分布式编码中广义伽马分布相关噪声模型研究

Research on correlation noise modeling of a generalized Gamma distribution in distributed coding

针对现有相关噪声模型—Laplacian模型不能精确描述相关噪声,导致分布式视频编码(DVC,distributed video coding)系统的率失真性能不高的问题,提出像素域DVC中广义伽马分布相关噪声模型。首先分析了相关噪声的统计特性,发现Laplacian分布的峰值比实际相关噪声分布的低,然后采用广义伽马分布对相关噪声进行拟合,并给出在线估计广义伽马分布参数的方法。实验结果表明,提出的模型能较精确地描述相关噪声,且有效地改善了系统的率失真性能,并减少了解码端计算复杂度。

Due to the inaccuracy of modeling the correlated noise（CN） with Laplacian distribution, the rate-distortion performance of DVC is relatively low. CN modeling via generalized Gamma（GGA） distribution was proposed for pixel-domain DVC. Statistical features of the CN modeling were analyzed. It was found that Laplacian distribution peak was lower than actual CN distribution. GGA distribution was exploited to fit CN and the GGA distribution parameters were estimated online. The simulation results indicate that the proposed model can more accurately reflect the statistics features of CN, improve rate distortion performance, and reduce computation complexity.