Biometric identification uses something you are—like your fingerprints, face, iris or retina blood vessel pattern or voice—to confirm who you are. Unlike passwords or tokens, which can be stolen or replaced, physical traits like these are impossible to forge.
Any biometric system relies on multiple factors to achieve its goals of security, scalability, fraud reduction, cost or convenience. This report takes a systems perspective to understand these complex applications.
Fingerprints
The ability of fingerprint recognition to distinguish individuals is based on the unique physical characteristics of the friction ridges and valleys that occur on a person’s fingertips. They are nearly unique, difficult to replicate and durable over time, making them ideal for use as a means of personal identification. Fingerprint recognition systems scan a user’s fingertips and compare them with a stored template to identify the person. The technology uses two key algorithms: minutiae matching and pattern matching. Minutiae matching compares specific points on a fingerprint for comparison with other known fingerprints to find similarities. It also uses a set of features to verify the identity of an enrolled individual, such as the thickness, density and curvature of the surface of the skin. Like all biometric identification systems, fingerprint recognition has certain limitations that should be taken into account. This includes the fact that not all members of a population will be able to provide an acceptable fingerprint sample. Additionally, the quality of a fingerprint may be too low to accurately capture and analyze. There is also the possibility that a biometric characteristic can be intentionally modified to circumvent or deceive the system. These factors, along with differences in feature extraction and matching algorithms, contribute to the performance variations seen between different fingerprint recognition systems.
Iris
Iris recognition uses a high-resolution video camera to capture images of the complex patterns visible in the colored portion of the eye (the donut-shaped area around the pupil). The image data is converted into a digital template using mathematical and statistical algorithms. This template is compared against a database of enrolled iris scans in order to identify an individual or to prevent someone from impersonating them. Like other biometric characteristics such as faces, voices, and fingerprints, the iris is unique and nearly impossible to replicate. It also remains largely unchanged throughout a person’s lifetime, unlike other body parts that can be altered by injury, disease, manual labor, or age.
Its advantages over other modalities include being less susceptible to false positives than face recognition and being easier for users to enroll and authenticate with — as long as the quality of the captured iris image meets certain minimum requirements. A high-quality iris scanning device that can perform quality and compliance checks on enrolled images is required to achieve the highest possible accuracy. While systems that rely on iris recognition tend to have lower false acceptance rates than other biometric technologies, they are not immune to attack. A presentation attack involves presenting a fake object that resembles the target’s iris, such as a printout or cosmetic contact lens. Developing countermeasures for these unknown attacks requires a lot of work, but Czajka and his team have already made some headway in the field.
Voice
Imagine you’re James Bond trying to get into a secure laboratory to disarm a deadly biological weapon. You need more than a key or password to get through the security system — you need the villain’s fingerprints, iris and voice. In a biometric voice recognition system, the sound of someone’s voice is captured with a microphone, then converted into a digital signal. Software analyzes the signal and creates a template of that person’s unique voice characteristics. This is stored in the database and compared to a sample from a new person during log in attempts.
The software compares the new sample to the stored template and decides whether it is authentic (the speaker matches) or not. If the match is not authentic, a false acceptance or rejection error occurs. To improve accuracy, the system adjusts the acceptance threshold. As with any form of identification, the quality of a biometric system depends on both within-person and between-person variability. The latter refers to the extent to which a specific human trait changes over time, such as when a person gets older or is under stress.
Handwriting
Handwriting biometric recognition is a type of behavioral biometric that uses the unique way someone writes to identify them. It’s a more recent technique with lower reliability ratings but has the potential to grow alongside improvements in other biometric technologies. Like fingerprints, faces, voice and iris structure, handwriting is difficult to lose or forget. And it’s hard to forge, which is one reason more businesses and governments use it. The technology behind handwriting recognition is based on the fact that every time you write, a series of synchronized neuromotor orders are fired in your brain to produce the sequence of letters you draw. This information can reveal a lot about the person who wrote it, including their health status and mood. There are two types of biometric systems: one-to-one (or 1:1) and one-to-many (1:N). The former compares a person’s biometric characteristic against a sample of others with the same characteristics. The latter compares a person’s biometric feature against all of the other characteristics in a database.
Regardless of the type of system, it’s important to understand that any biometric recognition application is embedded in a larger system. It may involve other technologies, environmental factors or appeal policies shaped by security, business and political considerations, which can reinforce or vitiate the performance of any biometric recognition technology. Furthermore, cultural or religious issues might limit an individual’s ability to enrol in a biometric identification system.