Abstract

The potential of optical processing techniques for high speed pattern recognition and vision systems was demonstrated over thirty years ago by Vanderlugt [1]. In his early experiments Vanderlugt prepared linear filters with transfer characteristics matched to an object of interest using holographic techniques. Once inserted into a coherent optical processing system these elements could be used to interrogate an input transparency, clearly marking each occurrence of a given object. Since this type of filter is matched to a single signal, however, it is sensitive to small changes in structure, scale and rotation which occur, in general, if the system is required to identify a particular class of object. In order to alleviate this problem, computer synthesis has facilitated the fabrication of other linear filters capable of classifying objects according to rules obtained from sets of training images [2].

© 1996 Optical Society of America