Being one of the most common techniques to attack different systems, spoofing in biometrics occurs when synthetic biometric samples of some valid user are generated in order to authenticate an impostor as a legitimate user. In this publication, the authors proposed an algorithm for detecting face spoofing attacks that takes advantage of noise and artifacts added to the these synthetic biometric samples during their manufacture and recapture.
To do so, the approach uses time-spectral features as low-level descriptors, which gather temporal and spectral information in a single feature descriptor. To handle several types of attacks and to obtain a feature descriptor with a suitable generalization, the authors also proposed the use of the visual codebook concept to find a mid-level representation from time-spectral descriptors. It shows that capturing spatio, spectral and temporal features from biometric samples can be successfully considered in the spoofing detection scenario.
This research work is a partnership between UNICAMP and UFMG. The complete paper can be obtained from IEEE Transactions on Image Processing, Vol. 24, No. 12, December 2015.
The adopted dataset is available here.
Pinto, A.; Pedrini, H.; Robson Schwartz, W.; Rocha, A., “Face Spoofing Detection Through Visual Codebooks of Spectral Temporal Cubes,” in Image Processing, IEEE Transactions on , vol.24, no.12, pp.4726-4740, Dec. 2015. doi: 10.1109/TIP.2015.2466088