On the detection of signals embedded in natural scenes

Abstract

In this paper we consider the processes by which observers can detect and recognize signals embedded in natural scenes and images. Although our results do not strongly support detection processes based strictly upon the cross-correlation (template-matching) of the signal and image luminance profiles, they do support a version of cross-correlation based upon the comparison of their "structural similarities." This latter correlational measure may be separated from the energy component in the cross-correlation function. Of particular importance to the structural similarity between the signal and image is their characteristic edge components, and we also show that recognition performance was consistent with the cross-correlations between the signal and image edge-only versions. This finding, again, exemplifies the importance, for the analysis of pattern detection and recognition, of isolating structural components from energy detection processes per se. © 1986 Psychonomic Society, Inc.