«

Bruce Fraser's white paper at

http://www.adobe.com/digitalimag/pdfs/linear_gamma.pdf

makes the case that half the information in a digital capture is in
the fifth, or highest, stop. And some digital cameras can capture and
image that is overexposed, thereby driving image information into that
often little-used last stop. The highlights, apparently blown out, can
be recovered in Photoshop's raw image converter.

Using a Canon EOS 10D, I've tried this with some amazingly good
results and some poor results. I'd like to hear about your 10D
experience in this area.

--David

"David Ellis" <> wrote in message
news:...
> Bruce Fraser's white paper at
>
> http://www.adobe.com/digitalimag/pdfs/linear_gamma.pdf
>
> makes the case that half the information in a digital capture is in
> the fifth, or highest, stop. And some digital cameras can capture and
> image that is overexposed, thereby driving image information into that
> often little-used last stop. The highlights, apparently blown out, can
> be recovered in Photoshop's raw image converter.
>
> Using a Canon EOS 10D, I've tried this with some amazingly good
> results and some poor results. I'd like to hear about your 10D
> experience in this area.
>
> --David

I quite often adjust the gamma of the photos, from 10D or other cameras,
usually between 1.2 and 1.5 in PSP. Can really bring out the detail but not
sure about the accuracy of the result compared to the original.

David Ellis wrote:
> Bruce Fraser's white paper at
>
> http://www.adobe.com/digitalimag/pdfs/linear_gamma.pdf
>
> makes the case that half the information in a digital capture is in
> the fifth, or highest, stop. And some digital cameras can capture and
> image that is overexposed, thereby driving image information into
that
> often little-used last stop. The highlights, apparently blown out,
can
> be recovered in Photoshop's raw image converter.

I've always wondered how this could be possible. 12 bits of
information, if highlights are blown out then surely that means the
level is at 4095 and can go no higher. How could you possibly 'recover'
extra information from this?

However ... after seeing this I finally understand why:

"... so cameras show the histogram of the image after processing using
the camera's default settings. Most cameras apply a fairly strong
S-curve to the raw data so that the JPEGs have a more film-like
response, with the result that the on-camera histogram often tells you
that your highlights are blown when, in fact, they aren't."

I'm a little miffed that my very expensive camera cannot accurately
inform me of truly blown highlights.

>The highlights, apparently blown out, can
>be recovered in Photoshop's raw image converter.

If only one channel is clipped then any RAW converter can salvage some
data and reconstruct something. If two channels are clipped they can
still salvage a little bit but with more guesswork and less accuracy.

If all three channels are clipped then there's nothing any RAW
converter can do about it.

On 9 May 2005 12:06:48 -0700, wrote:

>David Ellis wrote:
>> Bruce Fraser's white paper at
>>
>> http://www.adobe.com/digitalimag/pdfs/linear_gamma.pdf
>>
>> makes the case that half the information in a digital capture is in
>> the fifth, or highest, stop. And some digital cameras can capture and
>> image that is overexposed, thereby driving image information into
>that
>> often little-used last stop. The highlights, apparently blown out,
>can
>> be recovered in Photoshop's raw image converter.
>
>I've always wondered how this could be possible. 12 bits of
>information, if highlights are blown out then surely that means the
>level is at 4095 and can go no higher. How could you possibly 'recover'
>extra information from this?
>
>However ... after seeing this I finally understand why:
>
>"... so cameras show the histogram of the image after processing using
>the camera's default settings. Most cameras apply a fairly strong
>S-curve to the raw data so that the JPEGs have a more film-like
>response, with the result that the on-camera histogram often tells you
>that your highlights are blown when, in fact, they aren't."
>
>I'm a little miffed that my very expensive camera cannot accurately
>inform me of truly blown highlights.

The 10D practice of showing a jpeg histogram, in spite of capturing in
raw format, makes using a +1 exposure bias tricky because at the time
of shooting the amount of information spilled into the sixth stop (if
you'll allow that technical leap) cannot be gauged by viewing the LCD.
I wonder if the $8000 EOS-1Ds Mark II offers better information.

With the 10D I've captured about 200 files using +1 exposure bias (+2
is too extreme) and evaluative metering. About 10% of those have
pixels where at least one channel has a value of 255. Bill Hilton
pointed out, "If only one channel is clipped then any RAW converter
can salvage some data and reconstruct something." When two channels
are blown out, I'm finding the PS RIC reconstruction produces lower
quality than I would like, but it would be quite acceptable for
vacation snapshots.

For the files where the +1 exposure bias worked as expected , the
extra stop of dynamic range yields highlight detail in a sunny outdoor
scene shot through a window while showing shadow detail for the
interior room without flash fill. I've only achieved this before by
layering and masking in Photoshop two precisely-aligned images, one
exposed for highlight, the other for shadow. A tricky business if one
is shooting a basketball game. :=)

However, as "dylan" commented ("...not sure about the accuracy of the
result compared to the original."), I'm seeing problems with loss of
detail in yellow highlights even when the red and green channels are
clipped in very few pixels.

