Fractal image compression using competitive neural network in frequency domain

A new method for fractal image compression is presented. A combination between the idea of nearest neighbor search in the frequency domain and the clustering property of the competitive neural networks is used in this method. In this paper we use two methods for fractal image compression. In the first method (which is the goal of this paper), the DCT transformed range blocks are classified into clusters using a competitive neural network and the centers of clusters are used instead of domains to evaluate the coefficients vectors. In the second method, we implement the two dimensional discrete cosine transform (DCT) of the projected codebook blocks, ±F(D), and ranges, F(R), in order to search for nearest neighbor in the frequency domain as suggested by Barthel et al (IEEE Image Proc. Conf., pp. 112-116, 1994). Comparison of the techniques is conducted.