›

Noise Reduction done a half dozen different ways in PI?

John Williams
in PixInsight
Maybe I missed the right video that may put some overview commentary about which type of noise reduction process works best? Since the software evolves, it may be that certain of the noise reduction methods are still there because of prior users but a newer version is now the better to use. I understand that four are the most popular. MLT, MMT, TGVDenoise and MureDenoise.  I know Adam covers how to use all of them in the fundamentals videos. But, which one do you think is best to use for example in stacked and integrated linear single filter files before doing any combinations or before creating starless versions? Or, which one is best to use after creating an RGB stretched file?

Comments

  • BlockHeadBlockHead
    John,

    You will certainly find people who will answer your above question with a definite opinion- but I will not.
    The answer is it depends. How much noise? What is the size of the noise/grain/blobs? Is the noise mostly in the color? Is the noise mostly part of the sky (background) or is it part of the object as well? Will the reduction you do involve a mask? (I am getting tired of typing... lol)

    There are constraints of course. MureDenoise works on integrated (linear) data- and is a despeckle operation instead of a smoothing one. 

    My method has always been to leave a majority of the noise reduction to the non-linear images. MureDenoise is a (wonderfully) special exception. The BEST noise reduction is obtained by gathering long exposures and more images. :)

    If I *had* to answer your specific examples:

    1. Stacked images...MureDenoise... especially on the color data. Save big smoothing for later. If you smooth too much too early...you commit yourself to smooth patches that will not further smooth when non-linear later.
    2. Noise reduction doesn't really affect starless image production. The noise of the background (whatever method) stays the same. However the starless image you create..may indeed need aggressive noise reduction before blending it with other images.
    3. Given when I said above... you can guess my answer to the RGB question. MureDenoise if possible when linear...and then other things when non-linear. TGVDenoise is really hands downs better than MLT or MMT in most aspects for common noise reduction. 

    -the Blockhead
  • John Williams
    This is such a very helpful reply Adam! Thank you for putting this context to the processes. I expected your initial comments, "it depends" but your knowledge and experience with them comes through in such an important way to me as I'm new to this kind of photography post processing. Those subtle differences even in how and what they are attempting to do is excellent!

    Thank you very much for you guidence,
    John
  • John Williams
    Adam, in your detailed TGV video, near the end when you work on the image in linear form and show that local support is very useful especially in linear mode, I noticed you didn't use MLT first.  Does MLT work the same in linear mode for the one pixel level? Or would that task be already accomplished using MureDenoise instead?
  • BlockHeadBlockHead
    Yes, MLT works the same on linear images. It is just a "scale size" that is relatively independent of brightness. 
    MureDenoise could very well do the heavy lifting as far as an initial pass of noise-reduction. It depends on the data- but I suspect you are right.
    -the Blockhead
Sign In or Register to comment.