Here is an alternative method to get rid of those star artefacts. It builds on an effect that I already discovered during the original development of the process
http://pixinsight.com/forum/index.php?topic=3283.msg22572#msg22572 : Images after histogram transform have fewer of those star artefact problems. The underlying idea is this:
- In astrophotos, we are usually more interested in the faint features of the sky, not the bright stars. This is why we often do a non-linear histogram transform during the processing.
- While the data is still linear, bright stars have much higher values than faint features. In Kerry's example, stars have values of up to 0.99, while the nebula is around 0.002, and the background around 0.0007. Due to their high values, even minor brightness differences in stars cause relatively major adjustments to brightness during GMM. Example: a 1 percent calibration error of the nebula causes a difference of 0.00002, while a star causes 0.01. This is the root cause for the star artefacts.
- If we do a brightness transformation similar to HistogramTransform on the images before GMM, and transform them back afterwards, star artefact should become much less of a problem.
The HistogramTransform used by PI is discussed in detail in
http://pixinsight.com/doc/tools/HistogramTransformation/HistogramTransformation.html#introduction_002 (section Midtones balance). The underlying midtones transfer function MTF has a number of nice properties:
- it transforms range [0,1] into range [0,1], and transforms 0->0 and 1->1
- MTF(1-m,x) is the inverse function to MTF(m,x)
- It is implemented in PixelMath.
So here is the procedure to produce artefact free GMM results without manual fixes:
- Determine the midtones parameter m. This can be done by looking at one of the contributing images, and shifting the mid slider (not the other two) of STF until we see a decent value. The actual m value is displayed as a bubble, and can also be retrieved using the bottom left button to display the STF parameters. The exact value of m is not critical. Screenshot 1 show the STF on the top right, the parameter dialog below, and the STFed image on the top left.
- We use this midtones parameter to transform all images. Open PixelMath, enter "MTF(m,$T)" as the formula (screenshot 1, center), and deactivate Rescale. Apply it to all those images you are going to process in GMM, for example by using an ImageContainer (screenshot 1, right). One of the transformed images without an applied STF can be seen on the bottom left.
- Apply GMM to the transformed images (screenshot 2, top right). Re-transform the resulting image using PixelMath with formula "MTF(1.0-m,$T)" (screenshot 2, bottom right). The resulting image can be seen on the left: No star artefacts at all.
This procedure works very well, but so far has only been tested with Kerry's data. If others can confirm that it is helpful for their mosaics, we could implement it as part of the GMM process (and maybe as an alternative to the Feather Radius parameter). Let me know your experiences.
Georg
