DPReview.com is closing April 10th - Find out more

Canon EOS R3 Baked in Raw Noise Reduction Revisited

Started Jan 21, 2022 | Discussions
hjulenissen Senior Member • Posts: 2,442
Re: Canon EOS R3 Baked in Raw Noise Reduction Revisited

bclaff wrote:

hjulenissen wrote:

So do this camera appear to apply noise reduction irrespective of ISO setting?

Yes, that's what I said in the initial post.

Ok, I did not quite get that.

Is not the green curve (farthest back) more pronounced non-flat than the ones closer to the center?

-h

OP bclaff Forum Pro • Posts: 13,922
Re: Canon EOS R3 Baked in Raw Noise Reduction Revisited

hjulenissen wrote:

bclaff wrote:

hjulenissen wrote:

So do this camera appear to apply noise reduction irrespective of ISO setting?

Yes, that's what I said in the initial post.

Ok, I did not quite get that.

Is not the green curve (farthest back) more pronounced non-flat than the ones closer to the center?

The highest ISO setting seems to have a lot of signal processing; but that's not so unusual.

-- hide signature --

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

hjulenissen Senior Member • Posts: 2,442
Re: Canon EOS R3 Baked in Raw Noise Reduction Revisited
1

So an analysis of the psf/mtf of the max ISO vs some other ISO might give some insight into the nature of their noise reduction. Is it a per color channel blur? Median filter? Edge preserving?

One might then speculate that noise reduction at lower ISO possibly is of a similar nature, only backed off a bit in degree, kernel size etc.

EthanCam Regular Member • Posts: 111
Re: Canon EOS R3 Baked in Raw Noise Reduction Revisited
1

hjulenissen wrote:

So an analysis of the psf/mtf of the max ISO vs some other ISO might give some insight into the nature of their noise reduction. Is it a per color channel blur? Median filter? Edge preserving?

One might then speculate that noise reduction at lower ISO possibly is of a similar nature, only backed off a bit in degree, kernel size etc.

I've tried out a psf/blur kernel parametrization vs. noise for my canon r6, flat field files at different exposures: https://www.dpreview.com/forums/post/65857338

Note that on the plot, the noise (x-axis) is the observed noise, i.e. after NR, and the NR scaling (y-axis) gives how much the noise has been reduced. So you have to divide by this number to find out what the original noise value was.

It's just an hypothesis at this point, but from that particular test, for the r6, it seems the higher the noise, the weakest the NR (except for the highest ISO value, where NR probably is applied no matter what).

The camera could estimate the noise from the knowledge of the gain and the read noise (optical black pixels) , but it could also be estimated from local areas of the image for example.

Cheers, Ethan

J A C S
J A C S Forum Pro • Posts: 20,521
Re: Canon EOS R3 Baked in Raw Noise Reduction Revisited

hjulenissen wrote:

So an analysis of the psf/mtf of the max ISO vs some other ISO might give some insight into the nature of their noise reduction. Is it a per color channel blur? Median filter? Edge preserving?

I doubt that the camera would do such processing and still preserve the high fps. Also, I do not see any trace of softening of the image itself. In any case, the blur kernel at high ISO is negligeable, if that really is a convolution.

I posted something more about high ISO here:

More about high ISO: Canon EOS R Talk Forum: Digital Photography Review (dpreview.com)

One might then speculate that noise reduction at lower ISO possibly is of a similar nature, only backed off a bit in degree, kernel size etc.

Actually, it is much stronger at low ISO, and low signal.

Keyboard shortcuts:
FForum MMy threads