Recently we took a look at Adobe Denoise, the company’s noise reduction technology which made its debut in mid-April. The new feature can be found in Lightroom Classic v12.3, Lightroom CC v6.3 and Photoshop’s Camera Raw v15.3. We found a lot to like in its results, but when last we wrote we’d yet to evaluate how it compared to its strongest rival.
In this article, we’ll pit Adobe’s results head-to-head against those from the DeepPRIME XD denoising engine featured in DxO PhotoLab 6.6.1. DxO’s algorithms represent the pick of the field for noise processing quality. Admittedly, they’re a bit slower than those in the oft-recommended Topaz Denoise AI, but they tend to yield the best image quality overall and so are best suited to give Adobe a run for its money.
Without any further ado, let’s get down to some comparisons!
The crops in the remainder of this article come from denoised versions of these five images using both programs’ default noise reduction levels, and are shown here as the original out-of-camera JPEGs.
Adobe’s approach to workflow is a bit less versatile
It’s worth noting off the bat that Adobe and DxO take a different approach to the processing pipeline, which has some implications for your workflow. With Lightroom or Camera Raw, Denoise processing must be performed separately from the rest of your image processing; it delivers an intermediate, noise-processed DNG file from which you’ll then perform the remainder of your processing.
DxO PhotoLab can do the same thing if you choose to manually output DNGs after performing noise processing alone. That could be useful if you want to process the same set of images differently for multiple output formats while retaining the same noise reduction. Both programs take quite a while to perform their AI-based noise processing, after all, so you won’t want to repeat the NR step unless you’re actually tweaking its settings.
By eliminating the intermediate file and extra step in your workflow, DxO PhotoLab can save you processing time and disk space.
But with PhotoLab, you can also choose to perform noise processing at the same time as lens, exposure and other corrections, outputting straight to JPEGs. By eliminating the intermediate file and extra step in your workflow, PhotoLab can save you processing time and disk space (PhotoLab’s processed DNGs are routinely about 1/4 smaller in file size than those from Adobe Denoise, should you need to take that route instead of JPEGs).
Test 1: Dog (Sony ZV-E10, ISO 3200)
As we begin our head-to-head contests, remember that you can click each crop to see the full-size denoised image.
 |
|
| Adobe Denoise | DxO DeepPRIME XD |
Comparing the two versions of this shot from our Sony ZV-E10 review side by side, it isn’t easy to see a lot of difference between the Adobe and DxO result in the dog’s face. Both are noticeably crisper in the fur than the out-of-camera JPEG, but there’s only a razor-thin margin between the two denoised variants.
But if you look more closely you’ll spot some differences elsewhere, specifically in the lower-contrast regions. Look for example at the fine woodgrain patterns in the in-focus areas of the floor or the finer fur in the dog’s paws or near the tips of its ears and you’ll discern just a little more detail and contrast in the DxO DeepPRIME XD version.
It’s a very close thing, though. The biggest difference between the two shots in terms of noise and detail levels is that Adobe retains a little bit of fine (and quite natural-looking) grain by default, whereas DxO yields an almost noise-free result. This is most visible in the well-blurred bokeh areas of the image.
Test 2: City (Canon EOS R10, ISO 6400)
 |
|
| Adobe Denoise | DxO DeepPRIME XD |
Next, let’s step up the sensitivity range a bit to ISO 6400 with a shot from our Canon EOS R10 review. Here, the differences are a little easier to find. Again we can see that Adobe retains a bit more of the fine grain, which is especially noticeable in the flat areas of the lighter-colored buildings like the one at the very center of the shot.
In this same building, we can also see that DxO’s DeepPRIME XD algorithm has restored more of the very fine lines in the paneling than did Adobe. The same is true for the patterns of the brick textures in the building in front of it, which were barely even visible in the out-of-camera JPEG.
You can also better make out the phone number in the Premiere on Pine building just above and left of center in the full image. Pretty much everywhere in the shot, DxO’s result feels crisper than Adobe’s version, thanks to greater local contrast. It also has a bit more saturation across the board, even if Adobe’s version still feels far less washed out than the out-of-camera JPEG.
Test 3: Portrait (Sony a7R V, ISO 8000)
 |
|
| Adobe Denoise | DxO DeepPRIME XD |
Our third sample comes courtesy of our Sony a7R V review and was shot at ISO 8000. In this crop DxO pulls back just a little more of the finest detail in the skin textures. You can see the same elsewhere too, especially in the subject’s forehead, nose and foreground cheek.
