On1 NoNoise Vs DxO DeepPRIME Noise Reduction

Since publishing my review of the recently launched On1 NoNoise AI software I’ve received more than a few emails asking how it compares to DxO DeepPRIME. If you aren’t aware, DeepPRIME is the noise reduction technology used by DxO in their PureRAW and PhotoLab Elite products. It’s extremely well respected and produces very good results. Given the level of interest and that I’m curious as well, I think it’s time for an On1 NoNoise Vs DxO DeepPRIME challenge.

Comparing On1 NoNoise and DxO DeepPRIME

Whilst comparing On1 NoNoise with DxO DeepPRIME may seem easy, there are a lot of variables that could skew the results. Because of this I want to set out my method so that you understand how the testing took place. I’ll break my approach down into three areas:

1. Software
2. RAW Files
3. Settings

Software Features and Settings

On1 NoNoise AI is a dedicated noise reduction application. Compare this to DxO DeepPRIME and you’ll find that it’s a technology used in a couple of DxO products (currently PhotoLab and PureRAW). Probably the nearest comparable of these to On1 NoNoise AI is DxO PureRAW.

PureRAW is a RAW file pre-processor which removes noise and corrects optical problems in RAW files. It’s also possible to replicate it’s features using DxO PhotoLab, which is what I will be using for the testing. If you want to understand this further, I’ve published an article explaining how.

When using On1 NoNoise AI it’s possible to vary the level of both colour and luminance noise reduction. Rather than try to fine tune each image, I’ll be using the auto option in the software. I also won’t be using any of the advanced masking options but instead apply the adjustment to the entire image. The other feature in NoNoise AI is that it applies Detail enhancement and sharpening to the image. I’ll therefore also leave these at their auto and default settings.

With respect to using DxO PhotoLab and DeepPRIME, I don’t have much control. It’s either on or off although there is only a single Luminance slider. This seems to always default to 40% when using the auto option and I’ll have more to say about this shortly as it’s quite important. The other options in PhotoLab are the Lens Correction to remove distortion and Lens Sharpness to correct lens softness. It seems reasonable to use these given the features of On1 NoNoise that I’ve chosen to use. The Lens Correction option is either on or off, but the Lens Sharpness settings can be adjusted. Again, I will leave these at their default settings.

RAW Files

It’s also worth mentioning RAW files used for testing as this could be important in the results.

To get the best image quality using On1 NoNoise AI and DxO DeepPRIME you need to be processing RAW files. Images that have already been processed to a TIFF or JPEG produce noticeably inferior results. In the Case of DxO PhotoLab, I can only use the DeepPRIME and Lens Correction options with RAW files.

A further limiting factor in the DxO software is that there need to be supporting modules for your camera and lenses. This is a problem if you use an unsupported camera like the Fuji X Series and means that PhotoLab isn’t for you.

The final point to consider is that there could be variation between different manufacturer RAW files. For this article I’ll be using RAW files from a:

1. Panasonic LX5 pocket camera.
2. Panasonic GX1 micro 43 camera.
3. Sony NEX5 APSC camera.

I’ve selected these because they are all old RAW files shot around 10 years ago when digital technology wasn’t as advanced, and the cameras are all very noisy.

Flawed Settings

Now back to the point about the Luminance slider in DxO PhotoLab.

Initially I was going to set the DeepPRIME option to the automatic level which seems to be 40% Luminance reduction. After running a few tests of a very noise RAW file from a Panasonic LX5, I realised how much difference this setting made to the results as seen here (click the image to enlarge).

Sample 1 shows the default 40% noise reduction. When viewed at 200% magnification it’s easy to see Luminance noise, although it’s not unpleasant. At 100% noise reduction used for sample 3 there is no visible noise at 200%. But what I don’t like about the result is that it appears a little too smooth, especially in the sand of the beach. At 60% noise reduction in sample 2, there is a hint of noise at 200% magnification, but the image is more appealing. I’ll therefore be using a 60% setting in the testing BUT that could be too strong for some images.

If you want the best results from DxO PhotoLab DeepPRIME you probably need to refine the level for each image. The same may also be true of the Lens Sharpness settings.

Feature Summary

Whilst preparing to test On1 NoNoise AI and DxO PhotoLab I realised how much difference there is between the products. The things I’ve noticed are

Both packages apply noise reduction (Colour and Luminance) globally but with On1 NoNoise you can use masks and layers to build localised adjustments. That said, I’m not yet convinced this is a real benefit. The performance of both packages is very good when applied globally so why go to the additional effort which is time consuming.

Both packages offer control over the levels of noise reduction but with DxO PhotoLab, you only have a Luminance slider to control DeepPRIME. But if I were using PureRAW rather than PhotoLab, I wouldn’t even have this control. In DxO PureRAW, DeepPRIME is either on or off. What I don’t know is if PureRAW employs some clever routine to determine the correct level of noise reduction without user intervention.

When it comes to optical correction, there are variation. Here’s an example.

Compare the differences between image 1 and 2, especially along the edges and into the corners. You will see that image 1 which was produced in Photoshop has areas cropped off. It isn’t a big difference but it’s noticeable when you place one image over the other.

