Where did this noise come from?

[litoj]

Hi,

recently, I took an 8 minute shot with my X-T2 on ISO 200, MS.


extracted jpg from fuji RAF with still very visible noise

I didn't notice it until editing the photo, but there is a lot of noise that, first of all, I didn't expect to be there at minimum ISO at all, let alone to be so sharp and pronounced.

Second thing - I have no clue how to remove it without desaturating the car lights on the bridge. I use darktable but neither hot pixel nor denoise of the two finest levels were sufficient to remove the sharp noise signifficantly.

RAF file

Do you have any suggestions to get rid of the noise or ideas of the probable cause of this?

Cheers,
Josef

 

[DNBush]

Opened in Adobe ACR and checked the denoise box. Raised shadows a touch and pulled blacks back down. Then used Topaz Photo AI to do a second pass of denoise and sharpen.

Morris

We got charts, we got graphs…. and then, we have a person who used some state of the art editing tools, and some common sense, to make a decent image from the raw file. Bravo.

Rand

Thank you Rand

Morris

That was rather refreshing. I mean, sure, there's certainly truth and value in all the details that have been discussed here but really, isn't it the results that matter? I'm far more likely really to go your route Morris.

 

[Dem Bell]

OK. So is it then the case that looking at the Read Noise graph/data alone, your statement about read noise being greater at low ISOs and less at higher ISOs is not accurate

It is 100% accurate, as I've demonstrated in this thread and many others.

but, that graph taken alone does not portray the ISO/noise relationship accurately. One needs to dive into shadow improvement to get the full picture. Am I correct?

What relationship is that? It's a fact that ISO isn't a source of noise. Given that, what is the "relationship" you're referencing between "ISO/noise"?

The graph shows the relationship between Read Noise (Y-axis) and ISO (X-axis) such that, with the exception of the dip that occurs at ISO 800, noise increases as ISO increases.

That's because the Y axis in the first graph is a function of "Digital Numbers" that scale with ISO - for the same amount of light, DN will be higher at higher ISO. Wrong units to show "the decrease".

First, "Read Noise" is not an issue here at all. Second, the statement "read noise being greater at low ISOs and less at higher ISOs" is true in absolute units when people talk about "input referred read noise" which is measured as the number of photoelectrons that would produce this level of signal. So what Bill said is factually correct but was not supported by the right graph and is not relevant for your case anyway. Probably best to move on.

Edit: I see people came to this conclusion already

 

[DNBush]

OK. So is it then the case that looking at the Read Noise graph/data alone, your statement about read noise being greater at low ISOs and less at higher ISOs is not accurate

It is 100% accurate, as I've demonstrated in this thread and many others.

but, that graph taken alone does not portray the ISO/noise relationship accurately. One needs to dive into shadow improvement to get the full picture. Am I correct?

What relationship is that? It's a fact that ISO isn't a source of noise. Given that, what is the "relationship" you're referencing between "ISO/noise"?

The graph shows the relationship between Read Noise (Y-axis) and ISO (X-axis) such that, with the exception of the dip that occurs at ISO 800, noise increases as ISO increases.

That's because the Y axis in the first graph is a function of "Digital Numbers" that scale with ISO - for the same amount of light, DN will be higher at higher ISO. Wrong units to show "the decrease".

So it seems that what I'm taking away from all this is that the Read Noise graph on Photons to Photos is misleading at best and wrong at worst.

First, "Read Noise" is not an issue here at all. Second, the statement "read noise being greater at low ISOs and less at higher ISOs" is true in absolute units when people talk about "input referred read noise" which is measured as the number of photoelectrons that would produce this level of signal. So what Bill said is factually correct but was not supported by the right graph and is not relevant for your case anyway. Probably best to move on.

Yeah, I have. Especially after Morris' reply.

Edit: I see people came to this conclusion already

 

[Bill Ferris]

So it seems that what I'm taking away from all this is that the Read Noise graph on Photons to Photos is misleading at best and wrong at worst.

That would be woefully incorrect.

