NEWSLETTER
Vol22-Newsletter
Vol22-Newsletter
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
Paper Summary<br />
3) Rank-level Fusion: After computing the match scores of each test signal, each biometric matcher (ECG or LDV)<br />
outputs a rank for each of the classes (enrolled individuals). We used techniques including the highest rank, Borda<br />
count, and the weighted Borda count (logistic regression) [5], to obtain the fused rank.<br />
4) Decision-level Fusion: The decision made by a biometric matcher can be an accept or reject, in authentication<br />
mode, or an identity or reject, in identification mode. Example techniques we utilized to fuse the decisions from both<br />
biometric systems, include, ?AND? and ?OR? rules [6].<br />
Fig. 1: A 5s segment of a subject?s simultaneously recorded ECG and LDV velocity waveforms.<br />
Result s:<br />
Using the score-level fusion techniques gave the best multibiometric performance improvement for both<br />
authentication and identification modes of operation. The authentication performance, as measured by EERs,<br />
improved from ~6% for each of the unimodal ECG and LDV biometric systems, to ~2% for the multibiometric system.<br />
Moreover, in the identification mode, the rank-1 accuracy improves from ~80% for each unimodal biometric system,<br />
to ~92% for the multibiometric system.<br />
Conclusions:<br />
We performed a large scale cardiovascular biometric experiment on 258 individuals from three sessions. This<br />
represents a significant extension of the results of uni-variate modeling [1], [2] to a multidimensional case. We<br />
performed a comprehensive study of the performance of the multibiometric system at different levels of the<br />
biometric pipeline, using standard fusion techniques. In one measure of performance, namely EERs, we showed that<br />
performance improves from about 5.3% to about 2.1%. Based on our analysis, we made recommendations on the<br />
methods that we found to work best in the authentication and identification modes of operation.<br />
References:<br />
[1] M. Chen, J. A. O?Sullivan, N. Singla, E. J. Sirevaag, S. D. Kristjansson, P.-H. Lai, A. D. Kaplan, and J. W. Rohrbaugh,<br />
?Laser Doppler vibrometry measures of physiological function: evaluation of biometric capabilities,? TIFS, vol. 5, no. 3,<br />
pp. 449?460, 2010.<br />
[2] I. Odinaka, P.-H. Lai, A. D. Kaplan, J. A. O?Sullivan, E. J. Sirevaag, and J. W. Rohrbaugh, ?ECG biometric recognition: A<br />
comparative analysis,? TIFS, vol. 7, no. 6, pp. 1812?1824, Dec. 2012. [3] A. K. Jain and A. Ross, ?Multibiometric systems,?<br />
Commun. ACM, vol. 47, pp. 34?40, Jan. 2004.<br />
[4] A. Ross, K. Nandakumar, and A. K. Jain, Handbook of Multibiometrics. Springer-Verlag New York, Inc., 2006.<br />
[5] T. K. Ho, J. Hull, and S. Srihari, ?Decision combination in multiple classifier systems,? PAMI, vol. 16, no. 1, pp. 66?75,<br />
Jan 1994.<br />
[6] J. Daugman, ?Combining Multiple Biometrics.? http://www.cl.cam.ac.uk/_jgd1000/combine/combine.html, 2000.