考虑异种帧相关性的VBR MPEG视频源混合回归模型

Composite Regressive Model of VBR MPEG Video Sources which Considering the Correlations Between Different Types of Pictures

MPEG作为一种国际动态图像压缩标准，其应用越来越广泛。为在通信网络上有效地传输和控制MPEG可变比特率（VBR）数据流，第一个关键问题就是如何建立它的统计模型。目前已有的视频源横型都没有考虑不同类型帧之间的相关性，并不能很好地模拟VBRMPEG视频源。本文首次提出了一种考虑异种帧相关性的VBRMPEG视频源统计模型，混合回归（CR）模型，及其参数估计算法。实验结果表明，和传统的自回归（AR）模型相比，CR模型能更好地反映视频源的统计特性，CR模型产生的模拟MPEGVBR视频源对缓冲区排队系统的影响能够很好地符合真实视频源对缓冲区排队系统的影响。

Recently the MPEG coding technology received considerable interests as a international standard.To achieving the effective transmission and control of the MPEG variable bit rate video services,there are some problem to be conquered,the first one to be faced is how to modeling MPEG VBR video sources statistically. This paper presents a novel statistical model of MPEG VBR video sources,called Composite Regressive (CR)model,which take account the correlation between the different type pictures in the same group of pictures (GOP). Our expefiments show that CR model can give better statistical fitness of the characterizes of the MPEG VBR video sources than Autoregressive (AR)model and that the simulation sources generated by the proposal model have almost the same effects on buffer queuing system an the real MPEG VBR video sources.