I like your technique. But what you are trying to do is very difficult for several reasons.
1. The original image is very blurred and it's difficult to make out any facial features.
2. The enhanced version is very pixelated and the human mind is programmed to pick out faces from noise
3. Because you are trying to match up one small part of the image with another small image any errors of scaling are amplified.
A different and larger scaling reference would be more useful.