Two stages in the time-course of natural scene gist perception

I Groen1, S Ghebreab2, V Lamme1, H Scholte1

1Cognitive Neuroscience Group, University of Amsterdam, Netherlands
2Intelligent Systems Lab, University of Amsterdam, Netherlands

Contact: i.i.a.groen@uva.nl

The ability of the visual system to process natural images at remarkable speed may be mediated by a global "gist" percept. It is unclear, however, 1) how gist is computed by the visual system and 2) when and under what circumstances it is extracted. We addressed these questions using regression of single-trial EEG on scene statistics. Subjects judged one type of gist, "naturalness", for a large set of natural images. Using a neurophysiologically plausible contrast filtering model, we derived two statistical parameters for each scene: contrast energy and spatial coherence. Behaviorally, contrast energy correlated with reaction times, whereas spatial coherence correlated with perceived naturalness. In EEG, contrast energy and spatial coherence predicted differences between single-trial event-related potentials (sERPs) both early (90-150 ms) and later (> 200 ms) in time. In a follow-up experiment where we manipulated task relevance, early effects on sERP amplitude persisted when an orthogonal task was performed, whereas late effects were present only when gist categorization was required. These results suggest that scene gist 1) can be derived from responses to local contrast, for example present in LGN/V1 and 2) is computed bottom-up but can be selectively ‘read out’ if relevant for the task at hand.

Up Home