The Third Dimension

Oct. 27, 2008
Three-dimensional technology may be the key to legitimizing face recognition devices for identity management and access control

For years, face recognition technology has fought against its poor reputation for inaccuracy and lack of reliability. But a new approach to face recognition — capturing and matching 3D images designed specifically for identity and access management applications — may be the technology that can address the historical concerns. 3D face recognition can be accurate, fast and simple.

Face recognition has been gaining popularity as a biometric identity verification solution, with revenue expected to exceed $1 billion annually by 2012, according to the Biometric Market and Industry Report 2007-2012 from the International Biometric Group (IBG) .

3D face recognition systems are being used by financial institutions, Fortune 500 companies, biopharmaceuticals, casinos and organizations in the transportation sector to enhance security without introducing complex access control procedures that make it difficult for employees to gain entry. In organizations with a large workforce, speedy access control is a must, and with 3D face recognition, companies can quickly authenticate thousands of employees as they begin a shift.

The Advantages of 3D

While two-dimensional (2D) face recognition is being used by law enforcement for identification, it is not significantly deployed for verification systems. Identification involves comparing an existing image against a large image database in order to identify the individual depicted.

In verification solutions, a biometric identifier is compared either to another template called up through a PIN number or smart card, or against a small database of other templates. However, because 2D face recognition systems are affected by lighting conditions and the pose of the individual's face, they are generally not ideal for access control. In U.K. trials for biometric passports, for example, only 69 percent of able-bodied volunteers and 48 percent of disabled participants were correctly authenticated. Some 3D facial recognition systems combat this problem by using their own near-infrared light source, which enables accurate matching even in poor lighting conditions.

3D facial recognition technology also has other advantages. It is possible to collect more data points with 3D, and the types of data points collected are more valuable. While a 2D system might make a match using such data points as the distance between the eyes, 3D uses data such as the curvature of the forehead. The latter is more useful because it allows 3D systems to make matches even if a face changes as a result of a scar or other visual features.

How 3D Works

There are two main approaches to 3D facial recognition: the stereo approach and the structured light approach. Stereo 3D systems create 3D images by synthesizing two or more 2D photos. This computing-intensive (and thus, expensive) approach adds an unnecessary layer of complexity to 3D face matching and, like 2D, is hampered by poor lighting conditions.

Structured light 3D face readers, on the other hand, shine an invisible, near-infrared, grid-shaped light on a user's face and a camera takes a picture of the distortions in the grid caused by the face, collecting approximately 40,000 data points. Because structured light 3D face scanners use their own light source, they can work in poor lighting conditions. They can also accommodate various facial positions, making it easier for users to confirm their identity as they pause in front of the face reader for identity verification.

3D face readers simplify data entry for end-users, who can confirm their identity using only their face. Users just walk up to the reader and pause while the reader either searches the database for a face that matches or checks the face against a template called up by a proximity card, smart card or token — making a match in under a second.

The enrollment process is also simple, taking just a few minutes. Again, the enrollment reader projects a near infrared light pattern on the user's face. The light pattern is modified by the surface geometry of the face and the camera precisely records the pattern distortions caused by the face. This modified pattern is input into a 3D reconstruction algorithm in order to create a 3D mesh image of the face using triangulation. The face geometry can be measured in millimeters.

The reconstructed image is not stored in the database; instead, a biometric template is extracted from the 3D facial geometry and the numeric template is stored in the database. A matching algorithm then checks the face presented at the entry point to restricted areas against the template already stored.

Security Thresholds

With any biometric implementation, organizations have the ability to set different thresholds for security. At a high threshold, the number of data points that must match in order for someone to be granted entrance is raised, thus there will be fewer false accepts (the authorization of unauthorized personnel). But that also creates an increase in false rejects (denying access to someone who is actually authorized). At lower thresholds, the number of false accepts increases while the number of false rejects decreases. Companies are free to assess the levels that are appropriate for their security requirements. At very high thresholds, they can be sure to keep intruders out, but some employees may have to be more exact in how they present themselves for verification. At low thresholds, authorized employees can easily gain entrance, but someone without authorization might also be permitted through.

With 3D face recognition, organizations such as financial institutions and airports — those that cannot tolerate any false accepts — can set high security thresholds without a significant increase in the number of false rejects. This means their employees can get into the areas they need access to while unauthorized personnel are kept out.

Like other biometric solutions, 3D face recognition uses who a person is, rather than what they know or what they have for identity verification — significantly increasing security without sacrificing convenience. Passwords and PIN numbers (what a person knows) can be stolen, forgotten, lost or lent out. Smart cards, proximity cards and tokens (what a person has) can also be stolen, misplaced, forgotten, forged or borrowed. A person's face or fingerprint (what a person is), however, is always with them and cannot be stolen or copied.

Using face recognition alone or in conjunction with another authentication factor for dual- or multi-factor authentication, therefore, can significantly increase security.

Matthew Bogart is vice president of marketing for Bioscrypt Inc. and has spoken extensively about biometrics. He can be reached at [email protected].