How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

Started 3 months ago | Discussions
Currantos Regular Member • Posts: 377
Re: Correction

Mika Y. wrote:

Currantos wrote:

Since most of the noise is in the light you capture, this proposed method won't show you most of the noise of a typical photograph.. It will only show you the small portion of noise that is added by the camera.

Not sure what you mean "noise in the light"?

Noise is only a property of the imaging system. It is not a property of the original object being imaged. There can be variations in value of the object being imaged but that is not noise.

Look at the black sky. Do you see bright noise, the white pixels? These are not noise, they are stars. That is what the astroguys are trying to take pictures of. Now take a look at their pictures. You will see both stars, the original "variations in the object" and the superimposed noise of the photo equipment.

To summarize. ALL of the noise in the photograph is from the camera(lens/sensor/etc). None is from the "light" that comes in.

Unless I'm mistaken, he means shot noise: https://en.m.wikipedia.org/wiki/Shot_noise

Correct, there is shot noise, not sure how that would play into the whole equation, sounds even more complicated now.

MrBrightSide
OP MrBrightSide Contributing Member • Posts: 789
The Real World Is An Unsympathetic Place

That's a deeper question. So which is more useful to the working photographer? I produce photos within an ecosystem in which I and my clients use Macintosh computers and Adobe software almost exclusively. There's a very predictable workflow in other words.

So which would be more helpful for me in fine-tuning my exposure decisions? To know the effects on noise before going through the imaging chain or to know what happens after all the various programs have done their math?

And why jpegs? As we continue to converge video and stills, jpegs are becoming obsolete because they can't carry transparency information. Would the results be different with  TIFFs and PNGs? Do they have different characteristics?

bclaff wrote:

My suggestion is that if you shoot raw or raw+jpeg then examining the raw files (with something like RawDigger) is the real "scientific" approach, otherwise the tone curve hopelessly muddles the results.

-- hide signature --

Sit!

57even Forum Pro • Posts: 14,140
Re: How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

Sorry, I was distracted by your use of the word 'structure' which normally means patterns like banding or the wormy artefacts in Xtrans/Adobe raws.

Looking at DxOMark, the normalised ratings for each camera at a nominal ISO25600 are 18.5dB and 21.3dB, which is nearly 3dB or a full stop. 25600 is max ISO on the mkii, and tails off a lot. Still looks like a big difference, but mainly in the red channel. Other channels don't look that different.

tbcass wrote:

57even wrote:

I'm still waiting for you to explain what characteristics you are referring to.

Chroma vs luminance? Banding?

I have compared thousands of images from different cameras over the years, and as long as the images are not heavily processed (for instance if we use the default Adobe profiles) the noise structure is basically similar at a pixel level.

Here's a comparison between 2 sensors that according to DXO are within 1DB of each other but the noise characteristics are completely different.

-- hide signature --

"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away." Antoine de Saint-Exupery

 57even's gear list:57even's gear list
Fujifilm X-Pro2
Currantos Regular Member • Posts: 377
Re: Correction

photonut2008 wrote:

Currantos wrote:

To summarize. ALL of the noise in the photograph is from the camera(lens/sensor/etc). None is from the "light" that comes in.

That's not correct. Photons arrive at the sensor randomly, and the lower the exposure the more deviation there is to that randomness. See: https://en.wikipedia.org/wiki/Image_noise#Shot_noise

yes, that's correct

J A C S
J A C S Forum Pro • Posts: 15,686
Re: Correction

Currantos wrote:

There is perfectly clear notion what noise is.

There was at least one long thread in the PST forum about that and the bottom line was that it was not clear.

It is the erroneous signal present in the image that was not present in the original object introduced by the imaging device.

What is erroneous signal? To define that, you need to define signal (without errors) in the first place. Shot noise is a part of the signal. Is it noise then? Is an under or over exposed image noisy? Are geometric distortions noise? What is "the original object introduced by the imaging device"? The object is typically in front of that device.

It has two components, systematic and random.

This has been discovered at the time the normal distribution was discovered and statistics was born. All measurements were found to have "noise", ie. variations from ground truth state.

They were? All of them? Who did that?

