Abstract

Recently Fourier transforms of fractional order have drawn special attention with their unique features in both spatial and frequency domains.^1,2 By extending Fourier transforms into fractional orders in Vander Lugt correlators, a new type of filter was proposed to incorporate both the frequency and position-dependent filtering, and its analogy to neural networks was developed.³ Although neural network implementations have been limited by the availability of high resolution optical devices,⁴ by virtue of the simple optical architectures for the fractional Fourier transforms, the new neural networks is easy to implement in large-scale. In this paper we report learning and recognition characteristics of the new neural networks.

© 1995 IEEE