Performance and confusion effects for gist perception of scenes: an investigation of expertise, viewpoint, and image categories

Abstract

Human object recognition often exhibits viewpoint invariance. However, unfamiliar aerial viewpoints pose challenges because diagnostic features are often obscured. Here, we investigated the gist perception of scenes when viewed from above and at the ground level, comparing novices against remote sensing surveyors with expertise in aerial photogrammetry. In a randomly interleaved single-interval, 14-choice design, briefly presented target images were followed by a backward white-noise mask. The targets and choices were selected from seven natural and seven man-made categories. Performance across expertise and viewpoint was between 46.0% and 82.6% correct and confusions were sparsely distributed across the 728 (2 × 2 × 14 × 13) possibilities. Both groups performed better with ground views than with aerial views and different confusions were made across viewpoints, but experts outperformed novices only for aerial views, displaying no transfer of expertise to ground views. Where novices underperformed by comparison, this tended to involve mistaking natural for man-made scenes in aerial views. There was also an overall effect for categorisation to be better for the man-made categories than the natural categories. These, and a few other notable exceptions aside, the main result was that detailed sub-category patterns of successes and confusions were very similar across participant groups: the experimental effects related more to viewpoint than expertise. This contrasts with our recent finding for perception of 3D relief, where comparable groups of experts and novices used very different strategies. It seems that expertise in gist perception (for aerial images at least) is largely a matter of degree rather than kind.