Color Constancy

University Of Pennsylvania, Philadelphia PA

Investigators

Linked publications & trials

Abstract

Project Summary We use color to help identify and select objects (e.g., pick the ripest fruit, find our car in a large parking lot). For color to be useful in this way requires that it support selections that correlate well with object-intrinsic surface reflectance. This is not trivial, because the spectrum of the light reflected from an object confounds surface reflectance and illumination. The illumination dependence leads to ambiguity: changes in the spectrum of the reflected light can arise from changes in object reflectance, changes in illumination, or both. Ambiguity of the general sort that underlies object color perception is also characteristic of other aspects of perception, as well as of higher levels of language and cognitive processing. Studying object color provides a model system for making advances that may have applicability beyond the immediate content domain of color. Human color vision is thought to exhibit considerable color constancy. Classically, this is conceived of as the partial stabilization of object color appearance across changes in illumination. In quantitative appearance-based studies of constancy, observers typically adjust the color of one stimulus so that its appearance matches that of another, across a change of illumination. The study of color appearance provides the foundation for the field's current understanding of constancy. Despite the value and intrinsic interest of appearance measurements, however, it is critical to recognize that characterizing the stability of color appearance does not directly tell us how color supports selection: an object could look different across a change in illumination without the difference impairing accurate selection. This would, for example, be the case if a person also perceived the illumination and combined information about the illuminant with object color appearance, perhaps similarly to the way in which we can recognize a bicycle across changes of pose while still apprehending that it looks different in each pose. Understanding constancy requires going beyond appearance and measuring directly how color is used to select objects. The proposed experiments will measure color- based selection across changes of illumination in naturalistic tasks and i) test whether performance is predicted by the visual system's stabilization of color appearance; ii) measure the relative weight placed on color for object selection when other cues are available, and test whether this weight is adjusted dynamically according to the relative quality of the color information; and iii) measure learning in response to feedback provided during color-based selection, and test the generality of this learning.