Talk:Image noise

This article needs work

Much of what this page says is wrong or unclear. And it needs sources. I'm going to start by putting sources and cleanup tags, and removing the image unless someone can explain it including a source that justifies it. Dicklyon 07:19, 18 January 2007 (UTC)

OK, I took out what I considered to be the worst of the hot air. It would be good to put something back about noise types or sources, but if we have names like "salt and pepper noise" they need to be backed up with credible sources. Dicklyon 07:37, 18 January 2007 (UTC)

The part I took out started with "salt-and-pepper noise (also known as random noise or independent noise), pixels in the image are vastly different in color from their surrounding pixels." I'm pretty familiar with image noise statistics, but I don't see how "random", "independent" and "salt-and-pepper" can mean the same thing, and certainly none of them imply "vast" differences. Hence, the need for something verifiable instead. Dicklyon 07:45, 18 January 2007 (UTC)

I've corrected some confusing terminology: "sensitivity" in place of "film speed" (it was clearly misplaced with reference to digital cameras) "definition" in place of the very ambiguous "resolution" ; "detail" in place of "pixels", and so on. The whole page is still crying out for a major overhaul, maybe even a merge. Most of the noise reduction section is verbatim from the noise reduction page ~ mikaul^talk 18:50, 1 May 2007 (UTC)

"Film speed" refers to ISO on both digital and film cameras. Even though film is not used in digital cameras, it still refers to sensor sensitivity, as film speed on film cameras refers to film sensitivity. Furthermore, "sensitivity" is not quite as precise, since it can refer to things other than sensor sensitivity. Finally, image noise occurs with both digital and film cameras. "Film speed" is appropriate wording. Althepal 17:17, 9 May 2007 (UTC)

I realise that's the way the "beginners guide" photography books introduce it, but there's really no proper correlation between ISO (film speed) and ISO (electronic device sensitivity) - if you don't believe me, see the ISO website. I'm not sure which other thing "sensitivity" could be confused with vis a vis digital capture, and "noise" is a term common to electronics; "grain" is the equivalent for analogue capture. If you have an appropriate citation to support all of this, by all means go ahead and post up the changes. You might want to do so at Film speed as well, as there's nothing to support what you're saying ther either, as far as I can see. mikaul^talk 20:07, 9 May 2007 (UTC)

I simply pointed out that ISO is popularly referred to as film speed for both digital and film cameras. If you don't believe me, you can check out photography forums and articles, and you will see that ISO is described as or referred to as "film speed" without regard for camera type. (Even in the previous version of this article it was referred to as such.) Film speed always means film/sensor sensitivity, while sensitivity refers to how sensitive anything in the world is. I'm not making changes, just pointing out facts. Althepal 21:44, 9 May 2007 (UTC)

That's ok, I'm with you. It's definitiely in regular colloquial use, I see it used it all the time in this context. mikaul^talk

IIRC, the ISO speeds are specific scientific standards based on optimum exposure for silver halide-based images. A digital camera has what are considered ISO-equivalent "speeds", but it must also be remembered that in fact the sensor only has one true sensitivity - the rest is essentially the software, and would be considered the same as lifting in the print or perhaps force processing - past a limited point, all of these reach a threshold at which the signal to noise ratio drops greatly. Girolamo Savonarola 00:44, 10 May 2007 (UTC)

You do not recall correctly. The film speed is based on the minimum amount of light needed to get a certain quality threshold. Similar standards exist for digital, though they define 3 different methods of computing different ISO speed ratings for digital cameras. These are not generally applicable to the sensor in isolation, though logically they should be. The ISO setting or exposure index is something else again, as you say. Dicklyon 04:46, 10 May 2007 (UTC)

This is an article about 'Noise'. Film does not have 'Noise'. Film has 'Grain'. The speed, or sensitivity of a film is a characteristic of the grain. Film with larger grains is more sensitive to light than film with smaller grains, but larger grains also reduce the resolution of the film. It is this reduction in the resolution of film that seems to have been incorrectly associated with 'Noise'. A more correct association to 'Grain' in digital photography would be 'Pixelation'. In 'Film', there is no corrolary for 'Noise'. Any reference to 'Film' should be removed from this article. --TCav 21:08, 30 August 2008 (UTC)

Why in fact there is! Just like with digital, the image recorded on the film medium is a "signal", or information. Noise is random variation in that signal, or random variation in the image information - pixels that are randomly brighter or darker, or randomly colored a different way (which is really randomly brighter areas in the different color channels). You're describing another aspect of high-ISO film, which is the reduced resolution of larger-grained films. However resolution is irrelevant to noise. Tejastheory (talk) 21:52, 30 August 2008 (UTC)

