I've made a new noise reduction filter and wrote a blog post about it. I can't post the link, it tells me it is too spammy. So I've posted the text from the post below. I hope you like the filter.



With noise reduction software it is a case of picking your poison. Well I hope this poison at least offers a pleasant taste!

Iain’s Noise Reduction is a filter for the G’MIC plug-in for GIMP 2.8. It is currently found in the Testing->Iain Fergusson folder.



Control descriptions

Luma NR

This adjusts the amount of luma noise reduction. If the control is set to 0 this step is skipped and processing will be faster.

Chroma NR

This adjusts the amount of colour noise reduction. If the control is set to 0 this step is skipped and processing will be faster.

Despeckle

This reduces the amount ‘salt-and-pepper’ noise. It is performed just before the luma noise reduction, so adjustments made here will affect the luma noise reduction.

If the luma noise reduction is set to 0, the despeckle step is also skipped.

When set to level 4, the main luma noise reduction is adjusted to skip removal of fine noise because it should have been taken care of by the despeckle algorithm. This should speed up the denoising.

Adjust noise reduction by tone

This allows you to adjust the amount of noise reduction in the shadows and highlights.

Recover Details

The detail recovery step works with the information removed by the luma noise reduction and recovers details that are stronger than a certain threshold. There are three options recovering details:

Do not Recover Details
Recover Details
Recover Details as separate layer – this will output the recovered details to a separate layer that can be combined with other noise reduction layer using ‘grain merge’ blending mode in GIMP will produce the final image. This option allows the user to apply masks and further processing to the details in GIMP.
Recovering details is the slowest part of this filter, you will have to have patience.

The recovery amount slider sets the overall recovery amount. At low values only the strongest details are recovered. As the slider is raised more subtle details are revealed, but there is a risk that the recovery algorithm will see false details in noisy areas. Setting the slider to 10 will recover all details including noise.


Noisy image before processing.

Details on the red feather are recovered well in the lower right corner, but some false details appear in the shadow area of the feather.
Adjust detail recovery by size

These controls allow the user to adjust the detail recovery based on the size of the details. This is useful to control the appearance of false details

Detail Emphasis

This allows the user to increase the strength of the details so that they are more obvious. This often allows the user to use a lower recovery threshold and reduce false details.

Other considerations

Although the detail recovery is quite good, it can produce artefacts. False details has already been mentioned, another artefacts is detail bleed. Sometimes you will see that recovered pattern actually extend past the boundaries of the original pattern and ‘bleeds’ into other areas. This is because the detail recovery looks for patterns in blocks, and if a pattern is found, with certain settings, the whole block is filled with the pattern, even if the pattern did not fill the entire block originally.

Another artefact is ‘ringing’. If the original luma noise reduction softens strong edges too much the detail recovery algorithm can produce parallel lines running along side the edge.


Noisy image.

Ringing is visible as parallel lines running beside the edges.
Sharpen edges

This sharpens the luma channel, but does not sharpen the details.

Preview Shows

Normal Output
Luma Noise – shows an amplified view of what is being remove from the luma channel
Chroma Noise – shows an amplified view of what is being removed from the chroma channel
Details – shows an amplified view of what is being recovered in the detail recovery step.
Notes on using the filter

Successful detail recovery is a balance of the luma noise reduction setting (including despeckle) and the detail recovery settings. The detail recovery section works with the information removed in the luma noise reduction section, so adjusting the luma noise reduction will affect the amount of details recovered. Generally the luma noise reduction should be set to slightly over-smooth the image so that the detail recovery section has enough information to detect the details. Small adjustments of the luma noise reduction amount can improve the detail recovery, so a bit of fine tuning is required for the best results.

thanks Iain !

i guess this is the correct link ?

http://iainisbald.wordpress.com/2013/02 ... /#more-619

Excellent noise reduction filter! Thanks Iain

Yes that's it. Thanks!

