Renewable Minutiae Templates with Tunable Size and Security Bian Yang, Christoph Busch, Davrondzhon Gafurov and Patrick Bours Norwegian Information Security Laboratory at Gjøvik University College Gjøvik, Norway {bian.yang,christoph.busch,patrick.bours,davrondzhon.gafurov}@hig.no Abstract—A renewable fingerprint minutiae template generation scheme is proposed to utilize random projection for template diversification in a security enhanced way. The scheme first achieves absolute pre-alignment over local minutiae quadruplets in the original template and results in a fix-length feature vector; and then encrypts the feature vector by projecting it to multiple random matrices and quantizing the projected result; and finally post-process the resultant binary vector in a size and security tunable way to obtain the final protected minutia vicinity. Experiments on the fingerprint database FVC2002DB2_A demonstrate the desirable biometric performance of the proposed scheme. Keywords-renewable biometric template; fingerprint minutiae; template protection; random projection; tunable size and security I. INTRODUCTION For security and privacy concerns, renewability [1,2] was required for biometric template protection mechanisms to generate irreversible and unlinkable biometric templates instead of their plain-text counterparts for identity verification. Many schemes [3-7] have been proposed to protect the ISO standards compliant minutiae-based fingerprint templates. As a general method [9,10] for biometric template protection, random projection provides good distinguishability (thus desirable for template diversification) and the distance preservation property for the transformed templates by projecting a fix-length biometric feature vector onto an orthonormal random matrix. However, it is not obvious how to apply random projection to minutiae templates because of requirements such as geometric pre- alignment and fix-length feature vector extraction. In addition, the non-full-rank matrix based projection still needs to be investigated in terms of irreversibility and unlinkability. We analyze the security of random projection in Section 2, regarding the diversification case, which was thought to be a merit of random projection. An enhanced random projection is proposed in Section 3. Section 4 presents experimental results on the database FVC2002DB2_A. Section 5 concludes this paper. The proposed scheme is to achieve there targets: a. tunable size and security level for the protected binary templates; b. desirable biometric performance; c. reliable geometric alignment without needing global geometric reference such as core or delta. II. THE LINKAGE ATTACK Infinite solutions for a non-full-rank linear equation system underlies the irreversibility achieved by random projection [8,9]. For a p-dimensional biometric feature vector b = (b 1 , b 2 , …, b p ) and a non-full-rank orthonormal random matrix R ∈ R p×q (q < p), the binary projected vector is t = Q(kR T b) (1) where k is a constant, Q(.) is a quantizer to output an q-bit string t = (b 1 , b 2 , …, b m ) as the protected template. The irreversibility achieved by the non-full-rank projection can however be easily compromised if an attacker knows the linkage among I protected templates t i diversified from the same biometric feature vector b by projection onto different R i with the following condition satisfied: Rank([R 1 R 2 … R I ]) = p (2) where [R 1 R 2 … R I ] is the matrix horizontally concatenated by the I different R i . Despite the information loss caused by the quantization, the attacker can exploit the leaked information to invert the genuine biometric feature vector b or at least to shrink the searching space for b. III. PROPOSED SCHEME Similar to [7], the proposed scheme is based on local minutiae quadruplets (called vicinity). However, each minutia vicinity is transformed to a fix-length binary hash vector. Figure 1 illustrates the procedure to extract a size and security tunable vicinity hash H i . A. Minutiae Ordering and Geometric Alignment For each minutia m i (i=1, 2, …, L) in the original template consisting L minutiae, 3 closest neighboring (in Euclidean distance) minutiae are found around m i and the quadruplets (including m i ) are defined as a minutia vicinity V i . Denote the 3 neighboring minutiae as c ij (j = 1, 2, 3) indexed in an ascending order of distance from m i , 6 orientations O k (k = 1, 2, …,6) can be defined between minutiae pairs (e.g., pair m i and c i2 defining O 2 in the Figure 2) and along each orientation the remaining minutiae pair (c i1 and c i3 in the Figure 2 example) can be geometrically-aligned 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.221 882 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.221 882 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.221 878 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.221 878 2010 International Conference on Pattern Recognition 1051-4651/10 $26.00 © 2010 IEEE DOI 10.1109/ICPR.2010.221 878