I understand that digital images can and do suffer from noise. What you are referring to is what happens in digital imaging. That type of thing doesn't happen in film images. Film images are constructed from grain which correspond to the pixels in digital images. The difference is that grains on film are randomly positioned within the emulsion, while the pixels on the digital image sensor are in a pattern. The random variation you refer to is random errors in the pattern of signals from the digital image sensor. Nothing like that happens in film. You are confusing the randomness of a film's grain with the randomness of a digital image sensor's noise. A film's grain corresponds to a digital image sensor's resolution. When there is not enough resolution, on film it's called 'graininess', and on digital imaging it's called 'Pixelation'. There is nothing in film that corresponds to the noise from a digital image sensor. --TCav 23:37, 30 August 2008 (UTC) —Preceding unsigned comment added by TCav (talk • contribs)

No, I think you are misunderstanding what the term "noise" means. It is random variation in the image - say you took an image of a blank wall and expected it to be a certain uniform X brightness. If you used a high-ISO sensitivity on a digital sensor, or a high-ISO film, you would notice that in both cases you'd get some random variation - individual pixels or individual points on the film would be darker or brighter than the uniform X brightness you expected. You might say that film noise has a more "grain" like feel to it, since the pattern is more random than a digital image's pixels, but that's completely irrelevant to the fact that there is noise, which just means that there is some random variation in the image. Tejastheory (talk) 10:37, 31 August 2008 (UTC)

'Grain' is 'Grain' and 'Noise' is 'Noise'. I think perhaps you are misunderstanding what the term 'Noise' means. 'Noise' is the bad part of the 'Signal to Noise Ratio'. Noise obscures the signal. Noise is the stuff we don't want. In film photography, the 'grain' is the 'signal'; the 'grain' is NOT the 'noise'. We want the 'grain'. The individual grains in a film image are the equivalent of the pixels in a digital image. The fact that the individual grains are in a random pattern while the pixels are in a grid is immaterial. 'Noise' in a digital image is when a pixel goes awry. Individual grains don't go awry, therefore 'Noise' (the way you describe it here) doesn't occur in film photography. --TCav —Preceding unsigned comment added by TCav (talk • contribs) 01:23, 1 September 2008 (UTC)

Okay this isn't even related to the article so I'll leave it with this and if you're not really interested in hearing about it then I don't care anymore. Here's an example of an image taken on FILM:

[1]

Notice that there is a lot of "graininess" all over the image? That is, you'd expect a smooth gradient at the top of the image, but instead you have lots of speckles - dots that are brighter or darker than what they should be in the smooth gradient. This is random variation in the brightness of different areas of the image. What's the definition of noise? Random variation in the information. Thus this film image has noise. Tejastheory (talk) 06:29, 1 September 2008 (UTC)

The random nature of 'Grain' does not make it 'Noise'. —Preceding unsigned comment added by TCav (talk • contribs) 18:58, 1 September 2008 (UTC)

Film Grain has an article. That article never refers to 'Grain' as 'Noise'. If 'Grain' were 'Noise', wouldn't the article on 'Grain' say so?

Film Speed has an article. That article never refers to 'Grain' as 'Noise'. In fact, it doesn't even talk about 'Noise' until the section on digital photography. And the only place where it talks about 'Grain' AND 'Noise', it makes a distiction between them: "However, this loss is visible as image noise rather than grain."

Not even on the Discussion pages for these articles is 'Grain' ever called 'Noise'. This is the only article where 'Grain' is called 'Noise'. That is an error. 'Grain' is not 'Noise'. If 'Grain' were 'Noise' then 'Grain' wouldn't need its own article. --TCav 17:54, 1 September 2008 (UTC)

If anything, 'Grain' is a natural, random Halftone. The halftone itself is the signal, not the noise. Note that the Halftone article refers to film photography as "continuous tone imagery". It goes on to say "At a microscopic level, developed black and white photographic film also consists of only two colors, and not an infinite range of continuous tones." You may also note that, in the halftone article, the term 'Noise' is conspicuously absent. --TCav 01:29, 2 September 2008 (UTC)

If you'd like to help improve the articles here are some sources that should help. Dicklyon (talk) 04:37, 2 September 2008 (UTC)

WikiProject class rating

This article was automatically assessed because at least one WikiProject had rated the article as start, and the rating on other projects was brought up to start class. BetacommandBot 09:54, 10 November 2007 (UTC)

Dispute Resolution

I have pointed out on this Talk page that I believe the main page is incorrect; Film Grain is not Noise. Since this is an article on Image Noise, I believe it is inappropriate to mention Film Grain at all.

Film grain, while random when compared to the pattern of the photoreceptors on CCD or CMOS image sensors used in digital photography, constitutes Signal, not Noise. If anything, film grain can be compared to a random, naturally occurring Halftone. Therefore, a phenomenon that occurs in digital photography as an unwanted consequence of digital signal processing has no relationship to a phenomenon that occurs in film photography as the underlying photochemical reaction that makes film photography possible.

Tejastheory disagrees, and from his responses on this page, the basis for his disagreement is that film grain is random, and therefore, noise.

Two weeks after Tejastheory's last response, I edited the main article, removing any mention of Film. Three days later, Tejastheory undid my edit.

