Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster C in Poster Session C: Wednesday, August 5, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

Experienced dimensions of 3D scene structure provide organizing dimensions for scene-selective areas in the human brain

Mark D. Lescroart¹; ¹University of Nevada, Reno

Presenter: Mark D. Lescroart

Here, we explore how natural visual experience of scenes is related to the representation of 3D scene structure in the human brain. First, we define a model to quantify 3D scene structure in terms of the distribution of surface normals and distances in arbitrary images. To quantify natural visual experience of these 3D structural features, we apply this model to samples of egocentric video frames drawn from a large database of head-mounted video collected while human subjects performed natural tasks. We then use principal components analysis to discover commonly experienced patterns of 3D structural features. We find that the first dimension of variance in scene structure distinguishes open from closed scenes. To relate visual experience of structure to brain responses, we model brain responses in the Natural Scenes Dataset as a function of the same 3D structural features. Analysis of fit model weights reveals a gradient of selectivity from open to closed scenes across multiple scene-selective areas. Thus, the main dimension that varies in our visual experience of scenes is mapped across scene-selective areas in the human brain.

Topic Area: Computational Models of Vision & Visual Cortex