Putting a New Face on Security

Facial recognition may very well be the oldest known form of biometrics. Since the first human recognized another human, our species has used this technique to identify our friends and foes. And from prehistoric cave drawings to the portraits of Rembrandt to the modern drivers license, weve spent considerable human effort capturing the likeness of the human face and recording it for history. Yet, when it comes to using facial recognition in automated computer systems, the technology is still relatively new.

By Jennifer Allen Newton

The first serious techniques for large-scale automated facial recognition were developed in the late 1980s.  Today, facial recognition is the second most widely used form of biometrics, only slightly behind fingerprints.

Facial recognition technology is most commonly used in three areas, where it offers unique capabilities that make it more practical than other biometric modalities for certain applications:

• Large scale records management, such as motor vehicle licenses and passports;
• Automated surveillance with a particular emphasis on automated watch lists;
• Access control systems for secured buildings or areas  

Despite the sometimes creative myths around facial recognition technology -- the stuff Hollywood spy movies are made of -- biometric facial recognition holds tremendous potential for practical uses in private, commercial and government applications.  The technology is rapidly gaining in popularity and in technical sophistication.  And with the tools available today, facial recognition technology is both affordable and relatively simple to implement.

HOW IT WORKS

Facial recognition technology, put into the simplest terms, is asking a computer to memorize the characteristics of a human face (either through live enrollment or from a stored photograph) and later to be able to accurately recognize that same person when presented with a photograph or live image of that person. 

Like all other biometric systems, facial recognition systems build face templates for permanent storage in a database.  These templates are later used for comparison, one against another, in rapid succession.  In general a ‘winner’ or ‘match’ occurs when the comparison result between two templates is higher than a pre-set threshold.  

There are several major categories of facial recognition including facial geometry, skin texture analysis and facial thermography.  The latest advances in facial recognition incorporate 3 dimensional face modeling.  Some techniques ‘extrude’ 3 dimensional images from 2 dimensional photographs.  Others actually capture the face and head in 3D during enrollment. 

The one thing that all of these systems have in common is that they all rely on photographic images of the human face.  Other than this common starting point the methods vary greatly in practice. 

Because facial recognition systems rely on photographic images, the most important factor contributing to the reliability of the technology is image quality.  Image quality can be affected by many factors including lighting, resolution of the camera, focal depth of the camera lens and the speed of motion of the target among many other things.

Beyond image quality, the second most important factor contributing to a facial recognition system’s overall reliability is the quality of the facial recognition algorithm  the mathematical equation that is used to determine if one face matches another.

“Cameras and scanners are the eyes, but algorithms are the brains of every biometric facial recognition system,” said Algimantas Malickas, CEO of Neurotechnologija (www.neurotechnologija.com), a company that develops biometric algorithms and software development products for security system developers, integrators and other companies using biometric technology.  “Without a high quality algorithm, a facial recognition system will not be as reliable.  And if you are looking for a biometric application, chances are that reliability is extremely important to you.”

One of the most important factors in ensuring the accuracy of facial recognition technology is the human factor -- variations in how a camera is used, lighting and shadows on the subject can all wreak havoc on facial recognition algorithms.  Algorithm vendors are constantly looking for ways to reduce the human factor and many creative approaches have been developed just for this purpose.  Another challenge is making sure the appropriate camera and algorithm are used for a particular situational need.  Different applications require different technologies, and all cameras and facial recognition technologies are not created equal.

HOW FACIAL RECOGNITION IS USED

Facial recognition is being most widely used today in places where great effort and expense have already been taken to capture large databases of human faces.  In most cases these efforts were made without the thought of using facial recognition other than by human operators.  Perhaps one of the most common examples of the use of this technology is in the realm of driver’s licenses. 

In the United States, as in most countries of the world, people who wish to operate motor vehicles must first obtain a license to do so.  This ‘driver’s license’ becomes an important identity document and is commonly used by policemen to ensure that the bearer has a valid permit to drive.  When the technology became available to print the photos of the owner on the license, this was done with the thought that an officer might be able to quickly determine that the holder was in fact the owner of the document. 

Today, very large databases of facial images exist because of this systematic collection of facial images over many years.  But over the years this simplistic system has also led to great fraud and abuse, starting with forged licenses and escalating to new levels of identity theft.  So, today many US states have begun implementing facial recognition technology in order to detect duplicate records in their systems or to prevent someone from creating licenses under multiple names.   These systems are generally exception-driven, which means that if the facial recognition system detects a variance or similarity of great enough value, an exception report is created and a human operator intervenes to determine what the root cause of the exception is. 

