Abstract
Humans can perceive the content (gist) in a scene under pre-attention conditions. Previous work has shown that low spatial frequency information dominates this perception (Schyns & Oliva, 1994). In addition, colour pop-out shows that colour can be perceived pre-attentively, and coarse-scale colour is thought to contribute to scene gist, but only when colour is diagnostic (i.e. predictive) of a scene category (Oliva & Schyns, 2000). Here we investigate the spatiotemporal influence of colour on scene gist. Two types of image are used within our study: natural scenes where colour is diagnostic for scene category (e.g. mountains, coastlines), and man-made scenes where colour does not influence scene categorization (e.g. roads, buildings). Subjects performed a two-alternative forced choice task to identify the category of the presented scene for both natural and man-made scenes. Scenes were shown in three different chromatic conditions: normal chromatic, monochromatic, or inverted RG/BY chromatic channels. Chromatic condition and the duration of scene presentation was randomized within and between blocks. Our data show that for short presentation durations ([[lt]]150 ms), accuracy for natural scenes categorization is high for normal chromatic, but low for monochromatic and inverted chromatic channel scenes. In addition, reaction times are fastest for the normal chromatic condition. For the man-made images, accuracy and reaction times are equal in all three chromatic conditions. For presentation durations over 150 ms, accuracy is high and reaction times are low for all three chromatic conditions, irrespective on the type of scene used. Our conclusion is that although colour plays a role in defining the course scale spatial layout used within the initial onset of rapid scene gist perception of diagnostic natural scenes, other features are used in scene gist perception within non-diagnostic scenes categories, and at longer presentation durations.