Abstract
Human long-term memory for visual stimuli is substantial; however, it is still not entirely clear whether images are remembered on an image-based, perceptual level or on a more abstract, conceptual level. To disentangle the contribution of both components, we used a memory test paradigm in which we varied the similarity between target image and distractor. For each target image, three distractors were chosen. One (called “similar”) was conceptually and visually similar, one (called “gist”) only had the same gist, and one distractor (called “rand) neither shared the gist nor visual details. The image sets were validated by showing participants three images at a time and having them pick the one they considered as most dissimilar to the other two. 45 item sets were kept in which the random image was consistently chosen as least similar to the target, followed by the image sharing the gist. In the main study, 21 participants viewed a sequence of 184 images, each for one second. In the following test phase, participants performed a recognition task on 45 item pairs, each consisting of a target image and one of the three distractors. We analyzed recognition performance across the distractor classes. It was highest when a target image had been paired with a random distractor and lowest when paired with the distractor from the category “similar”. Significantly more errors were committed when the distractor image was most similar to the target image. After two weeks, six participants were tested again but on different target - distractor pairings. Performance remained virtually the same, both in overall performance and distribution across distractor classes. Our results show that long-term memory for natural stimuli is extensive and that visual features play an important role in recognition.