基于双目视觉特性的立体视频编码码率控制算法  

Rate control algorithm for stereoscopic video coding based on binocular visual characteristics

作　　者：蒋刚毅[1,2] 廖义[1] 郁梅[1] 邵枫[1] 王晓东[1] 李福翠[1] Yo-Sung Ho 

机构地区：[1]宁波大学信息科学与工程学院,浙江宁波315211 [2]南京大学计算机软件新技术国家重点实验室,江苏南京210093 [3]School of Information and Communications,Gwangju Institute of Science and Technology

出　　处：《光学精密工程》2014年第2期451-458,共8页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.61271270;No.U1301257;No.61271021;61311140262);科技部国家科技支撑计划资助项目(No.2012BAH67F01)

摘　　要：为了使立体视频中的比特分配更加符合人眼视觉感知特性,提出了一种非对称质量的立体视频编码码率控制算法。首先,建立了左右帧的码率分配比例与量化参数差值之间的立体指数RRQ(Rate-ratio Quantization)模型。然后,将码率控制算法分为SGOP(Stereoscopic Group of Pictures)层、立体图像对层和帧层等3个码率控制层。在SGOP层计算每个SGOP的目标码率和关键帧的量化参数;在立体图像对层根据剩余比特数和缓冲区饱和度计算每个立体图像对的目标比特;在帧层则通过分析双目视觉掩蔽效应,用一种适合于立体视频的率失真优化方法合理分配左右帧的目标码率。实验结果表明,本文算法的码率控制偏差平均值为0.21%;立体视频客观质量比对称质量算法和Wang的算法分别提高了0.23dB和0.06dB,且质量波动较为稳定。因此,该算法基本满足网络带宽传输要求。由于充分利用了人眼双目视觉特性,可满足人们对立体视频的视觉需求。To allocate bitrates more satisfactory with human visual characteristics,a new rate control algorithm was proposed for an asymmetric quality stereoscopic video coding.Firstly,an exponent model（Rate-ratio Quantization,RRQ） between the rate ratio and the difference between quantization parameters of left and right frames was established.Then,the algorithm was performed on three levels,namely Stereoscopic Group of Pictures （SGOP） level,stereoscopic image pair level and frame level.The target bits for each SGOP and the quantization parameter of each key frame were calculated in the SGOP level,the total target bits for each stereoscopic image pair were calculated according to the remaining bits in the SGOP and the current buffer occupancy in the stereoscopic image pair level,and by utilizing the binocular masking,a stereoscopic rate distortion model was used to allocate bitrates for the left and right frames more rationally in the frame level.Experimental results show that the proposed algorithm can control the bitrates error within 0.21% on average.Meanwhile,it has improved the stereoscopic video objective quality by 0.23 dB and 0.06 dB as compared with the symmetric quality algorithm and Wang＇s algorithm,respectively,so that it can provide more stable video quality fluctuation.In conclusion,the proposed algorithm has effective network transmission bandwidth,and can satisfy the requirements of human for stereoscopic video by utilizing the human binocular visual characteristics.

关 键 词：立体视频 码率控制 双目视觉冗余 非对称质量 

分 类 号：TN919.8[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]