Objective: To develop image compression systems using perceptual criteria and sub-band/wavelet transform vector quantization (VQ) and parallel algorithms for implementing the data compression systems on a MasPar MP-2.
Approach: Our approach involves developing a new model of perceptual masking functions for predicting the amount of distortion present in a given image that is not visible to the human viewers and then designing data compression systems that minimize distortions larger than those predicted by the model. We used a combination of sub-band/wavelet transform coding and VQ in our work. The algorithms for VQ and sub-band coding, as well as the new model for perceptual masking function, were implemented on the MP-2 system at Goddard Space Flight Center.
Accomplishments: (1) Developed a new model for the perceptual masking function that showed that larger amounts of distortion than previously thought possible can be hidden from the human viewers. (2) Developed a method to design filter banks that are tuned to the statistics of the input images in order to get the maximum performance out of the sub-band coders. (3) Developed a VQ algorithm that attempts to constrain the quantization errors to below specified threshold values as much as possible. (4) Developed very efficient algorithms for VQ on the MP-2.
Significance: Our algorithms provide better visual quality for compressed images at a given bit rate when compared with an image compression system that does not utilize our model of the perceptual threshold function. Applications that benefit from our work include browsing through large databases and easier access to databases over the Internet.
Status/Plans: Refine the components described in the Accomplishments and complete integration of the various components into the data compression system.
Point of Contact:
V. John Mathews
Department of Electrical Engineering
University of Utah
mathews@ee.utah.edu
801-581-5281
URL: http://ee.utah.edu/~mathews/hpcc.html