Bio-Metric, a known and widely used terminology we come across in our daily life at the present time. It has become increasingly important due to emergent needs of technology being more private and confidential. In the general sense, a biometric is a measurement of a biological characteristic such as fingerprint, iris pattern, retina image, face or hand geometry or a behavioral characteristic such as voice, gait or signature of human. The technology based on biometric data uses these characteristics to identify individuals automatically and help ensuring identity and thereafter the security at large. Ideally the characteristic should be universally present, unique to the individual, stable over time and easily measurable. There is no biometric characteristics which have been formally proven to be unique, although they are usually sufficiently distinct for practical uses. There are different biometric data which will be more suitable for different applications depending, for example, on whether the aim is to identify someone with their co-operation or from a distance without their knowledge. There is growing interest in the use of biometrics for security of amenities including buildings and IT systems and for use in digital access cards, national IDs etc.
Biometrics can be described as the “science of establishing the identity of an individual based on the physical, chemical or behavioral attributes of the person”. In biometric systems, the respective bio-metric data are used for automated, or semi-automated, identity recognition by comparing a trait captured in ‘real-time’ by a sensor against a copy of the same trait stored on a database. Comparison is achieved through the application of a matching algorithm and a match score is generated based on real-time processing. The match score indicates the degree of similarity between the two templates being compared: the higher the score the more certain the system is that the two templates belong to the same person.
Bio-Metric Identification and Verification
Bio-metric identification refers to the ability of a computer system to uniquely distinguish an individual from a larger set of individual biometric records on file with the help of biometric data. In this particular scenario a national biometric identification system could allow a citizen to prove who he or she is without recourse to any document with the existence as registered in the system. The presented biometric data would simply be compared with all other entries in the national database for a match, and upon a successful match the associated individual’s identity data would be released from the database. This is often referred to as a “one-to-many” match, and is used by law enforcement agency to identify criminals on watch lists, as well as by governments to identify qualified recipients for benefit-entitlement programs and registration systems such as voting, driver’s license and other applications.
The bio-metric verification and identification process in shown in the Figure 2. In biometric verification or authentication process, it involves “one-to-one” search whereby a live biometric sample presented by a person or individual is compared against stored samples previously given or recorded by that individual, and the match confirmed the identity. The eligibility of the person for the service or benefit has already been previously established. The matching of the live biometric to the sample is all that is necessary to authenticate the individual as an eligible user. It is not necessary that any search or matching to a central database, although a central database can still be used, provided that some other identification data is used. For example, a national ID number could be used to “look up” an individual in a biometric database, and the live biometric sample could then be matched against the sample stored on record to verify the individual as the rightful bearer of the national ID. In much simpler form, the person or individual could just type in his username which extracts the biometric template from the database for verification.
On the other hand, identification templates are always stored in a database which is secured and controlled by high level national authority. One-to-one templates can be stored either in a database or in a distributed medium carried by a user in the form of smartcard. For example, if a person or individual applies for a passport or ID card, his biometric samples enter a one-to-many search first. This is done to check his background and to make sure that the person or individual has not been listed in a criminal or suspect database before. Somewhere between “one-to-many” identification and “one-to-one” authentication lies “one-to-few” biometric data uses, where “few” is of an order of thousands.
Challenges in Bio-Metric Identification and Verification
There are several challenges and problems lies with the bio-metric identification and verification process. It is important to bear in mind that the collection of biometric samples and their processing into biometric templates for matching is subject to great variability. In simple form, the biometrics are “fuzzy” – no two samples will be perfectly identical as the mysterious creations. In case of facial recognition technologies, for example, are notoriously prone to variability due to different lighting conditions, angle, subject movement, and so forth. This is the reason, for example, that we are asked not to smile in our passport photos. Similarly, numerous factors affect the ability to obtain reliable and consistent fingerprint samples. Among the various biometric types, irises seem to be the most accurate and consistent.
