I Color Recognition in Outdoor Images

models of daylight illumination and hybrid reflectance, and predicts the color of objects b scene context. The second method shows that color can be nonparamet-rically " learned " th classification methods such under different Figure 1: Samples of objects in outdoor images, along with the variation of R G B color over 100 images. Top: two matte surfaces; Middle: camouflaged military vehicle; Bottom: shiny traffic sign. surface can be greater than the difference (in terms of color-space Cartesian distance) between two distinct colors (white and green, in this case). The variation in the apparent color of more realistic objects, such as a road surface and a camouflaged military vehicle (also in figure I), can be greater. Much of the work in computational color recognition under varying illumination has been in the area of color constancy, the goal of which is to match object colors under varying illumination without knowing the spectral composition of the incident light or surface re-flectance. An illuminant-invariant measure of surface reflectance is recovered by first determining the properties of the illuminant. While there have been many interesting advances in color constancy [7, 8, 16, 211, their applicability to unconstrained outdoor images has not yet been established. Indeed, as Forsyth [8] states, '(Experimental results for [color constancy] algorithms running o n real images are not easily found in the literature... Some work exists on the processes which can contribute to real world lightness constancy, but very little progress has been made in this area. " In the interests of brevity, this paper does not discuss the previous approaches in detail; Funt [9] and Forsyth [8] discuss the existing approaches, along with their