If you would like to have better visualization of what noise is take a picture of pure uniform gray paper and look closely at it. The pixels that stand out, i.e. that are darker and brighter than the overall background. The darker/brighter pixels were not present in the original paper and are the 'noise' of the photographic system. (yes, there is 'noise' in the paper, it is not perfectly uniform, but that noise is an order of magnitude below the noise of the photo equipment and is therefore disregarded here).

Hope that helps.

There are other definitions of noise depending on different industries and fields, like acoustics, etc but all boil down to the same thing and are studied with same statistical formulas.

Which ones?

MrBrightSide
OP MrBrightSide Contributing Member • Posts: 789
Re: How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

Seen this discussed before;  isn’t this why we have dithering?

, and in fact, too little can actually be a bad thing in smooth gradients, which will posterize into contour banding with too little noise.

-- hide signature --

Sit!

Video-vs-photo New Member • Posts: 18
Re: How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

So randomness could cause missing signal which is not exactly noise :). But distort original image.
To be able to get enough light you need to have enough light so the sensor will be saturated to produce useful picture. And if sensor is not saturated enough you will see pattern like grain.
But then we apply amplification to get more use of small portion of the light that we captured and this produce much more signal which does not exist in the image - noise.
So no way to cheat with missing light, we just could add more light. Or we can get bigger glass to catch more light. 
We could make our equipment better so at the end we will have small amount of added noise.
In this case shooting with cap on and shooting of gray uniform object will provide you of all possible information.

FingerPainter Veteran Member • Posts: 8,482
Re: How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

Video-vs-photo wrote:

So randomness could cause missing signal which is not exactly noise :). But distort original image.

No. Noise is variation in pixel values. How visible noise is depends on the Signal to Noise Ratio (SNR) - the higher the SNR, the less noisy an image looks.

Most of the noise in a typical digital photograph is variation in pixel values due to the variation that was present in the light itself even before the light was captured by the camera. When light is created, the timing of the release of individual photons, and the colour of each photon are both random. The camera adds additional variation. In most parts of a typical photo, the amount of noise added by the camera is small relative to the amount of noise in the captured light.

To be able to get enough light you need to have enough light so the sensor will be saturated to produce useful picture.

No. Saturation is the maximum capacity of the sensor. If you reach saturation on all pixels you will not have a usable picture. you will have a white picture.

And if sensor is not saturated enough you will see pattern like grain.

All photos that have not reached saturation on all pixels will have noise because of the (variation) noise naturally occurring in light.

But then we apply amplification to get more use of small portion of the light that we captured and this produce much more signal which does not exist in the image - noise.

No. When you apply amplification (which is only one of several ways of implementing an ISO increase), you don't get more light, and you don't degrade the SNR. Either the SNR remains the same or it is increased. Which of these two occurs depends on whether the camera adds any noise after the gain stage. Most cameras add some noise after the gain stage to so on most cameras that use amplification to implement an ISO increase, increasing the ISO reduces the noisiness (increases the SNR). On many modern digital cameras, the amount of noise added after the gain stage is small, so the improvement in SNR is not readily noticeable.

So no way to cheat with missing light, we just could add more light. Or we can get bigger glass to catch more light.

The effect of adding more light is to add both more signal and more noise. Since the noise in light is the square root of the signal, as you increase the light, the SNR goes up, and the image, despite having more noise, looks less noisy.

We could make our equipment better so at the end we will have small amount of added noise.

Yes. There are at least two ways to make the equipment better. On is to increase the portion of photons falling on the sensor that is actually captured (increase the quantum efficiency). The other is to reduce the amount of noise added by the camera.

In this case shooting with cap on and shooting of gray uniform object will provide you of all possible information.

No. Shooting with the cap on results in no signal and no noise in light. It will give you information about camera-added noise, which is usually a small subset of the noise in a usable image.

FingerPainter Veteran Member • Posts: 8,482
Re: Incorrection

Currantos wrote:

Since most of the noise is in the light you capture, this proposed method won't show you most of the noise of a typical photograph.. It will only show you the small portion of noise that is added by the camera.

Not sure what you mean "noise in the light"?

Shot noise, which you appear to agree does exist.

Noise is only a property of the imaging system.

