Re: With 12/16bit RAW do you still need an ND Grad

"Roger N. Clark (change username to rnclark)" <username@xxxxxxxxx> wrote:
Floyd Davidson wrote:
"Roger N. Clark (change username to rnclark)" <username@xxxxxxxxx> wrote:
Floyd Davidson wrote:
The "lowest usable signal" is determined by the Signal to Noise
Ratio, and with digitally sampled images that is generally
accepted to be a zone that has at least 8 levels. It is indeed
limited by the bit depth, and with JPEG that is 8 fstops. (It
is not directly determined by the 8-bit depth though, because
gamma correction is applied.)
This is not correct. First, this continued reference to "gamma"
correction is a misnomer.
It *is* correct, though you are certainly right that I'm not
using pedantically correct terminology to describe it. But if
we get tied up with terminology, people lose sight of how it
applies to the questions they asked...

No your statement
It is indeed
limited by the bit depth, and with JPEG that is 8 fstops.

is NOT correct.

It is absolutely correct. I am well aware that you have seen
the data which proves it many times...

Non linear encoding allows you to record a larger range
than 8 stops. What you sacrifice is precision between stops.

"Precision between stops" is defined, of course, by *noise*, and when
noise is too high the results are not useful.

For example, I can encode 26 stops by encoding integer log base 10,
so each bit is a factor of ten, or 10^8 -1. The non-linear encoding
in digital cameras typically encodes 11 to 12 stops.

JPEG encoding looks like this:

Range Number of Values
1 69
2 50
3 37
4 27
5 20
6 14
7 10
8 8 <=== Minimum useful number of values
9 6
10 4
11 3
12 2
13 2
14 1
15 1
16 0
17 1
18 1

Obviously if we arbitrarily say that fewer than 8 values per
zone is a higher noise level than is useful, the *useful*
*dynamic* *range* is 8 fstops. On the other hand if you want to
claim that any detectable signal change (i.e., SNR = 1:1) is a
valid point, you also have to accept that it means claiming at
least a 13 fstop dynamic range (and some would claim 18 fstops),
which is *clearly* *ridiculous*, given the resulting
posterization when viewing the results. It may have 13+ fstops
that can be measured, but that is not what can be pulled out to
provide a pleasant looking picture.

Indeed, look at the "great" looking shadows in your URL cited
below! Not exactly a lot of "useful" data down there... :-)

Or at least not useful if the purpose of the image is to be eye
appealing; but on the other hand for a star gazer it might be
*wonderful* data! That is why the noise criteria is subjective
and arbitrary.

The function for the "standard" transfer
curve in digital cameras matches closely the characteristic curve of print film.
It is a continuously varying gamma function, meaning if you applied
a gamma function, the exponent would have to be variable with scene
intensity. A gamma function plots as a straight line on log-log
plot. The transfer curve for digital cameras do not plot as straight lines.
e.g. see:
http://www.clarkvision.com/imagedetail/dynamicrange2
All true, but of no particular significance to the question I
had at hand, IMHO. However, your point *is* interesting and
probably of value to some readers, so I do think it is good to
bring up in a followup.

It is exactly to the point. It is the non linear encoding,
not gamma encoding, that allows digital camera images to cover
the range they do.

Call it what you like. As you can see from the encoding used
with JPEG, it provides 8 fstops of dynamic range (with an 8 bit
non-linear encoding).

Whether is is strictly speaking a gamma corrected non-linearity or
some other correction, the fact is that JPEG provides 8 fstops of
useful dynamic range.

The
standard definition is max signal / noise floor.
No it isn't. That is the definition for the dynamic range of
an analog electronic sensor. It is interesting, and is a useful
value to know about, but it doesn't necessarily translate
directly to the dynamic range of the digital data produced by a
codec. It certainly defines the upper limit for the digital
data though.

It is also the definition used for digital imaging data. It is the
definition used in the scientific terrestrial and planetary remote sensing
fields. It is the definition used by sensor manufacturers in
defining the dynamic range of their products.

It is the definition used for the analog dynamic range of the
*sensor*, and provides a very useful method for comparing them.
It is not the dynamic range useful in an image file produced
after the sensor data is quantized and formatted into a digital
image that will be printed.

In DSLRs, for example,
that is higher than 9-stops.
You mean that for CCD and CMOS electronic sensors it is higher
than 9 fstops. But that data is currently being digitized and
captured via a 12 bit codec, and images that result from such a
process have a *useful* dynamic range of 9 fstops, maximum.
That is a figure arrived at simply by the empirical observation
that if zone has fewer than 8 values it will appear to the human
eye to be posterized. It is entirely arbitrary and subjective
to choose 8 values...

So we are using your subjective definitions?

I did not invent them. You should not ignore them.

