Rate Control Algorithm For3D Video Coding Based On Visual Perception

Three dimensional (3D) video can provide viewers immersive sense of realism, however, the huge data putsforward high request to the bandwidth and storage, how to improve the perceived quality at the certain bits andcontrolthe bitstream to meet the complextransmission network are key problems in the development of3D video,rate control is an effective way to resolve the aforementioned problems. Since the3D video can be represented inthe form of stereoscopic video, multi-view video or multi-view video plus depth, the bits allocation among theviewpoints should be more considered in the rate control for3D video. Meanwhile, the evaluation approach ofPeak Signal to Noise Ratio (PSNR) in traditional3D video rate control methods does not well reflect the human’ssubjective quality.Thisthesis studiesthe rate control methodsfor3D video and the characteristicsof human visualperception,andthen proposesthefollowing threeratecontrolalgorithms:(1) To address two critical aspects of the traditional rate control method that are inaccurate estimation forstereoscopic image pairs complexity invoked by complicated change of scenes and disregarding human’sensitivity in the bits allocation for macro-blocks, a rate control method for stereoscopic video based on justperceptible characteristic is proposed. Firstly, the movement characteristic of macro-block is utilized to measurethe complexity of stereoscopic image pairs, based on which we allocate bits for stereoscopic image pairs.Secondly, a stereoscopic Just Noticeable Distortion (JND) model combined with the depth characteristics isestablished, and then the stereoscopic JND for macro-block is used as a visual sensitive factor to optimize the bitsallocationin the macro-blocklayer,thehighersensitivity ofmacro-blocksare,the morebits forthemare allocated.Experimental results show that the proposed method decreases the average rate control error by0.12%comparedwith that of the traditional rate control method, and it can improve the subjective quality of the visual sensitiveareaswithoutanydeclineonthe ratedistortionperformance.(2) To address the drawbacks of the existing rate control methods failing to utilize the binocular visionredundancy, a rate control method for stereoscopic video based on binocular visual masking effects is proposed.We firstly establish a Rate Ratio and Quantization (RRQ) model between the left and right frames by means ofstatistics and analysis, and the bits allocation for the left and right frames is optimized based on the RRQ modeland binocular masking threshold. Experimental results show that the proposed method can control the bitrateserror within0.21%, and increase the stereoscopic video objective quality by0.23dB and0.06dB compared withthat of symmetric quality method and Wang’s method, respectively, meanwhile, it can provide more stable videoqualityfluctuation.(3) To address that the rate control method in the multi-view video coding does not precisely predict theframe complexity and consider the human visual perception characteristics in the Rate Distortion Model (RDM).Firstly, the bitrates and PSNR of the two reference frames are utilized to measure the complexity of the currentframe in the frame layer bits allocation. Secondly, different macro-blocks are assigned different weights accordingto the foreground and motion features in order to optimize RDM in the macro-block layer. Experimental resultsshow that the proposed method decreases the rate control error by0.15%compared with that of JMVC-RCmethod,andmeanwhileitincreasesthePSNRofvisualimportantareasby0.12~0.54dB.