Multi-object intensity-invariant pattern recognition with an optimal processor for correlated noise

Normalized correlation provides a way to achieve reliable pattern recognition with images containing multiple target objects of unequal intensities without the need of image segmentation. We show that the optimum Bayesian processor for the detection of a target with additive correlated noise and disjoint background, introduced, has the form of the normalized correlation. In consequence it can be expressed with correlations and pointwise processing only--which is a condition for an efficient optical implementation. Moreover it may be applied to multi-object intensity invariant problems.