Pixel-Level Fusion Using 'Interest' Images

Abstract

A simple general-purpose means of representing image sensory data at the pixel level is presented in the context of target detection. One or more interest operators (i.e., feature detectors) are applied to one or more sensory images. The output of each operator is an interest image in which high pixel values are assigned to locations where features are found that are indicative of the object (or target) being sought. Any relevant sensory data that are transformed into interest images can then be fused in a pixel-wise manner by means as simple as computing averages, maxima or minima. The approach has been implemented in the detection module of an automatic target recognition system built at Lincoln Laboratory.