Noise Reduction Challenge

[freixas]

A short time ago, I posted a comparison of the noise reduction capabilities of three programs: DxO PhotoLab2, Capture One v12 and Adobe Camera RAW CS6. Of course, when working with a lot of programs that one might not be an expert in, the question arises as to whether one produced the optimal image.

So here's the challenge:

• Download the Canon CR2 file that I've placed on my Google Drive (if you click on the link, you will see a preview of the CR2).
  
• Using your favorite RAW tool, try to create a version that is sharp (without being over-sharp), avoids losing detail but has the least amount of noise you can manage.
• Post extracts roughly matching the areas that I include below.
• Include information on your exact version of the product and as much detail about the settings you use as you can manage.

Alternatively, you can provide a different RAW file to test with. If you do, you should also provide your best noise reduction results using at least one tool (and document your results as I did).

DxO PhotoLab2 v2.2.3 Build 23

Settings

I disabled everything except the settings below. If I don't mention an exact value for an enabled setting, then it is at the DxO default value.

• DxO Smart Lighting: Enabled
• Selective tone: H: -28, M: -39
• Contrast: Microcontrast: 23
• Noise reduction: PRIME, Luminance: 83
• Lens sharpness: Enabled
• Chromatic aberration: Enabled
• Unsharp mask: Intensity: 48, Radius: 4.45, Threshold: 4

Image 1

Image 2

Capture One v12.0.3.22

Settings

• I tried to set every adjustment off, then:
  
• Sharpening: Amount: 311, Radius: 1.2, Threshold: 1.5
• Noise Reduction: Luminance: 80, Detail: 61, Color: 50, Single Pixel: 40
• Clarity: Structure: 16
• Default lens correction

Image 1

Image 2

My Impressions

The Capture One (C1) image looked crisper and I can't say I ever matched its tonality with DxO PhotoLab2 (PL2). On the other hand, the PL2 image had less noise and more detail, even if it was slightly less sharp.

Compare the green area (Image 2) in PL2 with the C1 version. A lot of the detail maintained in PL2 is just gone in C1. On the other hand, the small fibers on that one strand cutting across the image look a little fuzzy in the PL2 version. In Image 1, the vertical ridges of the plant pot have almost no noise. The top of the pot looks a little blurry, yet there appear to be no details missing (vs. C1).

One problem with using PRIME noise reduction is that you can only judge the effect on the entire image if you export it. Within the program, you can only view the effect at 100% in a small capture of a portion of the image.

I have a PSD file that compares the entire image. It also includes versions of the image with all adjustments turned off. By toggling visibility on/off, you can quickly compare the differences.

 

[Derrel]

Too zoomed-in to really evaluate the efficacy as relates to a "picture". The full image does not appear to be "hurt" by noise, IMO.

 

[JonFZ300]

I agree with Derrel. I love the idea behind this challenge but I don't see a noise issue in the full size pic either. And if I'm totally honest, the pic is not interesting enough to me to make me want to play around with it.

 

[freixas]

I agree with Derrel. I love the idea behind this challenge but I don't see a noise issue in the full size pic either. And if I'm totally honest, the pic is not interesting enough to me to make me want to play around with it.

I don't think the picture needs to be interesting to evaluate noise reduction. However, if anyone has a better (worse?) RAW image that they'd like to donate, I'd be happy to process it with DxO PhotoLab2. Run it through your favorite tool and we can compare.

An ideal image would have both a lot of noise and a lot of fine detail, which is what I was trying for here.

 

[Derrel]

An ideal image would have both a lot of noise and a lot of fine detail, which is what I was trying for here.

Lots of texture here. But not a lot of noise...

 

[freixas]

An ideal image would have both a lot of noise and a lot of fine detail, which is what I was trying for here.

Lots of texture here. But not a lot of noise...

Did you download the PSD I linked to and look at the layers with NR disabled? Sample below; it looks to me like there's plenty of noise. It's an underexposed shot at ISO 3200 taken with my ancient Canon T1i. I believe it is sufficient to evaluate the effectiveness of any NR tool.

This is the second thread where I've tried to help others in comparing various NR tools. The two threads represent some considerable effort on my part and I don't plan to post any other sample image to work from.

If you don't like my image, feel free to donate your own RAW file and I'll promise to run it through PhotoLab2--if you also first post your best NR of that image using your favorite tool. If you don't want to use my image and you don't want to post one of your own, then you might as well move on.