See this post from over the weekend and the "Input-referred Read Noise" for documented evidence that read noise historically gets lower as ISO increases and that dual-gain sensors edit the same low read noise that works be generated at ISO 6400+ at ISOs as low as the 400 to 800 range.

Loading…

www.dpreview.com

 

[DNBush]

So it seems that what I'm taking away from all this is that the Read Noise graph on Photons to Photos is misleading at best and wrong at worst.

That would be woefully incorrect.

See this post from over the weekend and the "Input-referred Read Noise" for documented evidence that read noise historically gets lower as ISO increases and that dual-gain sensors edit the same low read noise that works be generated at ISO 6400+ at ISOs as low as the 400 to 800 range.

https://www.dpreview.com/forums/post/68343633

Bill,

Just looking at the Read Noise graph on Photons to Photos, the graph shows that read noise, the Y-axis, increases as the ISO, the X-axis, increases. Is that not what it's showing?

The line trends upward as you increase the ISO. I don't know how it could be seen any other way.

I'm only asking about what that specific graph shows.

So if then as you said, the chart to use to see read noise performance vs ISO is the input-referred read noise chart, what is the purpose of the Read Noise chart that I've been looking at?

 

[Bill Ferris]

So it seems that what I'm taking away from all this is that the Read Noise graph on Photons to Photos is misleading at best and wrong at worst.

That would be woefully incorrect.

See this post from over the weekend and the "Input-referred Read Noise" for documented evidence that read noise historically gets lower as ISO increases and that dual-gain sensors edit the same low read noise that works be generated at ISO 6400+ at ISOs as low as the 400 to 800 range.

https://www.dpreview.com/forums/post/68343633

Bill,

Just looking at the Read Noise graph on Photons to Photos, the graph shows that read noise, the Y-axis, increases as the ISO, the X-axis, increases. Is that not what it's showing?

An upward slope in the Read Noise in DNs plot can still reflect reducing or constant input-referred read noise. DNs are post-gain values. As ISO increases, the analog signal is amplified before being digitized. So even if the actual noise at the sensor (in electrons) stays the same, the amplified noise appears larger in DNs.

The input-referred read noise chart plots read noise vs ISO. The plot for the 1DX clearly shows read noise decreasing as ISO increases. This was standard performance for digital cameras prior to dual-gain sensor tech coming widely into use.

The input-referred read noise plot for a dual-gain sensor camera typically shows read noise abruptly improving (getting lower) at some ISO in the 400 to 800 range and staying low up through ISO 6400+. This is the benefit of a dual-gain sensor the same low read noise that was historically available at high ISOs is now available at much lower ISOs.

The shadow improvement chart plots photographic dynamic range in the shadows (the effects of read noise) vs ISO. The plots for the same cameras will show the same performance described above in a different way.

The 1DX profile shows PDR in the shadows gradually improving as ISO increases. This is the result of gradually reducing read noise as ISO increases. The reduction of read noise as ISO increases can be inferred from this plot.

The plot for a dual-gain camera will show an abrupt improvement to DR in the shadows at a given ISO and then constant DR up through much higher ISOs. Again, the constant read noise level can be inferred from the constant PDR in the shadows.

The read noise in DNs chart isn't your best option for understanding the read noise produced by a camera at specific ISOs (input-referred read noise), of the effects of read noise on the camera's performance (PDR shadow improvement). The DN charts don't contradict the performance documented in the other two. They also don't translate as directly if used as an indicator of read noise.

The photographic dynamic range (PDR) charts plots PDR vs ISO. This chart indicates the combined effects of shot noise and read noise. Shot noise is naturally occurring randomness in light and is the predominant type of noise we see in photos. It's determined by the total light energy used to make a photo, which is why this plot shows a steady decline - along with a bump in performance at a camera's dual-gain point - as ISO increases.

ISO is the setting we use to manage image lightness. As exposure decreases, ISO is typically increased to compensate. This the correlation between increasing ISO and reduction in the cameras dynamic range. It's the increased shot noise that's the culprit. ISO is thec canary in the coal mine. It's an indicator that a camera is become starved for light.