With that said, in Adobe’s rendering, the falloff of fine detail from blurred areas to defocused ones feels just a little more natural. In DxO’s version the falloff is much more rapid, likely thanks to the algorithms managing to claw back some detail in the more subtly-defocused areas and then suddenly reaching their limits. This is probably most noticeable in the subject’s far cheek.
Overall, DxO’s version feels noticeably more detailed due to greater per-pixel contrast and thus perceived sharpness. The difference is actually rather more subtle when compared very closely, though. (And again, the fine grain remains in Adobe’s version but is all removed in DxO’s variant.)
Test 4: Station (Sony a7 IV, ISO 12800)
 |
|
| Adobe Denoise | DxO DeepPRIME XD |
Our final two comparisons come courtesy of our Sony a7 IV review. The first of the pair is shot at ISO 12800. As you can see in the gallery at the top of this article, the out-of-camera JPEG is quite soft as a result of noise reduction, and its colors are a bit muted.
Both Adobe and DxO do a pretty good job here. Their improvements are most noticeable in the map beneath the Pioneer Square sign in the crop above, as well as the large grille towards the right of the crop.
As we’ve come to expect, DxO’s variant is the crispest and has the lowest noise levels, while Adobe’s actually shows just a little more luminance noise than the out-of-camera JPEG and sits between the other versions in terms of crispness.
But both AI-processed versions also show some artifacts, with some stair stepping and dithering being particularly visible in the grille. It’s still much less bothersome than the blotchy, mottled result from the camera itself, however. DxO’s quest for detail makes it a bit more artifact-prone, while Adobe’s less aggressive approach to denoising yields fewer artifacts too.
Test 5: Outlets (Sony a7 IV, ISO 20000)
 |
|
| Adobe Denoise | DxO DeepPRIME XD |
The second and final Sony A7 IV frame was captured at a high ISO 20000, and the story continues to be much the same as at lower sensitivities. We still see a little more noise in Adobe’s rendering, and a little more detail with significantly more sharpness from DxO. But both remain far preferable to the rather mushy and muted results from the camera itself.
This difference is particularly easy to spot in the scuffed-up plastic outlets and button panel beneath, but it’s also quite obvious in the wood grain, metal face plate and small placard as well. It’s in the plastic areas and placard where you will also note DxO’s slight detail advantage, with some of the fine scratches it has picked up being much harder to see in Adobe’s rendering.
Performance
Where there’s a much clearer advantage for DxO is in its performance relative to Adobe Denoise. The latter is almost glacially slow, consistently taking 3.8 to 4.2 times as long to deliver its final results as does DxO DeepPRIME XD. Adobe’s new denoising algorithms come pretty close to matching results from class-leader DxO, but they do so much, much more slowly.
To put some finite numbers on that, we timed the latest versions of both Adobe Camera Raw and DxO PhotoLab as of this writing. Performance testing was conducted on a mid-range Dell XPS 15 (9570) laptop using the Nvidia GeForce GTX 1050Ti Max-Q graphics processor and Windows 10 version 22H2.
Processing a batch of seven images ranging from 24 to 60 megapixels and totaling 314MB took about 22 minutes in DxO PhotoLab, regardless of whether processing to JPEG with lens corrections or to DNGs without them. The same batch of images required a whopping one hour, 24 minutes in Adobe Camera Raw, which was performing only the denoising with no further lens or other corrections applied.
Final thoughts
Overall, we have to say that the differences between Adobe’s and DxO’s results in this comparison have been pretty minimal and boil down largely to two things. Adobe’s results consistently have a little bit more luminance noise, which is nevertheless quite natural-feeling and unobjectionable, while DxO delivers noticeably crisper-looking fine detail.
 |
| Adobe’s new Denoise has a distinct advantage, since it’s included with every Creative Cloud subscription. |
DxO does occasionally extract just a little more actual detail from shots, especially when presented with details that have very low contrast. But in doing so, it makes itself a little more prone to artifacts. You have to be looking very closely to spot those differences. Most of the time, it doesn’t have more detail, just the feeling of more detail thanks to greater sharpness.
It’s the workflow differences, pricing models and performance that will drive the choice between them.
For most people, the decision as to whether they should use Adobe Denoise or DxO DeepPRIME XD won’t be made by those minor variations. Instead, it’s the workflow differences, pricing models and performance that will drive the choice between them.
If you already have a subscription to Adobe Photoshop and/or Lightroom Classic/CC, you can definitely live with the new Denoise algorithms alone, so long as you’re happy with the additional workflow step, and processing time isn’t a major concern.
But if time is money, or you want the simplest workflow and performance matters, you’d be well advised to stick with DxO DeepPRIME XD instead!
Author:
Mike Tomkins
Source: Dpreview
