Dynamic range testing of the Canon EOS R3 is complete

DBounce

Canon Eos R3
May 3, 2016
500
544
I have several objective tools at my disposal to make these decisions but as I indicated earlier it's hard to quantify the amount of signal processing (noise reduction) that is being applied. Ultimately the assignment of the down triangle symbol is a judgement call and I try not us use the symbol which is why the R3 results were delayed as I double-checked.
Ok, so you are saying you are guessing. Exactly what I thought. Canon makes it pretty clear in the manual that they are not applying noise reduction.
 
Upvote 0
Ok, so you are saying you are guessing. Exactly what I thought. Canon makes it pretty clear in the manual that they are not applying noise reduction.
That's not what I'm saying. And I don't care what is said in the manual; manuals are often inaccurate and I only care what is actually observed.

You can read about 2D Fourier Transforms and Energy Spectra at PhotonsToPhotos
snip.png

This 2D FT for the Canon EOS R3 at ISO 100 clearly shows noise reduction (otherwise the visualization would be uniform noise)
snap.png
This Energy Spectrum for the Canon EOS R3 at ISO 100 confirms noise reduction. Otherwise the curves would be more level rather than dipping.

If you don't understand the science that doesn't mean it isn't true.
 
Last edited:
  • Like
Reactions: 14 users
Upvote 0

Joules

doom
CR Pro
Jul 16, 2017
1,801
2,247
Hamburg, Germany
That's not what I'm saying. And I don't care what is said in the manual; manuals are often inaccurate and I only care what is actually observed.

You can read about 2D Fourier Transforms and Energy Spectra at PhotonsToPhotos
View attachment 201633

This 2D FT for the Canon EOS R3 at ISO 100 clearly shows noise reduction (otherwise the visualization would be uniform noise)
View attachment 201634
This Energy Spectrum for the Canon EOS R3 at ISO 100 confirms noise reduction. Otherwise the curves would be more level rather than dipping.

If you don't understand the science that doesn't mean it isn't true.
Thank you very much for what you do and the nice explanations on your site! That statement at the end is something surprisingly many people seem to ignore.

Just a heads up, your link there is broken due to missing a : near the beginning.

For those curious in why the noise reduction is NOT just a guess, here is the specific article that's being referred to I think:

 
  • Like
Reactions: 2 users
Upvote 0
Jul 21, 2010
31,094
12,857
Saying “no they don’t” doesn’t make it true. Page 152 in the Advanced User Guide.
Lol. Read this, from p152:

A1D4765F-D12A-4297-8887-4372A2935D15.jpeg

They call CRAW a RAW image. CRAW uses lossy compression, where in the AUG is that mentioned? If data are eliminated from the image file, how is it still a RAW image? The data are manipulated, but Canon still calls it RAW.

Well, Canon doesn’t say CRAW uses lossy compression in the manual, so it must not be true. :rolleyes:

Dig, dig, dig.
 
  • Like
Reactions: 1 users
Upvote 0
Thank you very much for what you do and the nice explanations on your site! ...
You are welcome.
Just a heads up, your link there is broken due to missing a : near the beginning.
Thanks. Edited and fixed.
For those curious in why the noise reduction is NOT just a guess, here is the specific article that's being referred to I think:

And also perhaps An Introduction to Energy Spectra for Sensor Analysis
 
  • Like
Reactions: 5 users
Upvote 0

Keith_Reeder

I really don't mind offending trolls.
Feb 8, 2014
960
477
63
Blyth, NE England
Meh...

DBounce is making a perfectly valid - if arguably poorly articulated - point.

The phrase "noise reduction", as it pertains to photography, has a very specific and unambiguous popular meaning, and Canon sensors do not apply noise reduction in that sense.

Even Bill Claff's own page (written by Emil Martinec, whose opinion on the subject I trust all day long - he is a scientist, and actually created the AMaZE noise reduction algorithm) has this to say about Canon sensors:

Correlated double sampling (sometimes referred to as "on-chip NR"): In preparation for a new exposure, the electrons liberated by photon capture from the previous exposure must be emptied from the sensels; each sensel must be reset to a state of zero exposure. If this process is not completely effective, there may be residual electrons present in the sensel prior to the next exposure, and these electrons will throw off the count of photons from this new exposure. The fluctuations in the number of electrons present after zeroing out the sensel constitutes the reset noise.

An advantage of CMOS sensors is their ability to read the state of the pixels non-destructively -- the state of the sensor (its count of electrons) can be read off without affecting that state. This feature can be used to eliminate the reset noise as follows. After the sensor is reset from the previous exposure, the state of the sensor is read, giving a count of the residual electrons of the reset noise while leaving those electrons in place; then, after the exposure, the sensor is read again, the result being the residual electrons plus those added via photon capture during the exposure. Taking the difference of the two readings gives the photo-electron count of the exposure while subtracting off the reset noise. This before-and-after reading of the sensels is known as correlated double sampling (CDS). On CCD sensors, the electrons in a CCD sensel must be extracted from the sensel and transferred to the edge of the sensor in order to do the readout; CDS can be performed in this per-column readout circuitry rather than separately within each sensel. The double read adds time to the readout when done serially at the edge of the sensor rather than in parallel at each sensel. CDS is a feature of most current CMOS DSLR sensors, see for instance the product literature by Canon and Sony.

