Abstract

A real-value implementation of circular harmonic expansion by a joint transformation correlation scheme to achieve rotation-invariant pattern recognition is presented. Both the real and imaginary parts of one component of the circular harmonic expansion of a template are used as reference functions. Based on the measurements of the two peaks of the correlation between input object and the two reference functions, rotation-invariant patten recognition is achieved. The rotation angle of the input object with respect to the template is recovered. By using a ¼ waveplate, the rotation invariance based on only one correlation peak is obtained while sacrificing shift invariance. The proper circular harmonic expansion center plays an important role in the circular harmonic expansion-based rotation-invariant pattern recognition. An efficient and effective way of finding the proper center is discussed. The ability of the reference functions to discriminate the object which contains the template from other objects is analyzed. Experimental results are included.

© 1988 Optical Society of America