PixInsight Forum (historical)
PixInsight => Wish List => Topic started by: Warhen on 2014 June 03 13:16:38
-
Whilst methinks I can hear Juan whispering back in my ear 'fast food, fast food'… there are at least two softwares that I'm aware of which have the feature of easily and quickly removing or splitting stars from an image: Straton and Images Plus. While I understand this can be accomplished in PI, it is not an automated or quick affair. I'd really enjoy having this feature. Thank you!
-
Me too ;)
Jokes apart, removing stars is a topic that interests me a lot. We have being doing this with different methods for a really long time (I remember a post by Wei-Hao in the APML, quite a lot years ago, that started this, at least from my perspective). I've being reading a lot these days about multiscale and HDR problems, and I have in mind a few algorithms that may help finding a good decomposition of stars and the image. My idea at the end is to be able to separate noise, stars and structures at several scales.
So, this is on my medium-term do-to list. Right now I'm working on incorporating poisson data models to the deconvolution problem, and hopefully achieve a better ring suppresion (with the tgv regularization). TGVRestoration right now works pretty well for gaussian noise (works wonderfully with planetary or lunar data), and have some deringing schemes included, but I'm aiming for something much better for deep sky targets.
-
Right now I'm working on incorporating poisson data models to the deconvolution problem ...
FYI, if you have not seen it, Luisier has a nice Poisson+Gaussian denoiser (http://bigwww.epfl.ch/algorithms/denoise/). I use it to denoise subframes (http://mschuster.zenfolio.com/blog/2012/12/subframe-denoising). Maybe this scheme would be applicable to decon? There are a few issues, one being this: registration, integration, and drizzle introduce correlations between pixels locally. This somewhat invalidates the Poisson+Gaussian per pixel noise model.
Mike
-
FYI, if you have not seen it, Luisier has a nice Poisson+Gaussian denoiser (http://bigwww.epfl.ch/algorithms/denoise/). I use it to denoise subframes (http://mschuster.zenfolio.com/blog/2012/12/subframe-denoising).
This is very interesting. I think we could integrate this denoising routine as an option in our ImageCalibration tool.
-
Thank you Carlos! I think that would be a welcome addition. Although I understand PixInsight fairly well, I'm always a fan of ease and whether accomplished semi-automatically or through manual steps, being able to treat an image sans stars, and conversely the stars alone would be very beneficial to me.
-
This is very interesting. I think we could integrate this denoising routine as an option in our ImageCalibration tool.
The Luisier denoiser was designed for single frames (life sciences, moving targets) when no multi-frame integration is possible. When integrating/drizzling denoising post integration is better of course. For single frames, when I am just surveying or framing a target, it works great. I can share my own Matlab implementation of Luisier. Sensor read noise and gain are required parameters. But an ImageCalibration option may make less sense for your Astro customers.
Mike
-
Hi Mike
Have you tried a comparison between that method, and using the MMT or MLT?
-
Hi Carlos,
Only quick comparisons, not enough for an answer. I think an expert with PixInsight denoising tools could do better than me. Subtracting the input and output to look for artifacts helps.
Mike
-
I'd really enjoy having this feature. Thank you!
+1 from me.