ISO 1600 vs correcting exposure in photoshop

[Daniel Browning]

Where do these statements come from?

3 stops was chosen because that's a typical amount of headroom for the autoexposure system to meter for.

Are they arbitrary figures, just so you can illustrate the differences? Or is "f/2.8 ISO 400: 3 stops..." (and the others) simply true, all by itself?

It's arbitrary in the sense that I chose the number that Canon uses for its AE system. But raw photographers are free to choose any amount of headroom they need for the shot, from 1 stop to 7 stops. The AE system will tend to choose between 3 and 4.5 stops, depending on the camera, settings, and other factors.


What's not arbitrary is the *relative* amount of headroom. When exposure is reduced by a stop, without changing ISO, headroom increases exactly 1 stop, every time. In a fixed exposure, ISO 800 always has 1 stop less headroom than ISO 400. That's why it went down from 4 to 3 in my example. But if I had chosen a different headroom to start with, say 5 stops, then it would have gone from 5 to 4.

Aren't headroom values dependent upon the dynamic range, which obviously vary from scene to scene? Or are these statements "constants" in some way?

You've got it. Headroom is chosen by the photographer based on the scene. A black lab against a black wall holding a black ball will need far less headroom than a white dress in sunlight against a black wall.

Sorry if I'm complicating this. I always think of Headroom as a safety margin of sorts, a tonal range that is available in the development process, existing only because the photographer didn't clip the HLs. How much headroom is a function of the exposure (and the scene's avail range of course) and then the process in PP (the creative intent).

I don't think that's the best view of headroom. I'll try to explain. The camera just captures linear values from clipped to noise. It knows that 1 stop less light means it is "darker". But it has no concept of "black". Black is just the point where we, as humans, say "this is too noisy, I'm going to call it black."


Raw files have no concept of "middle gray", either. Middle gray is just the point where humans say "make *this* certain brightness into the midtone". That could be 7 stops below clipping, which would result in a *lot* of highlight headroom. Or it could be 1 stop below clipping, which would result in very little highlight headroom. Cameras tend to meter for 2.5-4.5 stops of highlight headroom, and most raw converters are tuned for this as well. It can be quite difficult to get raw converters to use a different value. Film had 6 stops of highlight headroom.

I mean, I see the trend - higher ISOs squeezing the HL headroom - but how or why does it happen?

I don't know exactly what goes on inside the camera. I just know that it clips it.

[canoli]

This conversation is really revealing my limits of understanding - of the entire digital capture universe, and it's also extremely helpful.


Thank you!


So headroom really isn't some fluctuating range of tones, dependent upon the scene. It's really best defined as what the recording medium is capable of. Is that the right way to think of it? And so Canon can choose to make f/2.8 have 3 stops at ISO 400, or it could've chosen something different. The way it is, with RAW converters tuned similarly, is no accident...


On a related note - does any of this discussion apply to shooting on a tripod? I always use ISO 100 on a tripod. But there's that "long-exp. NR" Function setting, so I'm wondering when that comes into play...


I've read the manual of course. I also bought the Canon EOS 40D Guide to Digital SLR Photography by David D. Busch - and not to knock either one of them too harshly, but there is a lot of misinformation out there. Mr. Busch's book says, "Another way of adjusting exposure is by changing the ISO sensitivity..." and so on.


He describes the long-exp noise reduction as dark frame subtraction - I guess like what HotPixels Eliminator is supposed to do.


http://www.mediachance.com/digicam/hotpixels.htm


I've never used the long-exp. NR on my 40D. Do you think it's worthwhile using it? According to David Busch's book it's good for shutter speeds slower than 1 second, although you'll trade off some detail if you use it.


I don't know, I took a 30-sec shot the other night - was trying to open up shadows, and also see if the resulting image was full of noise and it really wasn't, it was okay. I was at ISO 100 and f/16. I still didn't light up the shadows the way I'd hoped to, but close enough. Eventually it'll be part of a composite. Here's a 100% crop from the upper RH side, out of ACR to convert to .jpg, nothing else changed.


Would I have done better with a higher ISO and a faster shutter speed?


(and at the risk of being redundant and appearing ridiculously obsequious, THANK YOU for all your help Daniel! [:$])


[img]/cfs-file.ashx/__key/CommunityServer.Components.UserFiles/00.00.00.29.05/_5F00_MG_5F00_8523_5F00_533x800_5F00_2.jpg[/img]

[Jon Ruyle]

I've never used the long-exp. NR on my 40D. Do you think it's worthwhile using it? According to David Busch's book it's good for shutter speeds slower than 1 second, although you'll trade off some detail if you use it.

