Biometric Challenges
As I write this I am sat watching my eldest son's karate class. There are about 40 children in the hall but I was able to instantly recognise him. The human brain is remarkably adept at this tye of pattern recognition. Here in the UK there are plans for a national identity card scheme that will include biometric information. In addition to the political implications this scheme poses some interesting IT problems.
I have just read a paper by Anil K. Jain, Sharath Pankanti, Salil Prabhakar, Lin Hong, and Arun Ross entitled "Biometrics: A Grand Challenge" from "Proceedings of the 17th International Conference on Pattern Recognition (ICPR’04)". In it they survey the current state of the art for biometric identification and note some of the major problems that it entails.
Uses of Biometric Information
To begin with they identify three uses of biometric information
- Positive Identification (“Is this person who he claims to be?”)
- Large Scale Identification (“Is this person in the database?”)
- Screening (“Is this a wanted person?”)
The first one is relatively simple. We check the biometric information of an individual and see if it matches the information we have stored against the individual they claim to be. We will use this in a a national database to verify the identity of an individual requesting an entitlement.
The next one is more interesting. We would use this as part of the registration process to check if someone has already registered under another identity - trying to weed out duplicate applications. This requires us to scan the entire database for anyone matching the provided biometric information. An alternative scenario might be a body is found with no identification, are they registered in the biometric database.
The third one is more of an anti-terrorist type system, is an individual going through an airport on the list of terror suspects. This has a smaller database - 100s rather than millions - but the biometric acquisition may not be so good, imagine facial recognition under less than ideal environment.
Some of the Challenges
There are many challenges and the paper goes into several of them. A few that caught my eye were
- Biometric matching is inexact - for example finger print matching may have a 2% error rate
- Computational complexity limits scaling - indexing of biometric data is still very difficult and when coupled with the computational complexity of a comparison of biometric data this limits scalability of the solution.
- Security - systems must prevent the injection of false information as biometric systems have no mechanism for credential revocation (plastic surgery excepted).
- Privacy - the inevitable privacy debate impinges on such a potentially useful tracking system as biometric data.
Moving On
Life is never as simple as the sci-fi movies where biometrics seem to be the primary form of identification. There are still some things we humans do better than machines, such as recognising my son at a karate class from any angle.
