Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction

Started 3 months ago | Discussions
bclaff Forum Pro • Posts: 10,490
Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction
35

So, bear with me, this chart is a bit dense ...

Best viewed "original size"

1) The red dot at 7 on the x-axis is ISO 400 and it's below the blue dotted line which immediately appears to be Noise Reduction (NR).

2) The red dots at 5 and 6 on the x-axis, ISO 100 and ISO 200, have a slope greater than 1; definitely NR at ISO 100 and perhaps at ISO 200.

3) The squares are 2-Dimensional Fourier Transforms (2D FTs) at 100% from the center of black frames. They should be fairly random.
They are shown from top to bottom, left to right, is ISO order.
So the 1st and 3rd, ISO 100 and ISO 400, show very clear NR.
While ISO 200 shows only a little and ISO 800 practically none.

Based on my experience with quite a few cameras I estimate the the NR gives ISO 100 and ISO 400 about 2/3 stop improvement in Photographic Dynamic Range (PDR).

NR always comes with loss of detail; whether that loss is visible is hard to say.

-- hide signature --

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

Canon EOS R5
If you believe there are incorrect tags, please send us this post using our feedback form.
Jonathan0007
Jonathan0007 Regular Member • Posts: 157
Re: Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction

bclaff wrote:

So, bear with me, this chart is a bit dense ...

Best viewed "original size"

1) The red dot at 7 on the x-axis is ISO 400 and it's below the blue dotted line which immediately appears to be Noise Reduction (NR).

2) The red dots at 5 and 6 on the x-axis, ISO 100 and ISO 200, have a slope greater than 1; definitely NR at ISO 100 and perhaps at ISO 200.

3) The squares are 2-Dimensional Fourier Transforms (2D FTs) at 100% from the center of black frames. They should be fairly random.
They are shown from top to bottom, left to right, is ISO order.
So the 1st and 3rd, ISO 100 and ISO 400, show very clear NR.
While ISO 200 shows only a little and ISO 800 practically none.

Based on my experience with quite a few cameras I estimate the the NR gives ISO 100 and ISO 400 about 2/3 stop improvement in Photographic Dynamic Range (PDR).

NR always comes with loss of detail; whether that loss is visible is hard to say.

Bill, first I wanted to thank you for the time you are putting into this testing. Greatly appreciated. A few questions:

1. How much detail do you believe is lost as a result of this NR? Would you consider it significant or negligible?

2. Do other high resolution cameras do this as well (e.g. Z7, A7riv)?

3. You said this impacts PDR. How significant is this impact?

Sorry for all the questions and thanks again for your time!

OP bclaff Forum Pro • Posts: 10,490
Re: Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction
5

Jonathan0007 wrote:

...

Bill, first I wanted to thank you for the time you are putting into this testing. Greatly appreciated.

You are welcome.

A few questions:

1. How much detail do you believe is lost as a result of this NR? Would you consider it significant or negligible?

It's hard to say since there's no way to see the raw data before the NR.
(That said ... I just had an idea ... but don't hold your breath!)

2. Do other high resolution cameras do this as well (e.g. Z7, A7riv)?

No.

3. You said this impacts PDR. How significant is this impact?

My estimate is that it improves PDR by about 2/3 stop.

Sorry for all the questions and thanks again for your time!

No problem.

-- hide signature --

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

Currantos Senior Member • Posts: 1,274
More interpretation from someone in the know?

So since you are probably more qualified to speak on this then the remainder of the people about to reply with wild baseless speculations, myself included, as is the habit of DPR, could you digest and interpret that for us before all kinds of "crazy" conclusions are drawn.

Fair to say that some "artificial" calculation is done to make noise appear less, and therefore gain some numerical value in DR rather than truly capture details? And by artificial I mean not directly tied to the signal coming at the sensor, rather more so arbitrary calculation on the available data?

Why does this remind me of the time when a car company after receiving a poor grade from Consumer Review for a particular directed bumper crash chose to 'improve' things by putting a special padding behind that exact spot only to improve the testing results and nothing to do with crash worthiness of the car. Consumer Review tested the car, get "better result", then disassembled the bumper, saw the padding and removed it and tested again  with similar conclusion published and the whole charade explained to the public. The company noticed.

Currantos Senior Member • Posts: 1,274
LOL
4

3. You said this impacts PDR. How significant is this impact?

My estimate is that it improves PDR by about 2/3 stop.

LOL

Just enough to become "king of the jungle" and outperform all other sensors. Hahah, crafty Canon engineers, who would have thought.

