Abstract
When studying visual memory for complex scenes, previous research has shown that humans can recognize a large number of previously presented photographs. However, what do we actually remember of each scene? Typical designs present participants with a large set of stimuli and query each scene only once. This approach limits the possibility to measure which individual details are retained in the memory. In our novel paradigm (denoted as intensive memory), we presented participants (N=33) with only three images and they studied each image intensively for three minutes. After learning, we queried each scene 21 times in a 2AFC manner showing two images (one was the original one, the other one contained one photoshopped change) with only a small area showing the original image. The rest of the image was desaturated and blurred to reduce the effect of relearning. This task was counterbalanced within-subject with a typical visual memory task (denoted as extensive memory) in which participants memorized 104 scenes and were tested in 2AFC with 52 pairs of scenes. The accuracy in both tasks correlated (r = .39). In extensive memory, participants reached high accuracy (mean=.89, SD = .08) compared to the intensive memory condition (mean= .64, SD = .08). Further rating of the changes showed that removing an object and changing color were the easiest to recognize and the size of the change was not related to recognition accuracy. Additionally, the difficulty of the task was not related to the perceptual difficulty to detect the change when the original and modified image were presented together. These results show that although it is easy to correctly mark a scene as previously seen in a recognition task, the amount of information people store about each picture is considerably lower, making judgements about scene parts difficult.