Help with ISO and S/Nr ratio

[tonyNZ]

Advertisement (gone after registration)

I'm doing an assignment looking at the ISO speed of a camera and comparing it to the signal to noise ratio.

 

Method as folllows:

I take an image of a grey card at ISO speeds (80, 160, 200, 400, 800, 1600), open in photoshop (with no post processing) and take a 64x64 sample from middle of the grey card. I record the mean, standard deviation for RGB, R, G, B channels from the histogram and create a table, then create a graph.

 

(I placed the grey card in a sunny spot outside, used a lightmeter to determine the right exposure (f4 @1/2000sec, ISO80) and then closed the shutter one stop for each increase in ISO. (ISO 160 and 200 use the same f/stop). I have done the test with Daylight, Auto and 5000 white balance)

 

What I find is that while the R, G, B values start with a high S/Nr (=100) and generally go down as the ISO increases, the RGB value sits at around 10 S/Nr regardless of the ISO setting. To me that is indicative of a high level of noise in the image, but where is it coming from? What should happen is that the RGB line follows the R,G,B lines - refer to graph.

(http://dl.dropbox.com/u/180981/ISO%20Leica.jpg)

 

I have done the same test with the NIkon D800 and the data turns out very well.

(http://dl.dropbox.com/u/180981/ISO%20Nikon.jpg)

 

Can anyone explain why there is this difference in the RGB S/Nr?

 

BTW the two red lines represent the first acceptable S/Nr speed (10) and the first excellent S/Nr speed (40). In other words, anything below 10 is considered unacceptable, between 10 and 39 acceptable, and above 40, excellent. (As an aside, even though the Nikon at the highest ISO produces an acceptable result, I would not call it acceptable looking at the noisiness of the image!)

 

Thanks

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[Lindolfi]

Interesting.

Any pixel has three values, which you have analysed on variability, that is clear, but what do you mean by "the RGB value"?

Edited October 19, 2012 by Lindolfi
Typo

[tonyNZ]

Im referring to the red, green, blue and combined RGB signal-to-noise ratios.

[Lindolfi]

How do you calculate " combined RGB".

That is important for understanding your constant curve compared to the declining curves of the separate R, G and B signal to noise ratios.

So it is of importance for helping you with your question.

[tonyNZ]

They are taken from the histogram. It will be the mean number divided by the standard deviation to get the signal-to-noise ratio. The histogram displays the combined RGB, red, green and blue mean numbers.

[Lindolfi]

Thanks Tony. I found the solution to your question: If you take the standard deviation of three RGB peaks and the peaks are not aligned, that standard deviation is larger than the standard deviation of each peak separately. When increasing the ISO, those peaks remain in their place and the standard deviation is still dominated by the position of those peaks. Therefore S/N ratio seems to be independent of ISO, as you have shown in your figure.

 

So if your white balance would be perfect, the S/N ratio would be the same as that of the individual peaks.

 

Suggestion: repeat your experiment with perfect white balance.

 

Below you can find the histograms of the RGB signal and red signal for non corrected white balance:

 

 

 

Below the histograms of the RGB signal and red signal with correct white balance:

 

 

 

As you can see, the standard deviation of the uncorrected white balance version is much bigger in the RGB signal than in the correct white balance version, because the positions of the peaks dominate (white balance) and not the width of the R, G or B histograms (true noise)

 

Hope this makes it clear.

 

PS another suggestion: if you take testshots for this purpose, defocus, so that possible unevenness in the gray card is not of importance (photograph at 1 meter, and focus to infinity), else you measure noise and unevenness.

Edited October 19, 2012 by Lindolfi
Added examples

[tonyNZ]

Advertisement (gone after registration)

Ha! (I feel so silly now!) You know I had considered that it was because of the white balance, and I am sure that I tried it but there was no difference. Anyhow after looking at your images I tried it and Hey Presto, my graphs look great. There is slight colour casting at the lower ISO settings otherwise the graph is to be expected. So, THANK YOU and thanks for taking the time to try this out for yourself, and providing a good explanation, it is appreciated.

 

(Incidentally, when compared to the Nikon D800 even at native ISO there is a large difference - Nikon SNr 160 compared to Leica Snr 80 - but does it make the Nikon a better camera? I would say the Nikon is more versatile especially at low light situations, it has a more accurate white balance control. Of course the CMOS sensor makes a difference and the fact that the Leica is older technology. However I reckon in 'normal' conditions both cameras would give a similar performance. Would this be your assessment, Bert?)

[Lindolfi]

Glad I could help, Tony.

Yes, noise is usually not so relevant. Depends a lot on proper exposure and kind of image. Sometimes I'm forced to go to 640 or 1250 ISO and usually not too much of a problem. I have no experience with that Nikon, but signal to noise ratio is just one of many variables that make up an image, so it is likely that there are more differences.