This is hard to be answered. In principle, size is irrelevant because it may be caused (elicited) by the size of a writing box or rectangle on the form. On the other hand, human signature-verification experts know that some individuals always write small, others always sign with a large signature pattern. So (homogeneous, not anamorphic) size normalization is good, but do not forget the original size of the ink pattern as a feature value for Bayesian modeling that is separate from, and additional to your major feature method. Similar problems occur in ink-trace thickness. In principle, the used writing implement is irrelevant to the overall signature shape (trace thickness should be normalized away) but human forensic document examiners will be surprised if a writer suddenly uses a thick felt tip pen instead of the usual thin ball point pen and will note that in their report.