Brainteaser: noise and ISO

Joined
Jan 15, 2008
Messages
4,553
Just for fun. :smile:

Three samples, all shot at the exactly same exposure. Same camera, same lens, same lighting, hand held. The only adjustment made in View NX is the equalisation of the final output brightness. No sharpening, no noise reduction, no curves, no picture controls, no ADL, no cheating, no bluffing.

One image shot at ISO 200, one at ISO 800 and one at 3200. (EXIF removed)

Brainteaser questions:
a. Which is which?
b. Why?


Number 1
Subscribe to see EXIF info for this image (if available)



Number 2
Subscribe to see EXIF info for this image (if available)


Number 3
Subscribe to see EXIF info for this image (if available)
 
Joined
Mar 4, 2006
Messages
4,830
Location
Newcastle, Wa
Appears to me like they are

1. 200
2. 800
3. 3200

Noise in the dark areas.

Is that a tape drive? Man, you need a computer upgrade :biggrin:
 
Joined
Jan 15, 2010
Messages
3,876
Location
UK
Only one exposure can be correct from either 200 ISO 800 ISO or 3200 ISO. Expecting the not so obvious so I am going to guess that:

Number 1 @3200
Number 2 @ 800
Number 3 @ 200

The increasing noise has been created during post processing?

I wonder if the Polaroid dust and scratch remover will shift the dust of the top of that tape drive :tongue:
 
Joined
Jul 9, 2008
Messages
1,418
Location
Tulsa, OK
I'd have to agree...

1 200
2 800
3 3200

The noise is visible on the left side of the picture moreso than on the actual CPU...
 
Joined
Jan 8, 2009
Messages
5,301
Location
San Jose, CA
Only one exposure can be correct from either 200 ISO 800 ISO or 3200 ISO. Expecting the not so obvious so I am going to guess that:

Number 1 @3200
Number 2 @ 800
Number 3 @ 200

The increasing noise has been created during post processing?

I wonder if the Polaroid dust and scratch remover will shift the dust of the top of that tape drive :tongue:
I'm with Tony. I think:

1. 3200
2. 800 (pushed)
3. 200 (pushed)
 
Joined
Jan 22, 2006
Messages
808
Modern cameras are reasonably clean at ISO3200, so I dont believe 3 is ISO3200 image. Image 3 is falling apart, it just looks like pushed image (and probably not from a D7k or some other camera with those sexy Sony sensors).

I'll go with the same:
1) 3200
2) 800 pushed
3) 200 pushed
 
Joined
Sep 15, 2008
Messages
1,413
Location
Sacramento
ViewNX will only allow 2 stops of push or pull from "as shot". Given the range of 4 stops from 200 ~ 3200, that would require that the 3200 was pulled down 2 and the 200 was pushed 2. Top is 3200 pulled 2 stops, bottom is 200 pushed 2 stops. Middle is 800 as shot.
 
Joined
Jan 15, 2008
Messages
4,553
Short version:

a) The series is 3200, 800, 200.

b) Increasing ISO lowers noise.

It is a phenomenon most shooters have realized intuitively but probably not necessarily put into numbers (evidenced by the common ETTR "rule"). As I wrote in the first post, this is just for fun, and this probably won't affect the shooting habits of anyone.


Long version:

Noise in digital signal processing is defined as random fluctuations of the signal. When capturing a digital image there are several sources of noise present. One is the photon shot noise, i.e. the randomness of the photons falling onto sensor. The lower the light, the more noise.

The second source of noise is read noise, caused by the electronics reading the photon count received by the sensor. The higher the ISO, the lower the read noise.

The other sources are ISO and light level independent, like the rounding errors at the AD converter.

In each shot the exposure was the same, i.e. the sensor received the same amount of light. All images were shot at 1/30 and f/1.4. This means that the different noise present in the images was not caused by different levels of light and thus not "seen" or "generated" by sensor, but appeared at a later stage of the image pipeline.

In the three samples the photon shot noise is constant, because of identical amount of light. The AD converter noise is also constant because the same camera was used.

What changes is ISO, and hence the read noise, which is entirely generated by the camera, but not the sensor. The lowest read noise of D3s is achieved at ISO 3200 and it stays almost constant above that. Not without consequences though, since each stop of higher ISO decreases the dynamic range by one stop.

Some cameras, like the D3, have a peculiar read noise characteristics. The read noise is at minimum at ISO 800 and it is higher everywhere else. Iliah Borg for example never goes beyond ISO 800 on his camera, because "1 stop underexposed" ISO 800 produces cleaner images than ISO 1600 shots. Similarly he uses ISO 800 -2EV instead of ISO 3200.

D3s does not share this feature, read noise does not increase beyond ISO 3200 either.

Before anyone falls asleep, the bottom line is that higher ISO decreases noise, to the contrary to the common wisdom. Typically we see more noise at higher ISO shots, but it is the wrong causality. Low light is what causes noise, and in low light we use higher ISO.

Our shots at high ISO would be even more noisy without this behavior. But as I wrote above, most shooters could not care less about the interplay of several noise sources and I am not pushing any ideas to anyone.

Summary: At the same exposure higher ISO decreases noise, but at the risk of blowing highlights and at the cost of reduced DR.

