I don't understand, though, why their approach ignores a particular sort of translational symmetry. One of their examples is (ascii-fied):
... AAAAAA ... ... YYYYYY ... OOOOOO AAAAAA YYYYYY OOOOOO AAAAAA YYYYYY OOOOOO
This has two kinds of "mirror signatures", which they denote **. There is a vertical mirror line between the letters, and a mirror line through the midpoint of the letters. But this doesn't describe the fact that each set of letters repeats vertically as well!
For example, we would expect a nonrepeating infinite vertical stripe to be denoted differently:
... 888888 ... ... UUUUUU ... 888888 888888 UUUUUU UUUUUU 888888 888888 888888 UUUUUU UUUUUU UUUUUU . . .
This has the same mirror symmetries but lacks the translation symmetry. (It's not that they ignore translation--- there is a type composed just of translations--- but don't seem to consider a repeating vertical stripe with two mirror symmetries different from a nonrepeating vertical stripe with two mirror symmetries. Maybe there is a technical reason when they get more formal later on.)
You can see the 17 planar symmetries here:
That page suggests the the technical difference is between finite and infinite "fundamental domains". But, ** still seems odd--- all the other groups extend naturally across the plane, but ** needs a translation tacked on.