CODING NON-HOMOGENEOUS FRACTALS USING NEURAL NETWORKS

Abstract

In this work we formed a neural network to coding homogeneous iterated function system. Our approach to this problem consists of finding an error function which will be minimized when the network coded attractor is equal to the desired attractor. Firstly we start with a given fractal attractor; find a set of weights for the network, which will approximate the attractor. Secondly we compare the consequent image using this neural network with the original image, with the result of this comparison we can update the weight functions and the code of (IFS). A common metric or error function used to compare between the two image fractal attractors is the Hausdorff distance. The error function gets us good means to measurement the difference between the two images. The distance is calculated by finding the farthest point on each set relative to the other set and returning the maximum of these two distances.