An image distance measure insensitive to amplitude and mean value variations. Application to data reduction through SVD

Abstract

The traditional m.s.e. distance evaluation, used commonly as a measure of degradation of images (or signals in general) through a given process, is very much affected by any linear amplification or mean value variation in the image. The significative contents of an image doesn't change with any one of the mentioned operations, so a fair distance measure should be independent from their effects. This paper describes a distance measure that fulfill this condition, and uses it to evaluate the evolution of the degradation during the Singular Values Dyadic Decomposition of some given discrete images represented by matrices. Results are plotted and discussed.