Noise, DR, and Bit Depth

[chauncey]

Help needed from the brighter ones out there... :confused:

I can't understand this http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html
What does it mean to me and my Ds Mk III?

[Grimes]

Wow, it looks like someone's dissertation on noise reduction and control in digital cameras.

To the final end user, it pretty much means nothing we don't appreciate already - try to expose properly and use the lowest ISO that gives you the results you are looking for. A lot of the things he mentions is intended for people designing imaging circuits and sensors, not really stuff a photographer needs to know intimately.

[rdenney]

The bottom line seems to be that noise is a function of photons, and photons are a function of format. With no change in format, the overall signal/noise ratio doesn't seem terribly dependent on the number and density of pixels. The take-away is that to increase the signal/noise ratio significantly, one will need a larger format to gather more photons in the first place.

It is tempting to compare a larger sensor with a smaller sensor using the same lens, which is, of course, not fair. I would need to compare a 50mm lens on my 5D with a 32mm lens on a 50D. To make the same exposure, both would be at the same f-stop and shutter speed. But the f-stop value is the ratio of the diameter of the aperture to the focal length, so the 50mm at f/4 (for example) would have an effective aperture of 12.5mm, while the 32mm lens at f/4 would have an 8mm effective aperture. Thus, to make the same image at the same exposure, the 5D will be admitting 2.44 times as many photons (the ratio of the area of a 12.5mm circle to an 8mm circle).

The benefit of the 5D over the 10D is the larger sensor--they both have pixels about the same size, and both perform similarly if you just compare a square millimeter on one with a square millimeter on another.

But comparing a 10D with a 50D is more interesting--both start with the same amount of photons and thus the same amount of information. The 10D divides that data into larger pieces, but then each piece has more data in it. The 50D divides it into smaller pieces, but each piece has less data in it. Thus, each pixel on the 50D might be more affected by noise, but when the images from both are viewed at the same size, the noise and therefore the signal/noise ratio will be about the same.

This conclusion gives credence to Wilt's abhorrence of comparing image quality using 100% crops, at least without being quite careful about what 100% means for the camera in question. A 100% crop from my 5D (or 10D) will always show better results than a 100% crop from a 50D, even with the same lens.

Rick "format is king" Denney

[chauncey]

"This conclusion gives credence to Wilt's abhorrence of comparing image quality using 100% crops, at least without being quite careful about what 100% means for the camera in question.
A 100% crop from my 5D (or 10D) will always show better results than a 100% crop from a 50D, even with the same lens."

Are you/Wilt suggesting that a better way would be to use a marquee tool and carve out a specific number of pixels for that crop example?

Aside from that, with regard to my Ds3, is it saying that if you shoot in RAW, using ISO 3200 is a bad idea?
And if 3200 is necessary, one would be better served by boosting exposure in PS?

[tzalman]

My practical experience (not with a 1Ds3, sigh) is that it is worthwhile doing the push in PP rather than letting the camera do it, simply because you can customize it more. You may not need exactly a one stop push, 0.77 or 1.22 might be better. or by using a tone curve you can apply a non-linear push, for instance to prevent the loss of highlight DR.. But this has nothing to do with scholarly research, more like gut feeling.

The one instance I can think of in favor of 3200 is if you need a brighter review image.

[jra]

Also, I find it interesting that the author is saying that the 14 bit is pretty much useless on todays cameras. I've heard his argument put forth before but many users seem to disagree. Personally, I think the 14 bit is still little more than a marketing gimmick after doing a bit of reading of more informed individuals.

Curiously, all the 14-bit cameras on the market (as of this writing) do not merit 14-bit recording. The noise is more than four levels in 14-bit units on all of these cameras (Nikon D3/D300, Canon 1D3/1Ds3 and 40D); the additional two bits are randomly fluctuating, since the levels are randomly fluctuating by +/- four levels or more. Twelve bits are perfectly adequate to record the image data without any loss of image quality, for any of these cameras (though the D3 comes quite close to warranting a 13th bit).

[tonylong]

Someone told me something in one of these "noise/high ISO" threads that I didn't know and had a hard time believing, and that's that as you up your ISO by a stop, the "read noise" (variable noise caused by different things) is acutally decreased. You are making the "collection" noise more visible when you raise the ISO by amplifying the signal, but the overall signal-to-noise ratio is improved over amplifying a lower ISO image.

In other words, say you shoot a scene at ISO 100, but in doing so underexpose the scene by 4 stops, then you take the scene at ISO 1600, correctly exposed. The ISO 100 would have more noise, due to whatever the ISO amplifying does to hold down the "Read Noise".

This baffled me, and it's in no way clear from any technical info I've heard of how this is accomplished, but assuming it's real, it is something to consider in this discussion. I think it may have been RDKirk, or some other of our technically knowledgable guys, who gave out that info.

But, it only applies reliably to whole stops. So, if you have to choose between underexposing at ISO 800 or properly or slightly overexposing (in RAW) at ISO 1600, use the 1600 for a cleaner image.

[chauncey]

You may be right Jason, but I get better images from my Ds3 than I got from a 5D or Ds2.
In all fairness though, I had those bodies when my skill level was somewhat...ah...in my formative period. :)

[chauncey]

Tony, I'm not sure that I understand your point. Are you saying this?

If you take the shot at ISO 100 and boost it to a needed 1600 in PS, you will have more noise than you would have by opting 1600 to begin with. If so I agree.

The link seems to say that shooting in 1600 and boosting to 3200 in PS is preferable to shooting at ISO 3200. Never tried this.

[tonylong]

Tony, I'm not sure that I understand your point. Are you saying this?

If you take the shot at ISO 100 and boost it to a needed 1600 in PS, you will have more noise than you would have by opting 1600 to begin with. If so I agree.

The link seems to say that shooting in 1600 and boosting to 3200 in PS is preferable to shooting at ISO 3200. Never tried this.

I believe that applies to cameras with no native ISO 3200 -- you gain nothing by shooting at a "fake" ISO 3200 excepting the visual aid in the LCD/Live View.