8
\$\begingroup\$

I have a question about image quality. Does noise in an image is depends upon "Megapixels" or on "ISO"?

\$\endgroup\$

7 Answers 7

15
\$\begingroup\$

Noise originates due to a number of factors, see:

What is noise in a digital photograph?

Increasing the number of megapixels keeping everything else constant (sensor size, technology etc.) will increase noise per pixel, but also has the effect of making the noise finer grained which is less objectionable.

ISO does not by itself increase noise, only if you combine increasing ISO with decreasing shutter time / closing aperture.

It's [probably] worth repeating this again here...

Increasing ISO whilst keeping shutter-speed/aperture constant does not increase noise:

Here is an example, as the ISO 100 shot was underexposed, raising the ISO to 1600 yielded a much less noisy result!

\$\endgroup\$
8
  • \$\begingroup\$ I'm not sure why this ISO topic always yields such a heated debate. \$\endgroup\$
    – rfusca
    May 16, 2011 at 15:57
  • \$\begingroup\$ @fusca: because people confuse the noise level with the signal to noise ratio, and because you don't need a degree in optical engineering to be a good photographer :) \$\endgroup\$
    – Colin K
    May 16, 2011 at 16:07
  • 9
    \$\begingroup\$ Because it's counter-intuitive, and because noisy images tend to have been shot with high ISO, by association they attribute this to high ISO. So when you come along and say "increasing ISO actually improves the signal to noise ratio (all else being equal)" people refuse to accept it, even when you provide the results of a simple experiment that proves this. \$\endgroup\$
    – Matt Grum
    May 16, 2011 at 16:33
  • \$\begingroup\$ @MattGrum I'm missing something here, how did you take the first photo with the same aperture and shutter speed as in the second one but with higher ISO and the first photo is not brighter than the second one? \$\endgroup\$
    – K''
    May 16, 2012 at 19:10
  • 1
    \$\begingroup\$ @Akram I applied auto levels to both images to normalize brightness, as I wanted to compare signal to noise ratio (which is much more important than the absolute noise level). \$\endgroup\$
    – Matt Grum
    May 16, 2012 at 21:37
3
\$\begingroup\$

When it comes to ISO, it's a misnomer that increasing the ISO increases noise. It's doing nothing of the sort. The reason high ISO produces noisy images is entirely due to the fact that for a higher ISO, you have to cut the amount of light (signal) hitting the sensor, ultimately increasing noise ratio before ISO amplification even takes place.

\$\endgroup\$
1
\$\begingroup\$

The short answer is both, but it requires some explanation.

Megapixels. Yes, but only if we're talking about 12MP vs 24MP for a given sensor size. The issue is not the absolute number of megapixels per se, but the size of each individual pixel. Buckets in the rain is the common analogy used. Essentially, if you have two buckets out in the rain, the bigger one with twice the area as the other, the bigger one will collect twice as much water (or photons).

This is all about signal-to-noise ratio.

Photon hits photodiode --> photodiode emits electron --> electrons are stored in a cell --> analogue charge converted to a digital signal by the analog-to-digital unit (ADU). Roughly, anyway.

enter image description here

The level of amplification that takes places at the ADU stage is determined by the ISO level. The issue is that along the way noise occurs. By noise we mean unwanted signal -- the level of charge stored in the well does not 100% reflect the light reading in a perfect world. There are various reasons for this (some of which are natural and cannot be avoided -- details [here][2]). At higher ISO levels the signal needs to be amplified even more increasing the impact of this unwanted noise as a percentage of the total signal.

The following numbers are arbitrary, but let's just say at ISO 100 you had 4 noise units out of a signal of 12.5 units collected and that at ISO 400 you had 4 units out of 10 units collected. Note...

  • the noise levels before amplification are independent of the ISO selected).
  • at ISO 400 1/4 as much light is collected (ISO 400 is two stops above ISO 100). Double the ISO, half the amount of light collected.

The amplification process needs to take the signal to 100.

