Human Object Recognition: Appearance vs. Shape

Abstract

In recent years, there has been a decided effort among many in the computer vision community to achieve object recognition by mapping images and regions of images directly onto basic-level object classes, such as chairs, cars, and cows (e.g., Serre et al. in Proc. Natl. Acad. Sci. 104:6424–6429, 2007; Ullman in Trends Cogn. Sci. 11:58–64, 2007; Oliva and Torralba in Prog. Brain Res. 155:23–36, 2006). In these efforts, termed “appearance based,” no attempt is made to make shape explicit or to distinguish shape from surface properties, such as color, texture, albedo, or direction of illumination. Whatever the success of such efforts in achieving classification into object categories, they do not appear to correspond well to either the behavior nor the neural coding evident in human and macaque object recognition. We will first consider a summary of the cortical stages mediating object recognition in humans and primates. We will then review research that indicates that the cortical loci critical for shape representation differs from those for the perception of surface properties, such as color and texture. Importantly, the tuning to shape can be largely engaged by a line drawing of the object. Last, we will consider the tuning properties of individual cells as well as fMRI cortical activity in those regions that do specify shape with respect to what do they tell us about how shape is represented. The extraordinary competence of humans to achieve object recognition can be largely understood as deriving from those properties.