View-dependent and view-independent properties in human object recognition

Psychophysical experiments were conducted in order to evaluate the possibility that both view-independent and view-dependent properties are involved in human object recognition. The results suggest that human prototypes of objects have descriptions that contain rich information on component structures but view-independent properties are found only in limited ranges. Therefore, after these prototypes are activated, a verification process by rotation is considered necessary to check correspondence to observed views. In recognition experiments using familiar objects, multiple regression analyses revealed that the degree of activation of the prototype affected the percentage of correct responses and that the time taken by the verification process mainly determined reaction time. For recognition of novel objects with small component variations corresponding to within-category differences, the verification process by rotation seems very important. In contrast, for recognition of novel objects with large component variations corresponding to between-category differences in natural objects, the prototype activation process plays a major role, while the precise verification process seems to be unnecessary. Thus, the proposed two processes can reconcile contradictory points between the psychophysical experiments which show view-dependence and view-independence.