Research of combination of fractals and neural network in image recognition

Abstract

This paper presents a method of image intelligent recognition with the combination of the fractals and the neural network. The values of the fractal dimensions of images are found with some fractal algorithms such as box dimension, correlating dimension, R/S analysis and power spectrum analysis and then the learning data set and the testing data set for the neural network are comprised with the fractal dimensions of images. Every network model is a 3-layer BP network. The all models of the neural network are constructed with the training images and their performances are tested with the testing images that are processed with different methods. The statistic results indicate that the high precision of the image recognition can be obtained when the images are made Fourier transformed after the median filtering.