Humans can spot familiar faces even in heavily blended images
• Humans can identify faces in blends of up to eight images.
• Familiar faces are easier to recognize than unfamiliar ones when cues are minimal.
• Possible applications in preventing identity fraud.
Face morphing generates a single blended image from two (or more) faces.
© Jan Rubinowicz via Wikimedia Commons
Humans have a remarkable ability to recognize faces despite variations in appearance. Regardless of aging, changes in weight or facial hair, or even cosmetic surgery, we can usually-but not always-identify individuals based solely on their facial features.
It has long been studied how well people can recognize faces when key identity features are blurred or reduced. Understanding this is important, as it may help prevent fraud with identity documents. However, past studies rarely pinpoint the exact level at which a face becomes too altered to identify.
Researchers of the Max Planck Institute for Biological Cybernetics (Tübingen, Germany) and the University of East Anglia (Norwich, UK) have now investigated this question using face morphing with three-dimensional faces. This process generates a single blended image from two or more faces, smoothly combining their features so the result appears like a halfway-between version of them all.
Testing the limits of face recognition
When shown images created by blending three different faces, participants were able to identify, on average, about half of the original faces in the mix. As more faces were combined, recognition accuracy declined. However, even when eight images were blended, participants still identified the faces at levels better than random guessing.
"This suggests that face identification remains possible with as little as an eighth of its identity cues,", says Isabelle Bülthöff of the Max Planck Institute for Biological Cybernetics, lead author of the publication. Beyond that threshold, though, recognition reaches its limit, as the team also showed: in blends of ten faces, participants' ability to correctly identify faces was reduced to chance level.
The team also found that given the limited face information in the morphs, participants were better at recognizing familiar faces, particularly those of people personally known to them, such as family and friends. Moreover, being able to see the original images, rather than relying on memory alone, also improved the performance.
However, it remains unclear whether the results apply differently to typical versus distinctive faces. Further research is needed to close the gap in understanding how facial uniqueness affects recognition limits.