The line trends upward as you increase the ISO. I don't know how it could be seen any other way.

I'm only asking about what that specific graph shows.

So if then as you said, the chart to use to see read noise performance vs ISO is the input-referred read noise chart, what is the purpose of the Read Noise chart that I've been looking at?

The "Read noise in DNs" chart isn't the one to use to see read noise output or the effects of read noise on DR in the shadows. The "Input-referred Read Noise" chart directly plots read noise. The "PDR Shadow Improvement" chart can be used to determine how invariant a camera is and to infer read noise performance. These are the charts to use a references.

--
Bill Ferris Photography
Flagstaff, AZ

 

[DNBush]

So it seems that what I'm taking away from all this is that the Read Noise graph on Photons to Photos is misleading at best and wrong at worst.

That would be woefully incorrect.

See this post from over the weekend and the "Input-referred Read Noise" for documented evidence that read noise historically gets lower as ISO increases and that dual-gain sensors edit the same low read noise that works be generated at ISO 6400+ at ISOs as low as the 400 to 800 range.

https://www.dpreview.com/forums/post/68343633

Bill,

Just looking at the Read Noise graph on Photons to Photos, the graph shows that read noise, the Y-axis, increases as the ISO, the X-axis, increases. Is that not what it's showing?

An upward slope in the Read Noise in DNs plot can still reflect reducing or constant input-referred read noise. DNs are post-gain values. As ISO increases, the analog signal is amplified before being digitized. So even if the actual noise at the sensor (in electrons) stays the same, the amplified noise appears larger in DNs.

The input-referred read noise chart plots read noise vs ISO. The plot for the 1DX clearly shows read noise decreasing as ISO increases. This was standard performance for digital cameras prior to dual-gain sensor tech coming widely into use.

The input-referred read noise plot for a dual-gain sensor camera typically shows read noise abruptly improving (getting lower) at some ISO in the 400 to 800 range and staying low up through ISO 6400+. This is the benefit of a dual-gain sensor the same low read noise that was historically available at high ISOs is now available at much lower ISOs.

The shadow improvement chart plots photographic dynamic range in the shadows (the effects of read noise) vs ISO. The plots for the same cameras will show the same performance described above in a different way.

The 1DX profile shows PDR in the shadows gradually improving as ISO increases. This is the result of gradually reducing read noise as ISO increases. The reduction of read noise as ISO increases can be inferred from this plot.

The plot for a dual-gain camera will show an abrupt improvement to DR in the shadows at a given ISO and then constant DR up through much higher ISOs. Again, the constant read noise level can be inferred from the constant PDR in the shadows.

The read noise in DNs chart isn't your best option for understanding the read noise produced by a camera at specific ISOs (input-referred read noise), of the effects of read noise on the camera's performance (PDR shadow improvement). The DN charts don't contradict the performance documented in the other two. They also don't translate as directly if used as an indicator of read noise.

The photographic dynamic range (PDR) charts plots PDR vs ISO. This chart indicates the combined effects of shot noise and read noise. Shot noise is naturally occurring randomness in light and is the predominant type of noise we see in photos. It's determined by the total light energy used to make a photo, which is why this plot shows a steady decline - along with a bump in performance at a camera's dual-gain point - as ISO increases.

ISO is the setting we use to manage image lightness. As exposure decreases, ISO is typically increased to compensate. This the correlation between increasing ISO and reduction in the cameras dynamic range. It's the increased shot noise that's the culprit. ISO is thec canary in the coal mine. It's an indicator that a camera is become starved for light.

The line trends upward as you increase the ISO. I don't know how it could be seen any other way.

I'm only asking about what that specific graph shows.

So if then as you said, the chart to use to see read noise performance vs ISO is the input-referred read noise chart, what is the purpose of the Read Noise chart that I've been looking at?

The "Read noise in DNs" chart isn't the one to use to see read noise output or the effects of read noise on DR in the shadows. The "Input-referred Read Noise" chart directly plots read noise. The "PDR Shadow Improvement" chart can be used to determine how invariant a camera is and to infer read noise performance. These are the charts to use a references.

