Researches On Rate Control Algorithm Based On Visual Perception Characteristics

In video communication, video information to be transmitted first shouldcarry out effective compression to reduce the total amount of data within thechannel bandwidth, especially the rate control is more necessary to regulate theoutput stream to fit the channel bandwidth when in the network or mobiledevices. The traditional rate control evaluation PSNR is used as a measure ofvideo objective quality distortion criteria, but studies have shown that the pastobjective quality have been unable to meet the human eye’s subjectiveexperience, so the rate control combination of human visual system (HumanVisual System, HVS) is a new research direction in recent years. In themultiview video, each frame reference relationship and the relationship betweenbit rate and the quantify parameters are more complex, the rate and qualitycontrolling needs to consider more sophisticated global optimization, it is alsomore difficult to control, and multiview video reference software has given norate control model. This paper mainly focuses on rate control algorithms formultiview video coding. And the contributions of this paper are:(1). The rate control G012algorithm in H.264/AVC achieves a betterbalance between the actual rate, image quality and the encoder buffer occupancy,but previous studies have shown that the objective quality indicators are notgood to reflect the image quality; this paper proposes a rate control methodbased on human visual properties by analyzing the human visual system. Framedifference ratio with respect to the complexity of current frame and its previousframe is utilized to represent the scene change and motion of the frames, and asthe guidance for bit allocation at frame level. Moreover, the bit allocation atbasic unit is based on the measured visual sensitivity; it makes the distortionlevel in different regions at the basic unit can consistent with the human visualsystem. The experimental results show that the rate control accuracy by theproposed algorithm is within0.12%~0.48%; PSNR fluctuation and standarddeviation are relatively lower compared with G012, the standard deviation reduction is up to1; thus the subjective quality of the reconstructed videosignals is more satisfied.(2). Reference software in multiview video has not yet given rate controlmodel, and the original rate control model for two-dimensional video can not bedirectly applied to multiview video coding because of disparity compensationand motion compensation. Meanwhile, scholars have established the justnoticeable distortion model based on the brightness contrast and masking effectof human visual properties, it gives the human eye’s maximum tolerabledistortion. In order to quantify HVS characteristics to be used in rate control formultiview video, this paper presents a macroblock layer rate control formulti-view video based on JND model. Firstly, JND model is established basedon the existing literature to strike the pixel JND map of color sequence, then onthis basis, the rate control algorithm based on JND model is proposed, which isperformed on four levels, namely view level, GOP level, frame level andmacroblock level. Especially in the macroblock level, MAD (Mean AbsoluteDifference) re-defined as MAPD (Mean Absolute Perceptual Difference) basedon pixel JND, and strikes the macroblock QP according to the redefinedquadratic model. Finally, the subjective quality and objective quality evaluationare both used to evaluate the proposed method. Experimental results show thatthe proposed method can obtain higher subjective quality while the ratecontrolling accuracy and PSNR remains unchanged, and to IPSNR as standard,the rate of the algorithm saves about10.31%~32.13%.(3). As I frame QP setting plan is too thick in JM and the specialhierarchical B pictures structure in multi-view video, this paper presents animproved key frame QP setting rate control algorithm for multi-view video,which is also performed on four levels, namely view level, GOP level, framelevel and macroblock level. In the view level, the target rates of different viewsare allocated based on a pre-statistical rate allocation proportion. In the GOPlevel, the total bits allocated and the initial QP of each GOP is set according tothe index R-QP model of key frames. In the frame level, a set of scaling-factorsapplied in bit allocation for the B frames at different temporal levels isintroduced. In the macroblock level, the target bits for current macroblock areallocated and the corresponding QP can be computed by the quadratic R-Q model. Experimental results show that the control accuracy of the proposedalgorithm is much less than1%. Compared with the original algorithm, PSNRand rate-distortion performance have improved, PSNR gain is up to0.8, so theproposed algorithm can not only achieve good objective quality, and can achievebetter subjective quality.(4). Rate control for multiview video coding based on statistical analysisand frame complexity estimation is proposed in this paper. The proposedalgorithm is performed on three levels, namely view level, GOP level and framelevel. In the view level, the target rates are allocated according to different typesof views with a pre-statistical rate allocation proportion. In the GOP level, thetotal number of bits allocated to each GOP is computed and the initial QP ofeach GOP considering the QP values of B frames is set. In the frame level, theproduct of encoded bits and QP indicates the image complexity of the currentframe, and the frame complexity is used to regulate the target bits for each frame.Experimental results show that the proposed algorithm can get high rate controlaccuracy and absolute error is less than0.42%, while the fluctuations ofaccuracy can achieve stability in a relatively short period time and PSNR is alsoachieved good results.