Here are the read noise (red dots) and dynamic range (blue squares) curves of D3s, by professor Robert M. Newman.

http://sensorgen.info/NikonD3s.html
 
Joined
Jul 30, 2006
Messages
11,564
Location
Southern California
Peter, help me to understand here. You said in your original post that each picture was taken at the "exactly same exposure." BUT, if all other aspects of the "light triangle" being equal, ie, your SS at 1/30th, and your aperture at 1.4, simply changing the ISO values from 200>800>3200 makes these pictures different exposures, no? Now, this does not probably affect your explantion of ISO and noise and it's causes (I'm not that technical :tongue:), but it does seem a bit misleading to label each picture as taken at the same exposure levels, when, according to how we understand photography, they were actually taken at 2-4 stops difference. Now the final exposures may have ended up the same being "pushed" in post, but then wouldn't that factor into the cause of the noise?
 
Joined
Mar 4, 2006
Messages
4,830
Location
Newcastle, Wa
I include ISO in exposure so in my definition, if you change ISO, either shutter speed or f stop must change also in exact same conditions. By keepimg shutter and f stop constant, exposure would be different for all 3 shots. It seems we have different definitions of exposure, your's only including shutter and f stop.
 
Joined
Jan 15, 2008
Messages
4,553
I'm :confused:

So the correct exposure is 1/30 and f/1.4 @ ISO 3200? Making the ISO 800 and ISO 200 image severely underexposed?
Exactly the same amount of light landed on the sensor, hence it is hard to call any of the shots underexposed per se. The sensor sensitivity never changes, digital is not film. "Underamplified", not "underexposed". All shots were exposed identically.

This is an exaggerated example, but one can deduct what happens in the shadows of a "normal" shot, when the shadows are 4 stops darker than midtones.
 
Last edited by a moderator:
Joined
Jan 15, 2008
Messages
4,553
I include ISO in exposure so in my definition, if you change ISO, either shutter speed or f stop must change also in exact same conditions. By keepimg shutter and f stop constant, exposure would be different for all 3 shots. It seems we have different definitions of exposure, your's only including shutter and f stop.
You are correct, I use the "narrowest" definition, which consists only of the scene lighting, aperture and shutter speed.

http://oxforddictionaries.com/definition/exposure

[count noun] the quantity of light reaching a photographic film, as determined by shutter speed and lens aperture.
 
Joined
Jan 15, 2008
Messages
4,553
Peter, help me to understand here. You said in your original post that each picture was taken at the "exactly same exposure." BUT, if all other aspects of the "light triangle" being equal, ie, your SS at 1/30th, and your aperture at 1.4, simply changing the ISO values from 200>800>3200 makes these pictures different exposures, no?
I define exposure as by Oxford dictionary, see previous post.

Digital is not film. Same exposure means the exact same amount of light landed on the sensor in each case. One would naively expect the resulting images to be identical noisewise, since ISO is only an after-the-capture-amplification. Unfortunately current cameras don't work that way (yet). Higher ISO reduces read noise.

Now, this does not probably affect your explantion of ISO and noise and it's causes (I'm not that technical :tongue:), but it does seem a bit misleading to label each picture as taken at the same exposure levels, when, according to how we understand photography, they were actually taken at 2-4 stops difference. Now the final exposures may have ended up the same being "pushed" in post, but then wouldn't that factor into the cause of the noise?
You have to remember the sequence of a digital image capture. First the sensor captures photons. Changing ISO does not affect this, but faster shutter or smaller aperture would decrease the number of photons.

After the capture the number of photoelectrons present in each pixel well is counted, this is called the "reading". The read transistor causes noise when it reads the pixel. Higher ISO decreases this noise by using higher analog gain (amplification) before the charge from the well reaches the read transistor.

Don't think the sample images as such. The images are rubbish, although they show the read noise. Think what happens in shadows which are 4 stops darker than the midtones in an image. Higher ISO would render cleaner shadows (with same exposure), but you might blow highlights and you will lose dynamic range.

I am not advocating any changes to anyone's habits. I think it is just cool to know how things really work. The common wisdom "high ISO means high noise" is not exact, since LOW LIGHT means high noise, but higher ISO means less noise (with current cameras).
 
Last edited by a moderator:
Joined
Jun 12, 2010
Messages
318
Location
Maryland, USA
What you have shown is that increasing the exposure in camera increases signal to noise ratio and gives better picture, while increasing exposure in PP amplifies noise along with the signal. It does not mean "higher ISO is less noise". If you took two images at the same ISO (say 3200), underexposed one and brightened it in PP, overexposed the other, then you will find the former has more noise than the latter.
 
Joined
Jan 15, 2008
Messages
4,553
What you have shown is that increasing the exposure in camera increases signal to noise ratio
The exposure stays the same.

... increasing exposure in PP amplifies noise
Cannot be done. Brightness can be increased in PP, but it is too late for exposure. You cannot modify the number of the photons that hit the sensor after the image has been captured. In every shot the same amount of photons landed in each pixel, because they were exposed identically. Sensor is not responsible for the noise, nor is the brightness increase in PP.

The noise "appears" at the reading phase, and is dependent on the ISO amplification.
 
Links on this page may be to our affiliates. Sales through affiliate links may benefit this site.
Nikon Cafe is a fan site and not associated with Nikon Corporation.
Forum post reactions by Twemoji: https://github.com/twitter/twemoji
Copyright © 2005-2019 Amin Forums, LLC
Top Bottom