The tale of a simple accurate MPEG video traffic model

This paper traces the development/evolution of three of our previously proposed MPEG video traffic models, that can capture the statistical properties of MPEG video data. The basic ideas behind these models are to decompose an MPEG compressed video sequence into several parts according to motion/scene complexity or data structure. Each part is described with a self-similar process. These different self-similar processes are then combined to form the respective models. In addition, the Beta distribution is used to characterize the marginal cumulative distribution (CDF) of the self-similar processes. Comparison among the three models shows that the latest model (called the simple models) is the most practical one in terms of accuracy and complexity. Simulations based on a real MPEG compressed movie sequence of Star Wars have demonstrated that the simple model can capture the ACF and the marginal CDF very closely.