videoprocessing

[kassandro]

The raw file for the first example was provided by Michael Bischof. It was shot with a Lumix FZ50 and ISO 400 in a low contrast environment. In such an environment contrast based denoisers usually destroy a lot of detail. All the images below were processed with Rawshooter (all parameters set to the default values) and JPEG quality high. First the output of the original raw file:

Then I denoised the raw file with "RawImageClearer -pass2 -v -v -v -mode 2 -radius 1 -radius2 3 -mode2 1 FZ50.dng" and got the following output:

Although the same compression parameters were used, the JPEG output of the denoised raw file is much smaller. That shows quantitatively how much noise was taken out.
Next I did iterated the second pass twice, i.e. I used "RawImageClearer -pass2 -v -v -v -mode 2 -radius 1 -radius2 3 -steps2 2 -mode2 1 FZ50.dng", and got the following output:

Next I used SmartMedian also for the second pass, i.e. "RawImageClearer -pass2 -v -v -v -mode 2 -radius 1 -radius2 3 -mode2 2 FZ50.dng":

That created to much aliasing without taking out significantly more noise than with the first configuration. It is not recommended.
Finally I tried "RawImageClearer -pass2 -v -v -v -mode 2 -pixels 5 -radius 1 -radius2 3 -mode2 1 FZ50.dng", i.e. the default value 6 for the pixels variable was replaced by 5. Usually higher pixels values make RawImageClearer more aggressive but odd pixels values are more aggressive than even values, i.e. -pixels 2*n-1 is always more agressive than -pixels 2*n, but -pixels 2*n -1 is less aggressive than -pixels 2n + 2. I got the following output

Again there is an increase in aliasing. All in all I prefer the recommended setting "RawImageClearer -pass2 -v -v -v -mode 2 -radius 1 -radius2 3 -mode2 1 FZ50.dng".
The FZ50 sensor has the unpleasant property that the green filter for the odd sensor rows is different from that of the even rows. That causes artefacts, if demosaicing ignores this sensor property. Obviously the Rawshooter demosaicing ignores the green difference and RawImageClearer amplifies the artefacts somewhat. With Rawtherapee VNG4 demosaicing, which takes into account green differences, should be prefered against VNG demosaicing. However, VNG4 extracts less details than VNG. My own demosaicing method, which will be part of my upcoming image processing workflow, is very robust against substantial green differences and does not sacrifice detail.

[michaelbischof]

Hallo kassandro,

Thanks a lot for these nice tests, kassandro! I will try my best to supply further examples for optimization of the options that RawImageClearer uses.

Regards,

Michael Bischof