Here I'll add an example of what's happening:
This is an artificial (CGI) image of the center of a resolution test target:
I open the file with PixInsight 1.8RC1, and disable Color Management. Color Managed images show the same issue, so it doesn't seem relevant in this case.
I also open the following image of a Gaussian Blur with a 1.0 radius PSF, also with Color Management disabled:
This Gaussian Blur PSF will be used to first Convolve the image with, and then to Deconvolve the blurred image. I use
Process|Convolution|Convolution and on the Image tab I select the 'PSF010-pointsampled' image as filter image, and Apply the convolution to the 'SinusoidalSiemensStar_500px' image. Sofar, all is well.
I now open the Process|Deconvolution|Deconvolution process on the External PSF tab,
and select the 'PSF010_pointsampled' image as kernel.
I used 100 iterations to improve the Richardson-Lucy restoration, and make the issue a bit more visible, although it also shows up at 10 iterrations. Since this is a CGI image source there is no noise other than quantization rounding errors, and therefore Wavelet Regularization was switched off. In regular camera images with noise, and Regularization switched on, the effect is the same.
This produced the following result, with the same magenta artifacts I see on other images:
Do note, that PixInsight 1.7 didn't produce this artifact with the same settings. The only way to avoid it in 1.8RC1 is to target the RGB/K components individually instead of the default Luminance (CIE Y), or first create a separate CIE L channel, deconvolve that, and recombine it with the original RGB image. Both 'workarounds' are less efficient, obviously.
Hope that helps to identify the cause.
Kind regards,
Bart
