sigh… wrote:
On Thu, 08 Oct 2009 18:36:37 -0800, (Floyd L. Davidson) wrote:
Hmmm… tell me what you make of these two histograms:
<http://tinyurl.com/493fyp>
<http://tinyurl.com/4l8lle>
Nothing can be determined from either without the accompanying photograph.
Didn’t this turn out to be indicative! Not one of the
folks who made so many claims about understanding
histograms can even come up with comments on the article I’m responding to, much less put their own analysis
forward. DRS at least figured out where the histograms
came from, and was willing to repeat the very simple
characteristics. Alan Browne went a little farther and
did manage to not make any mistakes either. But none of were anything past the most cursory look, and none
provided any *useful* information.
The article that I am replying to, despite starting with a grossly incorrect statement (above), which he then
contradicted by giving an analysis, at least was an
effort. There are several mistakes though, but it *is*
a good effort.
This is why I find histograms so useless. A much better option is a viewfinder that displays in real-time the under/over-exposed areas of your subject and its composition. Only then will you know what parts of your subject you can devote to which range of your sensor.
That is true! But misses the point too, because the
highlight display is derived directly from the
histogram, and anything that makes the histogram more or less accurate also makes the highlight display more
accurate. Hence this discussion directly relates to how accurately one can set exposure using the highlight
display to the same degree that it relates to histogram
accuracy.
Setting in camera contrast adjustments will *not* make
the highlight display more or less accurate. It also
will not do that for the histogram either.
Setting White Balance will have a dramatic effect of the accuracy of both.
Understanding histograms makes that statement rather obvious, and while DRS and others wanted to get into childish games of claiming they know histograms, note that none of them is able to prove it with an easy analysis of a couple of very neat histograms!
Since these are merely luminance histograms (not RGB) your first one only shows that you have saturated one of the RGB channels in the low-midtones.
It shows no such thing. The vertical scale of the
histogram is almost certainly adjusted with auto-ranging in the software, but what it means is that the entire
*area* of the image is taken up with those tones. They
are not "saturated".
But there is something interesting about the shape of
that peak! Note that it is a trapezoid, with an almost
squared off top. The indication that provides is that
we probably have a picture of a distant black sky (which explains the flat top and the angled sides). Almost
nothing else produces the shape. (Think about that for
awhile, and the reason is obvious.)
Most likely the blue channel. It says absolutely nothing about that being the proper exposure for the subject or not. The subject could very well be displayed best with that location and shape of histogram.
That is true! Unfortunately you seem to have missed the other values in the histogram, which do relate to
exposure. (Note too that this is an histogram generated in post processing, not in the camera.) There is a line of pixels at the bottom of the display 1 to 3 pixels
high, running up to something about 1 fstop below
maximum white. Those are the pixels to look at to set
exposure. If this were a camera generated histogram one would want to increase exposure to help reduce noise
back over there on the left side where the large peak
is. But this is actually a editor’s histogram, and the
desired white level was not at maximum white, so the
"brightness" was reduced with the effect that all levels, including noise were reduced.
That is, the camera was exposed to get maximum dynamic
range, and the brightness was reduced in the editor to
retain that dynamic range. If the camera exposure had
been reduced by that amount the noise would not have
gone down when the signal did, and a lower dynamic range would have been recorded.
That is the advantage of using Expose To The Right.
The second one it only shows that most of your image was lighter luminance values. Not if they were properly distributed or not. It could be a scene of a white shell on a sunlit beach with little to no shadows at all, nor required. A perfectly fine image, exposed properly. We’ll never know, because your histogram is not compared to a real image. Reading histograms alone is useless information.
Not really! You are of course very correct that it is a high key image. It is also set to slightly less than
maximum white, just as the other image was. Note also
the "comb" effects showing up on the brighter levels of the histogram… those pretty much indicate that someone has used the "contrast" control to stretch the tonal range for that part of the image.
To show exactly how much one can get *from* an
histogram, lets use an editor to *create* a pair of them that basically match the two above. The idea is that
this process removes those two specific images from the
information in the histograms. Rather than see that
information in the images, it can be seen as actions
within an editor.
The first one is very simple. Open a blank image,
select the entire image, and add a linear gradient fill
to it. The direction of the fill makes no difference,
but it must completely fill the area of the image. Pick a pair of colors that are relatively close (it’s easier
to use black and white, but it isn’t necessary). Try
#202020 and #404040 to get something very close to the
original histogram.
The distance between the two colors will determine the
width of the peak. The absolute values will determine
where the peak is located. If the entire area is
filled, the top will be flat. If a linear gradient is
used, the sides will be straight up and down. To make
the sides lean (for either a triangle or a trapezoid), a non-linear gradient is necessary. The top can also be
rounded by using the right non-linear gradient. The
exact location and width of the peak can be adjusted
using brightness and contrast. The contrast will,
however, honeycomb the shape if the tonal range is
expanded.
(And note that exactly the same procedure is used to
generate a starting point for either of the two
histograms.)
For the "moon shot", the next step is to select a relatively small area, say 1/10 the size of the whole
image. A square or rectangle will do (or any other
shape, as only the area is of any significance). If
that area is filled with a single color value a peak at
that value will show up on the histogram. If a gradient is used, it will be spread out as a line on the floor of the histogram. The ends of the line will be determined
by the two ending color values of the gradient, and the
height of the line will be determined by the percentage
of image area that it covers. Note that as the
percentage goes up… that first peak will go down.
With only that small area selected, it’s width and
position can be manipulated with the brightness and
contrast tools. (If the trapezoid is desired, use a
spherical gradient and draw it diametrically from one
corner to the opposite.)
The second image is only slightly more complex. Use the same technique to get a single peak, but it should be
roughly 1/2 stop below maximum white. (Rounding the top depends on what tools a given editor has for non-linear
gradients, so I can’t give a universal method to create
that.) To cause the top and right sides of the the peak
to become honey combed, use a rectangular selection that includes a little more than half the entire image,
selecting all of the brighter half. Use a brightness
adjustment to move it right and left, use a contrast
adjustment to get the honeycomb.
Obviously there *is* a huge volume of information
available in a histogram; and very little of it has
anything at all to do with setting exposure. E.g., the
*contrast* information does *not* help set exposure.
Just because somebody posted it on Luminous Landscape
doesn’t make it true. We might note that there are
dozens of web sites saying that ETTR will be more
accurate using a corrected White Balance, but only one
making this false claim about contrast. Also note that
while there are hundreds or thousands of sites
explaining ETTR, there are also a few that claim it’s
bogus.
But if one understands histograms, it isn’t really hard
to pick out the bogus information.
—
Floyd L. Davidson <
http://www.apaflo.com/floyd_davidson> Ukpeagvik (Barrow, Alaska)