Detecting machine-readable zones in passport images

This is very interesting; a human would OCR the page and the pick the region with many less than / greater than signs. I wonder whether that would work here -- you need to OCR after finding the region anyways.