Neural representation of the size of space and the amount of clutter in a scene

Park*, Konkle*, & Oliva

Estimating the size of a space and the level of clutter within a space is central to our interactions in scenes, for example when deciding whether or not to take a crowded elevator or how to organize furniture and objects in a room. Interestingly, the size of a space is independent from level of clutter: spatial volume is a property defined by the shape of the spatial boundary of a scene while clutter is a property defined by the contents within the spatial boundary. Here, we examined how neural areas respond to scenes that parametrically vary in both volume and the amount of clutter of depicted space.

Observers were shown blocks of indoor scene categories and performed a one-back repetition task while undergoing whole brain imaging in a 3T fMRI scanner. 36 scene categories were selected to fully cross the dimensions of scene volume and clutter. The size of depicted space across scene categories varied a 6-point log scale, from small and confined spaces such as closets or showers to expansive areas such as airport terminals or sports arenas. Scene clutter also varied on a 6-point log scale, from very empty spaces such as empty closets or garages, to very cluttered spaces such as full pantries or a full warehouses.

Using a regions-of-interest approach, we examined the multivoxel pattern activity across multiple higher-level visual areas. We used a leave-one-category-out method in which a classifier was trained with five categories per ranking and required to predict the volume or clutter of a new scene category, thus requiring generalization across semantic category. We found that while the patterns of activity in the parahippocampal cortex represented both the spatial volume and clutter information within a scene, retrosplenial cortex selectively represented the spatial volume dimension of a scene, and the lateral occipital complex selectively represented the amount of clutter within a scene. There was a significant interaction across region (RSC or LOC) and scene dimension (Volume or Clutter). Furthermore, a whole-brain group random effects analysis with parametric regressors for spatial volume increase showed converging evidence that the retrosplenial cortex and an anterior parahippocampal area had a parametric representation of spatial volume.

These data suggest that while scene information is represented in a distributed manner across multiple visual areas, spatial volume information of a scene is coded independently of the amount of clutter within a scene, consistent with previous results showing complementary but distinct neural representations of spatial boundary and scene content information.

Park, S. J., Konkle, T., & Oliva, A. Examining how objects of different real-world sizes are represented in ventral visual cortex. Program No. 326.4. 2010 Neuroscience Meeting Planner. San Diego, CA: Society for Neuroscience, 2010. Online.