基于合成视点感知失真模型的三维视频编码率失真优化及码率分配研究

3D video coding is aimed to efficiently store and transmit the large amounts of data for 3D video system. Rate-distortion optimization and bit allocation between texture and depth are two critical technology of 3D video coding. Currently, the works mainly utilize the traditional pixel based signal distortion metrics to solve the rate-distortion optimization and bit allocation problem. However, the previous studies proved that this kind of distortion metrics could not reflect the subject perceptual distortion of the human eyes. This project tries to study the rate-distortion optimization and bit allocation problem by using perceptual distortion metrics. First, the structural similarity distortion metric model of the synthesized viewpoint is built by using stationary random distribution analysis and autoregressive algorithm. Then, in order to increase the rate-distortion performance and the synthesis subjective quality of the 3D video coding, the rate-distortion optimization and bit allocation problem are solved by the above distortion metric model and Lagrange multiplier method. Finally, via the high correlation between the viewpoints, the just noticeable distortion model of the synthesized viewpoint which is proposed in the previous work is optimized to reduce the computational complexity. Also, it could be integrated into the rate-distortion optimization process of the depth map coding to significantly improve the coding efficiency and the synthesis subjective quality.

三维视频编码旨在解决三维视频系统巨大数据量的有效传输和存储问题。其中，率失真优化及纹理与深度间的码率分配是三维视频编码的两大关键技术。目前针对这两大关键技术的研究主要采用传统的基于像素点的信号失真度量，而研究证明基于像素点的失真并不能很好地反映人眼对视频主观失真的感知。本课题拟采用基于视觉感知特性的失真度量模型对三维视频编码的率失真优化及码率分配问题进行深入研究。首先，通过对三维视频虚拟视点合成过程的分析，采用平稳随机分布近似及自回归分析算法建立合成视点的结构相似性感知失真模型；其次，利用该模型及Lagrange乘数法求解三维视频编码的率失真优化及码率分配问题，以提高三维视频编码的率失真性能及合成主观质量；最后，利用视点间的高度相关性对先前工作中提出的合成视点的恰可识别感知失真模型进行优化，以降低其计算复杂度，同时利用该失真模型指导深度编码的率失真优化过程，以提高其编码效率和合成主观质量。

三维视频编码旨在解决三维视频系统巨大数据量的有效传输和存储问题。其中，率失真优化及纹理与深度间的码率分配是三维视频编码的两大关键技术。目前针对这两大关键技术的研究主要采用传统的基于像素点的信号失真度量，而研究证明基于像素点的失真并不能很好地反映人眼对视频主观失真的感知。本课题采用基于视觉感知特性的失真度量模型对三维视频编码的率失真优化及码率分配问题进行深入研究。首先，通过对三维视频虚拟视点合成过程的分析，采用平稳随机分布近似及自回归分析算法建立合成视点的结构相似性感知失真模型；其次，利用该模型及Lagrange乘数法求解三维视频编码的率失真优化及码率分配问题，以提高三维视频编码的率失真性能及合成主观质量；最后，利用视点间的高度相关性对先前工作中提出的合成视点的恰可识别感知失真模型进行优化，以降低其计算复杂度，同时利用该失真模型指导深度编码的率失真优化过程，以提高其编码效率和合成主观质量。