Need help with TGVDENOISE . . .

[Terry Danks]

I have had pretty disappointing results with TGVDENOISE. Considering how enthusiastically it has been received, I must be doing something wrong.
Attached are images of my attempts on a noisy image enlarged to show detail.
In order to preempt any feeling that I need more iterations, I upped the default 100 to 1000.
Other than that, I used the defaults. Playing with the parameters made some differences but I never got a result I would call satisfactory.
I have not used “Local Support” in the tool, because I really don't know how.
Assistance would be appreciated.
Full res. image at http://danks.netfirms.com/Compare.jpg

[NGC7789]

I too am trying to figure out the tool. Have you read Juan's processing example? Are you applying the tool at the linear stage or after stretching? The defaults are for stretched images. If using on linear they are too high by an order of magnitude. My understanding from various posts is that the "sweet" spot for iterations is between 300 and 500. 100 is good to get faster results while exploring gross settings.

[Carlos Milovic]

Hi Terry

The first parameter that you have to fine tune is the "edge protection". This may be the most critical parameter here. A good way to find a starting point is to measure the standar deviation of the background, and use this. If you see that true edges are lost, decrease the value. Otherwise, increase it a little bit. Then, play with the strength parameter. I believe that the current default parameter is too high, so I would try it in the 1-2 range first.
If you enable the local support, the first thing to do is to decrease the strength, working in the 0.1-1 range as a starting point. By default, if you do not specify an image to be used as the local support, the luminance is used.
I'm attaching a screenshot of some parameters that worked quite well (you have to fine tune a bit further) over your jpeg.
I also would try reducing the noise in the linear stage, and see if it is easier to preserv details inside the galaxy there.

[Terry Danks]

Many thanks 7789. I was indeed running it on linear images. 

I used your settings, Carlos and just so much better!  I seem to stumble often over what to run on linear data and what to use on stretched frames.

Many thanks . . . think I am on the right track now.

Terry

[NGC7789]

Carlos, forgive me if this should be obvious but what is the best way to measure the standard deviation of the background?

[Carlos Milovic]

One easy way it to create a preview on the background, and then use the Statistics process to measure the std dev. Probably it will be a higher than the std dev of the noise, since we have an underlying signal with its own variations (flux variations in the background, small stars, etc) but is a good starting point.

[Warhen]

I just wanted to comment on how very impressed I am with TGVD Carlos et al. I am having a blast watching mediocre/noisy 'final' images being rescued by this amazing process. Congrats!! I'm documenting the process this week at IP4AP. Well done!!!

[pfile]

yes, TGVDenoise is really awesome, thank you for this tool.

[Carlos Milovic]

Thank you I would love to see the presentation. Is it possible?