A novel discriminant multiscale representation for ear recognition