Put a whole other spin on the phrase "Canon caught up on DR", doesn't it now?

stevevuoso Regular Member • Posts: 125
Question on in camera noise reduction
1

Is this noise reduction different than noise reduction you apply in LR or C1? The reason I ask is whenever I apply noise reduction, it just kind of smears the image a bit to remove noise, but I don’t ever recall seeing better dynamic range from it.

dtibi Junior Member • Posts: 32
Re: Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction

Thank you for the investigation and for the measurements! NR looks a bit wierd in "raw" at ISO 100. As I know Canon raws are kind of lossless 12-14 bit JPEG files. Maybe they made a bit lossy to make smaller files but it looks like NR? Just an ideea to think about.

stevevuoso Regular Member • Posts: 125
Re: #1
2

I don’t know how much detail is lost, but in Tony Northrups video comparing the R5 to the Sony A7r4 the detail at ISO 100 was comparable, even when resizing the R5 files to match the Sony’s resolution. So that’s good news.

Pillar Lee New Member • Posts: 13
Re: Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction

bclaff wrote:

So, bear with me, this chart is a bit dense ...

Best viewed "original size"

1) The red dot at 7 on the x-axis is ISO 400 and it's below the blue dotted line which immediately appears to be Noise Reduction (NR).

2) The red dots at 5 and 6 on the x-axis, ISO 100 and ISO 200, have a slope greater than 1; definitely NR at ISO 100 and perhaps at ISO 200.

3) The squares are 2-Dimensional Fourier Transforms (2D FTs) at 100% from the center of black frames. They should be fairly random.
They are shown from top to bottom, left to right, is ISO order.
So the 1st and 3rd, ISO 100 and ISO 400, show very clear NR.
While ISO 200 shows only a little and ISO 800 practically none.

Based on my experience with quite a few cameras I estimate the the NR gives ISO 100 and ISO 400 about 2/3 stop improvement in Photographic Dynamic Range (PDR).

NR always comes with loss of detail; whether that loss is visible is hard to say.

Bill, did the Lumix S1R you previously tested perform similarly in ISO800 and below? thx.

 Pillar Lee's gear list:Pillar Lee's gear list
Canon PowerShot S70 Olympus C-5050 Zoom Panasonic Lumix DMC-FZ150 Olympus E-510 Panasonic Lumix DMC-GF1 +6 more
OP bclaff Forum Pro • Posts: 10,490
Re: More interpretation from someone in the know?
2

Currantos wrote:

...

Fair to say that some "artificial" calculation is done to make noise appear less, and therefore gain some numerical value in DR rather than truly capture details? And by artificial I mean not directly tied to the signal coming at the sensor, rather more so arbitrary calculation on the available data?

...

Typical Noise Reduction (NR) inspects nearby pixel values and may adjust the target pixel values based on that inspection.
This smooths out the data and softens edges so that detail may disappear.

Technical note, this consultation of nearby values sets up a mathematical correlation between a pixel and it's neighbors which is why the 2-Dimensional Fourier Transform (2D FT) exposes the process.

-- hide signature --

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

OP bclaff Forum Pro • Posts: 10,490
Re: Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction
2

Pillar Lee wrote:

bclaff wrote:

So, bear with me, this chart is a bit dense ...

Best viewed "original size"

1) The red dot at 7 on the x-axis is ISO 400 and it's below the blue dotted line which immediately appears to be Noise Reduction (NR).

2) The red dots at 5 and 6 on the x-axis, ISO 100 and ISO 200, have a slope greater than 1; definitely NR at ISO 100 and perhaps at ISO 200.

3) The squares are 2-Dimensional Fourier Transforms (2D FTs) at 100% from the center of black frames. They should be fairly random.
They are shown from top to bottom, left to right, is ISO order.
So the 1st and 3rd, ISO 100 and ISO 400, show very clear NR.
While ISO 200 shows only a little and ISO 800 practically none.

Based on my experience with quite a few cameras I estimate the the NR gives ISO 100 and ISO 400 about 2/3 stop improvement in Photographic Dynamic Range (PDR).

NR always comes with loss of detail; whether that loss is visible is hard to say.

Bill, did the Lumix S1R you previously tested perform similarly in ISO800 and below? thx.

The Panasonic Lumix S1-R 2D FT looks like this:

It's hard to quantify but I'd judge the NR to be about as strong but if you look closely the Panasonic pattern is more square whereas the Canon is more round; not better, just different.

