Visual categorization: The nature of category templates

Abstract

Visual categorization is a cognitive process by which humans effortlessly and accurately identify objects that belong to a category. Conceptually, it is believed that single categorization tasks involve the use of one category template pertaining to the task (for example, animal template). The category template is a coarse shape representation of category members (like animals), and reliable match to template enables rapid categorization performance. In this study, we investigate two questions pertinent to the nature of templates underlying visual categorization. The first question concerns the features involved in categorization. In classic categorization studies, is the category template orientation dependent? Does it involve internal details or does it store contour information? We investigated this by measuring human categorization performance in an animal detection paradigm by using inverted objects and object silhouettes. We report that while categorization performance decreases on inversion, the presence of contour information is sufficient to perform categorization. The second question relates to the processing of multiple categories by humans. In a multiple category task, are two templates activated or just one? To test the presence and use of more than one classifier, we conducted three categorization tasks on human subjects – 1) animal/non-animal categorization, 2) vehicle/non-vehicle categorization, and 3) animal/vehicle categorization. We hypothesized that subjects would employ single classifiers in the single categorization tasks, and use both animal and vehicle classifiers in the animal/vehicle categorization task. We provide evidence that humans indeed activate two classifiers in the dual categorization task; even though they could have performed equally well by activating one. Thus, this contributes towards understanding the representation of features and multiple categories in a fundamental task, like visual categorization.