Some measured dynamic ranges are shown as
a function of ISO in Figure 5 (~11.5 stops) at:
http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary
A 9-stop definition would mean a lower end signal-to noise ratio
of around 5.5. Note some high ISO film images do not have that high a
ratio. But people still view them as acceptable images.
And that is what causes so much confusion. It might well be
that for *some* particular use a zone with only 2 values will be
considered "useful"; and just as clearly that will not normally
be the case.

Incorrect. See below.

Except that it is quite correct.

An 8-bit RGB image with gamma adjustment actually has more potential DR
than a 12-bit linear RAW file. You don't see it, because of the noisy
JPEG files are limited to an 8 fstop useful dynamic range. 12
bit linear
data is limited to a 9 fstop useful dynamic range.
The data at
http://www.clarkvision.com/imagedetail/dynamicrange2
show over a 10 stop range for jpegs.
The data at

http://www.clarkvision.com/imagedetail/raw.converter.shadow.detail
show no clipped highlights at +2.0 stops in the sunset image to
shadow detail down to -7.6 stops. That is more than your 8 stops.
You are not addressing the same definition of "useful dynamic
range" that I am. All we have to do is assume that 6 levels is
satisfactory, and JPEG has 9 fstops of dynamic range. That is a
reasonable, and very arbitrary, assumption.
I have made it clear that the commonly acceptable number is 8
values, and that I am not going to argue that value. I will,
obviously, argue the significance of the value chosen though!
:-)

The following page is under construction, but has enough data
to illustrate what we are talking about. It is a test series
showing exposure latitude of a digital camera.

It ignores quantization distortion. Which means it is
interesting and helpful, but not definitive. You are getting a
pretty fair measure of the electronic sensor, which is very
useful for many things, but it doesn't equate to what a
landscape photographer will experience trying to pull out the
shadows of an underexposed image. (The best example I can think
of is trying to bit diddle an image of the Aurora Borealis,
where huge amounts of the image area are going to be way down in
the deep shadows. Trying to pull much detail out of the shadows
is a very nice demonstration of posterization caused by
quantization distortion.)

I did an exposure series
from -11.4 to +14 stops on both a 1D Mark II and some print film.
The page so far shows the 1D Mark II results.
The metered value = 18% gray card. The macbeth color chart, bottom
raw, forth patch from the left is 20% reflectance, so closest
to 18% gray. The brightest patch is +2.2 stops, the darkest -2.7 stops.
The idea is to produce the best image at each exposure.

You can look at the images and decide for yourself what is acceptable.
Exposures -5 to +2 stops produced excellent images, with
no saturation and excellent shadows. The meter +3 second from left
white panel is not saturated (59% reflectance) but the whitest
panel is, so that is +1.6 stops, so the real dynamic range is
+3+1.6 to -5-2.7 = 4.6 + 7.7 = 12.3 stops of excellent image quality!
The meter -7 is still pretty good, for a total range of 14.3 stops!
Each can decide on the lower limit to useful images. The meter -8
stops is also pretty good (better than some high speed film images)
and there is still information recorded in the -11 stops (label is 11, but
actual is 11.4). (Also note the -10 = 10.4 and -11 = 11.4 and were obtained
in shadow other in full sun, so light was blue sky and reds are lost.
I need to redo the low end).

Anyway, it is an impressive range, and a range greater than 12-bit linear
encoding. There is a scientific explanation too.

Yes there is! The scientific explanation is that your method is
limited and the analysis does not take that into account.

--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska) floyd@xxxxxxxxxx
.

Relevant Pages

  • Re: Free Landscape Photography Book!
    ... but I want mine to look like my perceived reality. ... It's impossible to make a high dynamic range image look exactly like ... for the cave and one normal one, then blended them using HDR. ... unrealistic images with it. ...
    (uk.rec.walking)
  • Re: With 12/16bit RAW do you still need an ND Grad
    ... and with digitally sampled images that is generally ... >>> limited by the bit depth, and with JPEG that is 8 fstops. ... What you sacrifice is precision between stops. ... defining the dynamic range of their products. ...
    (rec.photo.digital)
  • Re: With 12/16bit RAW do you still need an ND Grad
    ... and with digitally sampled images that is generally ... >>> limited by the bit depth, and with JPEG that is 8 fstops. ... What you sacrifice is precision between stops. ... defining the dynamic range of their products. ...
    (rec.photo.digital)
  • Re: Return to film... True!
    ... :> no real problem with digital images. ... I disagree in the belief that images need lots of dynamic range to look ... Negative films are thought to have more ... No other option exists except to not take the picture - try ...
    (rec.photo.equipment.35mm)
  • Re: Exposure problem with 420EX and 20D - please help :-(
    ... stops. ... The other alternative is to shoot RAW and use RAWShooter to develop the images. ... The only exception is when the scene has 60% or more of white and then you really need to under expose. ... David Littlewood ...
    (rec.photo.digital.slr-systems)
Loading