-- hide signature --

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

Hoka Hey
Hoka Hey Senior Member • Posts: 2,050
Re: Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction
1

Bill,

Since the R5 is basically ISO invariant, would it be fair to compare an ISO 400 image to ISO an 800 image that has had exposure reduced 1 stop in post? Would that be a fair comparison to see if there is visible loss of detail due to NR at ISO 400?

-- hide signature --

Best,
Joe

OP bclaff Forum Pro • Posts: 10,490
Re: Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction
2

Hoka Hey wrote:

Bill,

Since the R5 is basically ISO invariant, would it be fair to compare an ISO 400 image to ISO an 800 image that has had exposure reduced 1 stop in post? Would that be a fair comparison to see if there is visible loss of detail due to NR at ISO 400?

No; I don't think so.

-- hide signature --

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

Jonathan0007
Jonathan0007 Regular Member • Posts: 157
Re: Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction

bclaff wrote:

Jonathan0007 wrote:

...

Bill, first I wanted to thank you for the time you are putting into this testing. Greatly appreciated.

You are welcome.

A few questions:

1. How much detail do you believe is lost as a result of this NR? Would you consider it significant or negligible?

It's hard to say since there's no way to see the raw data before the NR.
(That said ... I just had an idea ... but don't hold your breath!)

2. Do other high resolution cameras do this as well (e.g. Z7, A7riv)?

No.

3. You said this impacts PDR. How significant is this impact?

My estimate is that it improves PDR by about 2/3 stop.

Sorry for all the questions and thanks again for your time!

No problem.

Thank you for your response, Bill!

Kenneth Almquist Regular Member • Posts: 119
Re: Question on in camera noise reduction
1

stevevuoso wrote:

Is this noise reduction different than noise reduction you apply in LR or C1? The reason I ask is whenever I apply noise reduction, it just kind of smears the image a bit to remove noise, but I don’t ever recall seeing better dynamic range from it.

It's basically the same.  Noise reduction reduces noise, but (as you note) it also smears the image.  It won't recover fine detail that has been obliterated by noise.

The difference is that programs like LR and C1 generally allow you to control the amount of noise reduction, whereas the R5 runs the noise reduction algorithm in camera before writing the "raw" image, and it seems there is no way to control or disable the noise reduction.

 Kenneth Almquist's gear list:Kenneth Almquist's gear list
Nikon D7200 Fujifilm X-H1 Nikon AF-S DX Nikkor 35mm F1.8G Nikon AF-S DX Nikkor 18-300mm F3.5-6.3G ED VR Fujifilm 16-55mm F2.8R LM WR +2 more
John Sheehy Forum Pro • Posts: 23,819
Re: LOL
4

Currantos wrote:

  1. You said this impacts PDR. How significant is this impact?

My estimate is that it improves PDR by about 2/3 stop.

LOL

Just enough to become "king of the jungle" and outperform all other sensors. Hahah, crafty Canon engineers, who would have thought.

Put a whole other spin on the phrase "Canon caught up on DR", doesn't it now?

Let's wait to see visual comparisons. There really should be DR resolution tests. We can look at the fine details in the DPR studio comparison tool compared to other cameras, if DPR gets around to including the R5 in the ISO Invariance or Exposure Latitude tools.

Lots of sensors get better ratings than they deserve in other ways; Sonys tend to have a lot of junk in their post-gain noise, for example; fine horizontal banding or Morse-code type effects that makes the RAWs fail to deliver what P2P or DxO promises about deep shadows.

Mika Y.
Mika Y. Senior Member • Posts: 1,505
Re: Question on in camera noise reduction

Kenneth Almquist wrote:

stevevuoso wrote:

Is this noise reduction different than noise reduction you apply in LR or C1? The reason I ask is whenever I apply noise reduction, it just kind of smears the image a bit to remove noise, but I don’t ever recall seeing better dynamic range from it.

It's basically the same. Noise reduction reduces noise, but (as you note) it also smears the image. It won't recover fine detail that has been obliterated by noise.

The difference is that programs like LR and C1 generally allow you to control the amount of noise reduction, whereas the R5 runs the noise reduction algorithm in camera before writing the "raw" image, and it seems there is no way to control or disable the noise reduction.

And for some specific purposes such as stacking a large number of exposures for astrophotography, it's particularly important to perform the noise reduction for the resulting stacked and calibrated image rather than for the individual exposures.

Now admittedly that's a somewhat marginal use-case, but I'd still like to have the maximum user-level control for that too.

