- Messages
- 3,267
- Name
- Graham
- Edit My Images
- No
This is a summary of my experiences testing several denoise programs and explains why I think you cannot rely on published test to help you make a decision.
We now have, at least three (four) AI denoising options ie Topaz DeNoise AI, ON1 NoNoise and DXO Deep Prime either in the form of Pure Raw, or as part of PhotoLab 5. I've used DXO Prime noise reduction in the past (Not DeepPrime), along with NeatImage, and more recently Topaz DeNoise AI.
Overwhelmingly positive tests and comments about DXO PureRaw, encouraged me to try it out. Not only for its denoising capability, but it's claims of improving the demosaicing quality of raw images at any ISO with claims that it could save you buying a new lens or camera because of how good it was. Not that I believed this, but it piqued my interest.
Introduction
My trial of DXO Pure Raw 2, didn't last very long because even with sharpening turned off, I was getting sharpening haloes, so I switched to a trial copy of DXO Photolab 5. A sharpening setting of +1 in Photolab 5is the same as the default in PureRaw, and and a sharpening setting of 0 in PhotoLab 5 is the same as switching sharpening off in PureRaw.
I used Photolab5 at sharpening settings of -1 and -0.5 for my tests.
While trialling PhotoLab 5, I bought a copy of ON1 NoNoise (half price offer, version 2022.5) and downloaded the beta version of Topaz Labs PhotoAI that combines Denoise AI, Sharpen AI and Gigapixel AI into a single program, which was getting rave reviews even in beta (I used the 0.6 beta). With NoNoise, I ended up using the "low" setting for noise reduction and switched the sharpening off entirely.
Topaz PhotoAI I abandoned as I couldn't get it to give consistent results, and I only used Topaz Denoise AI 3.7
I tested with eleven different files from ISO 200 to ISO 20,000. Cameras were Olympus EM5 MkII, Fuji XE1, Nikon D600, D750 and Nikon D500. Subject matter was landscape, flowers and birds. Raw processing was done by the denoise programs (except Neat Image), and exported as DNGs. Comparisons were made in Capture One 22, which allows subtle editing of multiple images at the same time, This it easier to match image quality than any other software I'm aware of. Neat Image was used a a Photoshop plugin with images round tripped from Capture One. Denoising was also done in Capture One for comparison purposes.
Where options were available in each program, I worked hard to get the best results, nothing was just "left at the defaults" as that rarely gave the best results. I spent several weeks, working at least part of every day comparing the results. Often re-running the denoising software, if I felt I had got something wrong.
At this point you may expect to see lots of photo comparisons, but firstly as I don't post images on line, sorting out posting lots of images is more work than I am willing to put into this. And secondly, I don't think they would mean very much, as my conclusions constantly changed as I went back and had another go at "improving" the results. I deliberately erred on minimising the effects of the denoising programs so I had some leeway to tweak (increase) sharpening and denoising in Capture One.
Not surpassingly there was no clear winner.
Results
With every image I came to a different conclusion about which combinations gave the best result.
