Optical pattern recognition based on normalized correlation

Abstract

We propose a method for optical correlation-based intensity invariant pattern recognition. Our approach relays on a normalization of the correlation signal applicable in conjunction with simple linear or nonlinear filtering of any type. The normalization is based on the Holders inequality if the input signal is segmented and on the Cauchy-Schwarz's inequality if more than one target object may be present in the input signal. We discuss the difference between our formula and the previously proposed simpler normalization, which is an approximation to the results obtained in the present paper. The performance ofthe normalized correlation with segmented and multiobject input is illustrated with numerical simulations.