Image databases, scale and fractal transforms

In contemporary image databases one finds many images with the same image content but perturbed by zooming, scaling, rotation etc. For the purpose of image recognition in such databases we employ features based on statistics stemming from fractal transforms of gray-scale images. We show how the features derived from these statistical aspects can be made invariant to zooming or rescaling. A feature invariance measure is defined and described. The method is especially suitable for images of textures. We produce numerical results which validate the approach.