Low ISOs (200 to 800)
With an 800 ISO Olympus file, all the denoise options, gave a slightly smoother, sharper and detailed result than C1, but at the expense of giving various artefacts and loss of colour gradation. C1 and DXO were close, but looking at the whole image, C1 clearly gave a nicer result. Interestingly, re-running NoNoise removed the texture artefacts it had created in the first run. Winner C1 or NoNoise
With the 200 ISO Olympus file, Topaz and DXO gave severe sharpening haloes (even with sharpening turned down in DXO) which I couldn't remove with the C1 halo slider. NoNoise and C1 also gave haloes, but these could be removed in C1. Not much to tell these apart at normal print size magnification. Winner NoNoise or C1
A Nikon D600. file at 200 ISO was just horrible with Topaz, which gave blocky textures and bad haloes. DXO, NoNoise and C1 were similar, all had halos that could be removed in C1. NoNoise had better detail and shadow contrast than DXO. Again CI had better colour. A second D600 file at ISO 320 gave similar results. Winner C1 for colour, NoNoise for noise and detail
Two Fuji XE1 files at 200ISO, gave very poor results but Topaz. Nonoise and DXO were very close, with NoNoise showing some artefacting. C1 has less detail, but as before still looks better in terms of colour and microcontrast. Winner C1, but DXO or NoNoise very close with more detail at the expense of colour
Medium ISOs (1600 to 6400)
No Olympus files from this range, but with Fuji XE1 files at 1600 and 6400 ISO, C1 started to really struggle and the slightly better colour from C1 ws more than offset by the lower noise and higher detail from the denoising programs. Again Topaz performed badly in comparison with the others and NoNoise was slightly better than DXO. Winner NoNoise
With Nikon D500 files up to 5000 ISO, with a bit of effort, I could get files that competed reasonably well with the denoise programs (i.e. comparisons at 50%) Topaz again trailed behind DXO and NoNoise. With ISOs above 5000, C1 couldn't seriously compete with the denoise programs. There was very little between DXO and NoNoise, NoNoise was maybe slightly better. Winner DXO or NoNoise
High ISOs (one file 20,000)
Nikon D500 file which C1 and DXO gave very poor results. This time Topaz Denoise gave similar results to NoNoise, but noNoise was still noticeably better. Winner NoNoise or DXO (see edit below)
EDIT 01/09/22: I have re-run the DXO test. My focus on avoiding sharpening haloes meant I had left DXO to auto-adjust the noise reduction bit,which seemed to work well. But having noticed that it seemed to always leave the noise setting at 40 I decided to tried to improve on my earlier results.
I ended up with the DXO noise reduction set to 75 and the sharpness at 0 rather than the 40 and the -0.1 I had previously used. The result was different to the NoNOise result, but with neither "overall" obviously better than the other. I probably preferred the DXO results, but as I tweaked the results, I kept on changing my mind.
Comments on the above.
Topaz
Topaz didn't do very well in the above, where it was used in the same way as DXO and NoNoise by Denoising the RAW file and exporting as a DNG. This is how Topaz recommend you use the program, but it wasn't how I had been using it before this test.
I repeated the tests using Topaz as a plugin for Photoshop on images demosaiced in C1 and round tripped from C1 to Photoshop for denoising in Topaz DeNoise AI. I didn't repeat the detailed comparisons described above, but in these comparisons Topaz did much better than the first run through. Where my results above suggest DXO and NoNoise are very close, I found that Topaz was also very close. I also noticed that the colour rendition and the microcontrast of files processed this way, were a better match for the C1 files than those processed in DXO or NoNoise.
DXO
Although NoNoise seemed to just beat DXO a few times, DXO seemed, overall, to be less susceptible to artefacts and gave better colour rendering than NoNoise.
EDIT (01/09/22) DXO TIP: DXO only displays the results of Deep Prime in a small window, making it difficult to assess how much it's affecting the image. But if you export the image as a JPEG (using the original image folder export option and keeping the file name the same as the original), it will appear (with deep prime applied) in the thumbnails alongside the original RAW.
In DXO you can then select the small disclosure arrow on the compare view icon, and choose the JPEG you have just created to get a split screen view showing the DeepPrime version in one half and the standard DXO noise reduction version in the other. The jpeg can then be deleted and different settings tried if necessary. It's not great, and you can't compare multiple JPEGS (at different deep prime settings), but it's a relatively easy way of seeing the effect of Deep Prime across the whole image in a single view.
Neat Image
I didn't make a full set of comparisons with the non-ai Neat Image 9 as it requires a lot of effort to use. It's been around for a long time and release 9 came out last year. There is some automation, but it mainly relies on the user to manually adjust settings at the colour channel and detail frequency level.
I think it gives more natural results at low to moderate ISOs than the AI programs do, but at cost of a serious learning curve and time needed on each file. I've never had much success, compared to the AI options with high ISO files.
Colour
I've mentioned better C1 colour (and the microcontrast) on a few occasions, as C1 has a reputation of doing both these things a bit better than other raw processors. Something I agree with, but it's very subtle and not something that will make or break a photograph. There was nothing wrong with the colours from any of the options tried. I did mean to repeat some of the comparisons with Lightroom, but I ran out of steam.
