Image identification using the segmented Fourier transform and competitive training in the HAVNET neural network

Abstract

An optical modeless image identification algorithm is presented. The system uses the HAusdorff-Voronoi NETwork (HAVNET), an artificial neural network designed for two-dimensional binary pattern recognition. A detailed review of the architecture, the learning equations, and the recognition equations for the HAVNET network are presented. Competitive learning has been implemented in training the network using a nearest-neighbor technique. The image identification system presented in this paper is applied to two tasks: the optical recognition of a set of American sign language signals and identification of grayscale fingerprints. Image preprocessing includes edge enhancement by histogram equalization, application of a Laplacian filter and thresholding. A segmented Hankel and Fourier transformation in polar coordinates is applied to the binary image giving a rotationally and translationally invariant image structure. This preprocessed image employs the HAVNET neural network for successful image identification.