I may not be enough of a scientist to correctly evaluate what's
happening, but this kind of dynamic-range gain is a most interesting
topic. Hopefully someone who has done similar experiments, or found
reports on the topic, will chime in.
--David

>David Ellis writes ...
>
>The 10D practice of showing a jpeg histogram, in spite of capturing in

>raw format, makes using a +1 exposure bias tricky ...
>I wonder if the $8000 EOS-1Ds Mark II offers better information.

I have a 1Ds and a 1D Mark II and I don't trust the histogram on those
either, for the same reasons. You eventually learn there's a 'fudge
factor' and only go so far to leave a bit of working room.

>With the 10D I've captured about 200 files using +1 exposure bias ...
>this kind of dynamic-range gain is a most interesting topic.

Here's another write-up on this which you may find interesting (I
haven't read the Fraser link yet so not sure how much is duplicated)
.... http://www.luminous-landscape.com/tu...se-right.shtml

>Hopefully someone who has done similar experiments, or found
>reports on the topic, will chime in.

I ran some tests trying to decide if it were better to shoot metered at
a low ISO or to shoot with exposure compensation at higher ISO (ie, ISO
100 vs ISO 200 @ +1 vs ISO 400 @ +2 vs ISO 800 @ +3), with the
overexposure compensated for during RAW conversion (note you can only
do this with low contrast scenes since with a full tonal range you
start to clip with the + compensation). My conclusion was it made
little difference in image quality whether you shot at ISO 400 @ +2 and
adjusted during RAW conversion compared to shooting at ISO 100 at 0
compensation. The problem with +2 as you note is that you get close to
the right edge of the histogram and can clip a channel, so as a
practical matter if I have exposure room on the histogram to go +2 or
+1 I would just as soon lower the ISO instead and shoot at 0 (ie,
metered) at the same shutter speed.

You can check your camera to see if you get the same results ... here
are two links showing what I got ...

http://members.aol.com/bhilton665/iso800_200.jpg (shot a gray card at
800, 800 -2 exposure +2 RAW conversion, 800 +2 exposure -2 RAW
conversion and at 200. All should give roughly equivalent exposures
but note the 800 -2 exp/+2 RAW is the noisest, 800 @ 0 second noisest
(as predicted by the "expose right" articles), and 800 +2 exp/-2 RAW is
pretty similar to the ISO 200 shot, noise-wise.

http://members.aol.com/bhilton665/iso800_100.jpg ... 800 +3 exp -2.2
RAW to get the exposure right vs ISO 100 metered ... looking at the
histograms you can see you are getting close to the clipping point at
+3 with nothing brighter than a gray card so any scene that would have
a decent range of tonal values would already be clipping. It's simpler
for me to just lower the ISO and shoot as metered since I get the same
end results (maybe a little smoother at 100 in this sample ... YMMV
depending on your camera and RAW converter).

Pretty easy to run your own tests with your camera and a gray card.

Bill

Bill Hilton wrote:
>
> I ran some tests trying to decide if it were better to shoot metered at
> a low ISO or to shoot with exposure compensation at higher ISO (ie, ISO
> 100 vs ISO 200 @ +1 vs ISO 400 @ +2 vs ISO 800 @ +3), with the
> overexposure compensated for during RAW conversion (note you can only
> do this with low contrast scenes since with a full tonal range you
> start to clip with the + compensation). My conclusion was it made
> little difference in image quality


It's not a big difference but I tried loading your test shots in
irfanview & boosting the contrast way way way up and there is a very
obvious differnce if you exaggerate it enough. But I agree, I can't tell
without that exaggeration, except of course the underexposed one. It may
be though that with a low contrast scene, you *would* want to increase
contrast significantly.

Here's my test on a D70 enlarged 200%:
<http://www.edgehill.net/1/?SC=go.php&DIR=Misc/photography/expose-right>


> whether you shot at ISO 400 @ +2 and
> adjusted during RAW conversion compared to shooting at ISO 100 at 0
> compensation. The problem with +2 as you note is that you get close to
> the right edge of the histogram and can clip a channel, so as a
> practical matter if I have exposure room on the histogram to go +2 or
> +1 I would just as soon lower the ISO instead and shoot at 0 (ie,
> metered) at the same shutter speed.
>
> You can check your camera to see if you get the same results ... here
> are two links showing what I got ...
>
> http://members.aol.com/bhilton665/iso800_200.jpg (shot a gray card at
> 800, 800 -2 exposure +2 RAW conversion, 800 +2 exposure -2 RAW
> conversion and at 200. All should give roughly equivalent exposures
> but note the 800 -2 exp/+2 RAW is the noisest, 800 @ 0 second noisest
> (as predicted by the "expose right" articles), and 800 +2 exp/-2 RAW is
> pretty similar to the ISO 200 shot, noise-wise.
>
> http://members.aol.com/bhilton665/iso800_100.jpg ... 800 +3 exp -2.2
> RAW to get the exposure right vs ISO 100 metered ... looking at the
> histograms you can see you are getting close to the clipping point at
> +3 with nothing brighter than a gray card so any scene that would have
> a decent range of tonal values would already be clipping. It's simpler
> for me to just lower the ISO and shoot as metered since I get the same
> end results (maybe a little smoother at 100 in this sample ... YMMV
> depending on your camera and RAW converter).
>
> Pretty easy to run your own tests with your camera and a gray card.
>
> Bill
>

