Adaptive appearance model tracking for still-to-video face recognition