How to deceive a face recognizer?

[joeblow]

http://citeseer.ist.psu.edu/cache/papers/cs2/219/http:zSzzSzwww.cmpe.boun.edu.trzSz~gokberkzSzbctp04.pdf/how-to-deceive-a.pdf

How to deceive a face recognizer?

Abstract

Many security systems depend upon face recognizers to identify a person. Many of these systems are passive and are deployed at places such as airline terminals. However, face recognizers are sensitive to deception attacks. Previous
studies suggest that hair regions are very crucial in face recognition and the success of a recognizer depends on the
success of a pre-segmentation stage which extracts the face region from the hair and the background. Deception attacks
which would change the hairstyle, apply make-up or occluding objects to the face would cause many systems to fail. In
this study, we study the effects of deception attacks on two basic face recognition systems: a PCA-based system and a
Gabor wavelet-based recognizer. We study the performance of the recognizers under different attacks and focus on the
selection of features so as to maximize performance under attacks.

Conclusion

In this paper, we analyze several deception attacks which use internal facial variations such as hair color change, expression variations, and occlusions by moustache and eyeglasses. Results show that both PCA-based and Gabor
wavelet-based face recognizers are sensitive to these variations, although the latter generally outperforms the first. In
hair color experiments, we see that PCA performance deteriorates drastically, while Gabor-based classifier is more
robust to color changes. In eyeglasses experiments, since a large portion of a face is occluded, both approaches perform
poorly. It is also shown that adding moustache does not effect the recognition rate significantly. After these observations, we propose a robust classifier which uses asymmetric trimmed distance measure. This distance measure is
suitable for modular representations. Therefore, a modular PCA algorithm is used to represent local facial regions. Our
experiments show that using asymmetric trimmed distance measure with modular PCA and Gabor methods significantly
improves the recognition performance when test images have considerable variations such as hair color change
and eyeglasses.