Hi Andrés,
I am sorry, I should have answered this post before. This is actually not a bug, but a feature of SA, so it won't be "fixed".
The problem here is in the origin of coordinates used by SA to refer star positions. Let's consider each pixel as a square of fixed dimensions. SA's star detection routine works under the assumption that the origin of coordinates is at the center of the pixel located at the upper left corner of the image. Y coordinates grow from top to bottom and X coordinates grow from left to right. Unfortunately, this origin is different from the origin used by the rest of the platform, where coordinates are measured from the top left corner of the top left pixel. However, it is necessary for SA in order to support specular transformations---as the ability to automatically correct for meridian flips---, so it can't be changed without introducing significant (and ugly) hacks in the whole SA process.
Of course, a suitable solution to this problem must be found. The solution that I have designed consists of adding two new floating point parameters to the StarAlignment process: inputOffsetX and inputOffsetY. The values of these parameters will be added to the reference X and Y coordinates, respectively, when the reference positions come from a CSV file. The default values will be zero. In your script, you'll just have to assign -0.5 to both parameters in order to adapt your positional data to SA's internal coordinate system.
Let me know if this solution is valid for you. These new parameters will be available in the next release of the ImageRegistration module. I can have it ready in a couple days.