There is variation in light before the imaging system. Are you tryign to draw a disticntion between variation and noise?

It is not a property of the original object being imaged.

It is a property of the light being reflected or emitted by the object begin imaged.

There can be variations in value of the object being imaged but that is not noise.

What do you call it?

Look at the black sky. Do you see bright noise, the white pixels? These are not noise, they are stars. That is what the astroguys are trying to take pictures of. Now take a look at their pictures. You will see both stars, the original "variations in the object" and the superimposed noise of the photo equipment.

Most of the noise in a typical photo is a manifestation of the variation in arrival rates and coulors of photons, not variation added by the camera.

To summarize. ALL of the noise in the photograph is from the camera(lens/sensor/etc). None is from the "light" that comes in.

Exactly incorrect. Most of the noise in a typical digital photo is due to variation present in the light even before it was captured by the camera.

FingerPainter Veteran Member • Posts: 8,482
Re: Correction

I suggest you refresh your understanding of black body radiation before you presume to correct people on the topic of noise.

sybersitizen Forum Pro • Posts: 15,214
Re: Correction
1

Currantos wrote:

Mika Y. wrote:

Currantos wrote:

Since most of the noise is in the light you capture, this proposed method won't show you most of the noise of a typical photograph.. It will only show you the small portion of noise that is added by the camera.

Not sure what you mean "noise in the light"?

Noise is only a property of the imaging system. It is not a property of the original object being imaged. There can be variations in value of the object being imaged but that is not noise.

Look at the black sky. Do you see bright noise, the white pixels? These are not noise, they are stars. That is what the astroguys are trying to take pictures of. Now take a look at their pictures. You will see both stars, the original "variations in the object" and the superimposed noise of the photo equipment.

To summarize. ALL of the noise in the photograph is from the camera(lens/sensor/etc). None is from the "light" that comes in.

I hope you have now abandoned the above position.

Unless I'm mistaken, he means shot noise: https://en.m.wikipedia.org/wiki/Shot_noise

Correct, there is shot noise, not sure how that would play into the whole equation, sounds even more complicated now.

Yes indeed, there is shot noise, or photon noise, which is always present in the form of random fluctuations in the photon stream. It's almost inconsequential when light is plentiful, but becomes increasingly important as the quantity of light reaching the sensor is reduced. That is not the camera's fault. Dark current noise and read noise are the camera's fault.

decaf14 Regular Member • Posts: 234
Re: How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

Noise is what we define it as. You have to know what you want in the image to know what you don't want.

Typically we remove "noise" by using a digital low pass filter. You could then quantify noise by examining the amount of high frequencies in an area of your photo you expect to be a single color. You could do this in MATLAB.

At a physical level, it's much easier to quantify. Typically, a transistors dark current (current when no light is shining) determines it.

I'm sure there are other metrics we use for noise that are far more clever than looking at a fourier, but I don't know them off the top.

 decaf14's gear list:decaf14's gear list
Fujifilm X-T10 Fujifilm X-T3 Fujifilm XF 35mm F1.4 R Fujifilm XF 18-135mm F3.5-5.6 R LM OIS WR Fujifilm XF 27mm F2.8 +2 more
bclaff Veteran Member • Posts: 9,240
Re: How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

decaf14 wrote:

Noise is what we define it as. ... Typically, a transistors dark current (current when no light is shining) determines it
...

Although it's in line with your first sentence; dark current is definitely not noise.
However, you can determine read noise in the fashion your propose.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

John Sheehy Forum Pro • Posts: 22,831
Re: How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

FingerPainter wrote:

No. Shooting with the cap on results in no signal and no noise in light. It will give you information about camera-added noise, which is usually a small subset of the noise in a usable image.

It is not a small subset where it really matters.

If one concentrates on the brighter tones of lower and medium ISOs, then it may seem that read noise is insignificant, but then again, in those tones sensors of the same size do not vary much.

If one starts looking at darker tones and/or higher ISOs, or situations where you need white balance, and one or two color channels are a lot weaker than another, then read noise becomes the greater factor.

