We are working on a project to automatically classify pain/non-pain faces in football players faces.  The idea is to build automatic classifiers to help the referees during world cup games distinguish "real" facial expressions of pain from "faked" expressions of pain, thereby helping to book players more fairly.  Our methodology was to gather expressions from world cup videos, and then assign to positive (real pain) or negative (faked pain) class based on whether the player in question stayed off the field/on the ground for more than 30 seconds.  Although we have found large numbers of negative class examples (see below for examples), we are having some difficulty identifying any positive class facial expressions.  We therefore seek input from the facial expression research community into how we can adjust our methodology while gathering this important dataset of facial expressions.  

Thank you for your help with this matter,

Jesse