I would like to resolve this dispute, and look forward to your contributions. --TCav (talk) 17:31, 20 September 2008 (UTC)

What's the basis of your dispute? Just your opinion? We should base this on sources. You failed to follow up the discussion above where I said "If you'd like to help improve the articles here are some sources that should help." (follow the link there). Dicklyon (talk) 19:46, 20 September 2008 (UTC)

The basis of the dispute is that film contains grain, and grain is signal not noise. If you read the articles I referenced on Signal and Noise, you will find that film grain fits the definition of signal and does not fit the definition of noise. Therefore, it is inappropriate for this article to say that film grain is noise.

The photosensitive particles make film photography possible, and no one ever thought to call it noise until digital photography when someone made the inappropriate comparison between the random nature of noise in a digital image and the random nature of grain in film, and there wasn't anyone around to correct the mistake. —Preceding unsigned comment added by TCav (talk • contribs) 03:22, 21 September 2008 (UTC)

I think you have got your semantics all mixed up. In one instance you're talking about grain as the physical crystals that make up a film emulsion, and then you start talking about grain as the random variation "graininess". You state "you will find that film grain fits the definition of signal and does not fit the definition of noise" and right after that specifically state "someone made the inappropriate comparison between the random nature of noise in a digital image and the random nature of grain in film". If something is RANDOM, it cannot be a signal! A signal is information, and "information" that is random really imparts no information at all, it is statistical noise.

Here's a really easy way to boil down this argument.

1. A film slide or negative contains information.
2. By definition, information is represented by a signal.
3. It is possible for all signals to have random variation (in our case, one point in an image may be brighter/darker than an adjacent point, if for no other reason than statistical variation originating from shot noise).
4. By definition, random statistical variation in a signal is called noise.
5. Ergo, it is possible for images (which is a piece of information/signal) to have noise. This applies regardless of what medium the image is contained in.

I've laid out the steps very logically, so if you still choose to think differently, please point out the fallacy in the above process. Tejastheory (talk) 09:50, 21 September 2008 (UTC)

The fact that grains of photosensitive material on film emulsion are oriented in a random pattern does not make it noise. Adjacent grains may react differently because of the random nature of the paths that the photons took. That's the shot noise (by definition). The grain on the film accurately represents the light that was reflected off a subject, and is, therefore, not noise. Your argument presumes that the film grain is the source of the noise, when in fact, film accurately records the noise inherent in light. If you were allowed to view a scene for only a short period, your retina and cerebral cortex would have recorded the random paths (shot noise) of those photons as well. --TCav (talk) 13:59, 21 September 2008 (UTC)

I think you need to figure out which definition of grain you are talking about. Are you defining grain as the individual crystals that make up the film emulsion? If yes, then no one is calling that noise - that'd be like calling individual photodiodes "noise", which makes no sense at all.

The "grain" referred to here is actually the variation in the image information. This is the definition used when someone looks at a high-ISO film image and declares it "grainy". Why do you think it is that the same person would declare a finely-grained low-ISO film to be "not grainy"? Do they mean that the low-ISO film has no grain? (i.e. the silver-halide crystals, as per the definition above?) Tejastheory (talk) 20:23, 21 September 2008 (UTC)

Tejastheory wrote: "If something is RANDOM, it cannot be a signal!"

How do you feel about fingerprints?

Sigh... your analogy does not work at all. If I told you, "Someone give me a model of the ideal human fingerprint", and all I got was samples of every individual's fingerprint, then YES, all of the deviations from my average/ideal fingerprint would be statistical NOISE. Tejastheory (talk) 20:23, 21 September 2008 (UTC)

Let us presume that you are correct, that film grain is noise. Since film grain is the visible effect of the reactions of photosensitive particles after being exposed to light, and photosensitive particles are all there is on film, then noise is all there is. Does noise just happen to organize itself into an image that looks striking like the subject? If film grain is noise, where does the signal come from? --TCav (talk) 14:03, 21 September 2008 (UTC)

Come back after you've settled on a definition of film grain. The crystals themselves are not noise - that doesn't make sense. The "graininess" is the noise. And graininess is random variation in the image information contained on the film. Tejastheory (talk) 20:23, 21 September 2008 (UTC)

I agree with Tejastheory. It's like in an image sensor that detects light by counting collected electrons; the discrete electrons are the source of shot noise, and they are all there is; that doesn't mean the electrons themselves are noise, or that they don't also represent the signal; it just means that their representation of the signal is noisy; and I have provided a book-search link to many books that back up this idea, if more refs are needed. TCav, do you have sources that say that film grain is NOT a noise source, or is not noise? Dicklyon (talk) 21:10, 21 September 2008 (UTC)

TCav noted that "no one ever thought to call it noise until digital photography;" this may be somewhat true. We find books talking about film grain noise in the 1970s and 1980s, when films were being scanned with microdensitometers and the measurements were being processed on computers to extract image information for medical and scientific purposes. Not exactly digital cameras, but digital techniques being applied to photography. The only other old applications of the term noise was in movie film sound tracks, where the noise due to the random nature of film grain was an audible noise in the sound. Anyway, even if film grain noise wasn't often called noise in the old days, it is a noise, the main source of image noise in film images, and has been widely recognized, discussed, and treated as noise. Dicklyon (talk) 21:15, 21 September 2008 (UTC)