It is myth that read noise has become insignificant, and that we are hitting the walls of "physics" and necessary shot noise. Read noise can improve quite a bit, and color filters on sensors are extremely inefficient, and typicall losing over 80% of the light in an exposure. Getting rid of that loss would both greatly reduce photon noise by a certain number of stops in each color channel, but also reduce any given electronic read noise relative to exposure signal by twice as many stops, without even reducing the read noise as measured in electrons.

bclaff Veteran Member • Posts: 9,240
Re: How Do I Measure Noise Using Photoshop? And What Is A Significant Change?

John Sheehy wrote:

FingerPainter wrote:

No. Shooting with the cap on results in no signal and no noise in light. It will give you information about camera-added noise, which is usually a small subset of the noise in a usable image.

It is not a small subset where it really matters.

...

Agreed.

While most of the noise in a normal photograph is shot (photon) noise, where noise is apparent the Signal to Noise Ratio (SNR) is low and those areas are not dominated by shot noise but rather a combination of read and shot noise.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

decaf14 Regular Member • Posts: 234
Off-topic Nerdy Dark Current Discussion

uhhh I think it's noise. Perhaps you meant it's not that significant? photons/time is much more important, but that doesn't change that dark current is an agreed upon metric for engineers to quantify phototransistor performance easily. Please see the two resources below.

https://www.photometrics.com/learn/imaging-topics/dark-current

"Electrons are created over time that are independent of the light falling on the detector. These electrons are captured by the CCD’s potential wells and counted as signal. Additionally, this increase in signal also carries a statistical fluctuation known as dark current noise."

If you have journal article access, check out this paper as well:

"Organic Photodiodes: The Future of Full Color Detection and Image Sensing" , Advanced Materials, 2016"

A quote: " the EQE and the dark current are the two primary parameters that influence secondary performance indicators such as the linear dynamic range (LDR) and specific detectivity (D*)."

 decaf14's gear list:decaf14's gear list
Fujifilm X-T10 Fujifilm X-T3 Fujifilm XF 35mm F1.4 R Fujifilm XF 18-135mm F3.5-5.6 R LM OIS WR Fujifilm XF 27mm F2.8 +2 more
bclaff Veteran Member • Posts: 9,240
Re: Off-topic Nerdy Dark Current Discussion
1

decaf14 wrote:

uhhh I think it's noise. Perhaps you meant it's not that significant? photons/time is much more important, but that doesn't change that dark current is an agreed upon metric for engineers to quantify phototransistor performance easily. Please see the two resources below.

https://www.photometrics.com/learn/imaging-topics/dark-current

"Electrons are created over time that are independent of the light falling on the detector. These electrons are captured by the CCD’s potential wells and counted as signal. Additionally, this increase in signal also carries a statistical fluctuation known as dark current noise."

If you have journal article access, check out this paper as well:

"Organic Photodiodes: The Future of Full Color Detection and Image Sensing" , Advanced Materials, 2016"

A quote: " the EQE and the dark current are the two primary parameters that influence secondary performance indicators such as the linear dynamic range (LDR) and specific detectivity (D*)."

That's a good resource but IMO they are careless to call that noise.

Variation implies up/down whereas dark current is an offset.

I think I could probably provide lots of cites that list noise sources and omit dark current.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

decaf14 Regular Member • Posts: 234
Re: Off-topic Nerdy Dark Current Discussion

I think that dark current still varies with time but I'm getting outside my expertise here. Please correct me if my understanding is mistaken as I'd like to learn more.

https://www.sciencedirect.com/topics/engineering/dark-current-noise

"Dark current noise is the constant current that exists when no light is incident on the photodiode. This is the reason it is called dark current. As shown in Eq. (4.2.8), this dark current is the same thing as the reverse saturation current Is because the photodiode is always reversely biased. Because of the statistical nature of the carrier generation process similar to that has been discussed for the shot noise, the dark current noise can also be treated as a white noise with the power spectral density."

 decaf14's gear list:decaf14's gear list
Fujifilm X-T10 Fujifilm X-T3 Fujifilm XF 35mm F1.4 R Fujifilm XF 18-135mm F3.5-5.6 R LM OIS WR Fujifilm XF 27mm F2.8 +2 more
Mark Scott Abeln
Mark Scott Abeln Forum Pro • Posts: 14,521
You can't always get what you want

MrBrightSide wrote:

That's a deeper question. So which is more useful to the working photographer? I produce photos within an ecosystem in which I and my clients use Macintosh computers and Adobe software almost exclusively. There's a very predictable workflow in other words.

