Advances in facial recognition technology are producing unprecedented levels of accuracy, speed and image quality. Using biometrics-based identity management, businesses and governments can better protect the public from terrorism, crime and theft perpetrated by fraudulent identities.
Facial recognition is one of the most predominant forms of biometric recognition today and among the best methods to prove a person is who he or she claims to be. Facial recognition can be better than human identification, with accuracy rates approaching 99 percent, according to the National Institute of Standards and Technology.
Adoption of facial biometrics as a means of identification continues to be driven by governments globally. Facial recognition systems are used in many credentialing programs such as passports and national ID cards. Customs and border agents in various parts of the world use facial recognition solutions for identifying and preventing unwanted individuals from entering the country. Military personnel and law enforcement also turn to mobile facial applications for real-time identification of subjects in the field.
As the technology continues to be proven within various government initiatives, commercial enterprises are beginning to implement this technology. Businesses around the world use facial recognition systems as a means of providing more secure access to buildings and restricted areas. The financial community uses facial recognition as an aid in identity theft investigations. Facial recognition is also used by retail environments to spot potential shoplifters and by casinos to identify known cheaters. Opt-in customer loyalty programs are another appealing application for facial recognition technology.
The appeal of facial recognition is that it is non-intrusive. Images can be captured from a distance and further away from the subject than any other biometrics. Facial images can also be captured while a subject is moving, making it an ideal application for identification in high-traffic areas, such as airports and border crossings. These factors, together with technological advances, are propelling this biometric modality into new directions.
Accuracy, Speed and Image Quality
Improvements in accuracy, speed and image quality have been dramatic over the past several years. Government sponsored technology evaluations and challenges are fueling advancements in the accuracy of facial recognition. This began with the Face Recognition Technology program in 1993 and today has evolved with the Facial Recognition Vendor Test (FRVT). According to the most recent FRVT testing, facial recognition algorithm error rates have been reduced 20-fold from 2002 to 2006.
Improvement in speed is also making it possible to read and match against a larger number of images than before. The primary drivers behind these improvements are accelerated hardware processing speeds and the use of graphics processor units. These faster processing speeds mean facial recognition is a more viable option for more complex and sophisticated identification scenarios involving mobile and real-time applications.
The progression of image quality also propels facial recognition forward. For example, there is no longer a requirement to look directly into a camera to capture an image. Technological innovations have improved the accuracy of the captured image, regardless of the position or pose of the face. Deblurring and focusing capabilities are increasing the suitability of any quality of image for matching.
While faces may often be captured off-center, improved modeling techniques that render a greater collection of face models can overcome this constraint. A 3-D image can be captured initially to eliminate this limitation. Alternatively, facial technology can render multiple off-angle or posed images from an original 2-D image with the help of an accurate 3-D face model.
Image preprocessing and enhancement is another important area where removal of image noise around a face in an image can make the face more pronounced, vivid and suitable for matching. For example, technology can lift a face from a blurry photograph and from holograms in an ID photo to produce a dramatically improved quality image. Additionally, a single higher-quality image can be constructed from several low-quality video frames.
The ability to fuse face with other biometrics is another important area driving face biometrics as a trusted method for accurate identification in a wider variety of scenarios.
Thousands of handheld devices are used today in areas of conflict that allow users to enroll and match subjects via iris, finger and facial images. These devices can store up to 200,000 records, depending on the configuration of biometric records. These multimodal devices offer unparalleled levels of speed and accuracy for real-time identification of subjects in any environment and under any operating condition. Moreover, Iris on the Move can increase facial recognition accuracy in identifying subjects by the iris of the eye.
Finally, the fusing of additional technologies, such as skin, offers an even greater probability for accurate matching using the face. Skin texture is inherently random and the inclusion of skin algorithms with facial recognition provides another means of proving facial images. However, it is important to note that this capability is predicated on the technological advances discussed in this article and requires an extremely high-quality image.
The advances in facial biometrics are indisputable, but it does not mean that facial recognition is always the best means of identification. Its suitability is determined by the application or scenario in which it is used.
For example, Japan's cigarette vending machine age verification system that uses 2-D facial recognition was fooled by holding up a magazine photo of an age-appropriate individual to the camera's lens.
The next generation of 2-D facial recognition systems is making these limitations a thing of the past. The more advanced facial recognition applications require that a person move the head left, right, up and down and assess the pose as a means of accurate matching. If the pose does not change, then the system immediately recognizes spoofing.
Innovations in facial technology are now better accounting for changing light conditions during image acquisition. Improvements in the ability to capture a high-quality image at a distance, regardless of atmospheric conditions and interference, are also underway. The ability to continuously capture images to facilitate unattended entry is another important area of advancement.
Together, these developments are making facial recognition a much more viable outdoor biometric indicator. Biometrics remains to be the least intrusive and most accurate means of identifying individuals.