BIOMETRICS-BASED SOFTWARE

Is It Viable for Authenticating Faces in Photographs?

by Joelle Steele

Many people who collect antique photos or who have large collections of old family photos are looking for quick and easy ways to make positive identifications of the people in their photos. They often want to be able to use all the latest high-tech programs that rely on biometrics to identify people. But biometric software is not an option -- at least not yet – for identifying people in old photographs.

Biometrics is a word derived from two Greek words: "bio" (life) and "metrics" (measurements). Most people do not know what it is or how it works. Biometrics is an ancient practice of identifying people by their physical features, and its use has been documented widely throughout the world. While there is evidence that handprints may have been used to sign cave drawings and that Babylonians used fingerprints as their "marks," it isn't until the 14th century that Portuguese historian and explorer João de Barros described in his writings how Chinese merchants stamped children's palm prints and footprints on paper to tell them apart, and how they signed their agreements with fingerprints.

And it was 500 years later, in the 1890s, that biometrics was pioneered and developed by French anthropologist Alphonse Bertillon for use as an aid to law enforcement in identifying criminals. His system was called "Bertillonage" or "anthropometry," the latter from another two Greek words: "anthropos" (human) and "metron" (measurement). It relied on Bertillon's observations that certain human body measurements did not change in adults, such as the size and shape of the skull, the length of fingers, etc. For many years into the 20th century, Bertillon's measurements were a standard in "wanted" posters. But, unfortunately, more details were necessary to rule out two people who shared similar measurements, so Bertillon's methodology ultimately failed and fell out of use.

Meanwhile, in the province of Bengal, India, fingerprinting had been in use for almost half a century after its introduction there by William James Herschel (son of the astronomer, John Herschel) after he discovered that fingerprints remained stable in form over time and that they were unique to each person. The local inspector-general of the police department in Bengal, Edward Henry, following correspondence with author and fingerprint expert Francis Galton, and with the assistance of others in his department, developed the system of fingerprinting that became known as the Henry Classification System. Henry's system eventually formed the basis for fingerprinting throughout the world which, until only recently, has begun to be replaced by the more effective ridge flow system of fingerprint classification.

Up until the 21st century, about the only other valid use of biometrics other than fingerprinting was ear biometrics, a forensic method of ear analysis used for identifying people. It was developed by Albert Iannarelli in 1949, and was used by law enforcement forensics experts, including the FBI, to identify suspected criminals who had undergone facial damage or plastic surgery to modify their appearance. Ears were rarely modified surgically because they were mostly cartilage and healed slowly and sometimes badly. Ear biometrics are still used today as a back-up to fingerprinting when a criminal has had his fingerprints  removed, or in identifying bodies when the fingers have been removed or destroyed. (Not everyone's DNA is on file in a database.)

But Alphonse Bertillon was on the right track when it came to identifying people with biometrics. His methods were just not adequate for use in law enforcement at the time. However, these days, biometric software has come to the forefront in security and law enforcement. So far, these programs work most accurately for fingerprinting and iris pattern recognition. There are also some facial recognition software programs, and they work in different ways to map the face and then compare it to other biometrically-mapped faces in databases to confirm an identity.

I first used a biometrics-based program for identifying people in photographs in 1990. I had bought my first scanner and simultaneously heard about a program called FaceShell. It was a shareware program that I bought and it examined digital images of faces and mapped the head/skull shape and proportions of the facial features to the head. This was a sophisticated program for its time, and it provided a great short-cut for me as far as measurements were concerned, but it did nothing else. I stopped using it in about 1994 because the tech support was poor and it was "buggy" as software frequently was in those days. I began using Photoshop instead and found that it had better routines that made it easier for me to make comparisons of faces. I have since experimented with other software designed for comparing facial features, but so far I have not found a program that is sufficiently thorough in accurately assessing faces without a significant amount of human intervention.

While widely available, facial recognition software still has disadvantages as far as its accuracy is concerned. This is primarily due to the fact that these programs do not size the irises, which is key to accuracy in identifying people. But, some software has improved over time, and the newer high-def two-dimensional and latest three-dimensional face recognition programs show some promise in overcoming the inaccuracies of the existing two-dimensional recognition software that currently dominates the market for security and law enforcement applications. But, as far as the accurate identification of people in old photographs is concerned, biometrics is not truly viable in either dimensional format.

The means of identifying people in old photographs cannot benefit from any system that relies on three-dimensional technology or that relies on databases of already biometrically-mapped faces. Those types of databases do not exist for people found in old photographs, many of which would lack sufficient details to be accurately mapped biometrically. Facial recognition software is also not exact enough to distinguish between or among people who just happen to resemble each other. And lastly, facial recognition software does not adequately handle the very slight variations in the facial expression or the direction in which a face is viewed. These are all areas in which a trained human eye is far more capable of discerning the subtle differences and drawing accurate comparisons.

Someday, someone will create software programs that can analyze and compare all of the many head/skull and facial features traits that are necessary to authenticate the identities of people in old photographs (I have personally mapped 89 items to be analyzed and compared). Until then, this work will continue to be done primarily by hand-measuring and a trained eye.