Abstract

An iterative technique can be used to synthesize pattern recognition filters that are invariant to image distortions. These distortions might be position (location), rotation, scale, and perspective. The image to detect may also appear at different intensities. In this paper, we present an iterative technique based on spectral iteration in angular Fourier space for the synthesis of rotationally, translationally, and intensity invariant optical pattern recognition filters. The iterative method allows full image information to be retained while simultaneously obtaining invariance to rotation and image location. The general method presented may be extended to synthesize filters with invariances to additional distortions.

© 1986 Optical Society of America