Will Michael revisit ETTR?

[joofa]

It is quite true that a SNR of 0 db is not useful for photography.  

Kindly define what exactly is 0 db SNR for a real image as in photography? if I give you an image of my cat how do you define a "noise floor" for that image?  

Sincerely,

Joofa

[hjulenissen]

You can see features below 0dB if they are large enough.  It is similar to SNR improving when you downsize an image.  Your visual system is actually surprisingly good at picking out large patterns masked by fine patterned noise.  In this case I meant 0 dB in the DxO context - pixel SNR of 0 dB in an 8MP image.  You won't see any fine pixel level detail at say -10 dB but if we have big feature, say one taking up nearly a quarter of the image, that will be visible.
...
Finally, just to add to the confusion, converting from 4-bits to 1-bit we'd add noise before converting to 1-bit to get the best detail in the final 1-bit image! (look up dithering for the details)

I believe that dithered "1-bit" prints will tend to have a SNR of ~0dB (or less), while still clearly having practical application.

-h

[ejmartin]

Makes no sense to me. One is one, and the avarage of one, is one. In a context of one level, any pixel is either black or white. However, in the deep shadows there can be no white, so one level is either black or near black.

Is this not correct?

I'm deducing that the black equals 0dB and the near-black is the one level.

The point is that noise acts to dither the levels.  Suppose the true signal is some value X between zero and one, over a patch of the image (we are going to ignore natural scene variation for the purpose of answering your question).  Suppose the noise is of strength N, for example let N be one level.  The noise adds a random number roughly between -N and N to X, so that the pixel wants to record some number between X-N and X+N.  Of course, the resulting signal plus noise is digitized so the output is either 0 or 1; if the noise is random (uncorrelated from pixel to pixel), the value of X is reflected in the percentage of 1's vs 0's in the patch -- a fraction X of the pixels will be 1 and the rest 0.  If we average the levels over a large enough patch, we recover the original signal, even though each individual pixel only recorded 0 or 1.

This is the basic idea that allows one to trade resolution for noise -- and why downsampled images look less noisy.  Note also that while the average is more finely graded than steps of one, that doesn't mean we buy anything by making individual pixels record values more finely spaced than the level of noise, because the individual values are jumping around randomly by an amount between -N and N.

[Guillermo Luijk]

I see! So you've deliberately created a set of artificial circumstances to demonstrate a point which may be valid only within that set of artificial circumstances. Is that correct?

No it isn't. I created a set of artificial circumstances to demonstrate 'that more levels not always mean anything useful'. Then added that this is what happens in a RAW file, making more levels thanks to ETTR irrelevant, and I told you to look at Emil's numbers for a real case applied to RAW. Why do you make me repeat you I already said? I find it silly and boring to other users.

Regards

[Guillermo Luijk]

Kindly define what exactly is 0 db SNR for a real image as in photography? if I give you an image of my cat how do you define a "noise floor" for that image?

SNR=0dB means the stdDev of Noise equals the level of signal (average value in a patch). To have an idea, you can create a (128,128,128) square patch in PS using some linear profile, and add the amount of gaussian noise needed to make its stdDev=128.

[kwalsh]

Here is a very low SNR image, and "below one level" to boot.  Taken with a GH-2, DxO gives a DR 11.3 EV and the RAW files are 12-bit.





So that's 14 EV difference in exposure from a camera DxO says has 11.3 EV of DR in a RAW file with only 12-bits.  The 4 second exposure RAW file has the highlights just under the clipping level.  In the 1/4000th exposure the difference between the highlights and shadows is 0.25 bits at best.  Based on the DxO data the SNR (by DxO's definition) for the 1/4000th exposure is about -15 dB.  Nonetheless the largest features are visible, and even the outer portions of the resolution star are still distinguishable.

Ken

[joofa]

SNR=0dB means the stdDev of Noise equals the level of signal (average value in a patch). To have an idea, you can create a (128,128,128) square patch in PS using some linear profile, and add the amount of gaussian noise needed to make its stdDev=128.

Unfortunately, real photography is not about patches. Is it?  Since this is a photography website I would expect you to apply your theory for an image that I have captured. Can you calculate the SNR of this image:



Sincerely,

Joofa

[hjulenissen]

Can you calculate the SNR of this image:

If you can provide a noise-free reference, then yes.

-h

[joofa]

If you can provide a noise-free reference, then yes.

-h

Do you take a noise-free reference image with your camera before acquiring a "real" image when you are out for photography? Remember, we must deal with real problems, not manufactured problems.

Sincerely,

Joofa

[Bart_van_der_Wolf]

Remember, we must deal with real problems, not manufactured problems.

Then why do you want to determine the noise level from an image that makes that determination more difficult?

Cheers,
Bart

[Ray]

Referring to the RAW files of my DR tests for the D7000, I see that the last shot in the series, beginning with a 4 secs exposure, was taken at 1/8000th exposure, which represents the 16th stop.

Surprisingly, there's still some broad detail visible through the noise. At least one can read the heading.

[hjulenissen]

Do you take a noise-free reference image with your camera before acquiring a "real" image when you are out for photography? Remember, we must deal with real problems, not manufactured problems.