Facial recognition systems such as the VeriLook facial recognition algorithm from Neurotechnologija are excellent for producing this type of automated tool because of their ability to scan through literally tens of thousands of face templates per second.  Prior to the existence of this type of technology it was practically impossible for states to implement face-based duplicate records checking.  Thousands of cases of abuse and fraud have been caught by systems of this nature.

FORMS OF BIOMETRICS

When making a decision about which form of biometrics to choose there are many factors to consider, not all of which are obvious.

“I’m often asked which biometric method is best, or why someone would choose facial recognition over fingerprint or iris,” said Ken Nosker is President and CEO of Fulcrum Biometrics, a biometrics technology consultancy and solution provider.  “I tell them all forms of biometrics have their specific purposes and implementations.  A better question would be ‘When should I choose facial recognition over fingerprint or iris?’  Each situation has a unique set of requirements to consider, including required accuracy, price, and ease of use by the subjects, “ Nosker explained.

According to Nosker, the issue of which modality to choose is directly related to the environment in which it will be used.  For example, one cannot do fingerprint recognition from 10 feet away, but one can do facial recognition where the subjects are removed from the sensor or image collector.  Therefore, unlike facial recognition, fingerprint recognition systems are not practical for automated surveillance systems because people do not randomly walk around touching finger image collection sensors.  On the other hand, fingerprint recognition is a more obvious choice for forensic applications because one cannot capture “latent” face prints at a crime scene.  Iris recognition systems generally require extremely high resolution or very carefully focused images in order to function reliably.  These types of images are not normally captured and stored in identity database systems, such as driver’s license or passport records. 

Nosker says computer-based biometric systems should be used to help solve security problems or to assist human operators do their jobs more effectively, and there are specific occasions where facial recognition is a good and natural choice.

“Facial recognition is an excellent choice for the introduction of a biometric into situations where the operation of the system requires little or no participation on the part of the subjects,” explained Nosker.  “This is called passive or uncooperative biometrics, and facial recognition is often very well suited to these types of applications, among others.”

The ability to automatically and rapidly identify a human face from a database of other enrolled human faces can be applied across a wide spectrum of applications.  While being able to accurately identify a face offers significant benefits, there is also great value in being able to rapidly and automatically determine that a particular face is not matched against any other face in a particular database.  Nosker provides two examples of simple face recognition applications that illustrate these two different uses: 

Example 1: Access Control Application

In this system, a wall-mounted access control system includes a camera, a keypad and a proximity card reader.  This system is controlled by a computer server that is managed by the human resources or operations department.  Employees are pre-enrolled into this system by having their pictures taken and each employee is issued a proximity badge which can be read by the wall mounted system.  Now, an employee who wishes to enter the building or any room in the building simply has to look at the camera and pass his or her badge in front of the proximity reader.  The facial recognition system looks at the person’s face and compares it to the face enrolled to the card number that was read when the proximity card was passed by the reader.  The facial recognition either could be done locally on the wall mounted unit or remotely on the server.  Now the security and personnel management of the company know that not only was an authorized badge number presented but the actual holder of the badge (not someone who may have stolen it) was standing at the door waiting to be allowed in.

Example 2: Security Check Point

In this system we have a series of high quality cameras mounted at security check points in an airport.  These cameras all feed back to a central monitoring room where individual face images are grabbed and processed in a high speed server.  This system is trained to take facial images of each person walking through security check points and compare their faces to a watch list.  As travelers step up to the gates and present their boarding passes and identifications, the cameras are taking full video of each person as they pass by and passing that data back to the frame grabbers and the high speed server.  In most cases, the person’s face is rapidly compared against the watch list and in almost all cases the person will not match and he or she will pass unhindered.  In a small number of cases the person may have features which closely resemble a person on the watch list.  The system automatically identifies these people and they may then be watched more closely by human security agents on location.  In a very few cases, the system actually provides a high quality match against a watch list person and appropriate action is taken.

In both of these systems cameras capture human faces in naturally occurring environments with very little extra effort on the part of the subjects.  The cameras pass the faces to high speed facial recognition servers which make automated decisions based on match characteristics or thresholds preset by the vendors or by the operators.  In both of these cases the computers provide a service that would be very difficult for humans to perform.  Similarly, in both of these cases a different form of biometric may have been chosen, but because of the nature of the systems, the preferences of the owners and the environments in which the systems are used, facial recognition was an excellent choice. 

     Security for Students And Staff

     Making Money with Video Surveillance