dynamic range

Paul Furman wrote:

Damn, I only have CS, not CS2. I need to do two conversions to get
everything. Huge difference!

Camera Raw 3 is itself enough reason to upgrade to CS2. It essentially makes
Photoshop a plugin for Camera Raw.

How does LAB help?

Whole different religion. LAB separates luminance from color, and makes some
editing moves easier (and makes some possible).

--
Jeremy |

"Paul Furman" wrote in message
t...
Could there be a filter to increase dynamic range? Kinda like a
polarizer that blocks bright light while letting in dim shadow detail?
As I understand, whatever the camera is capable of capturing before the
highlights blow is squeezed into the same file format between black &
white so a high DR camera is actually going to be a bit less contrasty.

With film there is a way to use a double exposure to do this. First shoot a
gray card (not 18% I think, but I don't know the details and don't have time
to look it up), then second exposure you expose for the highlights. This
pre-exposure is supposed to affect the film somehow that gives the shaddows
some help. I first read about this in one of Ansel Adams' books. I have no
idea if there is some trick that could simulate this, although I read some
blending techniques that work pretty well at Luminous Landscapes.

--
Mark

Photos, Ideas & Opinions
http://www.marklauter.com/gallery

A few things:

If dynamic range is your primary concern in photography, even more than
resolution, then you should look into Fuji's Nikon-mount S3 Pro. Each
pixel location is actually a large pixel and a small pixel, capturing
shadows and highlights. The camera has a relatively low resoluion at
around 6 megapixels, but it apparently has tremendous dynamic range.

Barring that, you could get Adobe Camera Raw, which has tremendous
highlight recovery capabilities. Check out "Camera Raw with Adobe
Photoshop CS2" by Bruce Fraser for some wonderful insights. You'll
start to convert RAW files in a quarter of the time and get 500% better
results. In any case, that's how I estimate my improvement after
reading that book.

Long story short, when exposing for RAW, you always want to do the
following two things:
1.) Expose low enough that nothing blows out that shouldn't be blowing
out (obviously, if you've got a light source in the shot, you might not
mind that it blows out; it would on even the best films if exposed
properly for the whole scene.

2.) Expose as high as you can with regards to (1.). Basically, you
don't want any blowout, but you want to be knocking on the door of
absolute white.

The reason for this is better explained in the book (and in the "Expose
Right" article on Luminous Landscape), but basically, a chip records
light in a linear fashion and this means that your hottest stop
contains half of your picture information, and your second hottest stop
contains one quarter (which makes perfect sense if you understand our
logarithmic perception of light). Adobe Camera Raw *basically* allows
you to distribute these bits more evenly throughout the image and
results in better conversions. Doing this, you'll find that most
cameras have terrific dynamic range.

As for specific DR ratings for cameras, Imaging Resource has started to
collect data on this. A recent example of their stats can be found on
the review page for the Canon EOS 1Ds MkII:

http://www.imaging-resource.com/PROD...DS2IMATEST.HTM

The methodology leaves plenty to be desired, so take it with some salt,
but they conclude that, independent of RAW conversion, the following
make the top 5 DSLRs in terms of DR:

1. Fuji S3 Pro : 7.94 stops at "high" quality
2. Nikon D50 : 7.36 stops at "high" quality
3. Canon 20D : 7.29 stops at "high" quality
4. Canon Rebel XT : 7.11 stops at "high" quality
5. Olympus E-VOLT : 7.07 stops at "high" quality

They only test RAW conversion on two cameras: the S3 and the 1Ds2, and
these both see an increase of a little over a stop because of it. So
you can conclude that alot of DSLRs, even cheap ones, can get 8 stops
with RAW conversion. That's nothing to shake a stick at.

Anyone know what the typical range of luminance is in natural settings
around the world? If we trust the "Sunny 16" rule, then you have
ambient light of around 6826 footcandles in bright sunlight. Which
suggests to me that the 14-stop difference between 1 footcandle and
8192 footcandles would get you nearly everything you want, if exposed
properly. Surely, then, a 16-stop range from 1 footcandle to 32,768
footcandles would be enough, right? And even then, that's if you wanted
full seperation between 1 and 2 footcandles. How long until we have
sensors that can give us that kind of range? At that point, do you even
rationalize exposure in the same way?

