How does ISO setting affect digital sensors?

GaryBakker · June 6, 2007

How does the ISO setting affect the photo quality on digital cameras?

I sort of understand with film how a higher ISO film could become grainy. But how does graininess occur with digital cameras?

The digital light sensor is what it is. There are only so many pixels* and the pixel is going to capture the light falling on it or it is not. It seems an either/or situation.

I guess I'm confessing that I don't fully understand what the ISO setting actually does in a digital camera. I'm going on an African safari in August and I want to make sure I get the best pictures possible.

*("pixel" is probably not the correct photography term, but you know what I mean -- what are they called? -- sensor elements?)

luke_church · June 6, 2007

Green Fleurdelis wrote:

How does the ISO setting affect the photo quality on digital cameras?

Short answer, as I'm kind of in a hurry...

ISO on digital cameras works by turning up the gain on the amplification circuits.

Low ISO -> low gain, high ISO -> high gain.

Imagine you've got a really loud radio station, you can turn the volume right now. Because of this, all you've got is the station.

However, if you have a queit station, you've got to turn the amplification (gain) up high. Because of this, you're also amplifying the hiss in the background.

Well it's pretty much the digital sensors. High ISO -> You're amplifying the signal a lot, but you're also amplifying the noise that comes from, e.g. heat or other sources. Conversely, on a low ISO, such noise is 'drowned out' by the image.

Low ISO therefore gives generally better image quality, but I'd rather have a high-ISO image, than no image at all. Particulally if you're using a larger/wierd sensor camera (DSLR, or R1, or some degree F30,F40), here the noise has much less of an impact.

Does that help?

luke_church · June 6, 2007

Green Fleurdelis wrote:

*("pixel" is probably not the correct photography term, but you know what I mean -- what are they called? -- sensor elements?)

There in lies a religious war. Pixel element is not strictly speaking correct, I believe it's a contraction of 'picture element', and if we're going to be seriously nit-picky that's a bit generous for most current cameras which have to do cross-site interpolation to produce 'pixels'. It's down-right imaginative for the Fuji cameras, which have an output that bears little geometric relation to its sensor. (And seemingly a good thing too :-))

So, technically, sensor element is better, but then if we want to be clear, we have to start identifying the difference in 'output resolution', which is kind of fudged into being a sensor element count, even when it shoudln't be. And we have to worry about multi-sensor site cameras, such as the Fuji super-CCD, and stacked sensors: Sigma.

Personally, I don't really give a monkey, call it what you like, and we'll disambiguate on the basis of context. :-) (One of the rare advantages of being human)

GaryBakker · June 6, 2007

luke_church wrote:

ISO on digital cameras works by turning up the gain on the amplification circuits.
<snip>
Well it's pretty much the digital sensors. High ISO -> You're amplifying the signal a lot, but you're also amplifying the noise that comes from, e.g. heat or other sources. Conversely, on a low ISO, such noise is 'drowned out' by the image.
<snip>
Does that help?

Yes. That is exactly the type of information I was looking for. Now if I were an electrical engineer I might have a much clearer idea as to what exactly "turning up the gain" really means, but your radio metaphor provides me with enough to give me an intuitive sense of what is going on.

Thank you.

Stan · June 6, 2007

Two very fine posts, very succinct
Thanks Luke

Stan

claudermilk · June 7, 2007

Green Fleurdelis wrote:

Yes. That is exactly the type of information I was looking for. Now if I were an electrical engineer I might have a much clearer idea as to what exactly "turning up the gain" really means, but your radio metaphor provides me with enough to give me an intuitive sense of what is going on.

Thank you.

It's really that simple. The radio volume knob analogy is perfect as that is turning up the gain. Same with the signal-to-noise ratio. Digital noise is the same exact thing as the background hiss you get in audio; the difference is seing it or hearing it.

GaryBakker · June 7, 2007

claudermilk wrote:

It's really that simple. The radio volume knob analogy is perfect ...

I find it fascinating that for completely different reasons, the end result for both film capture and electronic capture is the "same" -- as the ISO goes up, the quality goes down.

ziggy53 · June 7, 2007

One additional important point is that the dynamic range of the imager decreases with an increase in ISO, so that colors may appear more "synthetic" or "blotchy" when you increase the ISO, depending upon the scene.

Proper exposure becomes much more critical as the ISO increases.

