The first versions will probably be quite crappy. Give it 5 years and it will improve leaps and bounds, plus the amount of data to work from will also increase as more photos are uploaded.
Imagine this: the software will recognize that you change haircuts every couple of months. From this it can derive that you are more fashion-conscious than others and send you targeted marketing or offers. Also it can automatically set up appointments with hairdressers, complete with 3d renders of new, proposed haircuts that you can choose from. That would be awesome, this is taking personalization to a whole new level.
For example, auto-tagging your pictures from 1000 photos is easy. auto-tagging from 10,000 photos is easy. Hell, auto-tagging yourself from billion photos is easy, and the difficulty doesn't increase with the size of the verification set.
But when it comes to "auto-checking in", accuracy is really important. Out of the entire world population--6,697,254,041--how many people do you think would look similar to you? And it only gets more difficult with all those little details such as hairstyle, makeup, etc.
The Minority report model of using retinal scanner works because the accuracy is really high, and is statistically and scientifically proven. Same goes for your fingerprint. But no matter how good the algorithm gets, it is impossible to create an auto-check in system by automatically figuring out who you are based on just image recognition.