(I've fixed the link to the Canon paper).

In short, this isn't noise reduction in the common use of the phrase, but noise filtration prior to the RAW file being written.

Not the same thing at all - a fact he (or rather Emil) explicitly acknowledges:

In-camera filtering of raw data: All the above noise reduction methods involve manipulations at the level of individual pixels; never is a pixel value compared to or mixed with neighboring pixel values in attempting to eliminate noise. These methods are thus quite different from what is conventionally called noise reduction in image processing.

It's the pejorative, lazy and inaccurate terminology that rankles.

And yes, it's still RAW data - noise being excluded from the data written to file means that all of the meaningful image data is retained. This is perfectly consistent with any reasonable definition of "RAW", such as that on Wikipedia (with my emphasis):
A camera raw image file contains minimally processed data from the image sensor of either a digital camera, a motion picture film scanner, or other image scanner

Nothing about Canon's approach falls outside of this definition. It's not semantics, it's simply not true to say that Canon files have "baked-in noise reduction" - this phrase has a specific (negative) implication, and is fundamentally inaccurate on its face.

I mention the "Canon paper" referenced (by a broken link) on Bill Claff's page.

It adds this to the equation relating to random noise:

Canon’s highly effective method for random noise suppression is called complete electronic charge transfer, or complete charge transfer technology. Canon designed the photodiode and the signal reader independently to ensure that the sensor resets the photodiodes that store electrical charges. By first transferring the residual discharge — light and noise signals — left in a photodiode to the corresponding signal reader, the Canon sensor resets the diode while reading and Light Reset switch As residual noise charges N1 and N2 are not equal, some noise will inevitably remain. N1 S N1 S N2 S = optical signal N1, N2 = noise Reset 1 Optical signal is read Reset 2 Noise read Signal & noise accumulation ( S + N1 ) – ( N2 ) = S + N The difference between fixed-pattern and random noise Previous functionality Fixed-pattern noise: appears on the same pixels even at different times of day. Example: Caused by dark current leakage, irregular converters each pixel, etc. Suppressed by noise reduction and on-chip noise reduction technology. Random noise: appears on different pixels at different times of day. Example: Flickering light, thermal noise, etc. Cannot be reduced with noise reduction and on-chip noise reduction technology, so the noise itself must be prevented. V. WHY CMOS? 19 holding the initial noise data. After the optical signal and noise data have been read together, the initial noise data is used to remove the remaining noise from the photodiode and suppress random noise, leaving a nice, clean signal.

Again, we're talking about a method which filters out noise prior to the RAW file being written, not noise reduction per se, as it's commonly meant.

It is perfectly legitimate to consider what results to be a RAW file. In the absence of any definitive definition, it is entirely reasonable to consider the file that results - being only data, and needing to to be converted into an image format - as a RAW file.

The problem with this thread isn't whether Canon does on-chip "management" of noise - that's never been in doubt (the Canon paper above is from 2008, AFAIK) - but whether there's even the slightest legitimacy in calling it "baked-in noise reduction".

There isn't, and it isn't. The phrase as it's commonly used implies that something useful has been lost, and there's simply no basis for that assumption.




OMB
 
Last edited by a moderator:
  • Like
Reactions: 3 users
Upvote 0
Jan 27, 2020
826
1,796
Yes, but this is where the inability to quantify the amount of noise reduction hampers our ability to estimate the smoothing (loss of detail).
People seem happy with the results so it's probably academic.
Perhaps the most important statement in this entire discussion...

"If people are happy with the results it is probably academic." As usual on these forums, people seem so concerned with the process and ignore the results. "How dare they cook the RAW files" is often indignantly shouted! Constant arguments about compressed or uncompressed RAW. Even over-exaggeration about the slightest differences in DR, that have been demonstrated to be undetectable by the human eye are constantly discussed.

Yes, it is good to know something about the technology and the scientific results. But if it doesn't actually effect the end result - and what the photo looks like - then that is the most important knowledge that we gain from the information.

Too many folks seem to be hung up on the nuts and bolts of the information. It is what we learn from the information that is important, in my opinion.
 
  • Like
Reactions: 3 users
Upvote 0
The phrase "noise reduction", as it pertains to photography, has a very specific and unambiguous popular meaning, and Canon sensors do not apply noise reduction in that sense.
Yes they do.
Nothing about Canon's approach falls outside of this definition. It's not semantics, it's simply not true to say that Canon files have "baked-in noise reduction" - this phrase has a specific (negative) implication, and is fundamentally inaccurate on its face.
I have clearly demonstrated that a neighboring pixel algorithm has been applied. That is baked-in noise reduction.
I mention the "Canon paper" referenced (by a broken link) on Bill Claff's page.
Do I have a broken link? Can you supply some details so I can fix it?
Again, we're talking about a method which filters out noise prior to the RAW file being written, not noise reduction per se, as it's commonly meant.
Filtering at the pixel level is one thing (like CDS or noise shaping) but consulting neighboring pixels is another.
 
