Biometrics face off

According to MarketsandMarkets research, global biometrics market revenues are anticipated to reach US$20 billion by 2018. Increasing security requirements for public security such as border control management, national identity cards, e-passports, Internet and network access, and financial transactions are acting as growth drivers. As of now, fingerprint is the most commonly adopted form of biometrics, but face recognition will most likely become its successor in the years to come.

The global biometrics market is expected to grow at an estimated CAGR of 22.9 percent, as compared to the facial recognition market growth of 27.7 percent during the period of 2013 to 2018. Over the next six years, facial recognition is predicted to become highly pervasive, ubiquitous across its ecosystem, and penetrating the market to a huge extent, according to MarketsandMarkets.

Facial recognition, one of the oldest forms of biometrics, had been slow to gain widespread adoption due to the problems in accuracy and reliability often found in its algorithms. However, the dynamics revolving around the use of facial recognition is changing, as government officials and commercial sectors are starting to realize the convenience in using facial biometrics for various purposes. Its appeal stems from the contactless, noninvasive nature when capturing and recognizing an individual, but also from its similarity to how humans recognize each other — through the face.

Because of its enhanced accuracy, flexibility of being used in all environments, and the public's higher tolerance for it, the speed of adoption shall only accelerate.

Main Purposes
For one-to-one identification, face images are used in combination with video surveillance in a controlled situation. Ideal sources of controlled environment for image capture include motor vehicle agencies, visa and passport agencies, mug shots, background checks, and surveillance cameras placed at “choke points.”

For one-to-many identification, facial recognition algorithms have experienced noticeable improvements through continuous attempts to address commonly associated problems in uncontrolled environments.

“Facial biometrics is one of the most promising technologies to be widely adopted and more generally affordable in the short future, given that capturing of samples can be done at relatively long distances and without any participation on the subject's part,” said Gary Lee, International Business Development Manager at Herta Security. With the ability to operate from afar, facial recognition is used to conduct passive recognition where no real cooperation is needed from subjects to detect and collect their faces in a real-time surveillance video — and start the match against databases of unwanted personnel or the “blacklist.” Areas with large crowds, heavy traffic and high throughput will be more effective if a separate mode of recognition can be incorporated into the surveillance solutions to further ensure maximum accuracy.

When it comes down to identifying an individual against an entire or multiple databases, facial recognition drastically enhances the chances of locating a match. Database will continue to expand, not only because of the likes of the FBI's billion-dollar next-generation identity program, but with the help of social media and retail sites where users upload images for a virtual makeover. This allows operators to access dozens of photos of individuals from varying angles and settings. The growing computational powers ameliorate the process of scanning these massive databases.

Challenges and Limitations
As with all technology, using biometric devices and solutions has challenges and limitations, whether it is due to the algorithm itself or operational errors. Carles Fernández Tena, R&D Project Manager of Herta Security, mentioned some improvements on the way. “One will be the ability to process very high-resolution imagery in real time. This will result in higher image quality for identification, more opportunities for matching the short apparition of a subject against the database, increased bandwidth capacity for processing either a greater number of channels or larger frames with the same resources, and the development of more sophisticated algorithms that are not currently possible due to the existing computational limitations.”

Some other problems include the cost of employing facial recognition devices or software. The technology in search applications usually faces more challenging conditions such as lower resolutions, variability in pose and expression, changing illumination and larger occlusions, which result in higher costs. “Depending on the reliability and functionalities of access control systems, their price range is typically between hundreds and a few thousands of dollars,” Lee stated. According to Alf Chang, Senior Consultant at a&s, current cameras can detect faces up to six or seven meters. Identifying individuals from a long distance can be problematic if the cameras do not have high enough resolution. If users wish to detect or identify individuals from farther away, they must invest in cameras with higher resolutions.

2-D vs. 3-D
3-D recognition is the newest form of facial recognition to have emerged over recent years; however, the debate on its use continues to exist. By employing 3-D recognition, it is able to address some of the common problems faced by regular 2-D recognition, such as lighting and facial angle, and provides additional information to facial analysis. In turn, this could lead to more accurate recognition.

"The basic idea with 3-D facial recognition is that a biometric template based on unique geometry of a person's face can be readily stored on a database, for access control, and compared with a ‘live' analysis to identify the person in question,” said Anna Stebleva, VP of Business Development at Artec Group. “3-D facial recognition is fast, contactless and accurate, and this combination of features caters fully to the needs of the access control market today.”

As of now, 3-D facial recognition is still in the research stage for the most part. “Very few applications are actually incorporating the use of 3-D facial recognition. Capturing and storage of 3-D templates are more complicated than with 2-D technology. It is also an expensive approach for access control or any other applications, so it still remains a technology in search of a true application event,” according to Jim Slevin, Aviation Business Unit Manager at Human Recognition Systems, who thinks 3-D can be extraneous for regular access control and one-to-one verification, but remains attractive for forensics and postevent analysis of surveillance footage.

“The main limitation of 3-D technology is the very high cost and limited working range of the sensors required to make it accurate enough,” Fernández said. “This breaks with some of the traditionally attractive characteristics of 2-D facial biometrics: long-distance operability, multiple identifications in crowds, and relatively cheap deployments in distributed architectures, given that cameras have become a commodity.”

Some also believe that 2-D and 3-D can coexist. “In uncontrolled environments, 3-D can address some of the problems. 2-D, with some of the advancements we've had, like something as simple as IR-based images, has already advanced a lot and are already doing well,” said Mizan Rahman, founder and CEO of M2SYS. “We may not need to replace all of the 2-D systems, and they will continue to exist in some capacity. 3-D is more effective because it is not constrained by end-user training; 3-D systems are able to handle unexpected environmental conditions.”