They take I think 30 pixels across known positions on the screen and that’s apparently enough to recognize content without being able to look at your confidential data.