A new information measure for natural images

Abstract

Although natural images are a very small subset of all images, the direct computation of their block densities is not possible. On the other hand, the success of some image processing methods, most particularly, fractal compression, indicates that they somehow are able to capture at least part of the natural image statistics. In this work we shall show how a concrete procedure, hash based fractal image compression, can be used to derive quite precise mean-and-variance normalized block statistics. We shall use them to define an image entropy measure and a an image representation and discuss their relationship with other widely used image information measures. © Springer-Verlag Berlin Heidelberg 2003.