Sincerely,

Joofa

A lot of the maths and physics allowing you to use your camera in the first place was developed without a camera to experiment on - a lot was developed for totally different applications. Does this mean that you rip the corresponding parts out of your camera?

I dont understand your angle. I have real problems (and real joy) when using my camera. Trying to analyze why it behave like it does (and not always how I would like it to) is a real benefit to me, and I believe that my images are better as a result. SNR is just a tool that measure something, dont measure everything, and let me more quickly understand what other people are talking about. I am fully aware that not all noise looks the same, even at a given average level below the signal. Just like I know that a 200HP car may not be the same as another 200HP car. But I still feel that horsepower is relevant to some applications, just like SNR is relevant to some applications. If you disagree, that is ok.

There is of course the possibility that I "invent" a problem that does not manifest itself in my photography. But I can suggest of equally undesirable flaws that stems from blindly observing "real-world" image results without connecting the observations to a sound physical framework and an understanding of how humans work. One is observer-expectancy: people tend to believe that any "tweak" improves the IQ even when it does nothing at all. Another is that in-camera noise-reduction might cause many images to look better, one might think that this camera is less "noisy" than others. But for some images the flaws of the NR might be very visible. And compared to a seemingly equally noisy camera based on superior sensor tech, one should expect the latter to provide better starting-point for raw images in the long term.

-h

[joofa]

A lot of the maths and physics allowing you to use your camera in the first place was developed without a camera to experiment on - a lot was developed for totally different applications. Does this mean that you rip the corresponding parts out of your camera?

I dont understand your angle. I have real problems (and real joy) when using my camera. Trying to analyze why it behave like it does (and not always how I would like it to) is a real benefit to me, and I believe that my images are better as a result. SNR is just a tool that measure something, dont measure everything, and let me more quickly understand what other people are talking about.

-h

I find it rather interesting that there is tons of theoretical stuff like SNR, DR, noise floor, etc in this thread, but no method to calculate the SNR for a real image that I acquired? Seems like our theory has some disconnect when applied to photography. Right?

Sincerely,

Joofa

[hjulenissen]

I find it rather interesting that there is tons of theoretical stuff like SNR, DR, noise floor, etc in this thread, but no method to calculate the SNR for a real image that I acquired? Seems like our theory has some disconnect when applied to photography. Right?

Sincerely,

Joofa

No, I just think that you are inventing a problem. I think that you are trying to force this discussion in some unknown direction by claiming irrelevant requirements for any topic in digital photography. We know that SNR affects all images, some more visibly than others. The fact that it cannot easily be estimated from a single image is totally irrelevant. This is an internet discussion forum. It is practically impossible to force other people to stop discussing a topic, and if you try you will hopefully be banned.

-h

[Alan Goldhammer]

I find it rather interesting that there is tons of theoretical stuff like SNR, DR, noise floor, etc in this thread, but no method to calculate the SNR for a real image that I acquired? Seems like our theory has some disconnect when applied to photography. Right?

Sincerely,

Joofa

You need a reference point and a single image won't supply that.

[degrub]

i once heard the following observation -

"All models are wrong, but some are useful."

SNR may not be possible for a single image, but it has been useful for improving capture devices and thus the images we get from them.

Frank

[joofa]

No, I just think that you are inventing a problem. I think that you are trying to force this discussion in some unknown direction by claiming irrelevant requirements for any topic in digital photography. We know that SNR affects all images, some more visibly than others. The fact that it cannot easily be estimated from a single image is totally irrelevant. This is an internet discussion forum. It is practically impossible to force other people to stop discussing a topic, and if you try you will hopefully be banned.
-h

So you are the moral police here now. My original query was not even addressed to you. You jumped in and are now hoping to get me banned. My advise to you is that if problem is too hard for you, sit back and enjoy and ride.

Sincerely,

Joofa

[Guillermo Luijk]

Unfortunately, real photography is not about patches. Is it?  Since this is a photography website I would expect you to apply your theory for an image that I have captured. Can you calculate the SNR of this image:

Sure I can, the RAW file is needed. Of course SNR will be different according to the different areas (higher on higer exposed areas).

It's a matter of locating each area of the scene:



in EV zones:



Each of those EV zones will have a different SNR according to the sensor SNR response (SNR curves can be empirically measured, like DxOMark does).

Regards

[joofa]

i once heard the following observation -

"All models are wrong, but some are useful."

SNR may not be possible for a single image, but it has been useful for improving capture devices and thus the images we get from them.

Frank

A basic extension of the usual model will reveal that one can have at least a model of SNR of a single image with some assumptions placed. But the intention was not to go into that model. Rather pointing out the fallacy of statements regarding SNR, and errors in SNR calculation when doing certain operations, say downsampling, as pointed out in this thread and various links provided here.

Sincerely,

Joofa

[joofa]

Sure I can, but I need the RAW file, and of course SNR will be different according to the different areas (higher on higer exposed areas).

Guillermo, when we quote noise statistics for a sensor, such as say, X electrons for read noise, etc., do we quote several noise figures, and do each of them apply to a different patch/area?

Joofa