Abstract
Purpose: We investigated the human visual memory for faces, and evaluated the effects of perceptual similarity on visual memory performance with varying sets of study and test items. By using realistic, computer synthesized faces as stimuli, we could systematically vary the perceptual similarity among the items to be remembered, thereby gauging inter-item similarity's effect on visual memory. Methods: In Experiment 1, Sternberg's recognition memory paradigm was applied to a set of 21 synthesized faces. On each trial, from 1 to 4, briefly presented Study faces were followed by a single Probe face. Subjects indicated whether the Probe had or had not been among the Study faces. To force reliance on episodic memory, Study and Probe items varied from trial to trial. In Experiment 2, the method of triads, followed by multidimensional scaling (MDS), was used to characterize subjects' perceptual similarity space for the faces. Results: Experiment 1 showed that recognition memory was strongly influenced by the number of faces comprising a study set, and by the recency of a face's occurrence on a trial. Expressing differences among faces in terms of distances derived from the MDS similarity space, we found that perceptual similarity among faces accounted for much of the variance in recognition memory performance. Finally, between-subject differences in the face-similarity space were relatively small. Conclusion: Inter-item similarity has powerful effects on episodic memory. We applied further analysis to examine the effects of similarity between the probe and the most similar lure, the similarity among all lures, and the similarity between the probe and all other lures. The fitness of visual memory models will be discussed in light of these results.