Dynamic Range Test

[John Sheehy]

Of course you need to substract.

... if you subtract the subject's dynamic range from the camera's dynamic range; that leaves the exposure lattitude for that subject. Or, subtract the exposure lattitude from the from the camer's dynamic range, leaving the subject's dynamic range.

Your logic is how some
experimentators are getting 12 stops for 20D.

The 20D has almost 12 stops of dynamic range, depending on your standards. what I have been talking abiut is the math after the standard has been set.

But to avoid that kind of useless discussions, you just make the
experiment right. Take a subject with 0 stops dynamic range. Then,
even if you add, you would get the correct result.

I don't necessarily disagree. I was working within the context of the information supplied. It all really depends on arbitrary standards. There is nothing clearcut about deciding whether differentiation has been achieved at the shadow end, with the low signal-to-noise ratio there.

As I already wrote earlier in this thread, the correct subject (0
stop DR) is a gray card with two texts, one lighter, another darker
and you need to be able to read both.
Isn't that easier than to make a dubious experiment and then
discuss endlessly?

I didn't say or imply, otherwise. I was working within the information given.

--
John

 

[John Sheehy]

I was generalizing. I had heard of the upper limit, but not the
lower, and didn't know the source of this info. I also don't know
what is constraining the signal. I can see that the lower limit
could be "thermal", either from the sensor or from the amplifier,
or just due to some signal bias. Also, I don't know the possible
ranges at the other ISOs. Is something limiting the input range to
the ADC or is there some software processing after the ADC that is
imposing the limit? Or do you really think that the ADC itself
won't go above or below these numbers??? And why? Could you point
to the source of this info or shed any light (pun intended) ?

The source is the RAW data itself; a blackframe of 1/8000s, ISO 100 results in a RAW histogram with values on a bell curve from about 120 to 130, most of which are 124 to 128, except for a few unresponsive or dead pixels.

A grossly over-exposed image results in mostly 4007 values, some in the 4000 to 4006 range, and a few dead pixels at very low numbers.
--
John

 

[EOSMan]

I found this article from DPReview:

http://www.dpreview.com/learn/?/Glossary/Digital_Imaging/Dynamic_Range_01.htm

--
EOSMan
http://www.pbase.com/eosman

 

[EOSMan]

http://www.dpreview.com/learn/?/key=histogram

--
EOSMan
http://www.pbase.com/eosman

 

[Doug Kerr]

Hi, B,

Thanks for the link to the Koren paper. It is very interesting and valuable.

Best regards,

Doug

 

[Porchini]

I was curious to see if your theory would apply in my test. Here is a sample containing lighter/darker text + texture.

It appears that I have lost ~ 2/3 of a stop from the previous experiment.

so you add that to the lattitude
for the dynamic range, totalling about 8 stops.

Why would you add? you need to substract, resulting in 6 about stops.

No; dynamic range of the subject PLUS the subject's exposure
lattitude equal the dynamic range; a two-stop subject with a
lattitude of four stops gives six stops of dynamic range:

• +
• +
• +

-

• +
• +
• +

Not that I believe that the exposure lattitude is really that low
at ISO 100.

Very realistic, in fact.

Did you see what his subject was? - Two tones, less than a stop apart.

--
John

 

[ronis]

I think you are quite close and actual sensor DR could be about 8 stops.
Texture was not necessary, you need a gray background.

And you need to fix the moment when one text disappears. Seems that you still have some headroom left

I was curious to see if your theory would apply in my test. Here
is a sample containing lighter/darker text + texture.

It appears that I have lost ~ 2/3 of a stop from the previous
experiment.

 

[John Sheehy]

I was curious to see if your theory would apply in my test. Here
is a sample containing lighter/darker text + texture.

It appears that I have lost ~ 2/3 of a stop from the previous
experiment.