If you expose B&W film to a large gray subject, the proportion of transparent grains to opaque grains will be the same, whether the film contains small grains or large grains (that is, whether you use a low ISO film or a high ISO film, and presuming proper exposure.) The proportion of transparent grains to opaque grains is the signal, and is unaffected by the physical size of the individual grains. Noise is something that affects signal (See Electronic noise: "... noise is an unwanted signal characteristic ...".) If the size of the grain doesn't affect the signal, then it isn't noise. The size of the grain is just an unrelated characteristic (ignoring variations in it's sensitivity to light.) --TCav (talk) 00:47, 22 September 2008 (UTC)

In the limit of large area, the average signal is unaffected, but in any finite area that grain is apparent as a random fluctuation, a noise. But arguing is pointless if you're not going to propose a source that agrees with you. Dicklyon (talk) 00:50, 22 September 2008 (UTC)

In *some* finite areas, the proportion of transparent grains to opaque grains will certainly be skewed toward transparent, but in *other* finite areas, the proportion of transparent grains to opaque grains will be skewed toward opaque. Just as in some finite areas, my fingerprints are, no doubt, identical to yours, but in others, they are entirely different. And again, the size of the individual grains is irrelavent. Your argument applies for film with smaller grains just as well as it does for film with larger grains. Therefore, there's no basis for saying that a film with larger grains is noisier than a film with smaller grains. It's just "grainier" which is an unrelated characteristic.

And as for sources, I've been quoting sources all through this and the prior discussion that agree with me. --TCav (talk) 01:13, 22 September 2008 (UTC)

I have not been able to find where you mentioned a source; could you please tell me where to look for support for your point of view on this? Dicklyon (talk) 01:43, 22 September 2008 (UTC)

This makes no sense. Taking your same example, I could take a picture of a large gray subject with a digital image sensor. If you average the pixel values of the entire image, it will turn out grey. So by your example, this digital image has no noise either? Tejastheory (talk) 01:18, 22 September 2008 (UTC)

It would be a lighter shade of gray than the subject. Noise in digital photography isn't randomly darker or lighter, thereby being averaged out. It is predominantly (if not entirely) lighter than the actual subject. Therefore, the digital image would not be the same shade of gray as the actual subject (again, presuming proper exposure.) --TCav (talk) 01:55, 22 September 2008 (UTC)

The only component of image sensor noise that is biased toward lighter is the dark-leakage noise in long exposoures. The others, shot noise, amplifier noise, quantization noise, etc., are generally zero-mean. I've added a bit more info and a bunch of citations linked to book pages, so you can learn more about this noise stuff. The film grain noise is included among those. Feel free to add "citation needed" tags to anything you feel is suspect or needs to be backed up better by sources. Dicklyon (talk) 03:48, 22 September 2008 (UTC)

From [Image Engineering's 6mpixel.org article on noise] (listed in the main article as an external link): "The term noise is related to artefacts in digital images caused by the capture and transfer of the signal. It appears as coloured or bright spots in the image."

Therefore, noise in a digital image would have the effect of increasing the average luminance of the image. Therefore, the digital image would have a lighter shade of gray than the actual subject, or the film image for that matter (again, presuming proper exposure.)

Signal plus Noise is *always* greater than Signal alone. Noise *always* increases the average luminance of an image.

And noise "appears as coloured or bright spots in the image"[[2]], which means that image noise is *not* the result of zero-mean effects. --TCav (talk) 13:16, 22 September 2008 (UTC)

If the link to non-authoritative content is giving you that wrong impression, we should probably just remove it. Dicklyon (talk) 14:49, 22 September 2008 (UTC)

What wrong impression? 'Signal plus Noise is *always* greater than Signal alone'? --TCav (talk) 20:32, 22 September 2008 (UTC)

Yes, that one. Noise is usually approximately zero-mean, just as much lower as higher, in images as elsewhere. Dicklyon (talk) 04:44, 23 September 2008 (UTC)

No. Signal and Noise share the same medium. Noise never reduces Signal. If Noise reduced Signal, then the term "Signal to Noise Ratio" would be meaningless.

Shot noise in digital imaging is when adjacent photoreceptors detect a different number of photons when an inspection of the subject would seem to indicate that they should have received the same number, and only occurs when the sampling period is too short (the shutter speed is too fast). Amplifying the medium (increasing the ISO setting) increases the difference between the photoreceptors, but the apparent 'Noise' is actually Signal. That is, the adjacent photoreceptors actually did receive different numbers of photons. In this situation, Shot Noise is a sampling error (Key Word: Error), not Noise.

Amplifier noise is noise that is introduced onto the medium by an amplifier. That is, instead of just carrying Signal, the medium now carries both Signal and Noise. The description of Amplifier noise on this very page starts with "The standard model of amplifier noise is additive ..."

