Fractal transforms and feature invariance

In this paper, fractal transforms are employed with the aim of image recognition. It is known that such transforms are highly sensitive to distortions like a small shift of an image. However, by using features based on statistics kept during the actual decomposition we can derive features from fractal transforms, which are invariant to perturbations like rotation, translation, folding or contrast scaling. Further, we introduce a feature invariance measure, which reveals the degree of invariance of a feature with respect to a database. The features and the way their invariance is measured, appear well suited for the application to images of textures.