ABSTRACT
A distortion-invariant filtering algorithm permits the use of a smaller set of filters to detect particular targets when the exact size, orientation, or illumination is unknown. Although many filtering algorithms have been proposed to detect targets in noise [1-16] this technique incorporates distortion invariance [17-20], the effects of environmental degradation, and the additive overlapping and nonoverlapping background noises in the design of the optimal filter using the minimum mean-squared-error criterion [11,21-23]. The distortion-invariant filter has the advantage of detecting a target with a predefined set of distortions. It saves processing time because only one filter has to be used to detect a distorted target instead of going through the entire training set. However, its performance is lower than that of an optimal filter synthesized for one particular reference target aspect [4,9]. Environmental compensation is included in the filter design because of the adverse effects associated with atmospheric propagation [22,24,25].