Conclusion
I'm not sure I have a conclusion, other than the need to try these programs out for yourself. They all do a good job, and they all have positives and negatives. What is best will depend on your preferred workflow, sensor size, subject choice, working ISOs, and final output for your images (and how much money you want to spend).
I'm back where I started. Using Capture One and round tripping from C1 to Photoshop to use Topaz Denoise AI or Topaz Sharpen AI as PS Plugins.
At the price (especially as I got it for £35) ON1 NoNoise was impressive. especially if you look at the masking and editing tools it provides to subtly and selectively control noise and sharpness across your image. It was also very fast compared to the other options. If I didn't have PS plus Topaz, Denoise would be a very attractive plugin for C1 to deal with higher noise images.
DXO partially lives up to his marketing of making your existing cameras/lenses better, as it's generally more reliable in results than NoNoise or Topaz, and comes with high quality lens and sensor profiles, that give very clean results.
The lens corrections cuts off far less of the image than the C1 lens corrections and, colour apart, most of the time gave slightly more detailed and smoother results than C1, even at low ISOs. It increasingly got better than C1 as ISOs increased.
My one test at a very high ISOs suggest that Topaz and NoNoise give better results at high ISOs than DXO, and some Youtube bird photographers are combining DXO PureRaw at the beginning of their processing with Topaz in Photoshop at the end of it.
If they add more sharpening options to DXO Pure Raw, it should become a useful option to do initial processing, especially as a plugin to Lightroom. As it is, the full Photolab 5 feels like a much better value alternative, which can still be run as a Lightroom Plugin. But seems like an overkill to just manage noise. It could of course replace Lightroom or Capture One (if you don't need cataloguing) and I did do some comparisons, but "overall" I still preferred the result from C1 over DXO, and DXO over Lightroom.
Well done if you have made it to the end of this, hopefully some of it was useful.
Below is about a quarter of a Nikon D500 (APS-C) frame at ISO 20,000 processed in Capture One (lower image) and ON1 NoNoise (upper Image). Neither are very good, (its not very sharp to begin with) but it gives an idea of the improvements that NoNoise can bring.
We now have, at least three (four) AI denoising options ie Topaz DeNoise AI, ON1 NoNoise and DXO Deep Prime either in the form of Pure Raw, or as part of PhotoLab 5. I've used DXO Prime noise reduction in the past (Not DeepPrime), along with NeatImage, and more recently Topaz DeNoise AI.
Overwhelmingly positive tests and comments about DXO PureRaw, encouraged me to try it out. Not only for its denoising capability, but it's claims of improving the demosaicing quality of raw images at any ISO with claims that it could save you buying a new lens or camera because of how good it was. Not that I believed this, but it piqued my interest.
Introduction
My trial of DXO Pure Raw 2, didn't last very long because even with sharpening turned off, I was getting sharpening haloes, so I switched to a trial copy of DXO Photolab 5. A sharpening setting of +1 in Photolab 5is the same as the default in PureRaw, and and a sharpening setting of 0 in PhotoLab 5 is the same as switching sharpening off in PureRaw.
I used Photolab5 at sharpening settings of -1 and -0.5 for my tests.
While trialling PhotoLab 5, I bought a copy of ON1 NoNoise (half price offer, version 2022.5) and downloaded the beta version of Topaz Labs PhotoAI that combines Denoise AI, Sharpen AI and Gigapixel AI into a single program, which was getting rave reviews even in beta (I used the 0.6 beta). With NoNoise, I ended up using the "low" setting for noise reduction and switched the sharpening off entirely.
Topaz PhotoAI I abandoned as I couldn't get it to give consistent results, and I only used Topaz Denoise AI 3.7
I tested with eleven different files from ISO 200 to ISO 20,000. Cameras were Olympus EM5 MkII, Fuji XE1, Nikon D600, D750 and Nikon D500. Subject matter was landscape, flowers and birds. Raw processing was done by the denoise programs (except Neat Image), and exported as DNGs. Comparisons were made in Capture One 22, which allows subtle editing of multiple images at the same time, This it easier to match image quality than any other software I'm aware of. Neat Image was used a a Photoshop plugin with images round tripped from Capture One. Denoising was also done in Capture One for comparison purposes.
