Purchase this article with an account.
Linda Henriksson, Marieke Mur, Nikolaus Kriegeskorte; Representation of scene layout in human OPA is fast and invariant to surface-texture. Journal of Vision 2019;19(10):250. doi: https://doi.org/10.1167/19.10.250.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Although the scene-responsive occipital place area (OPA) and parahippocampal place area (PPA) are well-established functional regions in the human brain, their computational roles in scene perception remain incompletely understood. In this study, we investigated how scene layout is encoded in these regions, using both functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG). We constructed all possible combinations of five scene-bounding elements (left wall, right wall, back wall, ceiling, floor) in three different surface-textures. The resulting 96 scenes were shown to 22 subjects. The regions-of-interest (primary visual cortex, V1; OPA; PPA) were defined based on criteria independent to the main experiment. The representations were characterized by applying Fisher linear discriminant analysis between each pair of scenes and constructing representational distance matrices (RDMs) from the results. In V1 and PPA, the fMRI response-patterns better discriminated the texture of the scenes than the layout, whereas in OPA, the fMRI response-patterns discriminated the layout better than the texture. Moreover, only in OPA, the layout discrimination generalized across surface-textures. Already the early MEG representations were similar to the fMRI OPA representation, suggesting that the scene layout is rapidly extracted from the retinal input. Finally, the representations were also characterized using a set of models. The fMRI- and MEG-RDMs were fitted by a linear combination of low-level image feature (GIST) and scene-element based models. Whereas the representation in V1 was well captured by the GIST model, the OPA representation was better captured by models based on the presence of the scene-elements. Correspondingly, the dynamic representation captured by MEG was better explained by including the presence of the scene-elements than by the GIST model alone. Taken together, our results suggest that the layout of a scene is encoded rapidly in human OPA and that this representation is invariant to the identity of the scene.
This PDF is available to Subscribers Only