Well, that's about all the accuracy you can expect with 1/3 stop increments. You are also involving the contrast of the viewer's monitor and heavy quantization by leaving the under-exposures dark. Personally, I would adjust RAW "exposure" so that all results were averaged to grey. It doesn't make any sense, IMO, to quantify the dynamic range of capture by letting the display run into its limitations.

--
John

 

[Bitplayer]

http://www.dpreview.com/learn/?/key=histogram

--
EOSMan
http://www.pbase.com/eosman

Thanks for the links. The first article is a good introduction. And it got me thinking. (Always a mistake...)

Why isn't the dynamic range of digital cameras tested and published? There is somewhat of a tradeoff between noise and dynamic range -- and the S/N ratio is tested. Wouldn't it be more meaningful to include a dynamic range figure, and to combine them in some Range/Noise figure of merit? For example, I believe the 20D has smaller sensor sites than the 10D/300D. It also has lower noise at higher ISO and more pixels. But what about the dynamic range? Was there a trade-off -- or did Canon manage to also improve the dynamic range while increasing resolution. And what about those prosumer cameras? Sure, we know that DSLRs have lower noise. But what about the dynamic range? The standard reply to the "should I get a prosumer or a DSLR" question begins with a statement that the DSLR has better noise performance. How come no one says "because it has a greater dynamic range?" (Do they?) And what about those Fujis and Sigmas with their special sensor designs?

OK, maybe I'm being a real gear-head. Maybe we should just look at the pictures and judge them. But they do have a lot of variables that makes comparison difficult. And this is a technical site.

Just wondering...

--
Bitplayer

To err is human, to post-process -- divine.

 

[Doug Kerr]

Doug

 

[MitchAlsup]

For a given low intensity of light and below, the output of the
sensor/amplifier/ADC will be 0. For a given high intensity of
light and above, the output of the sensor/amplifer/ADC will be 2* n

• 1, where n is the number of bits of the ADC. Please note that

the magnitude of n has nothing to do with the light intensities.
It is the difference, expressed in fstops or EVs, between the two
light intensities that is the dynamic range.

I'm going to add a couple more degredations:

The A/D is setup so that when no light has fallen on the sensor cell the A/D output is in the range of 50 counts to 129 counts. At the other end of the range, the sensor saturates below the A/D limit by 50 counts to 100 counts. So from a inherent range of 12 bits (4096 counts) only 3900 to 4000 counts are used. This causes a 0.2 stop degredation of the image data.

Worse still is the A/D quantiation error of 0.5 bits.

So, a 12-bit A/D is only delivering 11.3 bits of data.

But (the BIG but) even a perfectly exposed highlight that exactly matches the highest count the A/D can produce only has a signal to noise ratio of 240 (SQRT(cell capacity)); or slightly less than 8-bits of actual noise free data. Midrange data from the image will (almost) never have more than a S/N of around 100.
--
Mitch

 

[Al Pacheco]

What is being measured and evaluated in this thread is the dynamic range of the conversion software, not the camera's sensor. That is usually not very well understood. The 20D, likle the models before it, can easily exceed 12 stops of clean capture. Even the D30 can do that.

But the conversion software (or firmware if you shoot jpegs) is the determining factor as far as to the end result.

If you use Canon's software you will end up with one set of results. Capture One will render another set and aims to produce clean, noise-free shadows, sometimes at the expense of detail.

Adobe Camera Raw will "try" to stretch the most from the raw file and is good for that purpose.
It is truly the software that is being tested here...not the camera.

http://www.dpreview.com/learn/?/key=histogram

--
EOSMan
http://www.pbase.com/eosman

Thanks for the links. The first article is a good introduction.
And it got me thinking. (Always a mistake...)

