"Marvin Margoshes" <> writes:
>I have a lot of experience in making quantitative light intensity
>measurements with B/W photographic emulsions, and I've published in
>scientific journals on the topic. The dynamic range (the ratio of the
>highest intensity that can be measured to the lowest level) is typically
>between 50:1 and 80:1. depending on the emulsion's gamma; a ratio of 64:1 is
>equivalent to 6-bits of digital data. The 8-bits per color of a typical
>digicam is 4X larger. In either case, getting the full dynamic range
>requires accurate exposure.
What is your definition of "can be measured"? I suspect you're using a
different definition of dynamic range than photographers usually do.
If you limit yourself to the central portion of a film's characteristic
curve, where the slope is more or less constant, you will indeed have
only 5 or 6 stops of exposure range. But the film characteristic curve
has shoulder and toe regions where the slope is reduced but not yet
zero. Exposure in these regions produces *some* change in density for
a change in exposure, and a photographer would consider these zones
part of the dynamic range of the film because some information is
captured. Using this definition, the dynamic range of a negative film
can be 10 stops or more (greater than 1000:1)
>Each time I say this, dissenting views are expressed that seem to be based
>on opinions, not measurements.
Since you haven't told us how your measurements were made, how can
anyone do better? My own opinions aren't based on my own lab
measurements, but on published film data. For example, if you look at
Kodak's datasheet for Tri-X at
http://www.kodak.com/global/en/profe...Pubs/f9/f9.pdf
and go to page 8 where the characteristic curves are shown, you'll see
that the log E (X axis) range over which the slope is non-zero is 3.0 or
greater. This is a base-10 log, so a 3.0 log E range is an exposure
change of 10^3 = 1000. And these graphs don't show the ultimate drop in
contrast at the shoulder, either - the film has even more highlight
capture range than this.
Or look at the typical film curves in "Photographic Materials and
Processes" by Stroebel, Compton, Current, and Zakia. Their graphs cover
a log E range of about 3.2.
So I think you're understating the dynamic range over which film
captures some information.
Another issue is that, even if a scene has only a 64:1 dynamic range,
that doesn't mean you can get away with quantizing the intensity to only
6 bits. What a photographer really wants is an encoding that is fine
enough so that you can't see the discrete steps between adjacent codes,
even at the darker end of the intensity range. This requires more bits.
For example, suppose we're using a linear encoding of intensity into
pixel value. Also suppose that the eye can't see a step change that's
smaller than about 3% in the dark end of the dynamic range of the image,
which has a total range of 64:1. To keep the largest step change to
about 3%, adjacent pixel codes must not differ by more than that in
intensity. This requires about 5 bits to represent the dark end of the
dynamic range, so that adjacent values would be (for example) 31 and 32.
But the brightest part of the image is 64 times as bright, so its code
value must be 2048, which requires 11 bits total. Basically, of the 11
bits, 6 bits is needed for the 64:1 dynamic range, and the other 5 bits
are needed to keep the step size small. Actual images would use pixel
values in the range 32-2047, not ones below 32. If you used only 6 bits
for the same image, adjacent pixels would have to be assigned values of
1 or 2, a 100% change in intensity.
In practice, we normally use a so-called gamma corrected encoding, which
makes the math more complex. And the eye's ability to see a step change
ranges from something less than 1% at high brightness to many percent in
dark areas of an image. But the analysis of how many bits you need is
still much the same: you need to keep the step size between adjacent
codes small enough at the lower end of the useful dynamic range, and
that always requires more bits than simply encoding the dynamic range.
Dave
Similar Threads
