Fair point. Elsewhere in thread it looks like they are using a fixed resolution. I guess at that point it comes down to whether they've left enough mid-frequency content for an algorithm to identify the slightly darker/lighter patches of facial features, and whether that spacing can actually identify a person.