Where options were available in each program, I worked hard to get the best results, nothing was just "left at the defaults" as that rarely gave the best results. I spent several weeks, working at least part of every day comparing the results. Often re-running the denoising software, if I felt I had got something wrong.
At this point you may expect to see lots of photo comparisons, but firstly as I don't post images on line, sorting out posting lots of images is more work than I am willing to put into this. And secondly, I don't think they would mean very much, as my conclusions constantly changed as I went back and had another go at "improving" the results. I deliberately erred on minimising the effects of the denoising programs so I had some leeway to tweak (increase) sharpening and denoising in Capture One.
Not surpassingly there was no clear winner.
Results
With every image I came to a different conclusion about which combinations gave the best result.
Low ISOs (200 to 800)
With an 800 ISO Olympus file, all the denoise options, gave a slightly smoother, sharper and detailed result than C1, but at the expense of giving various artefacts and loss of colour gradation. C1 and DXO were close, but looking at the whole image, C1 clearly gave a nicer result. Interestingly, re-running NoNoise removed the texture artefacts it had created in the first run. Winner C1 or NoNoise
With the 200 ISO Olympus file, Topaz and DXO gave severe sharpening haloes (even with sharpening turned down in DXO) which I couldn't remove with the C1 halo slider. NoNoise and C1 also gave haloes, but these could be removed in C1. Not much to tell these apart at normal print size magnification. Winner NoNoise or C1
A Nikon D600. file at 200 ISO was just horrible with Topaz, which gave blocky textures and bad haloes. DXO, NoNoise and C1 were similar, all had halos that could be removed in C1. NoNoise had better detail and shadow contrast than DXO. Again CI had better colour. A second D600 file at ISO 320 gave similar results. Winner C1 for colour, NoNoise for noise and detail
Two Fuji XE1 files at 200ISO, gave very poor results but Topaz. Nonoise and DXO were very close, with NoNoise showing some artefacting. C1 has less detail, but as before still looks better in terms of colour and microcontrast. Winner C1, but DXO or NoNoise very close with more detail at the expense of colour
Medium ISOs (1600 to 6400)
No Olympus files from this range, but with Fuji XE1 files at 1600 and 6400 ISO, C1 started to really struggle and the slightly better colour from C1 ws more than offset by the lower noise and higher detail from the denoising programs. Again Topaz performed badly in comparison with the others and NoNoise was slightly better than DXO. Winner NoNoise
With Nikon D500 files up to 5000 ISO, with a bit of effort, I could get files that competed reasonably well with the denoise programs (i.e. comparisons at 50%) Topaz again trailed behind DXO and NoNoise. With ISOs above 5000, C1 couldn't seriously compete with the denoise programs. There was very little between DXO and NoNoise, NoNoise was maybe slightly better. Winner DXO or NoNoise
High ISOs (one file 20,000)
Nikon D500 file which C1 and DXO gave very poor results. This time Topaz Denoise gave similar results to NoNoise, but noNoise was still noticeably better. Winner NoNoise or DXO (see edit below)
EDIT 01/09/22: I have re-run the DXO test. My focus on avoiding sharpening haloes meant I had left DXO to auto-adjust the noise reduction bit,which seemed to work well. But having noticed that it seemed to always leave the noise setting at 40 I decided to tried to improve on my earlier results.
I ended up with the DXO noise reduction set to 75 and the sharpness at 0 rather than the 40 and the -0.1 I had previously used. The result was different to the NoNOise result, but with neither "overall" obviously better than the other. I probably preferred the DXO results, but as I tweaked the results, I kept on changing my mind.
Comments on the above.
Topaz
Topaz didn't do very well in the above, where it was used in the same way as DXO and NoNoise by Denoising the RAW file and exporting as a DNG. This is how Topaz recommend you use the program, but it wasn't how I had been using it before this test.