Why isn't the dynamic range of digital cameras tested and
published? There is somewhat of a tradeoff between noise and
dynamic range -- and the S/N ratio is tested. Wouldn't it be more
meaningful to include a dynamic range figure, and to combine them
in some Range/Noise figure of merit? For example, I believe the
20D has smaller sensor sites than the 10D/300D. It also has lower
noise at higher ISO and more pixels. But what about the dynamic
range? Was there a trade-off -- or did Canon manage to also
improve the dynamic range while increasing resolution. And what
about those prosumer cameras? Sure, we know that DSLRs have lower
noise. But what about the dynamic range? The standard reply to
the "should I get a prosumer or a DSLR" question begins with a
statement that the DSLR has better noise performance. How come no
one says "because it has a greater dynamic range?" (Do they?) And
what about those Fujis and Sigmas with their special sensor designs?

OK, maybe I'm being a real gear-head. Maybe we should just look at
the pictures and judge them. But they do have a lot of variables
that makes comparison difficult. And this is a technical site.

Just wondering...

--
Bitplayer

To err is human, to post-process -- divine.

 

[Bitplayer]

I don't know if this comment was directed at my call/question for dynamic range testing in cameras. I wasn't suggesting this particular test.

I haven't seen anything to suggest that the 20D or similar 12 bit cameras can record more than 12 stops of range. Of course, if the sensor response is non-linear, that is possible. The Fujis, with their two types of sensor sites, might work this way.(?) The link I gave earlier doesn't mention this condition. It seemed to be a pretty authoritative source. Can you give me a source so I can learn about this?

You are definitely right about the differing algorithms. They often trade dynamic range for contrast, as do we when we use custom curves. This doesn't preclude a test. In fact, one thing a test might encourage is a more "honest" default conversion -- one that doesn't punch up the contrast at the expense of dynamic range.

http://www.dpreview.com/learn/?/key=histogram

--
EOSMan
http://www.pbase.com/eosman

Thanks for the links. The first article is a good introduction.
And it got me thinking. (Always a mistake...)

Why isn't the dynamic range of digital cameras tested and
published? There is somewhat of a tradeoff between noise and
dynamic range -- and the S/N ratio is tested. Wouldn't it be more
meaningful to include a dynamic range figure, and to combine them
in some Range/Noise figure of merit? For example, I believe the
20D has smaller sensor sites than the 10D/300D. It also has lower
noise at higher ISO and more pixels. But what about the dynamic
range? Was there a trade-off -- or did Canon manage to also
improve the dynamic range while increasing resolution. And what
about those prosumer cameras? Sure, we know that DSLRs have lower
noise. But what about the dynamic range? The standard reply to
the "should I get a prosumer or a DSLR" question begins with a
statement that the DSLR has better noise performance. How come no
one says "because it has a greater dynamic range?" (Do they?) And
what about those Fujis and Sigmas with their special sensor designs?

OK, maybe I'm being a real gear-head. Maybe we should just look at
the pictures and judge them. But they do have a lot of variables
that makes comparison difficult. And this is a technical site.

Just wondering...

--
Bitplayer

To err is human, to post-process -- divine.

--
Bitplayer

To err is human, to post-process -- divine.

 

[Luther Foreman]

It is truly the software that is being tested here...not the camera.

Thank you for that statement. I have tried to get involved in these discussions in the past and have always ended up arguing distinctions between sensor dynamic range, the dynamic range of the A/D, or the range that can be recorded by the camera system as a whole, including any conversion software. For me at least, it is the last case that really matters. I would be interested in knowing what can be achieved via jpeg vs. RAW, and a comparison among the various RAW converters.

galleries at http://www.pbase.com/louforeman

Luther

Few things are harder to put up with than the annoyance of a good example ---- Mark Twain

 

[Dale Hutchinson]

I did not find your pbase account address to be able to download
your CAN YOU READ THIS test file.

If you will just reply to this so that I can download and give it a shot.

I've read a lot of the replies to your original thread and am ready to 'see for myself'.

