Abstract

Techniques for performing digital image recognition on speckled images are presented. Rather than performing speckle reduction, the speckled image is correlated directly with a reference function. Statistical properties of the correlation value are derived using the multiplicative noise model for image plane speckle. It follows that the correlation value is a Gaussian random variable. Statistical detection theory dictates that images are distinguishable when the standard derivations of the correlation values are small compared to the separation of the mean values. Several reference functions are considered for both intensity correlation and correlation for which the input speckled images are clipped. One reference function considered is the incoherent image of the desired object: another is the maximum-likelihood reference function. Experiments performed on planar rough objects indicate good agreement between theory and experiment. The best ability to discriminate is obtained using the maximum-likelihood reference function.

© 1986 Optical Society of America