fine prints in your digital darkroom
updated Feb. 15,
to monitor calibration and gamma
In this page we cover some background (reference) areas related to
why a new chart was needed (many
old ones had an error), then we present two additional charts: one for
luminance levels and one for three colors
(R, G, B). We illustrate what luminance steps
like for different gamma settings. We moved the Praxisoft
WiziWYG instructions to this page: since I put up QuickGamma,
WiziWYG is no longer my choise for monitor calibration. Finally I
pages by Timo Autiokari, which contain some
advice on gamma.
a new gamma chart?
gamma charts consist of an arrangement of black and white patterns and
solid tones. You estimate gamma by viewing the chart from sufficient
(typically one to two meters from the monitor) so you don't see the
details, then either locating the position where the average luminance
of the pattern matches the solid tone or operating a slider to obtain a
match. Many of these charts use checkerboard patterns, typified by the
pattern on the left: a portion of the gamma chart supplied with Epson
enlarged 3x to make the pattern
visible. This type of checkerboard
gives an erroneous estimate of gamma!
I owe a debt of gratitude to Pete
Andrews for pointing this out.
problem is the risetime of the video card and CRT monitor. In
cases the brightness cannot change from pure black to pure white in the
short distance of one pixel (or even two or three). Since CRTs are
horizontally, abrupt vertical features are affected.
The situation is illustrated in the box on the
right, which consists of four quadrants, each of which contains equal
amounts of Black and White. 1:
vertical alternating B and W. 2:
horizontal alternating B and W. 3:
Vertical B and W alternating every second line (same as 1
magnified 2x). 4:
Horizontal B and W alternating every second line (same as 2
If risetime were not a problem, all four
would appear equally bright when viewed from a distance. But on most
monitors, quadrant 1
(alternating vertical lines) appears quite different from 2
(alternating horizontal lines)-- usually darker. Quadrants 2
both containing horizontal lines, should be similar. Conclusion:
patterns with vertical features are sensitive to risetime and cannot be
used reliably to estimate gamma. Liquid crystal displays (LCDs) don't
this problem, but most have another, worse, problem-- sensitivity to
angle. For more details see Pete Andrews' Monitor
calibration and Gamma assessment page.
I was dismayed to discover that my monitor's
which measured around 2.1 with the old chart, was closer to 2.6. I
it using my video monitor's software (illustrated above
). I'll be recalibrating my printer and editing many images in
Because of the gamma error I set Contrast to +12, well above the
setting (0). I'll undoubtedly reduce it. Annoying and a little
but the end result will be improved print quality.
level and three color charts
level gamma estimation chart
The three level chart on the right enables
estimate display gamma for three normalized luminances: 0.25 (left),
(middle) and 0.667 (2/3) (right-- 0.75 was too light to be legible).
is estimated for each zone by locating the position where the average
across the zone is constant. The corresponding gamma is shown on the
The section for luminance = 2/3, on the right, is difficult to read.
I use this chart mainly as a diagnostic
to see if gamma is consistant at different tonal levels. If your
is working well, gamma should vary by no more than ±0.1
the zones. For normal gamma estimation I use the combined gamma/black
level chart, on the previous page. The solid areas are calculated
color gamma chart
A gamma chart with separate R, G and B
is shown on the right. It could be useful if you need to adjust gamma
for each color. The blue pattern is dark-- difficult to read. I don't
this chart very useful for diagnostics; if gamma is different for
colors, you'll see color variations in the gray chart. But it could be
useful in adjusting gamma in special cases, like my ATI Radeon video
where the software doesn't allow the gamma adjustment to be coupled for
the three colors.
in pixel level and luminance
The following table displays 21 luminance level steps, from minimum to
maximum, in two different ways. In the top row the pixel levels are
spaced. In the bottom row the luminance is evenly spaced (linear
if monitor is calibrated
for gamma = 2.2, The upper number in each box is the normalized pixel
(pixel level/255). The lower level is the normalized luminance for a
with gamma = 2.2. There are approximately 12.75 luminance steps per box
[These tables will not
display correctly if you are using Netscape and the Always
use my colors, overriding document box is checked (under Edit,
Preferences, Appearance, Colors). You must uncheck it.]
steps in pixel level (0-255; 0-1 in steps of 0.05, normalized)
level on top; screen brightness for gamma = 2.2 below.
even steps in screen luminance for gamma = 2.2 ( 0-1 in steps of 0.05,
level on top; screen brightness for gamma = 2.2 below.
