'Bottom-up' Cues Aid Perception of Ambiguous, Partially Visible Figures

摘要

For ambiguous images, it is commonly assumed that top-down information (e.g. familiarity) is needed to correctly perceive a partially visible figure in the image. We show that bottom-up cues (derived from the retinal image) also play an important role in such perception. In Experiment 1, we found that the visual system can exploit differences in contour properties between contours that belong to a target figure and those that do not. In Experiment 2, we found that region information also aids perception of partially visible figures, although its contribution may be secondary to that of contour-based cues. Finally, we formulated a computational model of the perception of partially visible figures. This model detects differences in contour properties between a target figure and other image elements and classifies contours belongingto the target figure on the basis of such differences. An important feature of the model is that it efficiently computes global (image-wide) contour statistics in order to perform local classification of image contours.