Abstract

An optical system using rectangular-to-polar coordinate transformation can be used for rotation invariant pattern recognition and classification. We have constructed an optical device to accomplish the coordinate transformation. The output of this device is imaged onto a liquid crystal light valve and Fourier transformed, then sent to a neural network and compared to stored invariant values. The coordinate transformation system uses a nonregular sampling array to minimize effects caused by the nonregular sampling resulting from a rectangular to polar transformation.

© 1991 Optical Society of America