--
Paul Furman
http://www.edgehill.net/1
san francisco native plants

Bill Hilton wrote:
> Here's another write-up on this which you may find interesting (I
> haven't read the Fraser link yet so not sure how much is duplicated)
> ... http://www.luminous-landscape.com/tu...se-right.shtml

That's where I originally learned the "expose to the right" technique,
but the article places a lot of faith in the histogram that the camera
presents:


"But, we all know (or at least should by now) that the worst sin in
digital imaging is to blow out the highlights - just as it was when
shooting slide film. Once they're blown (past the right-hand edge of
the histogram) it's bye-bye data."
[...]
"The simple lesson to be learned from this is to bias your exposures so
that the histogram is snugged up to the right, but not to the point
that the highlights are blown. This can usually be seen by the flashing
alert on most camera review screens. Just back off so that the flashing
stops."


That's what I've been doing - if I see the camera indicating blown
highlights after the shot I will usually take the shot again with a
slight negative exposure correction... unless it's something
difficult/obvious like the sun hiding behind a tree branch.

One suggestion I've seen on another forum is to set your in-camera
defaults to produce a flatter, lower contrast image which will show a
less aggressive histogram. I'm going to do some experimenting with this
today, modifying the in-camera settings and trying various levels of
overexposure to see what happens.

> I ran some tests trying to decide if it were better to shoot metered
at
> a low ISO or to shoot with exposure compensation at higher ISO (ie,
ISO
> 100 vs ISO 200 @ +1 vs ISO 400 @ +2 vs ISO 800 @ +3), with the
> overexposure compensated for during RAW conversion (note you can only
> do this with low contrast scenes since with a full tonal range you
> start to clip with the + compensation). My conclusion was it made
> little difference in image quality whether you shot at ISO 400 @ +2
and
> adjusted during RAW conversion compared to shooting at ISO 100 at 0
> compensation.

For normal exposure times it probably doesn't make much difference, but
I've found that if you're planning to have your shutter open for
minutes, longer exposures at a lower ISO have less noise than shorter
exposures at a higher ISO. Makes sense: I believe that higher ISO in
camera is achieved by cranking up the analog gain.

>Paul Furman
>Here's my test on a D70 enlarged 200%:
<http://www.edgehill.net/1/?SC=*go.php&DIR=Misc/photography/ex*pose-right>


These seem a good bit noiser at 200% than the 1D Mark II shots I posted
at 400%, at least to my eyes.

On 9 May 2005 20:03:14 -0700, wrote:

>Bill Hilton wrote:
>> Here's another write-up on this which you may find interesting (I
>> haven't read the Fraser link yet so not sure how much is duplicated)
>> ... http://www.luminous-landscape.com/tu...se-right.shtml
>
>That's where I originally learned the "expose to the right" technique,
>but the article places a lot of faith in the histogram that the camera
>presents:
>
>
>"But, we all know (or at least should by now) that the worst sin in
>digital imaging is to blow out the highlights - just as it was when
>shooting slide film. Once they're blown (past the right-hand edge of
>the histogram) it's bye-bye data."
>[...]
>"The simple lesson to be learned from this is to bias your exposures so
>that the histogram is snugged up to the right, but not to the point
>that the highlights are blown. This can usually be seen by the flashing
>alert on most camera review screens. Just back off so that the flashing
>stops."

I also learned "expose to the right" at luminous-landscape. But the
"Just back off so that the flashing stops" troubled me because
retaining the flashing is precisely what's required to gain the
benefits of using positive exposure bias in the camera and negative
exposure values in the raw image converter.
>
>
>That's what I've been doing - if I see the camera indicating blown
>highlights after the shot I will usually take the shot again with a
>slight negative exposure correction... unless it's something
>difficult/obvious like the sun hiding behind a tree branch.
>
It is the high-contrast scene that utilizes the expose-to-the-right
technique.
<snip>

Because of my limited experience with this subject, I had trouble
relating to the 18% gray test targets in the other article. A regular
photo is easier for me to follow.

At www.ellisisle.com/raw_image_tests see a single raw file, processed
a variety of ways. After exposure the 10D's Info LCD had lots of
flashing. I suggest you open all four .htm files on screen together,
each in a tab or its own window, for comparison purposes.

The images show the difference between dealing with apparently
blown-out highlights in Photoshop Curves (2.htm) and dealing with them
in the raw image converter (3.htm and 4.htm). The latter provides
highlight detail recovery in an apparently overexposed image that
cannot be achieved with Photoshop Curves.

The broad tonal range in 4.htm is dependent on having the 10D's
blown-out highilght indication flashing.
--David