Hi Robert,

thanks juan. we've seen several methods for background noise reduction: acdnr, wavelets and greycstoration. is there a rule of thumb for which one is applicable to what situation?

Just as a rule of thumb —which means that you can always find examples where this is not applicable:

GREYCstoration is an extremely powerful noise reduction tool (perhaps the best one IMO) for daylight images. For deep sky astrophotography it works well during the final stages of post processing, as a finishing tool to be applied cautiously, but I don't recommend using it otherwise. When applying it, watch for possible artifacts generated as a result of excessive anisotropic filter response. This algorithm (at least in its present implementation) doesn't work for linear data.

ACDNR has been our noise reduction workhorse for a long time. When properly applied, ACDNR yields excellent results for nonlinear deep sky images. However, there are many cases where applying it correctly isn't an easy or obvious task. This happens when the noise must be reduced at a wide range of dimensional scales. In these cases, several applications are necessary with increasing filter size and mask protection. This can lead to manual procedures requiring a lot of experience and trial error work.

I designed the current version of the ACDNR algorithm in 2005-2006. I think it's time for a serious revamping. New noise reduction tools are now necessary, with more efficient algorithms and implementations. So don't be surprised if ACDNR gets moved to the Obsolete category in a future release, in the short-medium term, and replaced with new, more efficient noise reduction tools.

Finally, we have ATrousWaveletTransform. I've just put a good example of what this tool can do for noise reduction. As usual, the multiscale paradigm provides new and powerful ways of facing old problems. In this case, noise reduction becomes simplified because wavelet layers allow us to isolate the noise at different dimensional scales in a natural way. With just two or three simple and easy-to-understand parameters we can attack the noise very efficiently, with the added bonus that it works equally well on linear and nonlinear data. I recommend trying wavelet-based noise reduction in the first place for deep sky images. When no good results can be achieved with wavelets —no algorithm or tool works in all situations—, ACDNR remains as a solid fallback option.