Improper Noise Reduction for Nikon D3100 RAW

  • 1
  • Problem
  • Updated 5 years ago
RAW files from Nikon D3100 are far noisier than they should be.
Attached link is a cropped comparison of an image processed with Nikon's ViewNX 2 and ACR 7.4. In both cases, the image was simply opened, and then exported with default settings.



As you can see in the example, images processed in ViewNX2 have a bit more contrast, a slightly darker appearance, and very significantly less luminance noise. (The eye looks almost hazy in the ACR version). I can, to a degree, adjust these in ACR as well, but the interpretation in NX2 is far superior by default. Would it be possible to address this?

Attached below, the RAW file used.
https://dl.dropboxusercontent.com/u/1...
Thanks for any insight, etc!
Photo of James Monaco

James Monaco

  • 31 Posts
  • 10 Reply Likes
  • Disappointed

Posted 5 years ago

  • 1
Photo of Steve Sprengel

Steve Sprengel, Champion

  • 2653 Posts
  • 340 Reply Likes
If your camera is capable and your lighting is sufficient to allow the camera to use a low ISO then there shouldn't be much if any noise, so Adobe is conservative and doesn't, by default, remove any noise and image detail. In other words, unlike Nikon, Adobe's default Luminance NR setting is zero.

If you're not shooting in near perfect conditions and/or not using very good equipment or for whatever reason are using a non-low-ISO, then you may need some noise-reduction applied and you are free to adjust the Lum NR slider higher than zero.

If you are rarely if ever shooting in optimal conditions and have little control over the lighting, then you may want to set your own default Lum NR setting that is higher than zero for every photo.

If you find yourself sometimes needing no NR and sometimes needing some, and sometimes needing more, then make a few presets with different amounts of NR and use different presets for different shooting conditions.
Photo of James Monaco

James Monaco

  • 31 Posts
  • 10 Reply Likes
Well, thanks for the reply, I do appreciate the attempt, but it's not really a reply to my report. (And yep, I do have a preset set up with lum and exp adjustments). I feel like I must say things wrong on here, because it seems that whenever I point out something that seems not to be working well, the responses I get are mostly along the lines of *handwave* "That isn't the result you were looking for." I seem to keep failing to get across the aspect of the problem that I intend.

I do understand ISO, though I appreciate that you couldn't have realized that off-hand and thus reviewed the concept just in case that was the misunderstanding.

But my report of a problem was that ACR does not handle noise reduction very well, not that my photos are noisy.

To illustrate the point, I started out with a side-by-side comparison of ViewNX versus ACR. I think you'd be inclined to agree that regardless of whatever automatic process is happening when the image is loaded into ViewNX, the result is that no additional NR attempt is needed--noise is reduced entirely satisfactorily. The result from ACR, on the other hand, when pulled in directly, is horrendous, and does require NR. (And yes, implicitly, this is only the case in examples where there is noise...if the conditions or kit were perfect, then there'd be no NR discussion).

And to that end is where my point was driving--I did make it the title of my post, and, for good measure, the first sentence as well, but it remains unclear. My point is that ACR does not handle the reduction of noise from RAW D3100 files "properly." By which I mean "well," or "in the manner that the camera manufacturer feels it should be handled."

Since further example is necessary, I'll point to the image below. Now, admittedly, you may all simply be better at ACR than me, but I've gone in and adjusted the Lum NR and Detail for a good several minutes until I got the closest match to the immediate result obtained from ViewNX, Look closely at the noise, and it should be clear that the grain in ACR is still more "defined" and "chunkier." It doesn't seem to fade into the image as gracefully as the Nikon alternative. It desn't get that "smaller" look to it and that dithered, minimal feel.


I'm ready to confess some of you might just have more skill than me, if you can obtain a better result than this, but the point, I think, remains; The camera manufacturer has some voodoo up their sleeves such that when NR is needed, it instantly, automatically, and with no fiddling whatsoever, removes noise very admirably. ACR seems to have no such magic with this image and performs a generalized sort of effect that, with much time and fiddling, does definitely improve the image, but even still cannot match the method used by ViewNX.

*That*, I'd say, is the improper handling, or more precisely, the "not as the manufacturer would recommend it" handling, that I am referring to. Sorry for any confusion.
Photo of Chris Cox

Chris Cox

  • 20280 Posts
  • 815 Reply Likes
what immediately jumps out from that example is that you have more sharpening in ACR, and NX is much blurrier. That could exaggerate the remaining noise, or could be that NX is blurring the image more to reduce the noise.
Photo of James Monaco

James Monaco

  • 31 Posts
  • 10 Reply Likes
Ah, replied in the wrong box. Long story short, you've effectively solved the issue as well as I could want it, Chris :) Longer reply below!
Photo of Steve Sprengel

Steve Sprengel, Champion

  • 2653 Posts
  • 340 Reply Likes
In your first post you said that ACR's default NR is inferior. ACR's default is 0 NR.