Food for thought,

Will

In message .com,
wrote:

If dynamic range is your primary concern in photography, even more than
resolution, then you should look into Fuji's Nikon-mount S3 Pro. Each
pixel location is actually a large pixel and a small pixel, capturing
shadows and highlights. The camera has a relatively low resoluion at
around 6 megapixels, but it apparently has tremendous dynamic range.

Yes, but mainly into the bright end. That's great for shooting in
daylight with dark shadows (which aren't really very dark in an absolute
sense), or scenes that contain lights in them, but this does nothing for
sensitivity, which is part of what is desirable about "dynamic range".
It doesn't help you hand-hold a shot in very low light, which is
something that *would* be the by-product of having tremendous dynamic
range in a single pixel. I like to think of those extra pixels as
"highlight helpers".

I'm curious now about how the data is used in RAW converters. Are the
two sets of data merged into 16-bit linear data, or are they kept
separate in the conversion process so the effect can be masked out where
it isn't needed? I'd like to find some RAW samples and see what the DNG
converter does with them (uncompressed DNG holds literal RAW bitmaps at
the end, which can be loaded into PS as ".raw", but get posterized by
the 15-bit + 1 format).
--


John P Sheehy

In message .com,
wrote:

Barring that, you could get Adobe Camera Raw, which has tremendous
highlight recovery capabilities. Check out "Camera Raw with Adobe
Photoshop CS2" by Bruce Fraser for some wonderful insights. You'll
start to convert RAW files in a quarter of the time and get 500% better
results. In any case, that's how I estimate my improvement after
reading that book.

At least for some cameras, ACR doesn't render colors well when
"recovering highlights" (they were never really lost, if you shot RAW).
Look at this grey scale, over-exposed by two stops and rendered with -4
EC in ACR:

http://www.pbase.com/jps_photo/image/55953848

There really should be nothing higher than middle grey in this render,
as there is only 4.5 stops above middle grey in the least sensitive
(red) channel, so -4 should bring the top of the scale to 1/2 stop
above middle grey. ACR "nails" the RAW highlights to 255 in the output,
even when you reduce the exposure drastically. This distorts the
highlights, and should be an option; not hard-wired into the code. It
would be much better if there was an "exposure" control that was purely
linear, that acted on the data before variable gamma curves are applied,
and they should be optional. Also, look at the green color in the 5th
grey rectangle from the left. The program does a good job of rendering
the brightest rectangles grey, but it gets confused somewhere in the
transition point. The rectangle rendered green is the darkest one in
which the RAW data starts to clip; only the green channel is clipping
there. In the 4th rectangle, both the blue and green are clipped. The
red channel provides the distinction between rectangles 2, 3, and 4. The
5th rectangle really should have been treated as greyscale, not color.

I have an old version of Capture One around (1.2), and it renders the
way you'd expect; it maintains grey highlights from one or two channels,
but leaves no color cast, and the RAW clipping point gets pulled down
below 255 in the output when you render with maximum negative exposure
compensation.

IMO, ACR is not the best tool for recovering highlights, at least with
some cameras.
--


John P Sheehy

wrote:

In message ,
Andrew Haley wrote:


Think, for a moment, about lens flare. A really good fixed focal
length lens has 0.6% flare, according to
http://medfmt.8k.com/mf/flare.html. So, if there is a very bright
area somewhere in the image, any very dark areas will be veiled in
flare -- even with a very good lens.

If the lens can't deliver an image with huge dynamic range, the sensor
won't record it.

It doesn't work like that. Flare does not veil anything; it adds to the
other signal.

If anything, flare increases dynamic range by decreasing contrast like
the filters discussed above. It seems unavoidable that it muddies up the
image but it does sort of increase dynamic range.

On my Canon 20D, when there is no exposure, the RAW data is 128,128,128.
That is called the blackpoint. If lens flare adds 100,200,150 (much,
much more than your 0.6%, BTW; this would be about 10%), then the
effective blackpoint (assuming you want to remove the flare) becomes
228,328,278. Any small shadow signal is *added* to this; not "veiled"
by it. You don't lose any shadows because of this; you lose a small
amount of highlights, as it takes less "real, desired" image brightness
to clip the RAW data, but you'd be losing something like 200 out of 3968
(already less than 4096 because of the original 128 blackpoint) possible
levels in the green channel, for 3768 usable RAW levels.

log(3768/3968) / log(2) = -0.0746, or a loss of 0.07 stops of dynamic
range. This is not an understatement; 200 is really much more than what
you'd expect with a moderately flare-prone lens in normal circumstances.
Even if you are shooting into a cave with the sun shining on the lens,
and the flare is most of the signal. it will be about 500 in the green
channel with auto-exposure (that is where middle grey generally lies),
and your shadows will still be intact. For such a shot, with such low
contrast, you should use +2 EC (or more, if you know where the RAW
clipping point is), so your signal is as trong as possible; keeping the
flare exposed lower does not help; as the "desired signal" is then also
lower. A high-pass filter at a very low frequency would eliminate and
difference in flare from center to edge.

