NONLINEAR PREPROCESSING FOR IMPROVED FILTER CAPACITY OF AN OPTICAL CORRELATOR

Optical correlation is an efficient parallel implementation of image c lassifiers. One formulation of the holographic filter in the optical c orrelator is the least squares discriminant function (LSDF), wherein t he shape of the whole correlation function is specified when training the filter. An issue of particular interest is the capacity of the fil ter, or the number of training images that can be used while still ret aining discrimination capability. To investigate the capacity of an LS DF filter, and to be more general than lust testing on a particular ty pe of image, the fractal dimension of the images was used as an object ive complexity measure of the images that the filter is trained with. Nonlinear preprocessing, specifically edge detection, was used to incr ease the filter capacity and the discriminatory capabilities of the co rrelator. Simulation experiments of these issues are described and the results are discussed. (C) 1996 John Wiley & Sons, Inc.