Use of the Informational Difference as a Target Conspicuity Measure

The Informational Difference (InDiff) is a measure of the difference between two image sets. Previous work has shown that it can be used to explain result of important target recognition experiments involving human observers. In this paper we present the results of investigations on the suitability of using the InDiff as a measure of target conspicuity. First, the InDiff is defined and adapted to measuring the difference between two images, one containing a target on a background and the other - containing only the background. Second, we present results of two experiments involving human observers: In one experiment, gray level images of complex scenes were presented to the observers; the second experiment involved color images. The response times for detecting and of recognizing targets in these images were measured and the InDiff values for the images were calculated. Correlation coefficients of 0.60 - 0.85 were found between the InDiff values and the following quantities: Detection speed in both experiments, recognition speed in the gray-level experiment. Significant relations were found between the probability of correct detection or recognition and a quantity based on the InDiff, in the gray-level experiment. Finally we discuss possible applications of these findings and suggest extension to the formalism.