That may be an okay rule-of-thumb, but thermal noise depends on a lot of factors other than shutter speed. It is much worse at higher temperatures (which is why the serious long exposure guys cool their cameras), and like all noise, is most noticeable when the signal is weak. I think your picture looks fine and does not need long exposure NR.


If you took a 5 minute or longer exposure of the night sky, thermal noise would be obvious. I find that in these cases, the camera's built-in long exposure NR works very well.

[canoli]

ah, gotcha, thans Jon. So then it's strictly thermal noise that the long-exp. NR function addresses? And it can be mitigated by not using Live View I'd guess, also, shooting in lower ambient temps...


(Just how many types of noise are there with digital sensors?)


So with a tripod, is there ever a reason to boost ISO beyond 100, short of long (5-min.) exposures? (or even then?)


Although come to think of it, even at ISO 1600, what was a 5-min. exposure at ISO 100 is still going to be pretty long...something like...let's see, ISO 100 to ISO 1600 is 4 stops, so 5 minutes plus 4-stops = 18.75 secs. (cross fingers once again) Is that correct?


So I guess this is when knowing how ISOs affect HL headroom comes into play. What would be the lesson here?

[Daniel Browning]

So headroom really isn't some fluctuating range of tones, dependent upon the scene. It's really best defined as what the recording medium is capable of. Is that the right way to think of it?

In the case of film, that's correct: the headroom is built right into the recording medium. For example, a certain emulsion might have 6 stops above middle gray (headroom) and 6 stops below middle gray ("footroom") and that's it.


But with digital, it's a little different. You can put middle gray anywhere you want. So with 10 stops of dynamic range, you can have 3.5 stops above middle gray (headroom), and 6.5 below. Or 8 stops above and only 2 stops below. It's up to the raw converter. In practice, raw converters tend to put out 3.5 stops above, and 4.5 stops below, and stop there (8 stops).

And so Canon can choose to make f/2.8 have 3 stops at ISO 400, or it could've chosen something different. The way it is, with RAW converters tuned similarly, is no accident...

Precisely.

On a related note - does any of this discussion apply to shooting on a tripod?

Yes. Usually when you're on a tripod it's possible to "ETTR" so that the histogram is as far to the right as you want it. So there's no need to move on to the next step (ITTR).

Mr. Busch's book says, "Another way of adjusting exposure is by changing the ISO sensitivity..." and so on.

That's disappointing.

I've never used the long-exp. NR on my 40D. Do you think it's worthwhile using it? According to David Busch's book it's good for shutter speeds slower than 1 second, although you'll trade off some detail if you use it.

Yes, I think it's great. It reduces hot pixels and pattern noise, but it increases random noise a little bit. I can't think of why it would trade off any detail. (Maybe he's thinking of Nikon's long exp NR, which does eradicate small details.)

Would I have done better with a higher ISO and a faster shutter speed?

No, you did it right. Always do "ETTR then ITTR", never the reverse. Doing ITTR first would result in a lot more noise. Exposure is more important than ISO.

[canoli]

So the 10 stops of DR you speak of - is that what digital sensors are considered capable of capturing? Or is that the range that the visible world has? (or both?)


And when you say "you can put middle gray anywhere you want" and then later say "It's up to the raw converter" I'm not sure how those 2 statements go together. The first one I figured meant when we meter a scene, we're choosing the gray point. But if it's "up to the raw converter"...?


Doesn't our metering system take care of dividing the dynamic range? I thought that's what they all did, establish the 18% gray point (actually closer to 13% from what I've read) and then the 1/4 tones, 3/4 tones etc. fall where they may.


re: the book I quoted, yes, I wish I'd read a more advanced book first. I guess he's trying to keep it simple and just describe the effects of changing ISO - basically darkening or lightening an image capture. He says many times that "higher ISO means more noise" and I've taken that as gospel, since I've seen it so many other places too.


You need to write a book Daniel! (if you haven't already)


In any case, thanks for continuing the discussion. I appreciate it very much.

[Jon Ruyle]

So then it's strictly thermal noise that the long-exp. NR function addresses?

Yes. Well, almost. I believe it just takes a dark exposure of the same length and subtracts. This is primarily to reduce thermal noise, but it also must mitigate read noise as well (though if one wants to reduce read noise in a low signal image, more standard practice is to take several very fast exposures, combine eg by averaging, then subtract)

So with a tripod, is there ever a reason to boost ISO beyond 100, short of long (5-min.) exposures? (or even then?)

If the subject isn't moving and the tripod is sufficiently steady, then I don't think so. Thermal noise is less than linear, so longer exposure means better snr.

