A Neural Network For Detecting Fractals In Spatial Patterns

Abstract

In this paper, a neural network approach for classifying given spatial patterns into fractals or non-fractals as presented. The proposed network is a hierarchically organized, multi-layer feedforward architecture which exploits the structural properties of artificially generated fractals. The backpropagation algorithm is employed for training the network. The network is implemented in parallel on a multi-transputer system. It is able to classify all non-noisy test patterns correctly; fractal patterns where 1/ f and random noise are added take longer for the network to converge, but up to a certain signal-to-noise ratio they are also classified correctly.