Again, all good information for sure. And again, the message seems to be clear, namely, that the chart that I keep referencing, is not sufficient (accurate?) for getting the picture of how noise and ISO relate in a modern sensor.

But...

The reason I keep harping on this is because that chart has appeared in countless threads here on DPReview and other forums where it has been used to depict the ISO/Noise relationship that it so clearly and simply shows. Indeed, in at least one camera review (which I can't recall now), the DPReview reviewer used it.

Now, if that chart, taken by itself, must be qualified by all the other information you have presented, then it seems odd to me that the folks at Photons to Photos haven't included qualifying information to explain what you've been, patiently and thoroughly I might add, explaining here. Because without that other information, a lot of people have been using that one chart incorrectly (including at least one DPReview reviewer).

That's all I'm getting at.

 

[Bill Ferris]

So it seems that what I'm taking away from all this is that the Read Noise graph on Photons to Photos is misleading at best and wrong at worst.

That would be woefully incorrect.

See this post from over the weekend and the "Input-referred Read Noise" for documented evidence that read noise historically gets lower as ISO increases and that dual-gain sensors edit the same low read noise that works be generated at ISO 6400+ at ISOs as low as the 400 to 800 range.

https://www.dpreview.com/forums/post/68343633

Bill,

Just looking at the Read Noise graph on Photons to Photos, the graph shows that read noise, the Y-axis, increases as the ISO, the X-axis, increases. Is that not what it's showing?

An upward slope in the Read Noise in DNs plot can still reflect reducing or constant input-referred read noise. DNs are post-gain values. As ISO increases, the analog signal is amplified before being digitized. So even if the actual noise at the sensor (in electrons) stays the same, the amplified noise appears larger in DNs.

The input-referred read noise chart plots read noise vs ISO. The plot for the 1DX clearly shows read noise decreasing as ISO increases. This was standard performance for digital cameras prior to dual-gain sensor tech coming widely into use.

The input-referred read noise plot for a dual-gain sensor camera typically shows read noise abruptly improving (getting lower) at some ISO in the 400 to 800 range and staying low up through ISO 6400+. This is the benefit of a dual-gain sensor the same low read noise that was historically available at high ISOs is now available at much lower ISOs.

The shadow improvement chart plots photographic dynamic range in the shadows (the effects of read noise) vs ISO. The plots for the same cameras will show the same performance described above in a different way.

The 1DX profile shows PDR in the shadows gradually improving as ISO increases. This is the result of gradually reducing read noise as ISO increases. The reduction of read noise as ISO increases can be inferred from this plot.

The plot for a dual-gain camera will show an abrupt improvement to DR in the shadows at a given ISO and then constant DR up through much higher ISOs. Again, the constant read noise level can be inferred from the constant PDR in the shadows.

The read noise in DNs chart isn't your best option for understanding the read noise produced by a camera at specific ISOs (input-referred read noise), of the effects of read noise on the camera's performance (PDR shadow improvement). The DN charts don't contradict the performance documented in the other two. They also don't translate as directly if used as an indicator of read noise.

The photographic dynamic range (PDR) charts plots PDR vs ISO. This chart indicates the combined effects of shot noise and read noise. Shot noise is naturally occurring randomness in light and is the predominant type of noise we see in photos. It's determined by the total light energy used to make a photo, which is why this plot shows a steady decline - along with a bump in performance at a camera's dual-gain point - as ISO increases.

ISO is the setting we use to manage image lightness. As exposure decreases, ISO is typically increased to compensate. This the correlation between increasing ISO and reduction in the cameras dynamic range. It's the increased shot noise that's the culprit. ISO is thec canary in the coal mine. It's an indicator that a camera is become starved for light.

The line trends upward as you increase the ISO. I don't know how it could be seen any other way.

I'm only asking about what that specific graph shows.

So if then as you said, the chart to use to see read noise performance vs ISO is the input-referred read noise chart, what is the purpose of the Read Noise chart that I've been looking at?

