Robust Feature-Level Multibiometric Classification
Ajita Rattani, Dakshina Ranjan Kisku, Manuele Bicego, Mássimo Tistarelli · 2006
This paper proposes a robust feature level based fusion classifier for face and fingerprint biometrics. The proposed system fuses the two traits at feature extraction level by first making the feature sets compatible for concatenation and then reducing the feature sets to handle the 'problem of curse of dimensionality'; finally the concatenated feature vectors are matched. The system is tested on the database of 50 chimeric users with five samples per trait per person. The results are compared with the monomodal ones and with the fusion at matching score level using the most popular sum rule technique. The system reports an accuracy of 97.41% with a FAR and FRR of 1.98% and 3.18% respectively, outperforming single modalities and score-level fusion.