Abstract
The capacity of long-term memory (LTM) for pictures is outstanding: observers distinguish thousands of distinct pictures from foil exemplars after seeing each item only once (Standing, 1973; Brady et al., 2008). In contrast, change blindness shows that, even in short term memory, two versions of the same picture are difficult to distinguish when they differ by only a few objects. Clearly, there are limits to the resolution of visual LTM. Here, we investigated the fidelity of LTM by using foil images representing similar versions of a scene. During a learning phase, 312 color photographs of different categories were displayed for 2 seconds each. Observers performed an N-back task to encourage sustained attention. Importantly, observers were explicitly informed prior to learning about the testing conditions. At test, they performed a 2-AFC task with one old image and a foil whose resemblance with the target was manipulated: the foil could be a mirror image of the same scene, the same scene zoomed in or out by 25 %, or a nearby scene cropped from a larger panoramic image. The control condition, a foil from a novel category, led to 93% recognition accuracy, as in related previous studies. The fidelity of memory was poorest (54%, chance level) when the foil depicted a “zoom-out” version of the old image. Participants performed well (84%) with foils depicting a translated non-overlapping version, and were moderately accurate (79%) with foils image overlapping by 50%, a zoom-in (69%) or a left-right mirror of the old image (72%). In a broader context, these results contribute to understanding the nature of stored visual representations. LTM representations have been shown to be sensitive to changes in scene viewpoint. Nevertheless, our results suggest that visual long-term memory is “open-minded” about certain kinds of viewpoint transformations: it does not mind a step backward.