USINGDISTRIBUTEDSOURCE CODING TO SECURE FINGERPRINTBIOMETRICS Stark C.Draper, Ashish Khisti, EminMartinian, Anthony Vetro, Jonathan S.Yedidia Mitsubishi Electric Research Laboratories (MERL)
Broadway Ave, Cambridge Ma · 2007
Wedescribe amethod toencode fingerprint biometrics securely for use,e.g., inencryption oraccess control. Thesystem issecure because thestored datadoesnotsuffice torecreate theoriginal fingerprint biometric. Therefore, abreach indatabase security doesnot lead totheloss ofbiometric data. Atthesametimethestored data suffices tovalidate aprobe fingerprint. Ourapproach isbased on theuseofdistributed source coding techniques implemented with graph-based codes. Wepresent astatistical modeloftherelationship between theenrollment biometric andthe(noisy) biometric measurement taking during authentication. Wedescribe howtovalidate orreject acandidate biometric probe given theprobe andthestored encoded data. Wereport theeffectiveness ofourmethod astested on adatabase consisting of579datasets, eachcontaining roughly 15 measurements ofasingle finger. Wethereby demonstrate aworking secure biometric system forfingerprints. IndexTerms-Biometric, fingerprint, distributed coding, SlepianWolfcoding, belief propagation decoding