DxO RAW Processing and Noise Reduction - What is going on?

#1

Hi there,

I was in Tokyo a couple of months ago, I used a Lumix G8/GH81, but forgot to turn on the AutoNoiseReduction for Long Exposure :roll_eyes:. Nevertheless, I came across a strange behaviour with DxO’s processing and Noise Reduction. Attached you will find a comparison of a photo (ISO 200, 4s) without any noise reduction at all, on the left is the DxO processed version, on the right is Capture One11. Why is the picture quality out of the box with DxO so bad in this case? Why are single pixels/dead pixels so damn pronounced?

Furthermore, DxO’s Noise Reduction, regardless what version I use (HQ or PRIME) produces white or at least brighter blemishes. It is not that much seen in CO11’s case. In addition, I can’t even remove every “dead pixel” in DxO PhotoLab, even when the slider is set to 100. With Capture One it’s not a problem.

This is all not an issue in daylight images with PRIME. I really don’t understand what is going on here.
Don’t get me wrong: I am a longtime user of DxO and I love the PRIME NR which in my opinion is far superior than that of any other program, but in cases of dark images it’s not usable when it produces results like that.

1 Like
(Platypus) #2

hmm, noise reduction needs to decide, which pixels are noise and which are images of whatever it might be - the software (let’s ignore adobe and google for the moment) has no idea of what is in the image. If NR is too aggressive, it might wipe a starry night sky clear of all the little lights that made you take the snap in the first place.

CaptureOne is (meant for and) mostly used on images taken in a studio. With clean rooms and precisely set up lights, you can assume that little bright spots are unwanted and can therefore be ironed over. DPL is meant for general use, tiny details might be important and should therefore be kept.

Different shoes for different feet!

#3

I don’t see why Noise reduction needs to decide. I am setting the values and using the sliders. OK, I see that that it might be hard to decide what is a dead pixel, but I am not able to remove all with DxO, since 100 is not enough.

In addition, that can’t be the explanation as why the dead or single pixels and the overall noise are so pronounced in DxO as compared to Capture One when there even is NO noise reduction applied at all to the image. Look at the 2nd comparison shot where I zoomed in. The white dots in the sky in the DxO example on the left are NOT stars. I can’t understand why DxO and Capture One even process unaltered photos in their RAW state that differently.

1 Like
(Platypus) #4

well, it’s the guys and gals who design the noise reduction algorithms who do that kind of thinking. I also see different approaches implemented in Lr vs DPL. Differences exist and it might rather be a matter of accepting this than understanding it…

#5

And what about the image quality without Noise Reduction? Why does the opened RAW file look that much noisier in DxO than in Capture One?

(Platypus) #6

Each photo software uses a different base processing, even if you tell the software to do nothing. Sometimes, base processing is changed between on version of a software and the next release.

(I had an early version of Capture One and loved its “analogue” colour treatment. Later releases had more digital (punchy) colour which kind of obsoleted C1 for me. To be clear, C1 is a powerful software and I love how it does some of the things. I never loved the GUI though.)

#7

Hmm, then I think DxO should definetely improve for a better base processing.

1 Like
(Svetlana G.) #8

Hello guys,

First, let me ask @wolf to reply you.

Second, could you (@MaxReebo ), please, provide us with your RAW images for the analysis (please, upload them on upload.dxo.com and let us know when ready).

Thank you
Regards,
Svetlana G.

1 Like
(Wolf) #9

Hi @MaxReebo,

Thank you for your feedback. Comparing the images without noise reduction is not so interesting, from my point of view, because the image is, in fact, noisy and what you really want is noise reduction that works well.

So I concentrate on the comparison Capture One vs PhotoLab, with noise reduction enabled and dead pixels slider at 100. And I agree that our result is not as good as it should be. From looking at your images above, I identify two issues: (1) some dead pixels remain and (2) the image in general and the sky in particular are not as clean as they should be, even where there are no dead pixels. To investigate the first, we’ll have to look at those dead pixels before any processing, directly in the raw file. The second reminds me of an issue that we discovered recently and that is related to long exposure times. Having one more example of an image that suffers from the issue will help to raise the priority of the issue. It would really help us if you could upload the raw file, as Svetlana suggested. Thank you.

