Humans possess a remarkable ability to recognize faces, even when those features are significantly altered. New research from the Max Planck Institute for Biological Cybernetics in Tübingen, Germany, and the University of East Anglia in Norwich, UK, has investigated how well individuals can identify faces that have been blended or morphed. The findings, published in the journal Cognition, highlight the thresholds at which facial recognition becomes challenging.
The study utilized a technique called face morphing, which combines features from multiple faces to create a new blended image. Participants were shown images generated by merging three different faces and were able to identify, on average, about half of the original faces included in the mix. This ability declined as more faces were blended together; however, even with eight faces combined, participants maintained a recognition rate better than random guessing.
Isabelle Bülthoff, the lead author of the study, noted, “This suggests that face identification remains possible with as little as an eighth of its identity cues.” Yet, blending more than ten faces significantly diminished recognition capabilities, leading to performance levels comparable to chance.
The researchers also found that participants were more successful in recognizing familiar faces, particularly those of family and friends. Access to the original images further enhanced their performance, indicating that memory alone is less reliable than visual references.
Despite these insights, the study leaves open questions regarding whether typical versus distinctive faces are recognized differently under similar conditions. Further investigation is necessary to understand how unique facial characteristics influence recognition limits.
The implications of this research extend to practical applications, such as improving security measures to prevent identity fraud. Understanding the boundaries of facial recognition can enhance the effectiveness of biometric systems used in various security protocols.
For more detailed insights, refer to the original study by Mintao Zhao et al., “How much face identity information is required for face recognition?” published in Cognition in 2025. The full paper can be accessed via DOI: 10.1016/j.cognition.2025.106175.