Now, I'm not saying that the RAW converters we have do all of this well,
right now, but that is what is happening in the RAW data, and how it can
be overcome.

--
Paul Furman
http://www.edgehill.net/1
Bay Natives
http://www.baynatives.com

wrote:

In message .com,
wrote:


If dynamic range is your primary concern in photography, even more than
resolution, then you should look into Fuji's Nikon-mount S3 Pro. Each
pixel location is actually a large pixel and a small pixel, capturing
shadows and highlights. The camera has a relatively low resoluion at
around 6 megapixels, but it apparently has tremendous dynamic range.


Yes, but mainly into the bright end.

That's still useful. Probably that's when I really want more dynamic
range. It looks like the Fuji is about the same $1700 price as a D200.
That's how I got into this discussion, thinking about what a D200 is
going to give me that I don't have in a D70. I'd be curious to test a
Fuji. I think sheer pixel count also improves dynamic range though & the
D200 has 11MP which are pretty clean noise-wise. I wouldn't really want
that huge body the Fuji has either & the D200 is just a tad bigger than
a D70. If I was a pro with the budget, the Fuji would be a useful second
body for particular uses.

That's great for shooting in
daylight with dark shadows (which aren't really very dark in an absolute
sense), or scenes that contain lights in them, but this does nothing for
sensitivity, which is part of what is desirable about "dynamic range".
It doesn't help you hand-hold a shot in very low light, which is
something that *would* be the by-product of having tremendous dynamic
range in a single pixel. I like to think of those extra pixels as
"highlight helpers".

I'm curious now about how the data is used in RAW converters. Are the
two sets of data merged into 16-bit linear data, or are they kept
separate in the conversion process so the effect can be masked out where
it isn't needed? I'd like to find some RAW samples and see what the DNG
converter does with them (uncompressed DNG holds literal RAW bitmaps at
the end, which can be loaded into PS as ".raw", but get posterized by
the 15-bit + 1 format).

--
Paul Furman
http://www.edgehill.net/1
Bay Natives
http://www.baynatives.com

In message ,
Paul Furman wrote:

wrote:

In message ,
Andrew Haley wrote:


Think, for a moment, about lens flare. A really good fixed focal
length lens has 0.6% flare, according to
http://medfmt.8k.com/mf/flare.html. So, if there is a very bright
area somewhere in the image, any very dark areas will be veiled in
flare -- even with a very good lens.

If the lens can't deliver an image with huge dynamic range, the sensor
won't record it.

It doesn't work like that. Flare does not veil anything; it adds to the
other signal.

If anything, flare increases dynamic range by decreasing contrast like
the filters discussed above. It seems unavoidable that it muddies up the
image but it does sort of increase dynamic range.

It doesn't decrease contrast by using a system where 1 stop is a ratio
of 1.7 instead of 2; it is additive, and decreases contrast only by
losing the black end. It seems to have value with film, and it may even
have some value with JPEG, but the RAW data gains no dynamic range by
having a blanket of extra light across the frame; It steals a little bit
of the dynamic range, as I explained in the part of my post that you did
not comment on:

On my Canon 20D, when there is no exposure, the RAW data is 128,128,128.
That is called the blackpoint. If lens flare adds 100,200,150 (much,
much more than your 0.6%, BTW; this would be about 10%), then the
effective blackpoint (assuming you want to remove the flare) becomes
228,328,278. Any small shadow signal is *added* to this; not "veiled"
by it. You don't lose any shadows because of this; you lose a small
amount of highlights, as it takes less "real, desired" image brightness
to clip the RAW data, but you'd be losing something like 200 out of 3968
(already less than 4096 because of the original 128 blackpoint) possible
levels in the green channel, for 3768 usable RAW levels.