The "Read noise in DNs" chart isn't the one to use to see read noise output or the effects of read noise on DR in the shadows. The "Input-referred Read Noise" chart directly plots read noise. The "PDR Shadow Improvement" chart can be used to determine how invariant a camera is and to infer read noise performance. These are the charts to use a references.

Again, all good information for sure. And again, the message seems to be clear, namely, that the chart that I keep referencing, is not sufficient (accurate?) for getting the picture of how noise and ISO relate in a modern sensor.

But...

The reason I keep harping on this is because that chart has appeared in countless threads here on DPReview and other forums where it has been used to depict the ISO/Noise relationship that it so clearly and simply shows. Indeed, in at least one camera review (which I can't recall now), the DPReview reviewer used it.

Now, if that chart, taken by itself, must be qualified by all the other information you have presented, then it seems odd to me that the folks at Photons to Photos haven't included qualifying information to explain what you've been, patiently and thoroughly I might add, explaining here. Because without that other information, a lot of people have been using that one chart incorrectly (including at least one DPReview reviewer).

That's all I'm getting at.

I suspect Bill Claff would disagree that he's not provided contextual articles explaining what "Read Noise in DNs" shows out how to interpret it. That said, it's not a chart that gets used or referenced much in the DPR forums. Most references I see are to either the "Photographic Dynamic Range" or "PDR Shadow Improvement" charts. The "Input-referred Read Noise" chart sometimes gets referenced. "Read Noise in DNs"...not so much.

 

[flektogon]

Hi,

recently, I took an 8 minute shot with my X-T2 on ISO 200, MS.


extracted jpg from fuji RAF with still very visible noise

I didn't notice it until editing the photo, but there is a lot of noise that, first of all, I didn't expect to be there at minimum ISO at all, let alone to be so sharp and pronounced.

Second thing - I have no clue how to remove it without desaturating the car lights on the bridge. I use darktable but neither hot pixel nor denoise of the two finest levels were sufficient to remove the sharp noise signifficantly.

RAF file

Do you have any suggestions to get rid of the noise or ideas of the probable cause of this?

Cheers,
Josef

Josef,

Either set on the long shutter shooting noise reduction (you may have an "Auto" setting there like the Pentax cameras have), or apply some "smart" noise reduction in PP.

Šťastný focení (anglicky to zní lépe ).

--
Regards,
Peter

 

[Erik Baumgartner]

So it seems that what I'm taking away from all this is that the Read Noise graph on Photons to Photos is misleading at best and wrong at worst.

That would be woefully incorrect.

See this post from over the weekend and the "Input-referred Read Noise" for documented evidence that read noise historically gets lower as ISO increases and that dual-gain sensors edit the same low read noise that works be generated at ISO 6400+ at ISOs as low as the 400 to 800 range.

https://www.dpreview.com/forums/post/68343633

Bill,

Just looking at the Read Noise graph on Photons to Photos, the graph shows that read noise, the Y-axis, increases as the ISO, the X-axis, increases. Is that not what it's showing?

An upward slope in the Read Noise in DNs plot can still reflect reducing or constant input-referred read noise. DNs are post-gain values. As ISO increases, the analog signal is amplified before being digitized. So even if the actual noise at the sensor (in electrons) stays the same, the amplified noise appears larger in DNs.

The input-referred read noise chart plots read noise vs ISO. The plot for the 1DX clearly shows read noise decreasing as ISO increases. This was standard performance for digital cameras prior to dual-gain sensor tech coming widely into use.

The input-referred read noise plot for a dual-gain sensor camera typically shows read noise abruptly improving (getting lower) at some ISO in the 400 to 800 range and staying low up through ISO 6400+. This is the benefit of a dual-gain sensor the same low read noise that was historically available at high ISOs is now available at much lower ISOs.

The shadow improvement chart plots photographic dynamic range in the shadows (the effects of read noise) vs ISO. The plots for the same cameras will show the same performance described above in a different way.

The 1DX profile shows PDR in the shadows gradually improving as ISO increases. This is the result of gradually reducing read noise as ISO increases. The reduction of read noise as ISO increases can be inferred from this plot.

