Shape recognition using FFT preprocessing and neural network

Abstract

The paper presents the application of neural network to the classification of the closed contours forming different shapes. The shape is represented by ‐ samples of complex numbers z_k = x_k + jy_k where x_k and y_k are the samples in the x‐y plane and j is the complex operator. The same shapes may vary in scale, be rotated and translated in arbitrary proportion and be distorted by the noise. To obtain the classification invariant to all these factors the preprocessing techniques based on the application of Fourier transformation of the samples have been applied. The Fourier coefficients form the input data to the neural classifier. Different shapes have been checked in numerical experiments and the results have proved good performance of the developed neural classifier and its relative insensitivity to the noise.

Citation

Nghia Do, D. and Osowski, S. (1998), "Shape recognition using FFT preprocessing and neural network", COMPEL - The international journal for computation and mathematics in electrical and electronic engineering, Vol. 17 No. 5, pp. 658-666. https://doi.org/10.1108/03321649810221017