"Quantization Noise" is a misnomer. The correct term is Quantization Error. Quantization Error is what happens when you put a Canadian coin into a coin sorter designed for US Coins. A Canadian coin will jam the coin sorter, a quantization error. In an analog to digital converter, Quantization Error is when the ADC rounds or truncates the analog signal so it fits neatly into a digital value. That's a systematic error (Key Word: Error), not Noise.

Thermal noise and Dark noise, like Amplifier noise, are always in addition to the signal.

Therefore, Signal plus Noise is always greater than Signal alone. --TCav (talk) 16:09, 23 September 2008 (UTC)

Every method I know of, or have ever heard of, for measuring noise, boils down to this:

Noise = ( Signal + Noise ) - Signal

This doesn't work if Signal plus Noise isn't greater than Signal alone.

TCav (talk) 00:57, 24 September 2008 (UTC)

Heh, there are such things as negative numbers. Tejastheory (talk) 08:31, 24 September 2008 (UTC)

These are DC voltages we're talking about here. I would be pleased if you could point to a single source of noise that reduced current. --TCav (talk) 12:31, 24 September 2008 (UTC)

Have you read any of the sources on shot noise, Johnson-Nyquist noise, salt-and-pepper noise, etc.? They all can reduce as well increase the signal level (usually a charge or a voltage, but would also be a current at some point in the system). Maybe you're thinking of noise variance, or power; these generally add; that is, adding a zero-mean noise increases the power or variance of the signal + noise, even though it doesn't affect the mean. Dicklyon (talk) 15:37, 24 September 2008 (UTC)

Here's an example - the scene is a certain brightness, but with a small sample of light, shot noise will mean that not every location on the sensor/film receives a uniform amount of light, even though the scene is. Some locations (pixels or film grains) will be lighter, some will be darker. Darker points actually reduce the signal, so you see not everything is additive. Tejastheory (talk) 18:11, 24 September 2008 (UTC)

I talked about "shot noise" before, but I'll try again. "Shot noise" is when adjacent photoreceptors receive a different number of photons, when one would have expected them to receive the same number. The difference may be very small, but when the signal is amplified, so is the difference, and the difference may appear in an image.

This is not noise! This is a sampling error! On a digital image sensor, the adjacent photoreceptors actually did receive a different number of photons! That's signal, not noise. The Wikipedia article on shot noise states "... shot noise is often only a problem with small currents or light intensities." Therefore, if instead of amplifying the minute signal (increasing the ISO setting), one uses a longer shutter speed or a larger aperture to intensify the signal, the shot noise goes away, or becomes undetectable after further signal processing (quantization error, for instance), and therefore does not appear in the image. Therefore, shot noise is a sampling error, not noise. --TCav (talk) 19:39, 25 September 2008 (UTC)

Johnson–Nyquist noise is, indeed, zero-mean noise, but only that part of the noise that travels in the direction of the DC signal would be detected. Johnson–Nyquist noise could occur in any part of a digital camera circuit, but in any part of the circuit after the amplified photoreceptors, the noise level would be insignificant. And that part of the noise that travels in opposition to the DC signal would reach the photoreceptor and be cleared when the image sensor was reset prior to the next exposure. So the only part of Johnson–Nyquist noise that would appear in an image would serve to make the image lighter. --TCav (talk) 19:39, 25 September 2008 (UTC)

Salt and pepper noise "is produced by corrupting the original image so that individual pixels are randomly flipped to black or white (0 or 255 for 8-bit gray-scale) with some low probability"[[3]]. That is, naturally occurring "salt and pepper noise" is the result of malfunctions in the signal processing system. "Salt and pepper noise" is the visible effect of a malfunction. It is a malfunction, not noise. (I purposely included "naturally occurring" because the most prevalent source of "salt and pepper noise" is applications that add the "noise" to an image so they can test methods to remove it.) --TCav (talk) 19:39, 25 September 2008 (UTC)

TCav, I acknowledge your strongly-held views of which noises should be considered "noise" and which should not. But please read WP:V and WP:RS, and then if you find that you have sources that support adding to or changing what the article says, go ahead and just do so. All this talk is getting us no place. Dicklyon (talk) 03:40, 4 October 2008 (UTC)

Dicklyon, I have tried very hard to illustrate the difference between Signal and Noise. I have tried very hard to show that film grain fits the definition of signal and does not fit the definition of noise. If film grain were noise then photography would have been abandoned long ago as a futile exercise. But it wasn't. Images (signal) appear on film in the form of grain. Every reference to film grain being noise states that coarse film grain is noisier than fine grain film, and in an attempt to prove it, talks about sampling and distributions that apply equally to coarse grain as well as fine grain, failing to draw a distinction. Calling coarse grain noisier that fine grain is like calling a image printed at 300dpi noisier than an image printed at 1200dpi. The differences between the two images would be in the amount of signal, not the amount of noise.

