Raw Converter Shootout Part #3 – Noise Removal
Noise removal is a key step in image processing and all the converters we are testing have at least one method for doing so. This test will see how well they perform.
For the previous entry in this series, which tests initial sharpening, click here…
What’s being tested?
The converters being tested are:
The test image:
Shooting at many ISOs causes the image to have noise in it. Noise is a nasty, unwanted, speckling on an image. The test image ws shot at ISO 400 and has a mild degree of noise particularly in the water by the rocks.
The challenge is to remove the noise without degrading the image quality or detail. Depending on the final use of the image, this step might be unnecessary. Aside from complex noise removal tools other common steps, such as downsizing an image, also reduce its noise. Another (quite unintuitive) method is to upsize the photo (in Photoshop) to 150%, apply a guassian blur with a radius of about 1 pixel and then downsize the image back to its original size. Crazy, but it works.
In theory, at least, the ‘smart’ algorithms that these RAW converters use should do a better job… Let’s find out.
DXO has two Noise reduction settings: HQ and Prime. Both analyse the image and suggest the settings and these suggestions were used in the test. Photolemur provides no settings at all. The Smart Photo Editor offers many different noise reduction options. I chose the Denoise and Best Denoise options and tested both. The Best Denoise came with such a weath of options that I just left it at the defaults, It certainly led to a very smooth image but I think I’d have to spend a lot more time with it to get the best results…
The other converters all had similar options but none suggested the settings to use. I adjusted the parameters until it looked like the noise had been removed without degrading the image too much. I accept this is a bit subjective but it was the best I could do.
As before, adding noise reduction added time, sometimes considerable time, to the processing process:
|Converter||Basic Processing (seconds)||Including Sharpening (Seconds)||Including Noise Removal|
|DXO||20||25||30 Standard, 110 Prime|
|Smart Photo Editor||30||50||45 (basic), 300 (best)|
Here’s a quick comparison of the results:
|None||Guassian Blur + Resize|
|Smart Photo Editor||Denoise|
|Smart Photo Editor||Best Denoise|
Raw Tests – Noise Removal
Assessing the results isn’t entirely simple as a degree of subjectivity is involved in deciding what looks ‘best’, whatever that means… However, from an ease of use point of view, Photolemur, DXO and Smart Photoeditor all make noise removal simple as they do it automatically. DXO and Smart Photo Editor allow for the process to be customisable, whereas with Photolemur you get what you get.
Comparing the results to the simple Guassian Blur noise reduction is instructive – most of the algorithms used by the converters did not do much better than my crude Photoshop process, which was: Enlarge the image by 50%, apply a Guassian Blur of 1 pixel and then resizze the image back to the original size. And I could probably improve my attempt to denoise in this way.
DXO’s Prime noise reduction is, I think, the best. I’m sure with a lot of tweaking Smart Photo Editor’s Best Denoise process could be made to produce better results. It comes with so many parameters – almost as many as a RAW converter’s:
Photolemur’s noise reduction, despite offering no opportunities to configure it, has done well. Some of its other processing steps might have increased the apparent noise in its result. Only after the other converters have done extra image processing can it really be compared with them. All of the converters have produced acceptable results. Bearing in mind that noise is only going to be visible, unless it is truly terrible) in large prints then I think all of the converters have done OK here. If I were printing at large sizes, especially if I had to upsize the photo, I would definitely use DXO Prime.
In the next post in this series, I’ll taking a good look at each converter’s options for image processing beyond these initial steps…