Abstract
Empirical evidence suggests a massive, perhaps even unbounded capacity of visual long-term memory (LTM), as thousands of images can be stored and subsequently recognized with high precision. However, most studies examined massive visual LTM using meaningful items, therefore, the existence of a pure visual memory that is independent of conceptual information remains to be determined. In three experiments, participants viewed hundreds of images and then completed a memory test. Three types of stimuli were used: images of real-world objects, lightly-scrambled images of the same objects, and fully-scrambled objects. The scrambled versions preserved the low-level visual statistics of the objects while reducing their semantic meaning. In the memory-recognition phase, LTM was evaluated using a 4AFC test that included an old image, a new image, and two mirror transformations of these stimuli. As expected, the results revealed a superior memory for meaningful compared to scrambled images. Importantly, there was no hint for a massive memory capacity of the scrambled items. Additionally, while participants were able to distinguish memorized items from their mirrored versions when tested with the real-world objects, memory for visual properties was much weaker and close to chance-level for the lightly-scrambled stimuli, and no different than chance-level for the fully-scrambled stimuli. Overall, these results suggest that a ‘pure’ visual LTM is quite poor and far from having a massive capacity. Rather, meaning might be critical for the long-term memorization of visual properties, as it may serve as a "conceptual hook" that enables more efficient encoding and storage of visual information.