The software under development by Prof Rudolph, engineer Amit Roy-Chowdhury and art historian Jeanette Kohl will try to put a name to these forgotten likenesses by grabbing data about defining elements of faces from portraits and comparing them to known depictions.
Early work on the project established that key parameters for facial recognition in portraiture include the position of the corners of the eyes and mouth, the width between the eyes, and the width of the mouth. Mapping these characteristics using a 27-point scheme captures sufficient information to make identifications, said the researchers.
 |
| Benjamin Franklin death mask. Source: UndyingFaces.com |
The BBC article, however, discusses the application of facial recognition technology to artifacts created before the invention of photography, which presents a whole different set of challenges.
Art historians interested in applying software to facial analysis might be wise to enroll death masks (also mentioned in the article, or life masks) as database images since they are as nearly as possible an actual recording of a person's face. Those could then be compared to painted images purportedly of the same person. Using this technique, art historians would quickly get an idea of how "accurate" portraitists (in general, or individually) were at painting images to the mathematical specificity that underlies facial recognition algorithms.
They would probably learn a lot from this basic exercise. How accurate did portraits tend to be? Were they supposed to be scientifically accurate? Or, were they more like the airbrushed and Photoshopped images that grace the covers of magazines today? Were certain portraitists consistently more accurate than others. Once these questions are well understood, analysis of painted or carved images for identification purposes might come within reach even where no contemporary physical impression of the face exists.
