The term “global” in this case refers to operators that use one average luminance for the entire image. Thisalgorithm was taken from Erik Reinhard’s Photographic Tone Reproduction for Digital Images. In his paper, the averageluminance Lw is taken as
where Lw(x,y) is the luminance in the scene, N is the number of pixels, and D is some small value to prevent youfrom taking the log of zero. The luminance at each pixel is then scaled by the key value a and the average luminance Lw.
The key value of a scene is a value between zero and one that indicates whether it is very light, very dark, oraround middle grey. For example, the key value of a scene of a white painted room is very close to one. However, most scenes havea wide range of brightness, so the key value is usually set to middle grey, or 0.18.
Finally, the luminances are scaled down to a displayable range between 0 and 1.
where Lwhite is mapped to the maximum luminance in the scene. For low dynamic ranges, this will alsoenhance the contrast in the image. For the majority of high dynamic ranges, the algorithm preserves the contrast for low luminanceareas while mapping high luminances to a displayable range. However, detail is lost in scenes that have very high dynamicranges.