How do RT's wavelet sharpening algorithms work?

rawtherapee

#1

In RT there are two ways to use wavelets to sharpen images. I am curious how they differ and what exactly each is doing.

  1. The “edge sharpness” panel. Here there are the settings strength, radius, detail, “First level”, and a local contrast method selection. I am not really sure what these mean without knowing what the algorithm does. In particular, the local contrast curve seems odd because by default it is zero on the right, meaning whites would be mapped to blacks.

What exactly are the details of this algorithm? I know it uses Daubechnies wavelets but beyond that I am in the dark.

  1. In “contrast by detail levels.” Here I’m guessing this just adjusts the strength of the coefficients in a difference of Gaussians/Mexican Hat wavelet decomposition, with some thresholding to remove noise and additional thresholding to protect skin tones.

Is this roughly correct? If so, what exactly does “reduce artifacts” do?

I tried to read the source code for each but quickly became very confused. :frowning: If someone could please explain I would be happy to update the Wiki with an explanation. Currently it does not describe this functionality.

(This is related to my other question about wavelets in pixel editors, but it seemed more appropriate for a standalone topic: https://discuss.pixls.us/t/how-best-to-use-a-wavelet-decomposition-to-sharpen/)


(Andrew) #2

nik sharpener huh? :joy:


#3

@RawConvert I don’t miss the Nik Collection but many do.


#4

I noticed that the lowest strength for ‘edge sharpness’ already is quite aggressive, especially if the contrast levels have been played with. Going from 0 to 1 is a huge change in the image. It seems to increase local contrast as well - by a lot!