Forensic facial specialists identify faces from one or two features

In an experimental study of forensic facial identification, specialists matched a hidden face to one of two similar faces with more than 90% accuracy after seeing only one or two facial features. Researchers led by Carina Hahn of the US National Institute of Standards and Technology found that ears were the feature specialists most often selected, while revealing the forehead was associated with the highest identification accuracy. The peer-reviewed study, published in PLOS One on October 21, 2026, shows that trained facial specialists can extract substantial identity information from very limited visible features, although the artificial laboratory conditions do not directly represent performance in real investigations.

For each trial, participants viewed photographs of two different people and a third face that began completely concealed. Their task was to determine which of the two target images matched the concealed face, or probe, while revealing as little information as possible. The interface offered individual features, including ears, eyes, forehead, neck and chin, which participants could request one at a time.

A points system encouraged selective feature reveals and discouraged random guessing. In the self-selection condition, specialists chose which feature to inspect. In the predetermined condition, the available features were fixed and the only available action was a general “Reveal” command.

The design allowed the researchers to examine both which features specialists preferred and whether they changed their strategy according to the images in front of them. Across the task, identity judgments averaged more than 90% correct, and specialists were generally able to identify the matching face after only one or two features became visible. Ears and forehead were associated with the highest accuracies, whereas necks and chins were the least helpful.

Ears were also the most frequently self-selected feature, demonstrating that specialists had a strong preference for information they commonly found useful. That preference did not always correspond to the feature’s measured effectiveness: the forehead was not frequently selected, but revealing it was associated with among the strongest identification performance. The results also showed evidence of adaptive behavior.

When specialists did not choose ears, they generally requested features that looked more different between the two target faces. Their choices therefore appeared to be influenced by the particular comparison at hand rather than following one unchanging rule. The study offers a detailed account of how forensic facial specialists prioritize and combine visible information during a controlled identification task.

Hahn said she was most surprised that specialists could accurately judge identity from a single facial feature most of the time. She also described their ability to adjust strategy according to the face being examined as a striking combination of reliance on features known to be useful, such as ears, with flexibility when another feature offered more diagnostic information. The authors said these findings may help inform the training of human specialists and the development of machine-learning algorithms used for facial recognition.

Several limitations prevent the experimental accuracy from being interpreted as a measure of real-world performance. The conditions were artificial and designed to address specific research questions. The authors also cautioned that the facial images analyzed might not generalize well to other image sets, and that the study’s demographic composition—primarily Caucasian and East Asian individuals—could constrain broader generalization.

The findings therefore establish high accuracy within this structured experimental paradigm, not the reliability of abbreviated feature inspection across every forensic setting, image collection or population. Additional testing would be needed to determine how consistently the observed feature prioritization and adaptive strategies translate to practical investigations and expert testimony.

Subject of Research: How forensic facial specialists prioritize features when matching faces from images

Article Title: Face specialists need only one or two facial features to identify individuals

Article References: Face specialists need only one or two facial features to identify individuals. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: forensic facial identification, facial recognition, facial features, forensic specialists, expert decision-making, feature prioritization, NIST, PLOS One, Face, specialists, need, only