38 Color Constancy

Vision is useful because it informs us about the physical environment. In the case of colm~ two distinct functions are generally emphasized (e.g., Jacobs, 1981; Mollon, 1989). First, color helps to segment objects from each other and the background. A canonical task for this function is locating fruit in foliage (Regan et al., 2001; Sumner & Mollon, 2000). Second, color provides infm' mation about object properties (e.g., fresh fish versus old fish; figure 38.1). Using color for this second function is enabled to the extent that perceived object color correlates well with object reflectance properties. Achieving such correlation is a nontrivial requirement, however, because the spectrum of the light reflected from an object confounds variation in object surface reflectance with variation in the illumination (figure 38.2; Brainard, Wandell, & Chichilnisky, 1993; Hurlbert, 1998; Maloney, 1999). In particular, the spectrum of the reflected light is the wavelength-by-wavelength product of the spectrum of the illumination and the object's surface reflectance function (figure 38.2B). The dependence of the reflected light on illumination leads to ambiguity about object reflectance because changes in the spectrum of the reflected light can arise from changes in object reflectance, changes in illumination, or both. Despite this ambiguity, color appearance is often quite stable across illumination changes (e.g., Helmholtz, 1896/2000; Katz, 1935; Brainard, 2004). The approximate invariance of object color appearance is called color constancy. VVhen color conc-stan<:yfails, as in the case of Uluru (figure 38.2A), it is prterwJmc:n<m to remark on. Ambiguhy occurs pervasively in perception (e.g., 1978; Helmholtz, 1896/2000; Wolfe et al., and at higher levels of processing (for example, l;m_gnag<o) (Foder, Bever, & Garrett, 1974; Miller, 1973). how biological information processing informational ambiguity is a priority for vision (e.g., Brainard, 2009; Knill & Richards, 1996; & Lotto, 2003; Rust & Stocker, 2010). Object perception provides a model system for moving characterization forward. chapter provides an update on the treatment of constancy that appeared in the first edition of this (Brainard, 2004). Our emphasis here is to provide a self-contained introduction, to highlight some more recent results, and to outline what we see as important challenges for the field. We focus on key concepts and avoid much in the way of technical development. Other recent treatments complement the one provided here and provide a level of technical detail beyond that introduced in this chapter (Brainard, 2009; Brainard & Maloney, 2011; Foster, 2011; Gilchrist, 2006; Kingdom, 2008; Shevell & Kingdom, 2008; Smithson, 2005; Stockman & Brainard, 2010).