I can present reliable sources with verifyable data that the Earth is flat, but they would still be wrong. I can only presume either of two possibilities. Either people misunderstand the references they quote here, or the references they quote here are mistaken. --TCav (talk) 11:10, 4 October 2008 (UTC)

Sorry. According to WP:V, if most sources were to say that the earth is flat, then that's the view we should primarily present in wikipedia, even if some of us were to be pretty sure it's wrong. We could however also present alternative viewpoints if they have reliable sources. Without sources, your arguments here amount to WP:OR and other non-WP:V content, which is what I'm trying to tell you. It is pointless to try to engage other editors in a debate on a technical dispute, when sources are available on only one side; provide sources on the other side and we'll have something to talk about. If you study up on this, I think you'll find that just as the number of electrons collected in image sensor photodiodes represents light intensity (the desired "signal") plus shot noise (the undesired but unavoidable statistical fluctuation due to quantum efffects), film grain represents both the light intensity (the desired signal) and noise due to the random discrete representation of that signal. Your logic that it is signal and therefore can't also be noise is not one that follows from any fact in evidence. But I'm wasting my breath trying to tell you that again, right? Dicklyon (talk) 16:35, 4 October 2008 (UTC)

I don't understand what you're trying to say at all. "Noise in digital photography isn't randomly darker or lighter, thereby being averaged out" - there isn't any randomness to digital noise, yet it averages out? "It is predominantly (if not entirely) lighter than the actual subject." meaning digital images are always brighter than the actual subject? Tejastheory (talk) 05:21, 22 September 2008 (UTC)

I think you mis-parsed his confused statement, but it's clear that he doesn't understand image noise. That's why I pointed him toward reading the sources instead of trying to carry on the argument here. Let's drop it until he does so. Any more comments that don't directly relate to the relationship of statements in the article to sources should just be ignored, I think. Dicklyon (talk) 05:54, 22 September 2008 (UTC)

Signal noise and sound

I've reverted the claim that visible "noise" terminology originates with analogy to audio noise. It's a term in electronics exactly the same thing as audio noise, ie distortion of a signal due to amplification. I don't see any analogical reference to audio noise (the word originates with the latin for seasickness, interestingly enough..) at the other electronic noise articles so I'm guessing this is anecdotal at best. If there's a good reference for it, I'd stand corrected. I do think a broader definition is called for in the lede though, along with a brief explanation of the parallel with other electronics applications. --mikaul^talk 21:59, 23 June 2009 (UTC)

This is palpably wrong. Noise is a word that is understood by everybody, and long precedes electronics. Signal noise is not just "analogous" to unwanted sound; it IS perceived as crackling and hiss when listening to a crystal radio. For example, from the 1922 version of the Radio Amateur's Handbook by A. Frederick Collins: "STATIC.--Also called atmospherics, grinders, strays, X's, and, when bad enough, by other names. It is an electrical disturbance in the atmosphere which makes noises in the telephone receiver." There are several mentions of noise in this book, but they invariably mean a sound; presumably it hadn't yet become customary to speak of electrical noise. I've changed the heading if this Talk section, it's more than analogy (although I myself described the usage as an analogy when I first edited the article).

I didn't add the note on audio noise as a remark to no particular purpose, but because in some writing about cameras there are unclarified remarks about noise that can be confusing (one problem with this DSLR camera with instant-return mirror and small sensor is noise). I also added "visible" before image for a similar reason, perhaps pedantic in this case: an "image" file is used for such things as an exact copy of a hard disk. Pol098 (talk) 22:19, 23 June 2009 (UTC)

I agree with Pol098. The sound concept of noise was adopted into electronics by telephone engineers (e.g. Johnson–Nyquist noise), and it spread to other information topics via Claude Shannon and his information theory, primarily; it's pretty much just analogy to sound, but made more rigorous. Dicklyon (talk) 03:42, 24 June 2009 (UTC)

In this 1917 AIEE paper, the electrical noise induced in a telephone circuit by a power line was called "Noise" in quotation marks, suggesting that the usage was novel about that time. Dicklyon (talk) 03:47, 24 June 2009 (UTC)

Ok cool, we have a source for the entrance of the term into electronics. Do we really want this in the lede section at Image noise? It's also entered into fields as diverse as computing, statistics and stock market activity, so I'm struggling to see the acute relevance, plus of course this isn't specific to image noise; it would be more relevant to one of the electronic noise articles, of which this is a subtopic. IMO it's much more interesting and relevant to mention the parallel to electronic audio noise here, or at least the relationship they share to amplification and signal fidelity, rather than give undue weight to an (interesting but less relevant) etymological point.
While I might concede that some mention should be made of non-audible, visible noise, it needs no more than that, just two words, which probably do belong in the lede. I notice the etymological info is back up there and I'm afraid it proves my point. --mikaul^talk 04:43, 24 June 2009 (UTC)

I changed it to a more concise statement, not in the lead paragraph but still in the lead, with a more specifically relevant source that directly supports the point. Dicklyon (talk) 05:04, 24 June 2009 (UTC)

Right, that's miles better but the whole section is pretty incoherent and doesn't address Pol's original points which I think had some weight. I'd also like to see the concept of "ideal photon detector" referenced, or explained somewhere on the page or (at least) linked to some quantum efficiency article somewhere.
I've given the whole thing a copyedit without substantially altering the meaning except to add in some clarification. I'd prefer to post it here first and iron out any objections rather than a load of edits on the page:

Image noise is the random variation in brightness or color information in images produced by the sensor and circuitry of a scanner or digital camera. Image noise can also originate in film grain and in the unavoidable shot noise of an ideal photon detector.