Now you're saying that ACR's NR cannot be made to work as well as Nikon's does out of the box. That is at least something to talk about.

Adobe and Nikon are separate companies so neither has access to what the other's raw rendering algorithms which would include demosaicing and noise-reduction and sharpening.

Adobe has only changed their rendering algorithms once, recently, with Process 2010, and that has been a few years ago, with the major improvements detail rendering, sharpening, and noise-reduction.

Nikon can optimize their algorithms for each new camera sensor and those sensor-specific treatments likely make their way into the Nikon-licenced products such as ViewNX.

By comparison, Adobe is doing the same thing with every camera's sensor data--at least for the very common Bayer-array sensors which your Nikon would have--not including Fuji XTrans and Sigma Foveon--so for a particular camera sensor Adobe may be doing things better or worse than the manufacturer.

I am at a disadvantage because I don't have a Nikon camera and don't have ViewNX so have to rely on the small crop you've provided to judge what it does.

In your side-by-side, looking at the crease of the lower eyelid and the definition along upper part of the yellow bill against the background, my opinion is that the Adobe rendering has more fine detail structure than Nikon's.

From what you've said, the Nikon treatment is smoother in the out-of-focus area where the only texture is noise. That may be true and may be what you prefer to see when looking at images zoomed to 100%.

Having worked with raw files for a decade, mostly in Adobe products, I prefer to see a little grain, that way I know there is maximum detail in the other areas. A little grain also helps keeps detail when things are reduced and sharpened for export. In other words, if things are overly smooth at 100% then they will be smoother when reducing down and sharpening for a screen-sized image. They will also be smoother when printed.

That's not to say that Adobe's algorithms could use some improvement, especially in the shadow noise department, and perhaps some day they will, but Adobe doesn't have the luxury of tweaking their algorithms for every new sensor the way a camera manufacturer does. If that's more what you want, you might look at DxO products: http://www.dxo.com/ which try harder to optimize by camera sensor and lens. Of course if they haven't supported your camera and lens combination then you're out of luck and you're paying for that extra level of attention.

Without being able to see your detail sliders, the other thing to mention is that, if you haven't already, turn up the Sharpen Masking slider almost all the way--somewhere between 90 and 100, holding down the Alt key as a guide while moving the slider. This will keep ACR from applying sharpening the noise grain in the out-of-focus areas. Fiddling with the Sharpen Detail slider also affects the fine grain structure.
Photo of James Monaco

James Monaco

  • 31 Posts
  • 10 Reply Likes
Thanks for the reply. I did provide a link to the NEF file used in this image; I agree that in the image shown in my post, the ACR version looks a little sharper, but this may be due to the image being reduced in size by this website's style. Click the image to see in full size--I thin there, they look just about equally sharp. Regardless, though, at full size, the difference in the way in which the noise is reduced should be clearer.

You come to Adobe's defense admirably, and I do understand that they haven't got the capability to exactly determine both the noise threshold and the noise reduction methods used by each of the many manufacturers they support for every camera model released.

On the other hand, and by the same argument, I can contend that if nobody points out that something doesn't look right, there isn't a fair way to expect Adobe to know about it and make efforts to improve it. It may be that the formula for acquiring RAW from a D3100 needs to be tweaked a bit, or it may be that the formula for performing Lum NR could use improvement or greater variability. Somewhere, an Adobe Engineer probably knows which is contributing most to the results I've observed.

It's because of Adobe's general eagerness to continuously deliver improvement that I feel it's worthwhile to bring up the inconsistent rendering in the first place!

I may take a look at the DXO trial as well--my cameras are both supported--though as you can imagine, I always prefer to stick with Adobe when possible.
Photo of James Monaco

James Monaco

  • 31 Posts
  • 10 Reply Likes
Oh--that's an extremely good point, Chris!
I usually don't touch ACR's default sharpening, which comes bumped up to 25 out of the box.

Dropping that down to about 5 gives me a result much closer to the ViewNX-processed image. The behavior is a little different--takes on almost a sort of water paper effect--but at that lower setting, I think I actually prefer the ACR rendition in the end.

Good call on the sharpen; I had assumed the 25 setting was a camera-calibrated baseline and had entirely overlooked the possibility that it was adding to the noise. I would call this solved, and a lesson learned. :)

For reference, final comparison below--the similarities are actually close enough now that I think jpg compression renders them nearly impossible to spot.