  • ISO 100 (x2): 8/100 = 8% noise.
  • ISO 400 (x8): 32/100 = 32% noise.

So an increase in ISO leads to more noise. If our individual pixels had been twice as large so that no amplification was needed (again, this is just to explain the concept) then the following would have been true.

  • ISO 100: 4 units of noise for 100 units of total signal (4%)
  • ISO 400: 4 units of noise for 25 units of total signal (16%)

... and noise would have been reduced.

\$\endgroup\$
0
\$\begingroup\$

Both of those factors can have an impact on noise. Of the two, the effect of ISO is the most immediately visible. All digital cameras exhibit image noise that rises as ISO increases. Depending on the camera, you may see little or no noise at ISO 100, for instance, and proportionally more as you approach the maximum ISO setting for your camera (typically 1600 to 6400 or more). There are plenty of examples of this for just about every camera on the market (reviews for each specific camera will often show examples).

Image size (megapixels) doesn't directly impact noise, but it indirectly impacts noise when you consider two sensors of the same size. In this case, all things being equal, the larger-megapixel sensor will show more noise as a side-effect of trying to "squeeze" more pixels onto a sensor of similar size. Although this can be shown in certain cases within a camera line, this factor can be difficult to pin down because cameras also often have upgraded sensor or processing technology that manages noise more effectively.

There are some great answers on this site already regarding specific aspects of noise. If you search for "noise", you'll find all sorts of great reading.

\$\endgroup\$
10
  • 1
    \$\begingroup\$ All digital cameras exhibit image noise that rises as ISO increases not true. In fact the only way noise increases is if you decrease the amount of light coming in. Raise the ISO whilst keeping the same shutter speed and aperture will cause noise to decrease (or the highlights to clip). \$\endgroup\$
    – Matt Grum
    May 16, 2011 at 14:52
  • 1
    \$\begingroup\$ @Colin: Raising ISO raises the noise, if the aperture and/or shutter speed are adjusted to reduce the amount of light getting to the sensor, as is typically the case in autoexposure modes. To some people, this assumption is so natural that it is too obvious to state. To others, the assumption is so abhorrent that they break out in hives. \$\endgroup\$
    – coneslayer
    May 16, 2011 at 16:02
  • 1
    \$\begingroup\$ @Matt Grum: But what did you have to do to take those photos? With the same shutter and aperture settings, you must have either used a lot of ND on the high ISO shot, or boosted the image brightness a lot in post-processing for the low ISO image (or some other manipulation of the light level). That means that your SNR on the sensor was actually much worse in the low ISO shot. That changed the actual SNR at the sensor, so anything you do after that to make the images comparable is going to amplify a lot of noise in the dimmer image. \$\endgroup\$
    – Colin K
    May 16, 2011 at 16:16
  • 1
    \$\begingroup\$ @Colin sorry I didn't read the bit about SNR, yes you get a higher absolute noise level when you increase ISO but you get a better signal to noise ratio. When I talk about noise I mean signal to noise ratio, as this is what determines the amount of noise you see in an image. To answer your question the ISO100 shot was underexposed. By boosting exposure digitally I amplified both the image noise and read noise giving a worse SNR compared to the analogue amplified high ISO image. \$\endgroup\$
    – Matt Grum
    May 16, 2011 at 20:20
  • 3
    \$\begingroup\$ @Matt: Then it was the proper exposure which gave you a better SNR. Applying gain to a signal (analog or digital) has, at best, no impact on SNR, and in practice actually reduces SNR somewhat. A higher ISO only gets you a better SNR when it allows you to get a better exposure. You demo was flawed in the sense that the 100 ISO shot was not properly exposed. If you repeat the test, but use ND filters or a change in lighting to make the low ISO shot exposed correctly, it will have equal or better SNR than the high ISO shot. \$\endgroup\$
    – Colin K
    May 16, 2011 at 20:28
0
\$\begingroup\$

The primary cause of noise in an image is due to Photon Shot Noise as a result of not collecting enough light... This noise is due to the randomness of light/photons. When very little light is collected it is more likely that sensels (pixels) will receive inadequate or even no light. Think of this as being exposed to rain... a short exposure to a light shower and you won't be entirely soaked/covered.

