Improve enrollment for imaging devices
Current implementation of enrollment is quite naive.
We do 5 scans in a row, extract minutiae information from all of them and store it. During verification scan is compared to each of 5 enrolled scans and if any matches we consider it a match.
Besides being quite inefficient (we're doing 5 comparisons instead of 1) it's also not very reliable on small sensors. There's no guarantee that user didn't scan his finger the same way 5 times in a row, and different way during verification (and we'll get false rejection in this case).
What we should do instead is merging minutiae information during enrollment and do as many scans as we need to get acceptable number of minutiae.