This thesis investigates the application of discrete-time statistically self-similar (DTSS) systems to modeling of variable bit rate (VBR) video traffic data. The work is motivated by the fact that while VBR video has been characterized as self-similar by various researchers, models based on self-similarity considerations have not been previously studied. Given the relationship between self-similarity and long-range dependence the potential for using DTSS model in applications involving modeling of VBR MPEG video traffic data is presented. This thesis initially explores the characteristic properties of the model and then establishes relationships between the discrete-time self-similar model and fractional order transfer function systems. Using white noise as the input, the modeling approach is presented using least-square fitting technique of the output autocorrelations to the correlations of various VBR video trace sequences. This measure is used to compare the model performance with the performance of other existing models such as Markovian, long-range dependent and M/G/(infinity) . The study shows that using heavy-tailed inputs the output of these models can be used to match both the scene time-series correlations as well as scene density functions. Furthermore, the discrete-time self-similar model is applied to scene classification in VBR MPEG video to provide a demonstration of potential application of discrete-time self-similar models in modeling self-similar and long-range dependent data. Simulation results have shown that the proposed modeling technique is indeed a better approach than several earlier approaches and finds application is areas such as automatic scene classification, estimation of motion intensity and metadata generation for MPEG-7 applications.
Library of Congress Subject Headings
Digital video; Video compression; Digital communications; Markov processes
Department, Program, or Center
Electrical Engineering (KGCOE)
de Queiroz, Ricardo
Narasimha, Rajesh, "Variable bit rate video time-series and scene modeling using discrete-time statistically self-similar systems" (2002). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus