Info »

tom@fladsrud.com
+47 41 52 84 94

MSc Tom Fladsrud 2005 »

Face recognition is a biometric authentication method that has become more and more relevant in the recent years. From being too inaccurate, it is becoming a more mature technology deployed in large scale systems like the new Visa Information System. From the earlier FERET evaluations to the more recent Face Recognition Vendor Test 2000 and 2002 evaluations we have observed significant improvements in face recognition. Systems based on 3D face recognition even claims to distinguish between identical twins. During our research we have seen that even though face recognition have greatly matured since the earliest forms, there still exists several possible attacks against this technology. Some of the attacks reviewed in this report are specific to face recognition, while others apply for all authentication methods. During the deployment process of face recognition, these attacks should be taken in consideration. As Kosmerlj stated in her thesis; there is still work to be done to improve face recognition before it can be applied in high security settings or applied in large scale applications. One method to reduce the number of people being falsely accepted is by combining the face recognition system with human supervision.

To survey the additional value of a human supervisor, we conducted an experiment where we investigated whether a human would detect false acceptances made by a computerized system, and the role of hair in human recognition of faces. The study showed that, on average, humans were able to detect almost 80 % of the errors made by the computerized system. More over, the study shows that the ability of an individual to recognize a human face is a function of hair: the false acceptance rate was significantly higher for the image-pairs where the hair was removed compared to where it was present. This indicates that there is in fact a substantial opportunity for an impostor to circumvent the human guard using simple and cheap methods. Hair is a feature that may be easily manipulated, and this is perhaps the easiest and cheapest form of non-zero-effort attack on a face recognition system.

Keywords: Face recognition, False Acceptance Rate, False Rejection Rate, Attacks on face recognition systems, Biometrics, Visa Information System, Human supervision, The effect of hair on human recognition of faces

Master's thesis at Gjøvik University College - Master of Science in Information Security - 2005

Tom Fladsrud. Face Recognition in a Border Control Environment: Non-zero Effort Attacks' Effect on False Acceptance Rate. MSc Thesis. Juli - 2005.

Tom Fladsrud. Face Recognition Software: Non-zero effort attacks' effect on FAR of face recognition products used in a border control environment. MSc Project Plan. December - 2004.

Downloads from the MSc »

Face Recognition in a Border Control Environment: Non-zero Effort Attacks' Effect on False Acceptance Rate. My MSc Thesis. Juli - 2005. download

The article that I presented in Hong Kong at the IAPR International Conference on Biometrics 2006 - Face Recognition Issues in a Border Control. download