TGVDenoise help

[blackdragon72]

I get very patchy background after TGVDenoise though the actual image looks good. What is the issue and how to decide the settings of edge protection and strength?

[Carlos Milovic]

Are you working over linear data, with an STF enabled?

[blackdragon72]

Yes.

I just want to know this is not artifact, since it will show up after HT too if I stretch a little bit more.

[Carlos Milovic]

No, this is the normal behaviour, if you set the incorrect parameters.

Non linear images (for which the default parameters were set) are very different from linear ones. The most critical parameter is the edge protection. Use the standard deviation of a sample of the background as a starting point, and play with the values until you see no data loss, while the noise is effectivelly supresed. Then change the strength parameter to balance the noise reduction and the original data. If you are using a local support (almost mandatory for linear images), people feel that copying the values of the STF to the support yields good results.

[blackdragon72]

"until you see no data loss, while the noise is effectivelly supresed"

This is a tough one.

When I start with stdDev, I see rings around stars, so I know I need to reduce edge protection. But I dont know when it is "right".
Do you consider the image in my first post as "right" edge protection , where I can play with strength?

[Carlos Milovic]

I think that you have to protect edges more. Also decrease the strength.
Use a small preview, and try the parameters there. Use few iterations there (100 or 200), to get the right parameters. Then, try a large number of iterations (500 or more). Just practice. See how the image varies with different settings.

[blackdragon72]

I did more experiments.

Here is my observation:
- they do seem real, because if I look at the original image, I can see dark patch here and there, and they correspond the patchy background after TGV well.
- I can improve background, but I can not get very flat background.
- I compare TGV and Atrous noise reduction. Atrous background is flatter, but it does show the similar pattern, and more details get lost. On the other hand, TGV preserve more details with similar (if not better) noise reduction, but background looks patchy.

So it looks like TGV does real good job. The issue could be with the image itself.

Is it common that CCD or DSLR has uneven background? Will dithering help in this case?

[Carlos Milovic]

To smooth the background more, increase the number of itereations. Try 1000.