I'm imagining they are using a data set of fingerprint points (statistical anomalies), and then using machine learning to get a good statistical model of it whe

I'm imagining they are using a data set of fingerprint points (statistical anomalies), and then using machine learning to get a good statistical model of it when you first set it, and every time after that. I doubt they're image matching directly every time. Add some matrix math for 360˚ transforms and you're in business.