My best guess is that it’s caused by the lens profile embedded in the RAW file. Photoshop is reading and applying that automatically when I created image 1. This baseline image has the Noise Reduction and Sharpening disabled but there isn’t a way to disable the profile. When I process the image in On1 NoNoise it seems to ignore the lens profile, hence the difference.

It also appears On1 NoNoise strips the lens profile out of the file when saving it as a DNG. When I open the DNG file in Photoshop the information panel doesn’t show there is a lens profile. What I do know for certain is that any optical corrections applied by NoNoise are much less effective than those applied by PhotoLab. This is also true if we compare NoNoise to DxO PureRAW.

On1 NoNoise Image Quality Comparison to DxO DeepPRIME

Now let’s look at the results of the two Packages.

Comparison 1 – Panasonic GX1 at ISO400

Here’s the first image comparing the unprocessed RAW file with On1 NoNoise AI and DxO DeepPRIME. The image was shot using a Panasonic GX1 at ISO400 during the day. There isn’t much obvious noise in the image, but I wanted to test how the well the fine detail of the rock was preserved (or not). You may need to click the image to enlarge the display.

This is a small section from the centre of the image at 100% magnification. Sample 1 is the original RAW without noise reduction of sharpening applied. Number 2 is from On1 NoNoise AI and number 3 is from DxO PhotoLab. Whilst it may be difficult to see in this article, both NoNoise AI and PhotoLab has produced very good results. PhotoLab does however appear the sharper and more detailed of the two. It may be easier to see in the following comparison at 200% magnification.

Here sample 1 is the image processed in NoNoise AI whilst sample 2 uses DeepPRIME in PhotoLab. What I noticed as I zoomed in further is that the detail breaks down quicker in the NoNoise image. I also noticed a faint pattern on the image which is something I’ve highlighted in the past as a possible weakness in AI software.

Comparison 2 – Panasonic GX1 ISO800

This comparison also uses a Panasonic GX1 image but at ISO800, shot in the early evening. The comparison shows a small section of the image at 100% magnification.

Sample 1 shows the unprocessed RAW file which is quite soft and has visible noise. Sample 2 is from On1 NoNoise AI and is noise free, but the detail remains soft. Sample 3 was produced using DeepPRIME. Not only has the noise been completely removed, but the detail is also much sharper.

This comparison is also interesting because it shows the difference in the optical correction. Notice the small boulder on the right side of the image. The three samples were taken from the same coordinates on the image, but you can see the difference the optical correction has made, shifting the boulder to the right.

Comparison 3 – Panasonic LX5 ISO3200

This comparison shows a Panasonic LX5 RAW file shot handheld at ISO3200, well after sunset. The sample area is from the centre of the frame at 100% magnification.

The original image, sample 1, is loaded with both luminance and colour noise. Sample 2 was produced with On1 NoNoise and shows excellent noise reduction and detail retention. The results are quite amazing but there is a problem. I can see traces of red colour fringing on the sand behind the post. Compare this to sample 3 produced with DeepPRIME in PhotoLab and there is no problem. Sample 3 is equally free from noise although the detail in the distant sand is slightly softer which I can’t really explain. It could be due to my arbitrary 60% Luminance reduction setting but I don’t know.

Comparison 4 – Sony NEX-5 ISO800

The final comparison is from a Sony NEX-5 RAW file, shot handheld at ISO800, at night from the top of the Empire State Building. The sample area is from the centre of the frame at 100% magnification.

As with the other comparisons, sample 1 is loaded with noise whilst sample 2 and sample 3 are noise free. Comparing sample 2 from On1 NoNoise AI with sample 3 from DxO PhotoLab, both have performed very well. Strangely, with this example, PhotoLab doesn’t appear quite as sharp as NoNoise at first glance. After zooming in beyond 100% this appears to be caused by faint sharpening halos around some of the bright lights. I also noticed that the NoNoise sample doesn’t have quite the contrast range as the sample from DxO PhotoLab.

Summary of On1 NoNoise AI Vs DxO DeepPRIME

Firstly, let me say that in these tests both technologies On1 NoNoise and DxO DeepPRIME performed well, leaving the images noise free. The results that are possible using these are amazing, especially at very high ISO levels. But as I must pick a winner, I’m going to select DxO DeepPRIME.

The reason for my choice isn’t so much to do with the noise reduction. It’s the noise reduction performance coupled with the sharpness of the resulting image. Whilst I may be able to improve the results for On1 NoNoise by experimenting with the controls, this adds time and overhead. Equally I could also improve the performance of DxO PhotoLab by tweaking the settings.

The points that make DxO the winner for me are things like the lack of lens correction in On1. The colour fringe I noticed in some of the On1 images and the unprocessed pixels around the edge of the frame that I mentioned in my previous review.

Overall, the performance from DxO feels more polished providing it supports your camera and lenses. But as always, I recommend trying the software for yourself given the variables involved.

You can download a trial of DxO PhotoLab or PureRAW from the DxO website. You can also download a trial of On1 NoNoise from the On1 site. If you decide to purchase from On1, remember to use the discount code: LENSCRAFT20 for a discount at the checkout.