Abstract
Most studies of the role of context in object recognition have focused on conditions in which the target object is fully identifiable when viewed in isolation, and the role of context is in ‘;facilitating’ the recognition process. The current study was aimed at identifying and quantifying the role of context in performing recognition of ambiguous stimuli; objects that cannot be identified successfully on the basis of their local properties alone. Subjects had to identify ‘;mosaic’; photographs of objects—images divided up into grids of equally sized, square checks composed of the average values of the constituent image pixels. The number of checks making up each image-i.e. the ‘;resolution’;— was manipulated for each object (starting with fewer, large checks and progressing to more, smaller checks) and was taken as a measurement of the information in that particular object image. We compared the amount of image information needed to perform accurate identification of the ‘;mosaic’; object images with and without context. In the context conditions, the object image was shown within a larger scene (unlike the target object, the scene image was unaltered) while in the no-context conditions the mosaic object was shown in isolation. There were two context conditions: ‘;Generic Context’;—in which the stimulus photograph consisted of an unfamiliar environment whose category was identifiable (e.g. ‘a kitchen’) and ‘;Expert Context’;—in which the stimulus photograph consisted of an environment highly familiar to the subject (e.g. his or her own kitchen). The average reduction in image information needed to perform the task in the different contextual conditions, relative to the no-context condition, was calculated as an estimate of the ‘;top-down’; information available from the context.