Abstract

It has been recently demonstrated¹ that fast, reliable image recognition can be accomplished using a commercially available 2-D photon counting detector and position computing electronics. In this paper the method of circular harmonic function expansion for coherent rotation-invariant image recognition² is extended to the case of photon-limited image recognition. Theory for rotation-invariant filtering with incoherent illumination is presented and applied to photon-limited image recognition. Low light level input images are cross correlated with the square modulus of a single circular harmonic component of a high light level reference image stored in computer memory. The mean value of the correlation signal is found to be invariant with respect to rotation of the input image. The experimental results for the correlation signals for various input images are presented. Histograms of the correlation signal are shown and compared with theoretical predictions for the probability density functions. It is demonstrated that reliable image recognition, independent of the rotational orientation of the input, is possible with as few as 5000 detected photoevents.

© 1985 Optical Society of America