The New Scientist reports that computers may be better at recognizing faces as visual averages–the sum of a number of images of a person, from a number of different perspectives in different lighting–rather than in a single passport or ID photo. Scientists at the University of Glasgow tested an advanced facial recognition program, FaceVACS, with individual and averaged portraits of celebrities. FaceVACS was supposed to match these portraits with celebrity images in its database.
When FaceVACS only had a single image to go on, its success rate was a lowly 54%. Trials with averaged images, however, resulted in success rates between 80% and 100%. Although the article doesn’t give results for similar experiments on humans, it does assert that we process faces the same way. (If so, that would corroborate what I wrote earlier this week about seeing distortion in Martin Schoeller’s images of Amazonian peoples.) To my knowledge, very few contemporary photographers have experimented with averaged images of individuals in search of some “essence.” Instead, photographic composites are mostly used to blend different faces. A popular example would be Conan O’Brien’s old segment, “If They Mated,” which blended attractive celebrities with disastrous results.
This form of composite photography actually originated in the nineteenth century, when the Victorian polymath Sir Francis Galton described it as a possible method for identifying “types” in society–murderers, madmen, consumptives, etc. Galton believed that these types might all share certain facial features (an idea rooted in theories of physiognomy), which would emerge when enough individual faces were blended. Arthur Batut experimented widely in this vein. More recently, Nancy Burson has used composites to critique cultural stereotypes about race and beauty.
Top image © Musée Arthur Batut. Bottom image © Nancy Burson.