SMS: Learning texton models for real-time scene context