Note the large relative luminance steps at low levels in the
lower row, particularly between the first and second box. This is what
you would get for evenly stepped pixel levels with monitor gamma = 1,
in banding in dark areas (below). In the top row, the relative
steps at low levels are relatively small; the second box should be
lighter than the first, but still very dark. On the whole, gamma = 2.2
is more perceptually uniform than gamma = 1. Gamma = 1.8 (the value
in Macintosh) has near optimal perceptual uniformity; it has a slight
over gamma = 2.2.
The middle (11th) box in the upper row (pixel level 127) has a
luminance of 0.218. In photography, the 18% gray card is considered
gray." Reflective exposure meters are designed to give correct readings
when the gray card is used; it is regarded as being close to the
density of the "average scene." The flip side of the gray card is 90%
white. The normalized reflectance for middle gray is therefore 18/90 =
0.2: very close to the 0.218 luminance of the middle box-- pixel level
127-- for gamma = 2.2.
The following table zooms in on the above table to display the worst
case banding. The top row, for highlights, has evenly spaced pixels
236 to 255 in steps of 1. The bottom row, for shadows, has evenly
luminance steps when viewed with gamma = 2.2, the normal setting for
and the Internet. It illustrates the banding that takes place with
case banding examples
20 even steps in pixel level (236-255 in steps of 1)
screen brightness for gamma = 2.2 below.
areas: 20 even steps in screen luminance (in normalized increments of
integral pixel steps (0,1,2,3,...) for gamma = 1 (when displayed at
gamma = 2.2 on top; normalized screen brightness (n/255) below.
I find the highlight banding to be barely preceptible, if at all.
the shadow banding-- what you would get with gamma = 1-- is obvious and
unacceptable. There would be no visible shadow banding with gamma = 1.8
Autiokari and AIM (Accurate Image Manipulation)
|Timo Autiokari's Accurate
Image Manipulation for Desktop Publishing website is beautifully
rich in content, and seems well organized at first glance. Timo thinks
for himself; he doesn't follow the herd.
But his page contains significant misinformation. He
states that image
editing introduces serious
errors unless gamma is set to 1. And indeed, problems can arise in
operations that involve strongly contrasting adjacent pixels.
Blurring, which involves averaging neighboring pixels,
is a good example.
If you blur two adjacent pixels with values 1 and 255, you get pixel
around 128, which appear darker than the average of pixels 1 and 255 if
gamma is larger than 1. (This is the basis of the gamma chart, above.)
This effect causes visible errors with extreme sharpening and heavy
examples), and rotation (see Helmut Dersch's page on interpolation;
site), but there are no problems with
on individual pixels such as color and tonal adjustments.
And none with reasonable (not extreme) sharpening-- only with severe
with exaggerated edge contrast.
For the most part, Timo is wrong about gamma.
reason for rejecting his assertion is the banding in shadow areas with
gamma =1, illustrated above. This banding would be visible on prints. You
are best off sticking with your system's standard gamma (2.2 for
1.8 for Macintosh) for image editing.
Timo doesn't comprehend the simplicity of the gamma
= (pixel level/255)gamma. In the Introduction
to System Calibration, he states,
all the hues on the screen are additive mixtures of the red, green and
blue primary colors it is easily understood that gamma produces
also when these tri-color codes are edited."
Absolute nonsense! Though his statement would be correct if RGB data
for gamma. In practice, the same gamma is assumed for each individual
in an RGB file; there is no hue shift when it is printed or displayed.
In editing operations that affect color and brightness, RGB images are
temporarily converted into a format such as HSV
or HSL where hue, saturation and lightness (or brightness) are
Again, there is no hue shift, though these operations are performed
best accuracy in 48-bit color.
Timo assumes that 2.5 is the standard gamma for PC's.
Wrong! The de
facto standard is 2.2. But Timo insists on using 2.5 for examples
his page. 2.5 is the average gamma of an uncorrected monitor,
I've seen values between 2.2 and 2.8. That's why gamma adjustment is
The more I grapple with Timo's "logic," the more
frustrated I get. His
examples look impressive from a distance, but they are obtuse and often
miss the point entirely. Charles
Poynton has responded to Timo's
savage, emotional attack (which I find to be so mathematically
as to be unreadable) with a cautionary
tale. Poynton knows
his stuff. As much as I like mavericks, I must side with the
on this issue. See also, Timo's
technical argument | Timo
and linear coding| Linear
and nonlinear coding | Gamma
to Monitor calibration and gamma
and text copyright © 2000-2013 by Norman
Norman Koren lives in Boulder, Colorado, founded Imatest LLC in 2004, previously worked on magnetic recording technology. He has been involved with photography since 1964.