Abstract
Human neuroimaging studies have revealed a dedicated cortical system for visual place, or “scene” processing. But what are the stimulus features that define a scene, and thereby engage visual scene processing? Here, we proposed a novel, stimuli-driven approach to identify a scene-defining stimulus feature. We hypothesized that if there exists a visual feature that humans use to recognize a scene, then such a feature will 1) be common in visual scene, but not non-scene, stimuli, and 2) selectively engage neural and behavioral visual scene processing. In study 1, we analyzed thousands of naturalistic images and found that, across most scenes, there is a vertical luminance gradient (VLG), with the top half of a scene image brighter than the bottom half; moreover, VLG is prominent in scene, but not object, stimuli. In study 2, using fMRI data from BOLD5000, we found a positive and significant correlation between the amount of VLG in naturalistic images and the response in the scene-selective cortical regions, but not in early visual cortex, hinting at a potential role of VLG in driving cortical scene selectivity. Thus, in study 3, using tightly controlled VLG stimuli, we directly tested the role of VLG in driving cortical scene selectivity. We found that the scene-selective cortical regions, but not early visual cortex nor an object-selective region, respond selectively to images of VLG. Finally, we tested for the relevance of VLG for behavioral scene recognition by disrupting VLG in scene images via image rotation and testing participants’ ability to remember these images (study 4). Indeed, we found that image rotation selectively impaired behavioral recognition of scene, but not object, images. Together, these results suggest that VLG is a visual feature that humans use to selectively engage visual scene processing, and call attention to the relevance of stimuli statistics for visual processing.