XFF: A simple method to eXtract fractal features for 2D object recognition

Abstract

Automatic recognition of objects from their visual representation is a hard and computationally expensive task, mainly because it is difficult to extract the necessary discriminant information from the raw data. The current approaches to image classification commonly exploit some geometrical models of the objects of interest. The classification process is based on the comparison between the image at hand and the models for the classes: the object’s class will be the one of the closest model. In this paper we present XFF, a new method for representing 2-D images, based on the extraction of a set of fractal features which exploits the approximation of an image with an Iterated Function System, a technique that is already at the basis of many successful image compression tools. One of the advantages of these features is that they can be used directly to train an adaptive classifier, without the need of any a priori knowledge.