this reads a noise profile and then runs
some variant of a-trous fisz-transform wavelets for noise reduction.
if you did not profile your camera for noise coefficients at the given iso value yet,
you can manually dial in the gaussian and poissonian part of the noise model
in the raw input module.
you can run denoise pre- and post-demosaicing.
the denoise module also takes care to crop off the black borders (noise and
black point calibration pixels) and removes the black point and scales to
the white point of the raw image file. it is thus an essential part of every
raw pipeline. set strength to 0.0 if you only want to crop/scale and no
denoising. it will employ a specialised noop kernel in this case.
this module is usually fast and denoises a full resolution 24 megapixel raw image in around 20ms on a lower end nvidia gtk 1650 max-q.
for extreme low-light cases if you have a chance to take burst photographs, you can use the align module for further noise reduction.
input the input image can be rggb or rgbaoutput the output image either mosaic (normalised with black point subtracted) or full rgb, depending on inputstrength the overall denoising strength. this directly scales the unbiased soft shrinkage threshold in the wavelet method.luma this blends back a portion of the original y channel after denoisingdetail protect what is detected as detail. use lower values for extremely heavy noise where detail detection fails.gainmap if the input file ships metadata about a gainmap, it can be applied herea bayer image after lens corrections, demosaicing after denoising (slide the little right corner for comparison):
an xtrans image: