Quick Noise Question

R

Radar

Well-known member
Gday again Pteam.

This may be another very noob question, but I was just curious as to what the best way to remove noise from a film image would be?

Pixinsight seems to have more than one way of doing this. SGBNR, Wavelets and SCNR.

In one of your tutorials -

"Divide and Conquer Noise Reduction Strategy:Wavelets + SGBNR"

The tutorial refers to a black and white CCD image. But can the tutorial still be applied to a colour film image?

I tried using the Wavelet function anyway on my colour image. I set the 'Count' to four (as in the tutorial) hoping that this would show me four noisy to less noisy layers (as in the tutorial) but I get a message when I hit the preview key saying "no preview defined".

I have a funny feeling that I'm way off with this, lol, but i'm enjoying the learning curve.

And finally, just an observation when I was playing with PixInsight.

Whilst experimenting, I took a close look at my image using the three seperate colour channels. I was surprised to see that the blue channel held most of the noise. So I pulled out the Histogram and reduced the blue curve slightly, and my image seemed to be much cleaner straight away. This is probably a very crude way of reducing noise. My question here is trivial but I'm just curious, does noise usually fall into the blue end of the spectrum for some reason or may this have been caused by the film I was using (Fuji)?

Thanks Again Pteam
 
Hi Radar

[...]
>The tutorial refers to a black and white CCD image. But can the tutorial still be applied to a colour film image?

Yes, of course. But, you can add SCNR as a previous step to the "noise killer combo" ;-).
I ussually set SCNR to Neutral Average, over the Green channel, or to Neutral Max if there are some real colors affected. I think that you'll find SCNR quite easy to use.

With Wavelets are few differences. The only advice with color images is to disable the "Use Luminance" checkbox, becouse the high frecuency noise usually lives in each RGB channel differently, and is not related to the luminance. Other than that, is just the same.
SGBNR algo works a bit different. We recommend to use the L|ab color space. I think that this was covered in deep in the documentation (help files). Again, other than that, for each channels the same is applied as in the tutorial.
The reason to use L|ab instead of RGB is that this time noise at large scales is more easyly removed working the luminance and chrominance with a different set of paramethers. The human eye has a very poor accuraccy to detect variations in chrominance, so it is not a very critical adjustement. Just remember to use a mask to avoid the loss of color saturation in the stars.

>I tried using the Wavelet function anyway on my colour image. >I set the 'Count' to four (as in the tutorial) hoping that this >would show me four noisy to less noisy layers (as in the >tutorial) but I get a message when I hit the preview key >saying "no preview defined".

You have to define a preview. Previews are "copies" of the image, were you can try the processes, without affecting them, and always working over the original data. Read the documentation to lear more about them. They are very rich objets, and will help you a lot, specially when you have to save time while fine tunning some paramethers.

>Whilst experimenting, I took a close look at my image using >the three seperate colour channels. I was surprised to see >that the blue channel held most of the noise. So I pulled out >the Histogram and reduced the blue curve slightly, and my >image seemed to be much cleaner straight away. This is >probably a very crude way of reducing noise. My question here >is trivial but I'm just curious, does noise usually fall into the >blue end of the spectrum for some reason or may this have >been caused by the film I was using (Fuji)?

I would say that probably it was the film, accompanied with some factor like development, and a higher intensity of this channel, so the noise had a larger amplitude.
I had found that with some films the noise pattern changes a lot between each channel. For example, Provia 400F had "black noise" in the blue channel (it means, in a pure background there were some black holes) while the red channel had some bright features. In some very weird cases I had to work each Lab channel separatelly to obtain good enought results.

Well, you know that each image is a whole new world. And life wouldn' be as funny if it wasn't that way ;-)
 
You must log in or register to reply here.
Back
Top