Ric Grupe · June 7, 2007

luke_church wrote:

disambiguate

I hope you do not have the propensity for using that type of language!

colourbox · June 7, 2007

claudermilk wrote:

It's really that simple. The radio volume knob analogy is perfect as that is turning up the gain. Same with the signal-to-noise ratio. Digital noise is the same exact thing as the background hiss you get in audio; the difference is seing it or hearing it.

The audio analogy also works in that the quality of ISO boost depends on the quality of the technology behind it. If you have an audiophile amplifier, you can turn up the music pretty far and it's still sounds beautiful and clean, whereas if you turn up the volume on a cheap stereo the sound distorts and gets ugly.

In camera sensors, ISO is boosted by the signal amplifier on the processing chip. If you have a cheap camera, your high ISO images are full of noise. If you have a high-quality processor in your camera like the Canon DIGIC processor, you get the much cleaner high-ISO images that the Canon sensors are renowned for.

GaryBakker · June 7, 2007

ziggy53 wrote:

One additional important point is that the dynamic range of the imager decreases with an increase in ISO, ... Proper exposure becomes much more critical as the ISO increases.

I didn't know that. Thank you.

colourbox wrote:

The audio analogy also works in that the quality of ISO boost depends on the quality of the technology behind it. ... If you have a high-quality processor in your camera like the Canon DIGIC processor, you get the much cleaner high-ISO images that the Canon sensors are renowned for.

Interesting. Do you know if that sensor is used on the Canon non-SLR cameras too? (The reason I ask is that I'm disappointed in the picture quality of the new Olympus 550 with the 18x zoom, and I might return it for the Canon S3 IS.)

colourbox · June 7, 2007

Green Fleurdelis wrote:

Interesting. Do you know if that sensor is used on the Canon non-SLR cameras too?

Not sure. There have been several generations of DIGIC. I think they're up to DIGIC III now but only on the newest models. You can always check the ISO tests at review sites like dpreview.com and dcresource.com and compare the tests for the cameras you are concerned with. dcresource tests high ISO at normal and low light levels.

GaryBakker · June 7, 2007

colourbox wrote:

You can always check the ISO tests at review sites.

Thanks for the tip!!

claudermilk · June 8, 2007

colourbox wrote:

Not sure. There have been several generations of DIGIC. I think they're up to DIGIC III now but only on the newest models. You can always check the ISO tests at review sites like dpreview.com and dcresource.com and compare the tests for the cameras you are concerned with. dcresource tests high ISO at normal and low light levels.

Keep in mind the DIGIC chips and the sensors are two different IC's on the circuit board. The sensor captures the photons and send the resulting signal to the DIGIC ship for processing. My understanding is that the sensors in Canon's P&S cameras are source from Sony (like Nikon). If you look, the EOS cameras use CMOS (Canon's own) sensors, while the P&S lines use CCD. However, they still use their own DIGIC chips.

IIRC Steve's Digicams does a lot of ISO-setting testing as well. It's another good place to read reviews at.

GaryBakker · June 8, 2007

claudermilk wrote:

Keep in mind the DIGIC chips and the sensors are two different IC's on the circuit board. ... IIRC Steve's Digicams does a lot of ISO-setting testing as well. It's another good place to read reviews at.

Thanks Chris. Good information to have.

colourbox · June 8, 2007

Should have mentioned that the high-ISO goodness of Canon is mostly with their SLR sensors. Canon point-and-shoots aren't nearly as good at high ISOs.

flyingdutchie · June 11, 2007

Most of the previous replies explain it.

The radio-volume dial is a good analogy;
If you have a weak station ('low light situation'), you'd have to crank up the volume ('increase ISO setting') to be able to hear anything. However, you'll hear a lot of noise ('noise, blotches, grain') as well. If the noise is too much, you have a hard time understanding what is said on the radio ('loss of detail in the images'). Also, if the noise is too much, you can not hear the soft parts - low volume parts - of the broadcast very well or not at all (loss of dynamic range).

More technically explained:

Say a photosite (pixel) is a bucket. The bucket can be filled with electrons. The more light falls on the bucket, the more it will be filled with electrons. Say an completely empty bucket has 0 electrons, a completely full bucket has 100 electrons.

Before a picture is taken, all the buckets must be emptied. However, this emptying is not perfect (due to various reasons). After all the buckets are emptied, they still may contain some stray electrons. Say, on average, about 3 electrons are still there. Then a picture is taken:

In low light, a bucket is filled with only 10 extra electrons. That means that there are 13 electrons in there: 3 too much ~ 30% of the actual value of 10. If you set the ISO higher to get a lighter picture, this is multiplied by 5, for example --> The value will be 13*5 = 65. This is 15 more than the desired value of 50.

