AIRVL     Research Projects: Image Processing

[ AIRVL Home] [ People ] [Research] [ Facilities]

In this work, the issues of time and compression efficiency in image transformation and compression are addressed. The work is focused on the Gabor image transformation, and two new methods for enhancing the efficacy of the image transformation process have been presented.

The first method is an efficient Gabor-QR decomposition scheme for computing the transform coefficients. The Gabor-QR decomposition is a parallel matrix-based method that computes the exact coefficients without a time-consuming iterative process as required by most other Gabor Transform algorithms. Furthermore, the Gabor-QR decomposition subdivides the whole transformation process into a one-time pre-processing step and an efficient transformation step for computing the coefficients. The proposed algorithm allows multiple images to be transformed efficiently, bypassing the time-consuming pre-processing step.

In addition, the Maximum-Variance Basis Selection scheme for constructing the incomplete basis sets for the transformation has been developed by Papanikolopoulos' students. The proposed basis selection scheme exploits the image statistics and the energy compacting characteristic of the transformation process. The resulting incomplete Gabor transform has greater time and compression efficiency than the complete transform. The utilization of image statistics is particularly important in the compression of images that have dominant high-frequency features and/or peculiar statistical structures. Most transform-based compression systems operate based on the assumption that the input images have mostly low-frequency features. These compression systems will not perform efficiently in some image domains in which the aforementioned assumption does not hold. It has been shown that the proposed basis selection scheme has the capability of capturing dominant image features across the entire frequency plane and produces better results than common transform-based compression techniques such as JPEG.

Copyright: © 1998-2001 by the Regents of the University of Minnesota
Department of Computer Science and Engineering. All rights reserved.
The University of Minnesota is an equal opportunity educator and employer.