Thanks,

Hutch

I have come up with a method of testing a camera's dyamic range.
It is not very sophisticated but it seems to work. The process is
as follows:

Darken your room. Display the 'test' file on your computer monitor
and find an exposure setting where the shutter speed is below 1/25'
and the Exposure is + - 0. Systematically take additional pictures
at 1 stop intervals until you can't read the text. Attached find
an example.

It seems rather crude but accurate. Any thoughts?





PS. Feel free to download the test file from my account at PBASE.
(not for commerical use without prior permission please) The test
file is 1600x1200 which should accomodate most monitors out there.

--



'If you always do what you've always done - you'lll always get what you always got'.

 

[Dale Hutchinson]

What is Porchini's Pbase address. He initiated this thread and said to feel free to download his test file but failed to give it's url.

I read all these replies/responses/rebuttales/etc and need to give his test a shot even though a monitor is not the best to use in 'any' equation. At least I'll see if I get the same results.

Thanks,

Hutch

 

[Dale Hutchinson]

I have come up with a method of testing a camera's dyamic range.
It is not very sophisticated but it seems to work. The process is
as follows:

Darken your room. Display the 'test' file on your computer monitor
and find an exposure setting where the shutter speed is below 1/25'
and the Exposure is + - 0. Systematically take additional pictures
at 1 stop intervals until you can't read the text. Attached find
an example.

It seems rather crude but accurate. Any thoughts?

PS. Feel free to download the test file from my account at PBASE.
(not for commerical use without prior permission please) The test
file is 1600x1200 which should accomodate most monitors out there.

--



'If you always do what you've always done - you'lll always get what you always got'.

 

[ronis]

You don't need to use monitor. Also, you don't need to download anything. All you need is a gray card with light and dark text (or whatever). Disappearing of one of those texts will give you end points to calculate DR.

What is Porchini's Pbase address. He initiated this thread and
said to feel free to download his test file but failed to give it's
url.

I read all these replies/responses/rebuttales/etc and need to give
his test a shot even though a monitor is not the best to use in
'any' equation. At least I'll see if I get the same results.

Thanks,

Hutch

 

[ronis]

no offense intended, but would be much better if you knew what you are talking about.

http://www.dpreview.com/learn/?/key=histogram

--
EOSMan
http://www.pbase.com/eosman

Thanks for the links. The first article is a good introduction.
And it got me thinking. (Always a mistake...)

Why isn't the dynamic range of digital cameras tested and
published? There is somewhat of a tradeoff between noise and
dynamic range -- and the S/N ratio is tested. Wouldn't it be more
meaningful to include a dynamic range figure, and to combine them
in some Range/Noise figure of merit? For example, I believe the
20D has smaller sensor sites than the 10D/300D. It also has lower
noise at higher ISO and more pixels. But what about the dynamic
range? Was there a trade-off -- or did Canon manage to also
improve the dynamic range while increasing resolution. And what
about those prosumer cameras? Sure, we know that DSLRs have lower
noise. But what about the dynamic range? The standard reply to
the "should I get a prosumer or a DSLR" question begins with a
statement that the DSLR has better noise performance. How come no
one says "because it has a greater dynamic range?" (Do they?) And
what about those Fujis and Sigmas with their special sensor designs?

OK, maybe I'm being a real gear-head. Maybe we should just look at
the pictures and judge them. But they do have a lot of variables
that makes comparison difficult. And this is a technical site.

Just wondering...

--
Bitplayer

To err is human, to post-process -- divine.

 

[Porchini]

Here is the URL from the PBASE account.

Target:

http://www.pbase.com/aelliot/image/37881373/original

Meetering Card:

http://www.pbase.com/aelliot/image/37841739

Once the image is fully visible in your browser, right mouse click on it and save the image to your desktop (or any other place)

Dim the lights and start shooting! (Once you are +2 or -3 EV from 0, take the pictures in 1/3 stop increments to make sure you can catch the transition to black or white)

Enjoy!

-Alan