 Mika Y.'s gear list:Mika Y.'s gear list
Canon EOS 40D Canon EOS 5D Mark III Canon EOS Rebel T4i Canon EOS 5D Mark IV Canon EF 50mm F1.4 USM +14 more
stoffer Regular Member • Posts: 275
Re: LOL

John Sheehy wrote:

Currantos wrote:

  1. You said this impacts PDR. How significant is this impact?

My estimate is that it improves PDR by about 2/3 stop.

LOL

Just enough to become "king of the jungle" and outperform all other sensors. Hahah, crafty Canon engineers, who would have thought.

Put a whole other spin on the phrase "Canon caught up on DR", doesn't it now?

Let's wait to see visual comparisons. There really should be DR resolution tests. We can look at the fine details in the DPR studio comparison tool compared to other cameras, if DPR gets around to including the R5 in the ISO Invariance or Exposure Latitude tools.

Lots of sensors get better ratings than they deserve in other ways; Sonys tend to have a lot of junk in their post-gain noise, for example; fine horizontal banding or Morse-code type effects that makes the RAWs fail to deliver what P2P or DxO promises about deep shadows.

Exactly. In short I would preserve judgement until we can do a proper eye test on comparable RAW files.

-- hide signature --

Best regards, Stoffer
May the Light be with you!

 stoffer's gear list:stoffer's gear list
Sony a7R III Sony FE 35mm F1.4 Sony FE 24mm F1.4 GM Apple iPhone X +1 more
davidthefat
davidthefat New Member • Posts: 5
Re: Some Technical Details on the Canon EOS R5 Low ISO Noise Reduction
3

bclaff wrote:

So, bear with me, this chart is a bit dense ...

Best viewed "original size"

1) The red dot at 7 on the x-axis is ISO 400 and it's below the blue dotted line which immediately appears to be Noise Reduction (NR).

2) The red dots at 5 and 6 on the x-axis, ISO 100 and ISO 200, have a slope greater than 1; definitely NR at ISO 100 and perhaps at ISO 200.

3) The squares are 2-Dimensional Fourier Transforms (2D FTs) at 100% from the center of black frames. They should be fairly random.
They are shown from top to bottom, left to right, is ISO order.
So the 1st and 3rd, ISO 100 and ISO 400, show very clear NR.
While ISO 200 shows only a little and ISO 800 practically none.

Based on my experience with quite a few cameras I estimate the the NR gives ISO 100 and ISO 400 about 2/3 stop improvement in Photographic Dynamic Range (PDR).

NR always comes with loss of detail; whether that loss is visible is hard to say.

Hey Bill, thanks for all the analysis and explanation.

My question would be, in context of the dual pixel raw, that does the noise reduction only show up in the primary subframe or does it show up on the auxiliary subframe. There's a post on Raw Digger that looks at the highlight recovery in 5D Mk IV with the dual pixel raw. They concluded that the main raw frame is generated in an additive fashion (Pixel A + Pixel B) instead of an averaging which would allow the highlight to artificially clip.

I want to see if you take that secondary raw data and do a fourier analysis on it that you see different results. If you do, I wonder if the NR is done by somehow averaging the two subframes in the dual pixel raw for the low ISOs and reverts to the regular additive scheme for the higher ISOs. Because for the NR to turn off at 800 ISO while on the high gain circuit makes me think it's not an NR on the hardware level, but software (I'd think if it's HW, it would cut off as it switches to the high gain circuit, or stay persistent for every ISO)

 davidthefat's gear list:davidthefat's gear list
Leica M Typ 240 Panasonic Lumix DC-G9 Fujifilm GFX 50R
stevevuoso Regular Member • Posts: 125
Re: Question on in camera noise reduction

Kenneth Almquist wrote:

stevevuoso wrote:

Is this noise reduction different than noise reduction you apply in LR or C1? The reason I ask is whenever I apply noise reduction, it just kind of smears the image a bit to remove noise, but I don’t ever recall seeing better dynamic range from it.

It's basically the same. Noise reduction reduces noise, but (as you note) it also smears the image. It won't recover fine detail that has been obliterated by noise.

The difference is that programs like LR and C1 generally allow you to control the amount of noise reduction, whereas the R5 runs the noise reduction algorithm in camera before writing the "raw" image, and it seems there is no way to control or disable the noise reduction.

I guess I still don't see how that  creates more dynamic range. The noise was still there when the image was captured. I would think the dynamic range of an image is determined once the raw is captured.

Keyboard shortcuts:
FForum MMy threads