The whole is created by the sum of which parts?: using prosopagnosia to determine the visual primitives used in human object recognition
Alexander O’Brien · 2007
Some contemporary theories of basic level object recognition posit visual object recognition through the use of structural descriptions (i.e., a form of representation in which objects are represented in terms of their parts and the categorical relations among the parts). Much empirical evidence exists supporting this suggestion, yet no empirical investigation has determined the exact visual primitives and categorical relations used in the structural descriptions that humans use for visual object recognition. The current study attempts to identify the exact visual primitives used in basic level object recognition by testing a prosopagnosic patient's ability to discriminate visual primitives. According to the Coordinate Relations Hypothesis (Cooper & Wojan, 2000), individuals with prosopagnosia should perform normally relative to controls when discriminating visual primitives that are coded distinctly from one another by the visual system, but should be impaired relative to controls when discriminating visual primitives that are coded the same way. Nine studies investigated a prosopagnosic's ability to discriminate visual primitives (geons, Biederman, 1987) relative to controls. The results indicate that not all of the features defining visual primitives proposed by Biederman are coded uniquely in the human visual system, suggesting that the alphabet of visual primitives used for object recognition is comprised of fewer than the 36 geons proposed by Biederman. In particular, the results suggest that the features of axis curvature, size change of the cross section (constant vs. expanding; constant vs. expanding & contracting), and cross section symmetry (reflectional & rotational vs. reflectional; reflectional & rotational vs. asymmetrical; reflectional vs. asymmetrical) are used to define the visual primitives in object recognition, whereas the features of cross section curvature, and size change of the cross section (expanding vs. expanding & contracting) are not.