ISO simply allows you to collect less light (smaller aperture/faster SS).

More MP's (sensels/area) simply divide the light that is collected farther (more affected by the randomness). But if the sensor sizes are the same and the images are displayed at the same physical size, then the images will have the same light per area and thus the same noise characteristics, regardless of the MP count.

\$\endgroup\$
0
\$\begingroup\$

Some nice answers already. My 2 cents. I am trying to make a complete list.

The factors that affect ISO are:

  1. The amount of light hitting the sensor. Less light more noise. You can at some extent control it manipulating your aperture and the shutter speed, or adding more light to the scene.

  2. The quality of the sensor. Speaks for itself.

  3. The area of the sensor. The bigger the lesser noise.

  4. The working temperature of the sensor. The colder it is the less noise it has.

  5. The area per sensel (pixel) here the Megapixels have a bit of impact, but this variable is compensated by the area of the sensor. If two sensors are of the same size, but a different megapixel count the lower megapixel will have less noise, but a similar result could be achieved resampling the bigger megapixel count to the lower one.

  6. The Magnification of the signal. See point 7.

  7. The computation made by the camera to compensate for it. This is where the ISO comes into play, the algorithm used to boost the ISO is calculated for the specific characteristics of the sensor. Here is one of the advantages of using more modern processors, besides more modern sensors. Another computation can be done is the averaging of a long exposure shot.

  8. The amount of manipulation the user makes to the file, referring to how much boosting of the shadows is made using for example curves.

  9. The post-processing done to a file, referring to a noise reduction algorithm.

  10. Some other noise reduction technique like combining different shots. This is similar to the amount of light hitting the sensor (point 1) but done artificially.

  11. Any resampling is done to the file. Resampling down a file with noise using an averaging algorithm like bicubic will smooth it. See point 5.


Every

\$\endgroup\$
-4
\$\begingroup\$

Response: ISO

Examples:

  • ISO 80 (low iso) -> image without noise

  • ISO 6400 (high iso) -> image with a lot of noise

If you raise the ISO level, you will five the photo more noise.

\$\endgroup\$
4
  • 1
    \$\begingroup\$ Counter example: ISO 100 (low iso) -> image with a lot of noise, ISO 6400 (high iso) -> image with little noise. Proof: mattgrum.com/ISOcomparison/ISO_100_vs_ISO_1600.jpg \$\endgroup\$
    – Matt Grum
    May 16, 2011 at 14:48
  • 2
    \$\begingroup\$ @Matt: You are certainly correct, but I think the way most photographers think about things is that when you increase ISO, you offset that increase with a faster shutter speed or smaller aperture, to produce (roughly) the same photograph. (In most autoexposure modes, this way of thinking corresponds to the mechanics of shooting.) I think we'd be better off encouraging people to state that assumption in their writing, rather than just saying they're wrong. \$\endgroup\$
    – coneslayer
    May 16, 2011 at 15:16
  • 2
    \$\begingroup\$ Ok so every noisy high ISO image would be even noisier if shot with a lower ISO. Likewise every noisy high ISO would be less noisy if shot with a longer exposure. So I don't understand why people call it "high ISO noise" and not "insufficient light noise". ISO doesn't create noise, which is exactly what statements like "all digital cameras exhibit image noise which rises as ISO increases" implies. If you instead say "all digital cameras exhibit noise which increases as incoming light decreases" you would be correct without the need to state any assumptions! \$\endgroup\$
    – Matt Grum
    May 16, 2011 at 15:39
  • 3
    \$\begingroup\$ "Ok so every noisy high ISO image would be even noisier if shot with a lower ISO." Only under your unstated assumption that shutter speed and aperture remain fixed. Which is not necessarily how most photographers actually think about things. \$\endgroup\$
    – coneslayer
    May 16, 2011 at 15:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.