Abstract
Research on object recognition has focused mainly on bottom-up processes that analyze the visual input; the potential role of top-down processes has been relatively neglected. We examined the hypothesis that object recognition is an iterative process in which bottom-up and top-down processes interact to discriminate between alternative object identities, with attention playing a critical role. Although similar ideas have been proposed by others, to our knowledge this is the first attempt to test these ideas empirically. In two experiments observers discriminated between sets of artificial fish. A secondary visual-probe detection task was used to measure changes in the spatial distribution of attention over time. The probe was displayed on 50% of the trials adjacent to a local fish feature for a short duration at several SOAs. In Experiment 1, the fish varied in several local distinguishing features. We assumed that only one feature could be attended at a time and hypothesized that attention would be allocated in the most diagnostic sequence: first to a feature that discriminated between two general classes of fish and then, based on its value, to a second distinguishing feature that would conclusively identify the specific fish. In Experiment 2, fish recognition could be based on either of two distinguishing features, one much more discriminable (perceptually) than the other. On some of the trials, the more discriminable feature was occluded. We hypothesized that attention would be directed initially, by default, to the more discriminable feature, and when this feature was occluded – redirected to the alternative feature. In general, the observed pattern of spatial allocations of attention over time, indicated by probe detection rate at the different locations as a function of SOA, accorded with the predictions, supporting the idea that – when necessary – object recognition is an interactive iterative process in which attention plays a crucial role.
Meeting abstract presented at VSS 2012