Abstract
The human visual system has a remarkable ability for rapid recognition and categorization of visual scenes. This perceptual speed might strongly rely on low-level feature dissimilarities, such as different spatial frequency contents between man-made and natural scenes (Torralba & Oliva, Network, 2003). Here we set out to test the contribution of such early visual processes during rapid categorization by using an adaptation paradigm. We hypothesized that adaptation to spatial frequency distribution matched to that of man made scenes might modify the perception of a subsequently viewed real world image.
We presented grayscale images of natural and man-made scenes for 12 ms terminated by a mask. The images were preceded by an adaptation sequence of rectangles approximating the statistical properties of man-made scenes. Observers were either asked to rapidly detect the location of a single man-made scene amongst two or four simultaneously presented scenes or asked to rapidly localize the single natural scene amongst the presented images.
Following the adaptation, observers' performance (reaction time and percent correct) in localizing the natural image among distracting images did not change significantly while the localization of man-made images was severely impaired. We suggest that this results from a recalibration of the visual system to the spatial frequency profile of the adapter, allowing detection of statistical deviations from this mean during rapid categorization. Our findings present strong evidence for an important role of low-level processes in fast image categorization because they could recover the “gist” of the visual input.
This study was supported by the SensoPrim EU Marie Curie Early Stage Training Program