XnView Software

Ximagic Denoiser is a new photoshop plugin which includes the most known algorithms for noise reducction. It includes the following filters

- Median
- Bilateral
- Wavelets
- Anisotropic diffusion
- Non local Means

Using masks and different filters the noise reduction could be adjusted to the image characteristics.

The plugin is available at http://www.ximagic.com

This plugin is also compatible with Paint Shop Pro, Photo-Paint, The Gimp, IrfanView, XnView and others.

Hello,
Thanks for the link brodriguez,
You can see here too for more Plug-Ins:
http://newsgroup.xnview.com/viewtopic.p ... sc&start=0

Replies from
http://newsgroup.xnview.com/viewtopic.p ... 77&p=76067

Drahken wrote:Personally I don't have much use for dithering functions, but I find your denoiser plugin extremely useful.

Thanks

Drahken wrote: There's one annoyance about it though, scrolling the preview window. It'd really be nice if you could drag the image around instead of using the scrollbars, but I would at least like to have the image scroll in real time when you use the scrollbars. The way it is right now you have to take a guess at the right position, then let go & see if it's right, then scroll again & stop, repeat ad nauseum.
Other than that, the denoiser is a great plugin, even capable of making a decent looking image out of an over-compressed jpeg.

I agree that the preview move control is far from the best, I'll try to enhance it.
There is a new version published (v 2.1.2) which have scrollbar and image movement synchronized in real time, as you ask for.

Drahken wrote:edit: One other thing, it'd really be good if you made it so that we can type values into the various fields instead of having to use the up/down buttons.

I added Drag and text editing to the TODO list.

eL_PuSHeR wrote:I have a question about the denoiser. Can it be used to demoire a scanned image? If so, what combination of filters should I use?.

I suspect it can be used for this purpose, but I never test it.
If you send me the image I'll give a try.

Any scanned image will do. Here, I have scanned one audio cover at 150dpi to save bandwidth. I am using lossless compression to avoid taking JPEG lossy compression into account.



On a first test I have done (I was pretty sure demoiring involved some sort of Gaussian filter, Bilateral Gauss gave me very good results (at standart stock defaults), way even better than Paint Shop Pro Gauss filter).

Another test (before I go to work). Two stages:

State one: Non Local Means: 2, 16, 5
Stage two: Bilateral Gauss: 50, 2 or 120, 20 (if you prefer more banding removal).

I have always used ALL CHANNELS, because I don't know how they work. I think one of the is Luminance and the other Chroma, but I am not sure. It won't serve me either.

If the "use ycbcr" box is unchecked, then one channel is for red, one for blue, one for green. If the box is checked, then the y channel is for the luminance & the other 2 are for color (not sure exactly which does what).
If you want to see if you're better off working on just 1 or 2 channels instead of all, just check the box next to "ch1", "ch2" "ch3" under "view channel", you'll see the channel you selected in the preview window. If you see that channel 2 has noise but channels 1 & 3 don't, then you're better off only working with channel 2.

I got good results each with;
bilatteral gauss spatial sigma 100, range sigma 20
wavelets dwt noise- 650 hard quality 2 (also got seemingly identical results with the cwt wavelets, same settings).
I got the best results though with dct: noise sigma 1650, cell size 4.
In all cases, I applied it to all channels, because the noise seemed to affect all of them equally (in both rgb & ycrcb modes).

With that DCT values, you manage to remove all patterns. The output is good but faces on the left side tend to lose too much detail.

Yeah, there's always a tradeoff. You can preserve more detail by using a lower setting for the noise sigma.

On the other hand, if you use an image editing prog (I noticed you mentioned PSP x in another thread, PSP7 is my fav editor), you can simply select the faces & invert the selection, then run the denoiser are everything except the faces. Then re-invert the selection & run the denoiser on just the faces, but using more detail preserving settings. (You could of course do the faces first & only invert the selection once, doesn't really matter.)

The problem in this image is that it was printed with "artistic" patterns. For example see the torch.

There are 3 band patterns. Those patterns are preserved by the more advanced denoise methods because those methods try to preserve everything but noise.

To remove this patterns with denoise methods you need to use it as if there is a very high noise and you'll lose details. In this situation, masks are the only way to preserve some areas (faces) from being blurred.

Because those patterns are band-like, can be removed with directional blurs (or movement blur) orthogonal to the bands. Of course this could remove some details but is by far better than using a 2D blur.

I apply 3 PSP movement blurs at 0º (intensity 5) 73º (intensity 7) and 107º (intensity 7) and focus after each blur.

I get the following result:



After that you could apply the denoise. For example NLM 1x7x3



CWT gives also good results.

The difficult part is to remove the patterns, after that the denoising is easier.

Hello. I have one question that has always intrigued me. Why the Denoiser plugin installation (from .MSI package) does a pause when it say "Please wait...". Both the Quantizer and Graydither install faster.

I suppose you mean the .msi of the MP version of the Denoiser, it take more time because it install the multiprocessor libraries.

Yes, sorry. I meant the MSI version. Thank you for clarify things up.

Regards.

Why bother with the installer at all? Just grab the 8bf file from the download page & stick it in xnview's plugins folder.

There is no standalone 8bf file for the MULTIPROCESSOR version.

Starting with version 2.9.0 the installer is no longer needed for the MP version.

Yes. That's true. I find it very convenient.