Pattern recognition over distortions, by human subjects and by a computer simulation of a model for human form perception

Human Ss, and a computer simulation program of a model for form perception, were examined for their behavior in learning to respond with the proper name for a pattern type over variant examples of the pattern. In the main experiment, five pattern types of five variants each were presented to all Ss (including the simulation program). Four "conditions" were examined: (a) unrelated patterns, (b) interrelated patterns; each presented as the first, and as the second, set. Differences between Ss, and between "passes" through the set (learning trials) were also examined. All of the above variables showed differences significant beyond the .001 level. In particular, the simulation of the model performed at a significantly lower error rate (beyond the .005 level) than did the group of human Ss. The effects, on human Ss, or varying (a) duration of stimulus presentation, (b) size of the set of patterns, and (c) complexity of patterns were also examined. Only complexity of pattern affected performance significantly. In addition, three experiments by other investigators who tested human Ss were replicated, to test the computer model. The model outperformed human Ss in all cases. (PsycINFO Database Record (c) 2006 APA, all rights reserved).