Purchase this article with an account.
Jiye G Kim, Sabine Kastner; Mapping natural and texture scene representations across the visual system. Journal of Vision 2014;14(10):1084. doi: https://doi.org/10.1167/14.10.1084.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Recent neuroimaging studies have shown that scene category information is represented in the patterns of responses in scene-selective areas. What remains poorly understood is what features of scenes contribute to this information and how scene representations differ across scene and non-scene selective areas. In a series of behavioral experiments, we demonstrated that higher-order image statistics extracted from natural scenes provide meaningful category information, suggesting these features as the basis for the neuroimaging findings. Here, we used fMRI to investigate the transformation of scene representations across the visual hierarchy elicited by natural and synthesized texture scenes that have preserved higher-order image statistics from the intact scenes (Portilla & Simoncelli, 2000). To test for the robustness of representation, we manipulated attentional demands by either instructing subjects to passively view or classify rapidly presented scenes. We examined category (beach, city, or forest) and format (intact or texture) information across 27 functionally defined regions including areas in the early visual, category-selective (e.g., PPA) and topographically organized fronto-parietal cortex (e.g., FEF and IPS). Early visual areas consistently showed no category or format information across different task demands. Rather, the pixel-energy and gabor-filter models best predicted their responses. Under passive viewing, responses in scene-selective areas (PPA, TOS and RSC) were best predicted by image format, and to a lesser extent, category information. With attention, these differences were markedly reduced. Responses in fronto-parietal cortex were also best predicted by image format under passive viewing. However, with attention, fronto-parietal responses were best predicted by task demands irrespective of category and format. Together these results demonstrate a gradual transformation of the representations of scene information across the visual hierarchy and that category information obtained from image statistics is dynamically represented under different task demands.
Meeting abstract presented at VSS 2014
This PDF is available to Subscribers Only