Advisor: Prof. David G. Daut
Department: Electrical and Computer Engineering. Rutgers, the State University of New Jersey, Piscataway, New Jersey, USA.
Date of Graduation: January 1998.
Rate-distortion function (RDF) for the nonstationary Gaussian autoregressive (AR) process and nonstationary digital image statistical models for image data compression and image transmission are investigated theoretically and experimentally.
In this study, we first investigate the RDF for the nonstationary Gaussian processes. This rate-distortion measure servedsd as a guide to develop an adaptive transform coding system. For computational simplicity, researchers have developed various wide-sense stationary image models to approximate real-world images. However, in order to accurately estimate a real-world image, it should be treated as a random field having nonstationary statistics. Based on knowledge of the time-varying Gaussian AR process, we have developed two nonstationary image models. Using the rate-distortion bound for the nonstationary Gaussian AR process developed in this study, we have examined the system performance obtained when transmitting a real-world image (which we assume to be described by the proposed image models) by an appropriate adaptive compression algorithm design.
The rate-distortion measure also serve as an ultimate performance bound on actual system performance. The proposed nonstationary image models are related to the design of this adaptive transform coding scheme. Coded images are then transmitted over realistic communication channels. Overall performance for this image transmission system based on nonstationary image modeling methods is studied.