Image noise is most likely to occur with low signal level, for example in underexposed images or shadow regions, and is generally regarded as an undesirable by-product of image capture. Although these unwanted fluctuations became known as noise by analogy with unwanted sound, they are inaudible and actually beneficial in some applications, such as dithering.

--mikaul^talk 09:06, 24 June 2009 (UTC)

I find "Image noise is most likely to occur with low signal level" to be misleading; what's clearly true is that signal-to-noise ratio is worse at low signal levels, so the noise becomes more objectionable there; but depending on what you measure, the noise may actually be less (e.g. if you measure input-referred noise). Take it out of the lead and expand on it appropriately later (it may already be there). Dicklyon (talk) 14:52, 24 June 2009 (UTC)

Done. I'm not sure where to add that, as it's not solely relevant to digital cameras or gaussian noise, to pick the most obvious candidates. Maybe you could suggest a suitable place; I was considering a short introduction to the Types section, for example. I've also rejigged the running order into something more logical and tagged "noise problems with digital cameras" for expansion, as I'd guess that's likely to be a popular reason for people visiting the page. --mikaul^talk 22:18, 24 June 2009 (UTC)

As the originator of this issue, I don't have any strong opinion; I just think that it should be said somehow that "noise" in this context doesn't mean what a non-technical person would understand. My mention of the origin in radio static was simply to make the technical use more comprehensible; if you step out of the box a reasonable question is "what on earth do specks on the picture have to do with making a noise? Does the rattling make the picture go funny?" Even something like "Image noise (a technical term which has nothing to do with ordinary noise that can be heard)..." would be OK. You'll notice that I tend to use simple language in the introductory section, assuming that the article will be read by the non-technical, as an encylopaedia to be used by anyone rather than a technical treatise. Again, just my opinion, I don't want to push it. Certainly simple language shouldn't mean an oversimplified presentation (Feynman is a good guide). Pol098 (talk) 11:46, 24 June 2009 (UTC)

That's reasonable, but I don't see any reason to describe an image as "visual". Links to scanner and digital camera in the same sentence are enough for anyone confused as to the kind of image in question. As indicated above, I do agree terminology of a highly specialised nature should be clarified (referring to the mention of ideal photon detectors) and I've made a point of mentioning that the article discusses non-audible noise, at the end of the section. I have to say it's abundantly clear from the off that this is a technical issue referring to visual phenomena; personally I see no harm in briefly affirming that but I'd be surprised if it wasn't flagged by another editor somewhere down the line. --mikaul^talk 22:18, 24 June 2009 (UTC)

Limited scope

Hi, this is an interesting article but I feel that it has a rather limited scope in terms of image noise. Is noise really limited to only "pixel noise", i.e., independent effects on each individual pixel (or equivalent for analog film). What about

• Geometric distortion caused by lens effects
• Motion blur caused by object or camera motions during the exposure
• Blurring caused by a lack of precise focus at all depth in the scene
• Saturation and bleeding effects in digital cameras
• Spatial aliasing in digital cameras
• Jpeg coding artifacts in most digital cameras
• Shading effects from the $cos^{4}$ law
• Dirt on the lens or scratches on the film, dead pixels

I understand that some of these may not fit the "noise is a random phenomena" paradigm, but they are still usually considered as noise based on the "noise is an unwanted distortion of the signal" paradigm. There are clearly cases when the above effects can be used as artistic effects in the production of an image, but normally they are unwanted and modern cameras are packed with electronics to compensated or reduce their effects, and a more expensive camera typically produce less of these effects than a cheaper one (leading to the "noise is what to get if you don't pay enough" paradigm :-). My point is: should this article only discuss "pixel noise" or can it be given a broader scope? --KYN (talk) 09:58, 25 July 2009 (UTC)

I'd say that affects of distortion and blur and shading are not usually considered to be noise; on the other hand, trying to correct blur or shading amplifies any noise that is present, so it could be relevant. Aliasing is also not noise, but could be mentioned as another type of unwanted fluctuation. JPEG artifacts are a type of quantization noise, so could fit pretty well. Dirt and scratches are a type of noise, and often treated as such. As usual, the key is to find reliable sources that connect these things to the topic of noise. Dicklyon (talk) 15:04, 25 July 2009 (UTC)

That sounds about right, esp regarding increased noise resulting from post-processing, re-compression, etc. The section on noise specific to digital cameras would be the right place for it. Definitely find a couple of good sources first though. --mikaul^talk 15:11, 25 July 2009 (UTC)