As a result, live biometric samples can be at some variance with stored reference samples, making comparison, matching and identification an inexact process. In other words, biometric systems do not have 100 per cent accuracy. When the biometric system cannot perform a proper match and rejects a legitimate user, this is called a false reject, and the user must typically resubmit one or more biometric samples for further comparison by the system.
Biometric system designers can and do take measures to lower the false rejection rate (FRR) of their systems so this variability is smoothed out and the system can function properly. Apart from controlling the conditions under which fresh samples are taken, and improving the processing algorithms, one way to do this is to lower the threshold for matches to occur. However, the difficulty with this approach is that this often increases the false acceptance rate (FAR) of the system, that is, the system will incorrectly match a biometric to the wrong stored reference sample, resulting in misidentification. Usually there is a tradeoff between FRR and FAR.
The FRR/FAR numbers quoted by biometric vendors are often unreliable. For example, biometric competitions organized by the U.S. National Institute of Standard (NIST), or International Fingerprint Verification Competitions (FVC2000/2002/2004), the FRR ranges from 0.1% to 20%, meaning that a legitimate user will be rejected from one out of 1000 times to one out of five times on average. FAR ranges from one in 100 in case of low security applications to one in 10,000,000 in case of very high security applications. Other challenges for a biometric system are speed as the system must make an accurate decision in real time and ensuring the biometric data security against attacks.
For example, a biometric identification system with a 0.01% FRR and 0.0001% FAR. That is, the system is able to consistently match a genuine biometric sample 9,999 times out of 10,000 attempts on average. As remarkably efficient as this system sounds, a single biometric sample, when compared against a database of 1,000,000 samples, will generate on average one false accept in addition to one exact match.
In case of a database containing 30,000,000 entries, each biometric sample would generate about 30 false accepts, each and every time! Clearly, this would be unacceptable for any real-time automatic identification system and would require significant human intervention in order to function. Consequently, biometric system designers have resorted to other techniques to overcome the inherent technological problems of one-to-many identification. One way to significantly improve accuracy is to collect and compare multiple biometric samples. Multi-modal biometrics, for example, can involve collecting and using two or more fingerprints instead of one. If one fingerprint generates dozens or hundreds of false accepts, then the likelihood that two fingerprints will falsely match others in the database diminishes considerably. This is the primary reason behind emerging international requirements for including two separate biometrics that includes face and finger for example in machine-readable travel documents such as passports.
In most cases, the privacy issue of bio-metric data involves the fact that more and more biometric samples of personal information need to be collected, transmitted, stored, and processed in order for the system to function properly as these data carries high variability. If networks somehow fail or become unavailable, the entire identification system collapses and data breaches results in huge loss. Recognizing this, system designers often build in high redundancy in parallel systems and mirrors as well as failure and exception management processes to ensure availability which also demands greater scalability and maintainability with required skills. However, this can have the effect of increasing the security risks and vulnerabilities of the biometric data.
While applying in large centralized databases of biometric data, hooked up to networks and made searchable in a distributed manner, this directly or indirectly represent significant targets for hackers and other malicious entities to exploit. There are also significant risks associated with transmitting biometric data over networks where they may be intercepted, copied, and actually tampered with, often without any detection.
In case of large-scale biometric identification databases such as the IAFIS not only collect and file multiple biometric samples but, in an effort to preserve maximum compatibility with other fingerprint identification systems, store the full and complete images of the biometrics involved in addition to the templates. Some international standards for biometric-enabled machine readable travel documents, for example, call for storage of the biometric images in the document rather than a structured reduction of the biometric into a unique template, in order to facilitate cross comparison and identification with other databases. In any case storing, transmitting and using biometric images only exacerbates the privacy concerns with large-scale identification systems, since a very important privacy protection afforded by templates is removed, namely, the inability to exactly reconstruct the original biometric image from the template. This will continue to grow with senses of security for bio-metric data.