log(3768/3968) / log(2) = -0.0746, or a loss of 0.07 stops of dynamic
range. This is not an understatement; 200 is really much more than what
you'd expect with a moderately flare-prone lens in normal circumstances.
Even if you are shooting into a cave with the sun shining on the lens,
and the flare is most of the signal. it will be about 500 in the green
channel with auto-exposure (that is where middle grey generally lies),
and your shadows will still be intact. For such a shot, with such low
contrast, you should use +2 EC (or more, if you know where the RAW
clipping point is), so your signal is as trong as possible; keeping the
flare exposed lower does not help; as the "desired signal" is then also
lower. A high-pass filter at a very low frequency would eliminate and
difference in flare from center to edge.
--


John P Sheehy

wrote:

Paul Furman wrote:

wrote:

Andrew Haley wrote:

Think, for a moment, about lens flare. A really good fixed focal
length lens has 0.6% flare, according to
http://medfmt.8k.com/mf/flare.html. So, if there is a very bright
area somewhere in the image, any very dark areas will be veiled in
flare -- even with a very good lens.

If the lens can't deliver an image with huge dynamic range, the sensor
won't record it.

It doesn't work like that. Flare does not veil anything; it adds to the
other signal.

If anything, flare increases dynamic range by decreasing contrast like
the filters discussed above. It seems unavoidable that it muddies up the
image but it does sort of increase dynamic range.

It doesn't decrease contrast by using a system where 1 stop is a ratio
of 1.7 instead of 2; it is additive, and decreases contrast only by
losing the black end. It seems to have value with film, and it may even
have some value with JPEG, but the RAW data gains no dynamic range by
having a blanket of extra light across the frame; It steals a little bit
of the dynamic range, as I explained in the part of my post that you did
not comment on:

Heh, I didn't comment because I didn't understand g. But I can get the
idea that flare doesn't help raw data and there are plenty of tools to
bring up the shadows in PS & ACR so I guess that's the way to go, not a
flare inducing filter.

On my Canon 20D, when there is no exposure, the RAW data is 128,128,128.
That is called the blackpoint. If lens flare adds 100,200,150 (much,
much more than your 0.6%, BTW; this would be about 10%), then the
effective blackpoint (assuming you want to remove the flare) becomes
228,328,278. Any small shadow signal is *added* to this; not "veiled"
by it. You don't lose any shadows because of this; you lose a small
amount of highlights, as it takes less "real, desired" image brightness
to clip the RAW data, but you'd be losing something like 200 out of 3968
(already less than 4096 because of the original 128 blackpoint) possible
levels in the green channel, for 3768 usable RAW levels.

log(3768/3968) / log(2) = -0.0746, or a loss of 0.07 stops of dynamic
range. This is not an understatement; 200 is really much more than what
you'd expect with a moderately flare-prone lens in normal circumstances.
Even if you are shooting into a cave with the sun shining on the lens,
and the flare is most of the signal. it will be about 500 in the green
channel with auto-exposure (that is where middle grey generally lies),
and your shadows will still be intact. For such a shot, with such low
contrast, you should use +2 EC (or more, if you know where the RAW
clipping point is), so your signal is as trong as possible; keeping the
flare exposed lower does not help; as the "desired signal" is then also
lower. A high-pass filter at a very low frequency would eliminate and
difference in flare from center to edge.

--
Paul Furman
http://www.edgehill.net/1
Bay Natives
http://www.baynatives.com

As I understand it (and that may not be very well at all), one of the
differences between highlight recovery in ACR and in other software is
that ACR will allow a single channel to blow out and also allow you to
recover the other two. Other software, I know, views a single blown
channel as a blowout in all channels at that pixel location. Such a
distinction, then, could explain the artifact you observe if the grey
chart was lit with light of a different color than the camera's actual
color balance (no idea what that might be, but it must have one). In
such a case, a grey chart is a test that disadvantages ACR and
advantages others, just as a test with many non-neutral hues plays to
ACR's strengths and other software's weaknesses.

If my understanding is correct (and I trust that you'll be able to
acurately tell me if it is not), then it is really a question of
taste... should a blown channel render itself illegible because the
color information cannot be made fully accurate, or should attempts be
made to make it as accurate as possible? I certainly think that I
prefer the ACR way for my subject matter. If I need to correct neutral
subjects that pick up an unwanted color cast, I can always desaturate,
even selectively (your grey chart desaturated beautifully on my system,
and it looks like you'd be hard pressed to fault ACR's luminance
rendering). In my real-world tests, I like the color rendering of ACR
more than I like Aperture or CaptureOne LE. But to each his own.

Will