Although come to think of it, even at ISO 1600, what was a 5-min. exposure at ISO 100 is still going to be pretty long...something like...let's see, ISO 100 to ISO 1600 is 4 stops, so 5 minutes plus 4-stops = 18.75 secs. (cross fingers once again) Is that correct?

That's what I get []

I can't think of why it would
trade off any detail. (Maybe he's thinking of Nikon's long exp NR,
which does eradicate small details.)

I assume the canon long exposure nr just takes a dark exposure of the same length and subtracts. Does Nikon do something different?

[Daniel Browning]

You're welcome.

So the 10 stops of DR you speak of - is that what digital sensors are considered capable of capturing?

Close to it. Engineers measure dynamic range from the clipping point down to where SNR reaches 1:1, which can be 11.5 stops on many cameras. But 1:1 is *very* noisy. Noisier than many photographers like to use. So how much dynamic range someone will use depends on their personal standard of how much noise is acceptable, especially pattern noise. You can find the "engineering" dynamic range for many cameras here:

http://dxomark.com/

And when you say "you can put middle gray anywhere you want" and then later say "It's up to the raw converter" I'm not sure how those 2 statements go together.

It's both. You choose where to put it by changing the parameters of the raw converter (if it lets you).

The first one I figured meant when we meter a scene, we're choosing the gray point. But if it's "up to the raw converter"...?

Basically, it's best to "expose for the developer". In other words, think ahead to how the shot will be converted, then expose for that. If you know that you can move middle gray down, so that you get 2 stops more highlights for a certain contrasty shot, then you'll "underexpose" (compared to the AE meter) by two stops, but in reality you'll get just the exposure you wanted (perfect exposure).

Doesn't our metering system take care of dividing the dynamic range? I thought that's what they all did, establish the 18% gray point (actually closer to 13% from what I've read) and then the 1/4 tones, 3/4 tones etc. fall where they may.

That's right.

Yes. Well, almost. I believe it just takes a dark exposure of the same length and subtracts.

I think so too.

This is primarily to reduce thermal noise, but it also must mitigate read noise as well

It does help the "fixed pattern" part of the read noise, which is very beneficial, and it also takes care of the hot/dead pixels, but the random noise actually gets a little worse, because the read noises add in quadrature. If Canon used the technique you mentioned, it wouldn't add noise:

(though if one wants to reduce read noise in a low signal image, more standard practice is to take several very fast exposures, combine eg by averaging, then subtract)

Unfortunately, none of the popular raw converters support dark frame or bias frame subtraction, so I tend to just recommend the long-exp NR.

I assume the canon long exposure nr just takes a dark exposure of the same length and subtracts. Does Nikon do something different?

They have two forms of long exp NR. One is the same as Canon (dark frame subtraction), and can be turned off. The other is an algorithm that looks for hot pixels, and cannot be disabled. This is of course bothersome for astrophotographers, since stars tend to look a lot like hot pixels. There is a cumbersome workaround, though: if you *enable* long exp NR, then turn off the camera in the middle of its dark frame, it will cancel the dark frame subtraction, cancel the hot pixel killing software, but still save the light frame. Turning off the camera after every frame is too much work for me, though.

[Jon Ruyle]

Thanks for the clarification, Daniel.

Unfortunately, none of the popular raw converters support dark frame or bias frame subtraction, so I tend to just recommend the long-exp NR.

Iris does it I think, and it supports 5DII raws.

One is the same as Canon (dark frame subtraction), and can be turned off. The other is an algorithm that looks for hot pixels, and cannot be disabled. This is of course bothersome for astrophotographers, since stars tend to look a lot like hot pixels. There is a cumbersome workaround, though: if you *enable* long exp NR, then turn off the camera in the middle of its dark frame, it will cancel the dark frame subtraction, cancel the hot pixel killing software, but still save the light frame.

That absolutely sucks, and is a deal breaker for me since I like to take multiple exposures with a timer. Glad I decided to pull my 5DII out of the trash []

[Daniel Browning]

Iris does it I think, and it supports 5DII raws.

Yeah, but it's not the kind of popular I was thinking of. It's also kind of basic on other features. Not a very advanced demosaic, etc. What I would really like is if programs could open raw files, make adjustments to raw values, and re-save a Bayer raw DNG file. I would love to be be able to:

• Bias frame and dark frame subtraction in an IRIS batch script
• Lens corrections in DxO Optics Pro (vignetting, distortion, CA, deconvolution, etc.)
• Demosaic and everything else in RPP

I can dream.

Glad I decided to pull my 5DII out of the trash [img]/emoticons/emotion-1.gif[/img]