Abstract
The human visual memory system can store a remarkable amount of information in long-term memory. However, it appears to be the peculiar fate of memories that they must fade. Much work has examined the structure of forgetting for general classes of stimuli, but an item analysis of individual visual stimuli is lacking. Do all images fade alike? Recent work has shown that there are large differences between the memorabilities of different images, and these differences are quite consistent across context and observer, suggesting that the differences are intrinsic to the images themselves. Here we ask: could these differences arise from different rates of forgetting, or are the differences stable over time? We ran a Visual Memory Game on Amazon’s Mechanical Turk in which participants viewed a sequence of images and indicated whenever they noticed a repeat. Each image repeated at most once for each participant, appearing at one of three possible delays: ~15 images back, ~100 backs, and ~1000 backs. We measured the memorability of each image at each delay as the proportion of times a repeat of the image at that delay was correctly detected. Strikingly, even after the shortest delay (10-20 images back; i.e. 24-48 seconds back), there were large memorability differences between the images, and these differences were remarkably similar to those at both longer delays (that is, image memorabilities at one delay correlated strongly with those at the other delays: r = 0.63, 0.68, and 0.55 for the three pairwise comparisons). Thus, it appears that the memorability differences we measured did not emerge slowly over time, but were already stable shortly after encoding. This suggests that differences in intrinsic image memorability reflect differences in how images are encoded and retrieved; in storage, all images may intrinsically fade alike.
Meeting abstract presented at VSS 2012