Extraction of Discriminant Features from Image Fractal Encoding

In this paper we face the problem of finding characteristic information about images of different objects, showing that the fractal encoding based on Iterated Function Systems, besides allowing very high compression rates, can be successfully applied also for capturing discriminatory features that can be exploited for non-fractal image classification. An original feature extraction algorithm was developed and applied to encode the hand-written digits data set. Then, different learning algorithms were applied and their performances were compared both to those obtained using a general purpose fractal encoder (enc by Fisher) and to the work done in the StatLog project on the same data set.