They actually have face-recognition algorithms that only show you pics where the face is clearly visible now.
When a user has to tag someone they are given standard sized box to drag over the person you want to tag while the face recognition systems selects only the face of the person no matter how big or small it is on the picture.