Digital multiple correlation for pattern recognition

Rotation-invariant digital pattern recognition using circular harmonic expansion

A new method has been developed for rotation-invariant pattern recognition. One component of the circular harmonic expansion of the target is used in the preparation of the reference. Correlations between the input and reference objects are accomplished by FFT and multiplication in the frequency domain. In an experience with targets from an image with 192 x 192 pixels, target orientations were detected with an accuracy of approximately 0.1 degrees . This method is also suitable for optical implementation.

Optical pattern recognition using circular harmonic expansion

Presented in this paper is an optical implementation of a shift- and rotation-invariant pattern recognition technique. A computer-generated hologram is designed to match with one of the circular harmonic components of the target. Experimental results with simple photographic objects show that targets with different locations and orientations can be simultaneously recognized by the optical system.

Incoherent method for rotation-invariant recognition

An optodigital hybrid system using an incoherent circular correlator that has features suitable for space and industrial applications is introduced. The incoherent circular correlator uses a fiber-optics sampler and carries out the correlations in white light. The trade offs involved for efficient detection of objects are discussed. The system has been built, and experimental results on a real air photograph are presented.