I repeated the tests using Topaz as a plugin for Photoshop on images demosaiced in C1 and round tripped from C1 to Photoshop for denoising in Topaz DeNoise AI. I didn't repeat the detailed comparisons described above, but in these comparisons Topaz did much better than the first run through. Where my results above suggest DXO and NoNoise are very close, I found that Topaz was also very close. I also noticed that the colour rendition and the microcontrast of files processed this way, were a better match for the C1 files than those processed in DXO or NoNoise.
DXO
Although NoNoise seemed to just beat DXO a few times, DXO seemed, overall, to be less susceptible to artefacts and gave better colour rendering than NoNoise.
EDIT (01/09/22) DXO TIP: DXO only displays the results of Deep Prime in a small window, making it difficult to assess how much it's affecting the image. But if you export the image as a JPEG (using the original image folder export option and keeping the file name the same as the original), it will appear (with deep prime applied) in the thumbnails alongside the original RAW.
In DXO you can then select the small disclosure arrow on the compare view icon, and choose the JPEG you have just created to get a split screen view showing the DeepPrime version in one half and the standard DXO noise reduction version in the other. The jpeg can then be deleted and different settings tried if necessary. It's not great, and you can't compare multiple JPEGS (at different deep prime settings), but it's a relatively easy way of seeing the effect of Deep Prime across the whole image in a single view.
Neat Image
I didn't make a full set of comparisons with the non-ai Neat Image 9 as it requires a lot of effort to use. It's been around for a long time and release 9 came out last year. There is some automation, but it mainly relies on the user to manually adjust settings at the colour channel and detail frequency level.
I think it gives more natural results at low to moderate ISOs than the AI programs do, but at cost of a serious learning curve and time needed on each file. I've never had much success, compared to the AI options with high ISO files.
Colour
I've mentioned better C1 colour (and the microcontrast) on a few occasions, as C1 has a reputation of doing both these things a bit better than other raw processors. Something I agree with, but it's very subtle and not something that will make or break a photograph. There was nothing wrong with the colours from any of the options tried. I did mean to repeat some of the comparisons with Lightroom, but I ran out of steam.
Conclusion
I'm not sure I have a conclusion, other than the need to try these programs out for yourself. They all do a good job, and they all have positives and negatives. What is best will depend on your preferred workflow, sensor size, subject choice, working ISOs, and final output for your images (and how much money you want to spend).
I'm back where I started. Using Capture One and round tripping from C1 to Photoshop to use Topaz Denoise AI or Topaz Sharpen AI as PS Plugins.
At the price (especially as I got it for £35) ON1 NoNoise was impressive. especially if you look at the masking and editing tools it provides to subtly and selectively control noise and sharpness across your image. It was also very fast compared to the other options. If I didn't have PS plus Topaz, Denoise would be a very attractive plugin for C1 to deal with higher noise images.
DXO partially lives up to his marketing of making your existing cameras/lenses better, as it's generally more reliable in results than NoNoise or Topaz, and comes with high quality lens and sensor profiles, that give very clean results.
The lens corrections cuts off far less of the image than the C1 lens corrections and, colour apart, most of the time gave slightly more detailed and smoother results than C1, even at low ISOs. It increasingly got better than C1 as ISOs increased.
My one test at a very high ISOs suggest that Topaz and NoNoise give better results at high ISOs than DXO, and some Youtube bird photographers are combining DXO PureRaw at the beginning of their processing with Topaz in Photoshop at the end of it.
If they add more sharpening options to DXO Pure Raw, it should become a useful option to do initial processing, especially as a plugin to Lightroom. As it is, the full Photolab 5 feels like a much better value alternative, which can still be run as a Lightroom Plugin. But seems like an overkill to just manage noise. It could of course replace Lightroom or Capture One (if you don't need cataloguing) and I did do some comparisons, but "overall" I still preferred the result from C1 over DXO, and DXO over Lightroom.
Well done if you have made it to the end of this, hopefully some of it was useful.
Below is about a quarter of a Nikon D500 (APS-C) frame at ISO 20,000 processed in Capture One (lower image) and ON1 NoNoise (upper Image). Neither are very good, (its not very sharp to begin with) but it gives an idea of the improvements that NoNoise can bring.
Last edited: