Abstract
Scenes are composed of multiple visual features, yet previous research investigating scene representation has often focused on the unique contributions of single features, such as spatial layout and texture. However, these features rarely exist in isolation. As texture can provide depth and contour information necessary for spatial perception, a dynamic and convergent relationship may exist between texture and spatial layout. Yet it has been unclear how this relationship would manifest in scene-selective cortex. Since texture is typically more diagnostic of place identity in natural compared with non-natural scenes, and spatial layout is typically less ambiguous in non-natural compared with natural scenes, we tested the hypothesis that PPA would show equal sensitivity to manipulations of texture and spatial layout in natural scenes, but would show greater sensitivity to layout in non-natural scenes. Using fMRI, we examined brain activity in areas of scene-selective cortex while participants performed a matching task when attending to either the layout or texture of four different scene categories which varied by spatial boundary (open vs. closed) and content (natural vs. non-natural). As predicted, univariate analysis in PPA revealed equal sensitivity to texture and layout in natural scenes, but less sensitivity to texture in non-natural scenes. Importantly, multivariate analyses revealed decoding accuracy significantly above chance in PPA for both spatial boundary (only in non-natural scenes, which is consistent with our univariate results) and content, replicating previous studies and validating the use of our stimulus set. Our findings demonstrate that scene representation in PPA is not based solely on spatial structure, and may be driven by interactions between diagnostic visual features such as texture and spatial layout. Therefore, we propose extending the study of scene perception to not only investigate which individual features are represented in scene-selective cortex, but how these features converge to define place identity.
Meeting abstract presented at VSS 2015