Good morning Davy,
with my DigitDia6000 I encounter exactly the same problem with ME. As you mention Cyberview I would conclude that you are also using a scanner from PIE. It is not the slide that is moving but the scan head is not able -- due to its mechanical / electronic design -- to accurately come back to the same scan position for the second scan. Therefore SF tries to align the images in software (v**s**n doesn't do that, CyberView for this very reason does not offer ME at all).
From my experience one gets better results when
REDUCING the allowed offsets, say to 10 or even 5 pixels. But even with this setting I do get double contours or unsharp results quite often in an extensive scan series. In the end it is trial and error since the deviating head position is random. Since the gain in dynamic range is only marginal I gave up with ME for good due to this problem.
It is unclear to me, why the SF alignment algorithm does not work reliably. Admittedly, one could imagine alignment difficulties with slides showing almost no contrast or motif (which should be a rare exception as it is not worth scanning such images in general). In other disciplines image alignment to a fraction of a pixel is a standard procedure (you might take a look at an example from medicine
https://sites.google.com/site/qingzongtseng/template-matching-ij-plugin#description or from astronomy
http://www.aanda.org/articles/aa/pdf/2010/05/aa12428-09.pdf. In the latter case the images making up the RGB-images had even been taken with different instruments with different image scales and optical distortions).
Best wishes
Hermann-Josef
