Intrinsic Images

Figure 7.36 RGB intrinsic images for a color Mondrian. The input image is shown in (a). The image in (b) was created by applying the extra light shift in the positive direction whereas the image in (c) was created by applying the extra light shift in the negative direction.

The intrinsic images obtained by the method of Finlayson and Hordley (2001 a,b) only depend on the reflectance of the object points provided the illuminant can be approximated by a black-body radiator. Since a green illuminant cannot be approximated by a black-body radiator, the result will not be an intrinsic image. That aside, here, achromatic patches are illuminated by four different illuminants. Finlayson et al. compute the logarithm for each channel and subtract the geometric mean. The resulting coordinates are denoted by p, for i e [r, g, b ... 

Finlayson GD, Drew MS and Lu C 2004 Intrinsic images by entropy minimization In Proceedings of the 8th European Conference on Computer Vision, Part III, Prague, Czech Republic, May, 2004 (eds. Pajdla T and Matas J), pp. 582-595. Springer-Verlag, Berlin. 

Tappen MF, Freeman WT and Adelson EH 2002 Recovering Intrinsic Images from a Single Image. Technical Report AI Memo 2002-015, Massachusetts Institute of Technology, Artificial Intelligence Laboratory. 

Weiss Y 2001 Deriving intrinsic images from image sequences Proceedings of the International Conference on Computer Vision, Vancouver, Canada, July 9-12, 2001. IEEE. 

---