Freed from the presumed need for simple analyses imposed by preattentive processing, the object recognition literature has hypothesized the existence and use of a wide range of visual features. The broadest cut through this literature separates structural-description theories, those that assume the composition of objects from simpler three-dimensional features (e.g., Biederman,
1987; Biederman & Gerhardstein,
1993; Marr & Nishihara,
1978), from image-description theories, those that assume the extraction of two-dimensional features from information closer to the retinal mosaic (Bülthoff & Edelman,
1992; Poggio & Edelman,
1990; for a review, see Tarr & Bülthoff,
1998). Within the realm of image-based theories, the features that have been suggested for object recognition are often far more complex than the simple features believed to underlie search guidance. This is most evident in the case of face recognition, where it is common to assume features that code the metrical spatial relationships between the regions of a face (Sigala & Logothetis,
2002), or even large-scale features that code these facial configurations directly (Zhang & Cottrell,
2005). Neurophysiological and neurocomputational work in inferior temporal cortex also suggests that the features used for object recognition are considerably more complex than the simple color and edge-based features found in earlier visual areas (Gross, Bender, & Rocha-Miranda,
1969; for extended discussions, see Rolls & Deco,
2002; Tanaka,
1996), with the suggestion that domain-specific neurons code features for specific objects or object classes (Grill-Spector,
2009; Huth, Nishimoto, Vu, & Gallant,
2012; Perrett et al.,
1984). It is perhaps not unfair to characterize the object recognition literature as converging on the assumed existence of features or feature weightings dedicated to the recognition of faces (Kanwisher,
2000), scenes (Epstein & Kanwisher,
1998), body parts (Downing, Jiang, Shuman, & Kanwisher,
2001), and various other categories of known (Gauthier, Skudlarski, Gore, & Anderson,
2000; Haxby et al.,
2001; Xu,
2005) and novel (Gauthier, Tarr, Anderson, Skudlarski, & Gore,
1999) objects, but with the specific information coded by these features still largely unknown.