I'm thinking about an article that discusses these and possibly other types of distortions, simply because it is relevant to understand that these effects exist, what their causes are, and how to reduce them (before or after the image is produced by the camera) in order to work with digital or film based images. It could be an overview rather than detailed description. Is this article (image noise) the right place to do that or should I start a new one? --KYN (talk) 09:17, 26 July 2009 (UTC)

Again, that depends on sources. I think they may not all fit in this article, and there are already articles on image distortion and motion blur. You'd need to find a topic that includes them all, as supported by sources. Dicklyon (talk) 15:21, 26 July 2009 (UTC)

You could look at expanding Image quality. I'd concentrate on fixing the current red links there before adding new ones; I'd also be looking to radically improve the unencyclopedic, conversational tone there. "How-to" type articles are not an asset, in fact they're a big problem, especially when they're very poorly sourced like that one. Good luck. --mikaul^talk 05:53, 27 July 2009 (UTC)

That might be a good place for it. I fixed the redlinks, but it still needs a lot of work. Dicklyon (talk) 06:09, 27 July 2009 (UTC)

OK, image quality seems to be a good beginning of what I talked about. Still, there are (at least) three articles that describe similar topics:

• Image noise is about random and independent effects on each pixel
• Image distortion is about the geometric effects
• Image quality is an overview of a range of distortion that can happen to an image

Is there some logic behind the apparently random distinction of related phenomena into different articles? For example: if distortion in terms of noise is what causes low quality, then why are these three concepts given three separate articles? --KYN (talk) 12:57, 27 July 2009 (UTC)

If you want to impose some logical order, I'd say make image quality be a WP:Summary style article, dispatching to the more specific topics like image noise and image distortion. That's pretty close to the way it is already, but it probably didn't develop with a lot of logic in mind. I'm unclear on what you refer to as "distortion in terms of noise". Dicklyon (talk) 20:46, 27 July 2009 (UTC)

What I mean is that image noise (in the way it is described in this article) can be a type of distortion of an image in the sense of "the ... unwanted ... alteration of the original shape" defined here, if we by shape mean the intensity function of the image. I understand that noise often is defined as a random phenomena and, as such, can be considered as distinct from distortion in general, but I havn't seen this distinction described in the literature a clear manner. If it exists, it deserves to be mentioned somewhere in these articles. --KYN (talk) 10:52, 29 July 2009 (UTC)

It sounds to me like you're making a synthesis, or interpretation of the definition, or of the word "shape", that goes beyond what is supportable in sources as applied to images (distortion of a waveform shape in sound is not at all the kind of shape distortion that we talk about as optical distortion). I don't know that we need a source to clarify the difference, or the relationship, but if you find such a source either way, we could use it. What you're more likely to find is something like a "distortion measure" that will measure noise, but won't have anything to do with optical (geometric) distortion; different use of the same term. Dicklyon (talk) 22:53, 29 July 2009 (UTC)

In general, image distorion can be defined as the deviation from the ideal image that we want to have and the actal image that we have. This can be measured e.g. in terms of a plain sum of squared differences between the two images (given that both are available) but other more human perception based measured are also discussed in the literature. For examples, see

• Wang & Bovik, A Universal Image Quality Index, IEEE SIGNAL PROCESSING LETTERS, VOL. 9, NO. 3, MARCH 2002
• E.R. Davies, Image distortions produced by mean, median and mode filters, Vision, Image and Signal Processing, IEE Proceedings - Volume 146, Issue 5, Oct. 1999 Page(s):279 - 285
• Heeger & Teo, A MODEL OF PERCEPTUAL IMAGE FIDELITY, ICIP 1995

In these examples, the deviation that causes the distortion can be anything from noise, geometric distortion, to artefacts from coding or filtering operations. --KYN (talk) 08:57, 30 July 2009 (UTC)

External links modified

Hello fellow Wikipedians,

I have just modified one external link on Image noise. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:

• Added archive https://web.archive.org/web/20080405214934/http://www.dpreview.com/articles/compactcamerahighiso/ to http://www.dpreview.com/articles/compactcamerahighiso/

When you have finished reviewing my changes, please set the checked parameter below to true or failed to let others know (documentation at {{Sourcecheck}}).

This message was posted before February 2018. After February 2018, "External links modified" talk page sections are no longer generated or monitored by InternetArchiveBot. No special action is required regarding these talk page notices, other than regular verification using the archive tool instructions below. Editors have permission to delete these "External links modified" talk page sections if they want to de-clutter talk pages, but see the RfC before doing mass systematic removals. This message is updated dynamically through the template {{source check}} (last update: 18 January 2022).

• If you have discovered URLs which were erroneously considered dead by the bot, you can report them with this tool.
• If you found an error with any archives or the URLs themselves, you can fix them with this tool.

Cheers.—InternetArchiveBot (Report bug) 16:31, 16 July 2016 (UTC)

A Commons file used on this page or its Wikidata item has been nominated for deletion

The following Wikimedia Commons file used on this page or its Wikidata item has been nominated for deletion:

• Noisy Smithsonian Castle.jpg

Participate in the deletion discussion at the nomination page. —Community Tech bot (talk) 00:37, 11 January 2022 (UTC)