Abstract
Background: Recent computational models of human visual search presuppose that the visual system has access to accurate estimates of visual detection ability for stimuli at different retinal locations (E.g. Najemnik & Geisler, Nature, 2005). To test this assumption, we designed a decision task that revealed subjects' estimates of their performance for different contrast levels at different retinal eccentricities. Methods: In a calibration session, we mapped the subject's probability of correct response as a Weibull function of retinal eccentricity for targets at each of three contrast levels (low, medium, high). The subjects also learned to associate a color symbol with each contrast level. In the subsequent decision part, we asked subjects to choose between two possible combinations of contrast and eccentricity. E.g. low contrast at 2 degrees or high at 10 degrees. They knew they would actually attempt some of their preferred choices at the end of the experiment and earn monetary rewards for correct responses. We used a staircase procedure to measure the point of subjective indifference between targets that differed in contrast, one fixed in eccentricity and the other varied in eccentricity. For each subject we used 12 such staircases (3 contrasts x 4 probabilities) and estimated the eccentricity that the subject considered to be equally detectible for the variable contrast. Eight naïve subjects participated. Results: Despite their calibration experience, all eight observers matched probabilities incorrectly, with 0.14 mean error over all observers, conditions (theoretical upper limit 0.35). The matching failures showed a common pattern of underestimating the difference between high and low contrasts. Conclusion: Observers exhibited little knowledge of their visual detection ability as a function of contrast and retinal eccentricity. We find no evidence that the visual system has access to accurate estimates of detection ability for different types of targets at different eccentricities.