Hi Iain,

just seen your new filter and tried immediately. It looks very good!

Input: old noisy image

2nd input: same image denoised


output of your new filter


nice result!
I will continue tomorrow with new tests.
Thanks, Iain

a new experiment, Iain:

first note: the input images were originally quite big (2592*4608)
with these dimensions I get an execution error (first step), which says:
GIMP Error
Calling error for procedure 'gimp-drawable-width':
Procedure 'gimp-drawable-width' has been called with an invalid ID for argument 'drawable'. Most likely a plug-in is trying to work on a layer that doesn't exist any longer.

GIMP Error
Calling error for procedure 'gimp-drawable-height':
Procedure 'gimp-drawable-height' has been called with an invalid ID for argument 'drawable'. Most likely a plug-in is trying to work on a layer that doesn't exist any longer.

GIMP Error
Calling error for procedure 'gimp-drawable-bpp':
Procedure 'gimp-drawable-bpp' has been called with an invalid ID for argument 'drawable'. Most likely a plug-in is trying to work on a layer that doesn't exist any longer.

tried many times, same result

hence, I decided to downscale both
the filter ran correctly, no errors

input original:


input denoised (anisotropic):


output new filter:


the comparison is "subtle" with this experiment, because I started with a more "acceptable" image, nevertheless we can appreciate how the face (look at the glasses) has been restored in many details; good filter, Iain!

Thanks for testing. I always worry it's not going to work for other people.

I'd also like to point out that the filter removes some of the artifacts from the smoothing. If you look at the white areas of the stained glass, the smoothed version sometimes has a subtle dark halo in the white area which my filter removes (by coincidence, not by design). It's most noticeable on the stylized clouds at the top.

I like your filter, Iain!
For me, it works and works well.
Thanks

(btw: did you check why I got that strange gimp error when using the original big images?)

Sorry, I got distracted by the results.

The filter has no output to gimp when analyzing the noise (pass 1) so you need to set the filter to 'new layers' to avoid the error. Or it tries to remove your layers.

At least that works for me when I tried and image of 2592*4608

Thanks, Iain!
Yes, with "new active layer(s)" as output the filter works without any error.

(still for me it remains a mistery while the interface gives that error only if the images are large, with small it worked without any error... boh!)

Just added MS_Patch_NR to my testing folder, which is an excellent noise reduction to do before Savenoiseprint. Have fun.

OK, Iain
Back to my attempts:

original image (taken only the bottom part, not resized, not edited)


outcome after applying the new MS_patch_NR (parms=6,12,18,6,12)


outcome after applying avenoiseprint (on the two images, denoised on top of the original)


as one can see the second filter has restored the brick structure that was partially "destroyed" by the denoising filter; hence: EXCELLENT

A few notes:
1-I selected those parameters on the denoising filter to get an acceptable van at the bottom
2-it would be nice to have one (better: three) presets (if one: to use on photos with average noise; if three: to use on photos having light, average, strong noise)
I know that trying to find the correct parameters values is almost totally dependant on the photo itself and on the user's taste, but IMHO a filter should have defaults, which the user can anyhow modify, but which can be reset after each attempt)

Thanks Iain for your always excellent filters!

My first attempt at SAVENOISEPRINT is part of the process of trying to make something of very poor quality family photographs of nearly 100 years old.
This is the original scanned photograph.
I applied a contrast mask, a curve adjustment to the base layer, then merged down.
Duplicated the base layer, then Iain's noise reduction, used savenoiseprint, reduced opacity to 59% and merged layers.
Here is the result. (I'm sure others can do much better!)

my attempt, David
applied many "personal" filters (G'MIC based) to create the noisy base; out=A
then applied IAIN's new denoise; out=B
then applied IAIN's new savenoiseprint on (B) over (A); out=C
then applied again IAIN's savenoiseprint on (C) over (A)
final outcome



IDK if somehow acceptable

Good result - but I did say I was certain others could produce better results than I!!

Very nice work David

Thanks David.

another attempt on using the two new Iain's filters on an image taken by myself.

original (simply reduced by cutting out the surrounding area, otherwise the image was too big; but no modification applied)


after applying the new denoising filter


after applying savenoiseprint on the two images above


details have been restored (look at the portal of the church); well done, Iain!

on the Iain's outcome, I then applied some personal filters -still using GMIC- to improve -to my taste- colours
here it is



but this has nothing to do with the topic, just for my pleasure

I like your outcome Dinasset. I find so much improvement on the photo in definition, colours, light... Well done!

thanks Issabella, glad to know your taste is close to mine
(in case we are in the near future again co-teamer for some new plug-in/script, it's even more important, hehehe...)

I made another attempt on removing noise-restoring details

the original is almost bad


as first step, I used G'MIC smooth-skin (+ removing red-eyes simply desaturating)


then I used G'MIC smooth-wavelets (instead of Iain new MS patch nr)


(continuing...)

finally I run the new Iain's filter SAVENOISEPRINT having the last two outcomes previously shown an input layers
here the result



my comments:
A) details on the hair have been restored (good!)
B) details on the pull structure have not

could this be a useful info for the author (Iain)?

Thanks, dinasset. I appreciate you taking the time to test the filters.

I've been testing the filter on very lightly processed raw images and your images seem to be out of camera jpegs. This means that they already have a little bit of noise reduction on them so some of the details are already missing. It's helpful to see the results for completely different images, even if those results aren't quiet as good as I would like.

Another problem to overcome is the different amount of noise in the different tonal ranges of the image. I have used 'gamma' to bias the noise reduction towards the shadows (or not). savenoiseprint will work best if the gamma setting matches the gamma setting of the original noise reduction. If no gamma setting exists, use gamma = 1.

Another thing to consider is that once you have used savenoiseprint you can use a layer mask in GIMP to limit the visibility of the detail recovery to certain areas, which might mean you can use a lower setting than you would have before.