Levels of feature analysis in processing visual patterns

In this paper, a revised Pandemonium-like model of visual-feature processing is formulated and a preliminary test of its feasibility is reported. The model differentiates visual-feature processing into a series of hierarchical stages organized by increasing complexity, with the output of each stage going both to the next higher stage, and directly to a more central processor. In the experiment, subjects sorted decks of cards into piles according to the presence or absence of a target stimulus which differed from nontargets in a variety of different features; detection of a feature was sufficient for detection of a target. The data generally supported the revised Pandemonium model, in that targets which differed from nontargets in features thought to be low in the hierarchy were processed faster than targets whose difference was in a high level feature. An extension of the revised model did somewhat less well in predicting the results of sorting for targets in which detection of any one of several features was sufficient for target detection.