The Current Challenges of Biometric Access Control
Each day it seems that there are more and more innovative ways to unlock your door, car, or even your smartphone. Most of these breakthroughs use biometric access control such as fingerprint and iris recognition. With great innovations, sometimes challenges can arise. Arun Ross, affiliate of the Center for Identification Technology Research, takes a look into the current challenges of biometric access control. View the video and transcript below to find out more.
My name is Arun Ross. I am an associate professor at the Lane Department of Computer Science and Electrical Engineering at West Virginia University.
I am also affiliated with the Center for Identification Technology Research, which is a National Science Foundation Center focusing on identification technologies. I work with my students and other collaborators on a variety of biometric technology problems.
Some of the problems include multi‑spectral biometric analysis, finger print indexing, face indexing, iris biometrics, and also, visual cryptography. All of which pertain to some core issues in biometrics. I see a number of interesting research areas.
One is biometric template prediction. People are concerned when they are provided biometric trades as to where they are located, where they will be stored, and what kind of protection, in terms of security and privacy, is accorded to those biometric templates. I see template security and privacy as one important area of research. Another important area of research is human identification at a distance.
Currently, most technologies can deal very well with individuals who are cooperative and who are in close proximity to the biometric sensor.
Now, there’s increasing interest in recognizing people at a distance, people who are not cooperative and in unconstrained environments where the lighting can change quite a bit. There are factors such as the human pose with respect to the sensor that can change rapidly and so forth.
Those represent highly unconstrained environments. The ability to perform human identification would benefit both law enforcement agencies as well as other federal agencies. I see a couple of areas that have developed rapidly.
One is in the field of iris biometrics. Again, as I stated before, there is interest in recognizing individuals at a distance, but using the iris entity. That’s a challenging problem, because the iris is a moving object within a moving object, the eye, within a moving object, which is the head, and then, within another moving object, which is the human body.
The ability to extract the iris from a distance and perform identification remains to be an interesting area of research.
Now, there are multiple research groups which are engaged in that. Other important areas include biometric fusion, which is how do you consolidate the evidence presented by multiple biometric traits?
If you have information to the face, fingerprint, and iris, how do you combine this information in order to assess the identity of a subject?
Biometric fusion is an interesting area of research and is also impacting the way in which biometric systems operate. We do have laptops already equipped with biometric sensors. Very soon, as we perform transactions over the web, it is going to become very important to establish a person’s identity in a reliable and robust manner.
Remote authentication may be based on face recognition over web cams or voice recognition over voice channels. That might become important as well as more practical in the years to come. Multi‑biometric systems have several advantages. I could list a few.
One is the potential for addressing the problem of non‑universality. What I mean by that is, let’s say you are using a fingerprint system, and let’s also assume that there is a small fraction of the population that cannot provide good quality prints, perhaps due to dry fingers or some other implicit problems with their finger.
In such a case, you do not want to exclude those users from accessing the biometric system. If you have a multi-biometric system, then, a person who is not successfully enrolled using a fingerprint sensor perhaps can be successfully enrolled using an iris system, there is this notion of non‑universality.
The second advantage is performance improvement. When a suitable fusion rule is employed, you can take the evidence of these two modalities, bring them together, and improve the overall recognition performance.
Of course, the trade‑off is between cost, performance, and time. Now, if you have multiple modalities, it is very likely that you’ll have to spend more time collecting the data from the subject prior to the subject accessing the resource that is protected. Having said that, again, performance improvement is another benefit.
The third benefit is robustness to noise. What do I mean by that? Consider a subject who is walking through an airport aisle. Let’s say, there are cameras trained on this aisle. As the subject walks through this aisle and you obtain the face image and, let’s say, also, the iris image, you now have two different modalities that can be adaptively used.
If this person walks to a different environment where it can only obtain the face, you can still perform matching, because in the first environment, you acquired both the face and the iris.
Here, the goal is to acquire as many traits as possible so that depending upon the context you can invoke one or two or a subset of the available traits. That’s another significant advantage.
Also, there are other types of advantages with respect to performance. One of which is called, as the increasing degree of freedom, or increasing the capacity of a biometrics system. What we mean by that is by incorporating multiple traits you can accommodate more users in the system in such a way the the overlap between subjects is reduced.
These are some of the advantages of fusion system. At the same time, we have to be mindful of the trade‑offs.
Again, there is computational requirements which can increase. There can be the time requirement, which can also increase. Again, putting all these things into one framework would allow us to benefit from the use of multiple traits.