i
i
i
i
i
i
i
i
23.2. Dynamic Range 601
A second class of operators is grounded in physics. Light interacts with sur-
faces and volumes before being absorbed by the photoreceptors. In computer
graphics, light interaction is generally modelled by the rendering equation. For
purely diffuse surfaces, this equation may be simplified to the product between
light incident upon a surface (illuminance), and this surface’s ability to reflect
light (reflectance) (Oppenheim et al., 1968).
Since reflectance is a passive property of surfaces, for diffuse surfaces it is,
by definition, low dynamic range—typically between 0.005 and 1 (Stockham,
1972). The reflectance of a surface cannot be larger than 1, since then it would
reflect more light than was incident upon the surface. Illuminance, on the other
hand, can produce arbitrarily large values and is limited only by the intensity and
proximity of the light sources.
The dynamic range of an image is thus predominantly governed by the illu-
minance component. In the face of diffuse scenes, a viable approach to tone re-
production may therefore be to separate reflectance from illuminance, compress
the illuminance component, and then recombine the image.
However, the assumption that all surfaces in a scene are diffuse is generally
incorrect. Many high dynamic range images depict highlights and/or directly
visible light sources (Figure 23.3). The luminance reflected by a specular surface
may be almost as high as the light source it reflects.
Various tone reproduction operators currently used split the image into a high
dynamic range base layer and a low dynamic range detail layer. These layers
would represent illuminance and reflectance if the depicted scene were entirely
diffuse. For scenes containing directly visible light sources or specular highlights,
separation into base and detail layers still allows the design of effective tone re-
production operators, although no direct meaning can be attached to the separate
layers. Such operators are discussed in Section 23.5.
23.2 Dynamic Range
Conventional images are stored with one byte per pixel for each of the red, green
and blue components. The dynamic range afforded by such an encoding depends
on the ratio between smallest and largest representable value, as well as the step
size between successive values. Thus, for low dynamic range images, there are
only 256 different values per color channel.
High dynamic range images encode a significantly larger set of possible val-
ues; the maximum representable value may be much larger and the step size be-
tween successive values may be much smaller. The file size of high dynamic