In good light, a bucket is filled with 50 extra electrons. That means that there are 53 electrons in there: 3 too much ~ 6% of the actual value of 50. Since the light was good, it is not necessary to increase the ISO.

Noise:
Good light gives an error of only 6%.
Low light (and then cranking the ISO up to make it lighter), gives an error of 30%.

Dynamic range:
Good light:
Range (signal : noise) is 100 : 3 = 33 : 1
Low light with an ISO setting that multiplies by 5:
Range (signal : noise) is 100 : 15 = 7 : 1

Rhuarc · June 11, 2007

flyingdutchie wrote:

Most of the previous replies explain it.

The radio-volume dial is a good analogy;
If you have a weak station ('low light situation'), you'd have to crank up the volume ('increase ISO setting') to be able to hear anything. However, you'll hear a lot of noise ('noise, blotches, grain') as well. If the noise is too much, you have a hard time understanding what is said on the radio ('loss of detail in the images'). Also, if the noise is too much, you can not hear the soft parts - low volume parts - of the broadcast very well or not at all (loss of dynamic range).

More technically explained:

Say a photosite (pixel) is a bucket. The bucket can be filled with electrons. The more light falls on the bucket, the more it will be filled with electrons. Say an completely empty bucket has 0 electrons, a completely full bucket has 100 electrons.

Before a picture is taken, all the buckets must be emptied. However, this emptying is not perfect (due to various reasons). After all the buckets are emptied, they still may contain some stray electrons. Say, on average, about 3 electrons are still there. Then a picture is taken:

In low light, a bucket is filled with only 10 extra electrons. That means that there are 13 electrons in there: 3 too much ~ 30% of the actual value of 10. If you set the ISO higher to get a lighter picture, this is multiplied by 5, for example --> The value will be 13*5 = 65. This is 15 more than the desired value of 50.

In good light, a bucket is filled with 50 extra electrons. That means that there are 53 electrons in there: 3 too much ~ 6% of the actual value of 50. Since the light was good, it is not necessary to increase the ISO.

Noise:
Good light gives an error of only 6%.
Low light (and then cranking the ISO up to make it lighter), gives an error of 30%.

Dynamic range:
Good light:
Range (signal : noise) is 100 : 3 = 33 : 1
Low light with an ISO setting that multiplies by 5:
Range (signal : noise) is 100 : 15 = 7 : 1

Wow, that is a fantastic explanation!! I think that should get put into the help/tutorial section for future users to benefit from. Thanks a lot!

GaryBakker · June 11, 2007

flyingdutchie wrote:

Before a picture is taken, all the buckets must be emptied. However, this emptying is not perfect (due to various reasons). After all the buckets are emptied, they still may contain some stray electrons.

Noise:
Good light gives an error of only 6%.
Low light (and then cranking the ISO up to make it lighter), gives an error of 30%.

Dynamic range:
Good light:
Range (signal : noise) is 100 : 3 = 33 : 1
Low light with an ISO setting that multiplies by 5:
Range (signal : noise) is 100 : 15 = 7 : 1

Ah. I now see the light. (pun intended)

While I understood the radio analogy regarding "static" and "turning up the gain", I was still a little perplexed regard the source of the noise. With a radio, the noise is stray radio signals and other deep space astronomical stuff.

With the sensor, I inferred that the noise came from stray photons due to imperfections in the lens (that didn't end up where they would if the lens were "perfect").

Your explanation of the photon noise makes sense in that it is related to limitations in the sensor technology itself (not clearing out completely).

I still think there is probably some exacerbating effect from lens imperfections too, that contribute to the noise. Is that true? But with enough light, the signal-to-noise ratio (as you point out) is large enough that these imperfections don't really matter.

Can you say more about the affect of this on the dynamic range? I don't quite understand what you are saying about that. How does the noise affect the dynamic range?

colourbox · June 11, 2007

Green Fleurdelis wrote:

With the sensor, I inferred that the noise came from stray photons due to imperfections in the lens (that didn't end up where they would if the lens were "perfect").

Digital noise doesn't come from the lens; otherwise it would also show up on film.

Digital noise comes from sensor impurities.

Rhuarc · June 11, 2007

I think stray photons from lens imperfections just show up as distortions / blurriness, etc... on the image. I wouldn't think it would actually cause noise. So if there was a nick (imperfection) in the glass then it would show up on the sensor as a dark spot, not as colored noise, etc...