The plot for a dual-gain camera will show an abrupt improvement to DR in the shadows at a given ISO and then constant DR up through much higher ISOs. Again, the constant read noise level can be inferred from the constant PDR in the shadows.

The read noise in DNs chart isn't your best option for understanding the read noise produced by a camera at specific ISOs (input-referred read noise), of the effects of read noise on the camera's performance (PDR shadow improvement). The DN charts don't contradict the performance documented in the other two. They also don't translate as directly if used as an indicator of read noise.

The photographic dynamic range (PDR) charts plots PDR vs ISO. This chart indicates the combined effects of shot noise and read noise. Shot noise is naturally occurring randomness in light and is the predominant type of noise we see in photos. It's determined by the total light energy used to make a photo, which is why this plot shows a steady decline - along with a bump in performance at a camera's dual-gain point - as ISO increases.

ISO is the setting we use to manage image lightness. As exposure decreases, ISO is typically increased to compensate. This the correlation between increasing ISO and reduction in the cameras dynamic range. It's the increased shot noise that's the culprit. ISO is thec canary in the coal mine. It's an indicator that a camera is become starved for light.

The line trends upward as you increase the ISO. I don't know how it could be seen any other way.

I'm only asking about what that specific graph shows.

So if then as you said, the chart to use to see read noise performance vs ISO is the input-referred read noise chart, what is the purpose of the Read Noise chart that I've been looking at?

The "Read noise in DNs" chart isn't the one to use to see read noise output or the effects of read noise on DR in the shadows. The "Input-referred Read Noise" chart directly plots read noise. The "PDR Shadow Improvement" chart can be used to determine how invariant a camera is and to infer read noise performance. These are the charts to use a references.

Again, all good information for sure. And again, the message seems to be clear, namely, that the chart that I keep referencing, is not sufficient (accurate?) for getting the picture of how noise and ISO relate in a modern sensor.

But...

The reason I keep harping on this is because that chart has appeared in countless threads here on DPReview and other forums where it has been used to depict the ISO/Noise relationship that it so clearly and simply shows. Indeed, in at least one camera review (which I can't recall now), the DPReview reviewer used it.

Now, if that chart, taken by itself, must be qualified by all the other information you have presented, then it seems odd to me that the folks at Photons to Photos haven't included qualifying information to explain what you've been, patiently and thoroughly I might add, explaining here. Because without that other information, a lot of people have been using that one chart incorrectly (including at least one DPReview reviewer).

That's all I'm getting at.

As Bill said, the read noise increasing DNs probably isn’t the best chart for understanding how ISO/image noise works. The most important problem with that chart is that it only looks at read noise which is really, really insignificant relative to shot noise or, in rare cases like this particular image, long exposure hot pixel noise.

The most important thing to remember is that the primary noise that you will encounter in most images is shot noise (aka photon noise) - which is directly related to exposure (or rather, lack thereof), not read noise.

There is no situation that I can think of where using a “less noisy” ISO value beyond the dual gain threshold will result in less image noise than increasing exposure at a “noisier” ISO value instead. Always expose as much as possible in low light - the slowest possible SS (without incurring motion blur issues, and the widest possible aperture (without incurring DOF/sharpness issues). For jpeg shooters, the correct ISO will be what it needs to be to achieve the desired image lightness.

So, for the least noise/greatest possible dynamic range…

Always (for jpeg shooters, anyway), use the most possible exposure (scene illumination + SS + Aperture), and whatever ISO that will deliver the desired image lightness and/or record the brightest important highlights just below clipping. Really very simple. Unless you are shooting at base ISO on a tripod, Auto-ISO really makes a lot of sense for most shooters (me included). If you choose to set the ISO manually, it should really be considered after maximizing exposure (unless it’s fixed at base ISO).

There are situations where choosing a “cleaner” ISO value above the dual gain threshold makes sense (or rather, not using one below it), but typically that’s for RAW shooters in very low light and or high dynamic range situations who will be post processing (and best covered in a different thread).