Arguably, it is typically only amateurs on a budget who worry about noise. The professional way is to avoid the problem altogether and not worry about it. Higher end commercial photographers and cinematographers throw huge amounts of light on their subjects, and buy or rent the best gear they can afford and use the largest format that is practical, and it has been that way for over a century.

The end result of this shooting is cleanliness which makes things easier for post-processing and editing.

So which would be more helpful for me in fine-tuning my exposure decisions? To know the effects on noise before going through the imaging chain or to know what happens after all the various programs have done their math?

If you shoot under ideal conditions, those with lots of light even if you use reasonable shutter speeds and apertures, then your work is easy: do you have clean shadow detail with a good exposure? The answer here is  either yes or no, and the reasons why may not be so important.

I think that it is valuable to know a little bit about basic imaging physics: that light has quantum properties that will introduce noise in all conditions, and how using good exposure will reduce the relative amount of noise. Also, having an idea of the amount of read noise that is introduced by a sensor helps considerably, especially if you want to use an antique digital camera for whatever reason. Also realizing how sensor size may influence the results can help also, as well as the trade-offs that switching sizes will require.

Getting deeper into the details is intellectually rewarding, and can be useful, but it does require lots of specialized knowledge and critical thinking skills as well as college-level math in many cases. Some details can be misleading: for example, much published noise data of cameras is only taken from the unprocessed green raw channels, but processing does change the noise characteristics which will vary with color and white balance.

And why jpegs? As we continue to converge video and stills, jpegs are becoming obsolete because they can't carry transparency information. Would the results be different with TIFFs and PNGs? Do they have different characteristics?

JPEGs have been adequate until lately. While it works well enough for final display, casual users these days are routinely doing strong edits on JPEGs (for example, "Instagram-style filters"}, and the bit depth is not quite good enough; also, higher bit depth, wider range monitors are becoming widespread. The HEIC/HEIF file format is gaining popularity as a replacement for JPEG: it supports more bits and transparency as well, and it is a video format as well (it is a container format, so it is flexible). As major vendors are supporting HEIC, it appears as though it has a good chance to be successful.

Bandwidth is still an issue: higher common data rates are being offset by higher resolution, and many web pages are littered with large numbers of images, while people are gaining an appreciation for high quality photography, so there is a need for good quality, high compression image formats.

TIFFs are usually huge, as they encode data which is not necessarily visible: particularly, high frequency detail and chroma information which typically isn't needed. Also, the bulk of TIFF data is false, interpolated data, at least with images that come from Bayer sensors. PNGs have some of the same faults, but its lossless compression works usually extremely well with line art and graphics.

 Mark Scott Abeln's gear list:Mark Scott Abeln's gear list
Nikon D200 Nikon D7000 Nikon D750 Nikon AF Nikkor 50mm f/1.8D Nikon AF-S DX Nikkor 35mm F1.8G +2 more
bclaff Veteran Member • Posts: 9,240
Re: Off-topic Nerdy Dark Current Discussion

decaf14 wrote:

I think that dark current still varies with time but I'm getting outside my expertise here. Please correct me if my understanding is mistaken as I'd like to learn more.

https://www.sciencedirect.com/topics/engineering/dark-current-noise

"Dark current noise is the constant current that exists when no light is incident on the photodiode. This is the reason it is called dark current. As shown in Eq. (4.2.8), this dark current is the same thing as the reverse saturation current Is because the photodiode is always reversely biased. Because of the statistical nature of the carrier generation process similar to that has been discussed for the shot noise, the dark current noise can also be treated as a white noise with the power spectral density."

Obviously there's some disagreement/inconsistency even among "professionals".

For me I distinguish between a drift or offset, like dark current; and a variation both up and down like photon noise or read noise.

-- hide signature --

Bill ( Your trusted source for independent sensor data at PhotonsToPhotos )

Keyboard shortcuts:
FForum MMy threads