Upvote 0

Keith_Reeder

I really don't mind offending trolls.
Feb 8, 2014
960
477
63
Blyth, NE England
"If people are happy with the results it is probably academic." As usual on these forums, people seem so concerned with the process and ignore the results. "How dare they cook the RAW files" is often indignantly shouted!
Yep - even when there's no actual cooking of the files in the first place.
 
Upvote 0

Keith_Reeder

I really don't mind offending trolls.
Feb 8, 2014
960
477
63
Blyth, NE England
Yes they do.
Despite your assertions, Bill, you've never proven it.

Unless your Fourier analyses are the only way, and are proven to be 100% accurate in every possible scenario, you're speculating based on the results you see.

You've pretty much admitted to that in this very thread.
 
Upvote 0
Jul 21, 2010
31,094
12,857
It's the pejorative, lazy and inaccurate terminology that rankles.

And yes, it's still RAW data - noise being excluded from the data written to file means that all of the meaningful image data is retained.
Is a lossy compressed RAW image still a RAW image? It is, according to Canon’s terminology.
 
Upvote 0
Despite your assertions, Bill, you've never proven it.

Unless your Fourier analyses are the only way, and are proven to be 100% accurate in every possible scenario, you're speculating based on the results you see.

You've pretty much admitted to that in this very thread.
The Fourier analysis is 100% reliable and is confirmed with the spectral analysis.
 
  • Like
Reactions: 1 user
Upvote 0
Jan 29, 2011
10,675
6,121
Meh...

DBounce is making a perfectly valid - if arguably poorly articulated - point.

The phrase "noise reduction", as it pertains to photography, has a very specific and unambiguous popular meaning, and Canon sensors do not apply noise reduction in that sense.

Even Bill Claff's own page (written by Emil Martinec, whose opinion on the subject I trust all day long - he is a scientist, and actually created the AMaZE noise reduction algorithm) has this to say about Canon sensors:



(I've fixed the link to the Canon paper).

In short, this isn't noise reduction in the common use of the phrase, but noise filtration prior to the RAW file being written.

Not the same thing at all - a fact he (or rather Emil) explicitly acknowledges:



It's the pejorative, lazy and inaccurate terminology that rankles.

And yes, it's still RAW data - noise being excluded from the data written to file means that all of the meaningful image data is retained. This is perfectly consistent with any reasonable definition of "RAW", such as that on Wikipedia (with my emphasis):


Nothing about Canon's approach falls outside of this definition. It's not semantics, it's simply not true to say that Canon files have "baked-in noise reduction" - this phrase has a specific (negative) implication, and is fundamentally inaccurate on its face.

I mention the "Canon paper" referenced (by a broken link) on Bill Claff's page.

It adds this to the equation relating to random noise:



Again, we're talking about a method which filters out noise prior to the RAW file being written, not noise reduction per se, as it's commonly meant.

It is perfectly legitimate to consider what results to be a RAW file. In the absence of any definitive definition, it is entirely reasonable to consider the file that results - being only data, and needing to to be converted into an image format - as a RAW file.

The problem with this thread isn't whether Canon does on-chip "management" of noise - that's never been in doubt (the Canon paper above is from 2008, AFAIK) - but whether there's even the slightest legitimacy in calling it "baked-in noise reduction".

There isn't, and it isn't. The phrase as it's commonly used implies that something useful has been lost, and there's simply no basis for that assumption.




OMB
I am not a scientist. I consider this discussion academic and interesting. I am very happy with the output of my Canon cameras.

What I don’t get from your comment is an explaination for the dips in the relevant graphs that Bill Claff points to as evidence of ‘noise reduction’. Presumably, using the ‘filtering’ techniques you mention above Canon applies that to all iso levels? So why are those graphs not consistent lines or curves?

Could it be because at different points in the iso scale different processes are applied and at some of those points a noise reduction process, as it is more commonly understood and Mr Claff refers to as a 'nearest neighbor process', is in fact applied?

Using one previously described process does not preclude sometimes also using another.
 
Last edited:
Upvote 0
Is a lossy compressed RAW image still a RAW image? It is, according to Canon’s terminology.
The question is not whether the file is "raw" but rather whether signal processing consulting neighboring pixels amounting to noise reduction has been performed. Such processing correlates pixels statistically with their neighbors.
 
  • Like
Reactions: 1 users
Upvote 0
Jan 29, 2011
10,675
6,121
Saying “no they don’t” doesn’t make it true. Page 152 in the Advanced User Guide.
You said "Canon makes it pretty clear in the manual that they are not applying noise reduction."
I said "No they don't"

Here is page 152 from the Advanced User Guide in it's entirety, at no point is the term Noise Reduction mentioned. Neither is Highlight Tone Priority (HTP) which is 100% known to 'cook' the RAW file.
Screen Shot 2021-12-12 at 12.47.39.png
 
Upvote 0