Monday, December 7, 2015

Better Noise Reduction with Darktable

In a previous post, I wrote about Darktable's profiled noise reduction with a set of sample images to compare the results. And while the basic profiled denoise algorithm still gets great results in Darktable, I have found another method that yields even better results (that incidentally, still utilizes the profiled denoise, but to a lesser extent). It may not show as much as I had hoped in the images below, but I have found this method especially attractive when dealing with photos of people. Steps and samples below:

First step is of course to import your images to Darktable. Next, we will run the Raw Denoise module. In most instances (especially with people) I have found .003 to be a good starting point. Typically the default .010 threshold can make images too soft. For the compared images of the kiddo toys, I used .014 for this setting.

Next step is to apply the profiled denoise option. The default strength of 1.000 is typically overkill after applying the raw denoise first. For my images I have found a value of less than .300 is a good balance to retain detail, typically I use .289. For the sample image, the value used was about .400.

Composition for Comparision
No Noise Reduction
No noise Reduction (Imported and then Exported in Darktable)

Profiled Denoise Only
Profile Denoise Only

Raw Denoise & Profiled Denoise
Raw denoise & Profiled denoise used together

For additional comparison I have added a new image to the original Flickr set that shows the processing done with my refined noise reduction routine.