Abstract
Purpose: We aim to test the class of appearance-based object recognition accounts that predict that an image identical to the previously seen necessarily gives rise to better recognition than does an image different from (and hence less similar to) the previously seen. We assumed that two identical images are more similar to each other than two different images; and that the more similar an image is to a representation in memory, the more likely this image will be perceived as previously seen.
Method: In an old-new memory task, participants first rated the attractiveness of 30 grayscale images of natural scenes, with 50% of each image area occluded by randomly distributed red rectangles. Participants then rated how likely each of 60 images (30 old and 30 new) had been seen in the previous attractiveness rating. Ten of the occluded old images were identical to those seen previously. Ten was less occluded (40% of image area) and 10 least occluded (30%). Reduced occlusion was achieved by reducing the sizes of the red rectangles. Occlusion of the new images was identically manipulated. A separate experiment used faces as stimuli.
Results: For both natural scenes and faces, the less-occluded old images were rated as more likely to be old than the originally-seen occluded images (p ≤ 0.01). This result could not be attributed to bias favoring less occluded images, because sensitivity (when new images were considered) showed the same effect (p ≤ 0.05). The least-occluded old images were rated as less likely to be old than the originally-seen (p [[lt]] 0.01). These results indicate that a visual scene is not remembered like a snapshot for subsequent recognition. Rather, perceptual completion behind occlusion was apparent. Such completion appears to be limited in spatial extent, since further removing occlusion did not facilitate recognition of old images.
Supported by NSF grant BCS 0617628