GaryBakker · June 11, 2007

Thanks Rhuarc and colourbox for the facts. I appreciate your responses. Shows what can happen when conclusions are drawn from faulty assumptions.

PS - the article is interesting ... especially regarding heat as a source of noise.

As long as we're on the topic, where does the noise come from with high ISO film? I've always thought it was simply related to the chemistry involved and the actual ability of the film to detect photons. Is that close to the truth?

flyingdutchie · June 11, 2007

Green Fleurdelis wrote:

Thanks Rhuarc and colourbox for the facts. I appreciate your responses. Shows what can happen when conclusions are drawn from faulty assumptions.

PS - the article is interesting ... especially regarding heat as a source of noise.

As long as we're on the topic, where does the noise come from with high ISO film? I've always thought it was simply related to the chemistry involved and the actual ability of the film to detect photons. Is that close to the truth?

Noise on sensor's has to do with imperfect 'buckets' and with quantum mechanics. Heat increases the noise (more dynamic systems).
http://learn.hamamatsu.com/tutorials/noisegraph/

High ISO film needs a coarser grain to make it more sensitive. Why...? I don't know... that has to do with chemistry and i'm not familiar with that at all

But the funny thing is that the results are a bit similar.
Noise, especially luminance noise, looks a lot like film grain.

GaryBakker · June 11, 2007

flyingdutchie wrote:

High ISO film needs a coarser grain to make it more sensitive. Why...? I don't know... that has to do with chemistry and i'm not familiar with that at all

Me either.

flyingdutchie wrote:

But the funny thing is that the results are a bit similar.
Noise, especially luminance noise, looks a lot like film grain.

Yes, interesting isn't it?

(By the way, I followed your blog links and read the entry on the Prisoner's Dilemma and how it might relate to SmugMug. Also very interesting. I've been fascinated with PD for decades, and it will be interesting to see how this plays out.)

Stu Engelman · June 11, 2007

Hi,

This is a long thread, so if some of this reply is redundant I apologize, but this additional information may help.

When you take a shot, light rays are mapped to little pits in your sensor (these pits are called photosites or photocells). The photosites always have a certain amount of "information" (electrons) that were not placed there by your exposure (i.e., noise).

Sections of your composition that are very bright map alot of electrons to their asociated photosites, and dark sections map very few electrons to their asociated photosites. The latter will have a higher noise-to-signal ratio.

When you boost your ISO setting, you magnify the information in your photosites. Since the darker photosites have a higher noise-to-signal ratio, the darker parts of your composition will more prominantly display the magnified noise (evidenced as random colored dots in the shadows, sometimes called "luminance noise").

Just how much ISO is too much ISO depends on the quality of your sensor and extent of dark areas in your picture, but going anywhere over 200 often adds grain. The best approach is to use as wide an aperture as possible given your DOF requirements for the shot, and then try to keep ISO under 200 (to avoid grain) and shutter speed under 1/30 second (to avoid chromatic noise, evidenced by splotchy patches in your picture). If these parameters don't deliver enough light, you could shoot anyway and reduce noise in post-processing, or alternatively use a flash.

Stu

claudermilk · June 12, 2007

My understaning of how the film works is the larger silver crystals react to light faster, thus giving a higher sensitivity & ISO rating.

Here's the wikipedia page, which basically says that: http://en.wikipedia.org/wiki/Film_speed

Rhuarc · June 12, 2007

claudermilk wrote:

My understaning of how the film works is the larger silver crystals react to light faster, thus giving a higher sensitivity & ISO rating.

Here's the wikipedia page, which basically says that: http://en.wikipedia.org/wiki/Film_speed

Hmm, but then becuase the crystals are larger they show up more, aka, the film grain.

claudermilk · June 12, 2007

Rhuarc wrote:

Hmm, but then becuase the crystals are larger they show up more, aka, the film grain.

EXACTLY!

GaryBakker · June 12, 2007

As a consequence of the discussion in this thread (thank you everyone) I have:
a) learned a lot
b) been motivated to do additional online research

As an aside, I suppose I could have (some would say "should have") done the research first, but asking the questions here on dGrin and getting pointed to some pertinent sites is much more efficient. And much more enjoyable.

I've learned about photosites and how photons falling on them excites electrons that travel to and get stored in tiny capacitors, to be read by the processor one line at a time, then erased as it is read (but not always completely, hence one source of the noise), then permanently stored for retrieval later. Quite a complicated process. Not to mention the gyrations that have to occur to figure out the color of the light falling on each photosite.

Cool stuff.

How does ISO setting affect digital sensors?

Comments