Abstract
Visual perception has traditionally been explained in terms of feed-forward processing. Demonstrations of ultra-rapid categorization (e.g., Thorpe, et al., 1996) have been interpreted in terms of feedforward processing because responses seem too fast (120–150 ms) for feedback to play a role. But is ultra-rapid activation sufficient for perception? An alternative is that representations of many objects potentially present in a display are activated in a feedforward processing pass; then a subset is selected for perception. We investigated whether fast categorization of familiar versus novel silhouettes is slowed when another familiar object is potentially present in the scene but is rejected by perceptual organization processes. On each of 80 trials, we showed subjects two silhouettes, one familiar and one novel, one above and one below fixation, and asked them to saccade to the location of the familiar silhouette as quickly as possible. Familiar silhouettes portrayed nameable everyday objects. Half of the novel silhouettes were experimental silhouettes, with portions of familiar objects suggested but not perceived on the outside of their borders. Critically, the outsides appeared to be shapeless grounds, as did the outsides of the other half of the novel silhouettes that did not suggest familiar objects there (control silhouettes). Correct saccades to familiar silhouettes were significantly slower when an experimental rather than a control novel silhouette was concurrently in the display (311 ms vs. 293 ms, p <0.01), even though participants were unaware of more than one familiar stimulus. These results show that activation of familiar objects that are potentially present in a scene, but are pruned by perceptual organization processes can affect fast categorization responses. Thus, fast access to category knowledge is not sufficient for perception. These results fit with models that assume a first pass of processing produces broad activation, but that integrative processing is necessary for perception.
Meeting abstract presented at VSS 2012