Abstract

In recent years, the use of joint-transform correlators for pattern recognition has become very popular, particularly because of improvements in spatial-light modulator (SLM) devices. These correlators exhibit shift-invariant pattern recognition but have high sensitivity to other parameters (scale, rotation, etc.). Harmonic expansion is one of the most promising wasy to add another invariant parameter to the already-existing shift invariance. For example, rotation invariance can be added by using a single circular harmonic as the reference object. Previously we presented an elegant way of allowing the use of joint-transform correlators for rotation-invariant pattern recognition. There, we used as reference object a single harmonic from the circular-harmonic decomposition, and we showed how recording of phase as well as amplitude information is achieved by encoding them on a grating (by means of Lohmann's phase-detour model). The input object, placed side by side with this reference, has to be multiplied by a grating too, so that the joint-transform correlation distribution can be obtained at the location of the first diffraction order. In this work we suggest a similar way to achieve projection-invariant (tilt invariant) pattern recognition in joint-transform correlators by using a single logarithmic harmonic as the reference image. In addition, based on the composite filter concept, we now present an equivalent approach, called "composite image," for use with project ion-in variant joint-transform correlators, thus extending the selectivity capabilities of this scheme. Some optical experimental results will also be presented.

© 1990 Optical Society of America