation prediction error(MCPE)and the frame level bit allocation to assign the bit rate more reasonably for every frame.Secondly,to minimize the whole distortion of a sequence,an effective way is to get the optimal RDO of every GoP(group of picture).Then,with the RDO process of the GoP level rate control,the optimal frame ? ratio can be obtained based on the relationship of the GoP ? and framel.Finally,with the RDO process of the frame level rate control,the average lvalue of all the CTUs should be equal to the frame ? and the optimal CTU lclipping method will approach the encoded CTU ? to the frame ? gradually.The experimental results show that the proposed algorithm improves the coding quality significantly with a good rate control accuracy.(4)A distortion propagation based optimal ? decision for Random Access(RA)rate control is proposed.Firstly,the main reference model is built based on the reference ratio from the experiment.Then,the distortion propagation model at RA configuration will be derived by the distortion estimation of current block to the following blocks.Secondly,with the RDO processes of the sequence level rate control,the GoP level rate control and the frame level rate control,the lrelationships between the sequence ? and the GoP ?,the GoP l and the framel,the frame ? and the CTU ? will be obtained.Finally,the optimal lproportions can be calculated for the frames and the CTUs.So the coding performance can be improved from the sequence level,the GoP level and the frame level respectively.The experimental results show that the proposed algorithm improves the objective and subjective coding performance with a good rate control accuracy.