Iain, After one more post and you'll be able to add files, links, etc.

I recently discovered this G'MIC filter and it is excellent. The edges and details are preserved quite well, something difficult to obtain with other noise reduction methods.
I'm not yet acquainted with the Recover Details options, but after applying the Iain Noise reduction filter, GIMP Unsharp Mask works pretty well too.
Thank you very much Iain!

Edit:
Ohhh!
Looking your filters folder in G'MIC Testing now I have found the filter "Fill Holes". With the "Fill Light colours" option enabled I can reduce scratches and hot dots noise in old photos and scanned negatives. It's fantastic!
Iain, you are a master in reducing noise issues!!!

For example, try "Fill Holes" filter with the original fluffybunny image here:
viewtopic.php?f=18&t=8017

Apply the filter with the next options:
Morph radius: 8
Edge radius: 50
Close radius: 0
Check "Fill light colours" option.

See the result in scratching on the floor. It's magical
Then if you want you apply GIMP unsharp mask for more detail.

i have almost no experience with photo manipulation, so i tried the filter on Fb's image - its indeed magical !
how did you find the values, YAFU ?
just playing around or is there a method ?

Hello Esper.
I have to repair hundreds of old negatives that I scanned, and I am researching different "automatic" methods of noise reduction. Manually fix each of them would be an endless job.
The main problems I have are related to the preservation of edges and details in noise reduction filters, and with scratches and hard/hot bright little stains/dots. Scratches and bright stains/dots are hard to automatically repair because the filter must smartly determine in the image what is scratches or noise dot, and what is not.
I researched different methods. Also "I borrowed" Photoshop to compare its filters with GIMP filters. I think GIMP filters are as good as those in Photoshop. But "Iain" filters in G'MIC outperform any filter I've tried. They are excellent!

Regarding how I found the values for that photo, I had sensed that this filter could be used to what I was needing, and just playing with values and watching the G'MIC "preview" I got the result. That's why it is so important the "preview" feature in filters/plugins!!!

so do you think understanding what each feature of a G'MIC filter does is unimportant ?

i mostly feel im very much in the dark, when i try to change values and options
its G'MICs greatest weakness and its great Rononours finally does his video tutorials

Of course it is also important to know what each feature does. I admit often I have no idea what it does some item/value in G'MIC filters, so I must experiment to know what it does. In some filters there is a small explanation of how to use it. As is the case with "Fill Holes" where you can read a little explanation.

I have seen that this filter can be useful in old or black and white photos (which is what I need). It is not so good with sharp colorful images.

Thanks for the feedback. I really appreciate it.

The recover details portion of the filter is very slow so if you get good results without it don't use it. The detail recovery tries to find patterns in the noise that was removed in the first stage. Noise is usually random, so a pattern is probably not noise.

Fill Holes was actually the result of a request someone had to refine image masks and not intended for noise reduction at all!

Hahaha. Would not be the first time in history that great things/usages are discovered by accident

So maybe you can redesign and improve the filter to remove scratches and bright spots of old photos or negatives. An automated method it is very useful and hard to find.

Automated processes are very difficult. I keep it in mind though.

Edit: if you upload a full res version one of your scans I might have a better chance of figuring it out.

I'm referring to the typical scratches and bright spots that appear in old photos or scanned negative. For example these:
http://www.digitalcameraworld.com/2013/ ... y-pictures
http://howtoscan.ca/photo-scanning-tip- ... photoshop/

I know it is very difficult, but it would be great a method that automatically detect scratches, select it and apply a smart fill method like this:
viewtopic.php?f=23&t=8020

I have sent you two of my digitized negatives to your email (the email that appears on your blog).

Thanks!

I got your photos. It looks really tricky, but I've put your pictures in a folder so I can have ago at auto scratch repair when I feel inspired.

Thank you for having this in mind. I know this issue related to scratches autodetection is not easy to solve.
Regards.