Best regards,
Wolf

2 Likes
#10

Dear Svetlana, dear Wolf,

I agree that the versions without any noise reduction don’t matter that much. However, it seems odd at first that - although it is the same raw file - the image seems to need more noise reduction in DxO than in Capture One since the noise/dead pixels are more pronounced in DxO - or are at least displayed in such a way that they seem to be more pronounced. But I have to admit that the results strongly depend on the use of the lens correction features of DxO, especially the lens sharpness feature.

Nevertheless, I uploaded the raw file with “MaxReebo” as support ticket number since I was unsure what to fill in.
Thanks in advance for your support in solving this issue.

Best regards,
Andreas

(Svetlana G.) #11

Hello Andreas,

Thank you, we’ve received the image and will do the investigation.

P.S.

This is the correct step :wink:

Regards,
Svetlana G.

(Wolf) #12

Hello Andreas,

Thank you for providing the raw file. My investigations confirmed that there are two issues, one related to dead pixels, some of which are not removed as they should, and the other related to the noise in the image, which is much stronger than we expect it to be at ISO 200, probably due to the long exposure time of 4 s. As a consequence, our denoising algorithms consider part of it as details and preserve it.
Unfortunately I don’t have an immediate fix for any of the two issues. I discussed both with our manager and we try to allocate some time for working on them as soon as possible.

Best regards,
Wolf

4 Likes
#13

Hello,

I have the same issue with raws from my Samsung NX1000:
I’m using two cameras for stereo photography (mostly night photography with long exposures), where hotpixels can be very disturbing. On images from my Canon EOS 100D most hotpixels are removed with Prime-standard-setting. But on images from the NX1000 many hotpixels stay even if the dead pixel slider in Prime is on 100 %. I know that it’s difficult for the algorithm to decide, if a pixel is a hotpixel or maybe a star. And it’s good that it preserves as many details als possible. Most of the hotpixels that stay visible are not a huge problem, but there are always some in darker areas which I have to remove manually:

Best regards,
Christoph

1 Like
#14

I Downloaded your file. Those dots will not even be recognised when you would print 2m wide. Do you print bigger?

(Svetlana G.) #15

Hello @muelli,

Could you, please, upload the RAW images to upload.dxo.com under your Forum name instead of the “support ticket number” and we can analyze them and add to the issue?

Thank you,
Regards,
Svetlana G.

(Josh Koerner) #16

As a recent owner of Photo Lab 2 elite I am seeking a manual for the elite features. Or, at least, instruction how to tweak PRIME noise reduction parameters to get a custom look and not simply the default.

(Svetlana G.) #17

Hello @jdkoerner,

Have you seen this tutorial - http://dxo.tuto.free.fr/ ?

  • I think we have a lot of guys on the Forum who can give some useful advice :slight_smile:

Regards,
Svetlana G.

(Alec Kinnear) #18

I’ve posted quite a bit on the right settings for Prime Noise Reduction. A good default is 12 not 40. For Fast Noise Reduction, 20 is a good starting point.

The DxO defaults for the smart features (Smart Lighting, ClearView are two others examples) are way too high. They are set like this to show off the feature and make it clear what it does.

You can permanently fix the defaults by altering your default preset to include the settings you prefer.

1 Like
(Josh Koerner) #19

Great page!:heart_eyes:

1 Like
(Peter) #20

A old thread but it covers the theory behind the sliders. At least i tried to reveal that.

i am lazy and have PRIME always on, with m43 default would be ok.

i suspect that if you test on different settings i advise to use Dfine2 as a kind of Tiff after noise reduction denoiser to see how much dfine2 is redoing again.
i found Dfine2 and essential HQ or prime default and Dfine2 almost the same as 70 on prime.
(night shots on 6400iso and 3200iso was my test images)

Its mostly personal perception of what you find too smutched, polished out detail, or to noisy.

sharpening/denoise is always a kind of “feel” of end goal. a grainy black and white high contrast and detail but noisy or a “plastic” colorfull clean image. and all in between.

Maybe its a idea to place some image here and we get a run with it trying to get the best you want. See which settings are popping up. :slight_smile:

1 Like