Image-Based Model for Visual Search and Target Acquisition

M. Snorrason, H. Ruda (Charles River Analytics), and J. Hoffman (Department of Psychology, University of Delaware)

Proceedings of Ground Target Modeling and Validation '99, Houghton, MI (August, 1999)

The overall objective of this project is to develop a computational model for predicting probability of detection during search for hard-to-see targets. This model is image based: it uses imagery for input, rather than estimated parameter values characterizing critical factors such as clutter and target detectability. Consequently, it generates probability-of-detection values that are functions of image content, rather than functions of subjectively estimated parameters. The input domain is infrared or visible-light imagery of distant vehicle targets in cluttered scenes. Such hard-to-see targets are generally only detected once they have been fixated. Hence, our modeling approach focuses primarily on factors influencing the choice of fixation points during visual search. A saliency map is constructed from bottom-up image features, such as local contrast. To account for top-down cognitive effects—such as bias towards the horizon—a separate cognitive bias map is generated. The combination of these two maps provides a Fixation Probability Map (FPM). Given the FPM, a sequence of fixation points is generated in a way that accounts for imperfect memory of past fixation locations. Results are presented comparing model-generated FPM’s with eye-tracker data collected from observers in visual search experiments.

Predicting fixation points from saliency and horizon bias

Request a Copy