Abstract
Boundary extension—a memory distortion in which observers misremember a scene as containing visual information beyond its edges—is widely regarded as a universal effect in visual memory, reflecting our brain’s automatic extrapolation of scene information. However, most studies of boundary extension exclusively focus on a narrow stimulus set of object-oriented images. In this study, we test 2000 participants on a recognition-based boundary transformation paradigm across a naturalistic set of images, consisting of 500 object-oriented images and 500 scene-oriented images. While the object-oriented images replicate previous boundary extension findings, the scene-oriented images show equal proportions of both boundary extension and the opposite effect of boundary contraction. The direction of boundary transformation is highly predictable from simple image properties: images with fewer, central, close and large objects tend to cause extension, while images with more, dispersed, distant, small objects cause contraction. Finally, these boundary transformation scores are highly consistent across paradigms (both rapid recognition and recall drawings), including a perceptual task with minimal memory load in which participants draw copies of images while viewing them with no time limit. Collectively, these results put into question boundary extension as a universal scene memory phenomenon, given that we observe boundary contraction for the most scene-like images, and that these effects occur even during minimal memory conditions. These findings also suggest alternate image-property-based accounts of scene transformation in the brain. Crucially, these results highlight the importance of revisiting already accepted psychological phenomena and the necessity to always consider broad, representative samples and images when making global inferences about the brain.