The PNG Guide is an eBook based on Greg Roelofs' book, originally published by O'Reilly.

Gamma Gotchas

Finally, as a prelude to the following sections, I'll note a few caveats. First, although I've referred to cathode-ray tube monitors (or CRTs) throughout the discussion so far, not all computers use them; in fact, notebook computers have long used liquid crystal displays, and LCDs are becoming increasingly popular on desktop systems as lightweight and space-saving alternatives to traditional monitors. Do the simple exponential (or power-law) transfer functions used earlier apply to LCDs as well? Yes, they do, but I need to qualify that answer. Raw LCDs are actually characterized by an S-shaped transfer function technically referred to as ``sigmoid'', for which the best exponential fit would have an exponent of 1.0. This is a lousy approximation, but fortunately, all real-world LCDs have corrective circuitry built in that makes them behave like monitors. So it is safe to use the same exponential transfer functions we discussed earlier. If the extra circuitry did not exist, the only reasonable-looking alternative would require support from both the encoding and decoding software. Specifically, an image editor running on an uncorrected LCD would need to include with the image a full International Color Consortium profile, which we'll discuss at the end of this chapter, and the decoder would in turn need to use it to correct the image on other display systems. Alternatively, the editor could precorrect the image samples to correspond to a normal CRT and include only gamma information, but this would be a lossy transformation of the image data.

A second caveat is that even when a monitor is the primary display device, other output devices such as grayscale or color printers are often used as well. Because of the vast differences in physics and technology between an image reproduced by emitting light directly from a monitor versus one reproduced as light reflected from printed paper, gamma correction is often of lesser relative importance than color correction. A full color management system may no longer be merely desirable but actually necessary. On the other hand, printers are sometimes calibrated to work properly with the local display, so an image that is gamma-corrected to look good on the monitor will also print properly.

A third caveat is that monitors are not perfectly described by exponential transfer functions, either. A better approximation is a combination of a linear function near zero and an exponential function elsewhere. But a simple exponential works well enough for most purposes.

The last thing to note is that even experts do not always agree, and the issue of what exponent to use to describe CRTs is one of those areas of disagreement. We've used 2.2 in the preceding discussion; that's the value used in the sRGB specification (more on that later) and the consensus of the color experts in the PNG Group. It is also the value used by manufacturers of professional, calibrated display equipment, such as Sony and Barco. On the other hand, Charles Poynton, one of the Web's leading color experts and the author of a number of technical papers and books, steadfastly maintains that 2.5 is more correct. At the time of this writing, things seem to be at an impasse, but there is hope for resolution as further test results become available in 1999.

In the meantime, Michael H. Brill has taken the initiative and written a poem that not only summarizes the gamma disagreement rather nicely but also does so with enviable wit and succinctness. It rhymes, too. The poem is entitled "Gamma and Its Bases" and may be found on Charles Poynton's web site:

Last Update: 2010-Nov-26