Hi Fabio,
Thank you for uploading the bias frames. We have two issues here: your bias frames are incorrect (IMO), and there is (was) a bug in my implementation of Winsorized sigma clipping.
Let's see the data problem first. Open one of your bias frames and apply the following PixelMath expression:
$T == 0
You'll be surprised to know that about one half of the total pixels are pure black. This happens with the whole set of 12 bias frames, and I think it denotes that something went wrong when you acquired them.
Now the bug. The Winsorized sigma clipping (WSC) algorithm doesn't work well when there are many zeros in a pixel stack. Instead of rejecting outlier pixels based on a simple yes/no criterion, WSC attempts to preserve more data by replacing extreme values with neighbor values at each iteration. However, the excess of zeros violates some assumptions about the statistical distribution in the pixel stack, which leads to very long iteration loops and hence slows down the whole process. Eventually, one of these loops can go crazy and cause a segmentation fault.
An easy way to check how this problem works is by enabling the
clip low range option. With this option enabled, WSC works fast and without problems with your biases.
The bug is now fixed, so ImageIntegration no longer crashes when WSC is applied to data with a "many zeros" problem. Later today I'll release an update with a new version of this module for all platforms. However, the slowdown problem can only be fixed by enabling "clip low range". I'll release also a new version of the BatchPreprocessing script where this option will be enabled by default for integration of bias and dark frames.
