Stage 3: Stage 3: Transform variable-size PGTQ to fixed-length vector
4.6 Implementation of Fuzzy Commitment
with the value 1 indicates the highest information rate. However, obtaining ideal information rate is nearly impossible in practice due to the inevitable dependency caused by the biometric feature extraction. Table 4.8 shows the estimated entropy in bits and information rate for FVC2002 and FVC2004 databases. From Table 4.8, it can be observed that the maximum and minimum entropies obtained by the proposed schemes are 256 bits over 280 bit-length string (FVC2002 DB2 for TKLSH-DQ) and 171 bits over 185 bit- length string (FVC2004 DB2 for TKPCA-DQ), respectively. The information rates achieved by the proposed schemes are approximately from 0.9 to 0.94.
These results have demonstrated the significant randomness of the binary templates compared to the results from (Zhou et al., 2011).
Table 4.8: Estimated entropy in bits and information rate on FVC2002 and FVC2004 databases.
Entropy &
Information Rate
FVC2002 FVC2004
DB1 DB2 DB3 DB1 DB2 DB3
TKPCA-DQ 178/0.94 (190bits)
171/0.92 (185bits)
172/0.93 (184bits)
192/0.93 (206bits)
205/0.93 (220bits)
206/0.93 (221bits) TKLSH-DQ 249/0.89
(280bits)
254/0.91 (280bits)
248/0.89 (280bits)
234/0.92 (256bits)
231/0.90 (256bits)
256/0.92 (280bits)
thus we compare our method with Nandakumar (2010) on FVC2002 DB1 and DB2. In this implementation, BCH error correction code is applied and TKLSH-DQ is used for demonstration. TKLSH-DQ is 280-bits representation for FVC2002.Yet, the length of the codeword for BCH is set to 511. So zero padding on TKLSH-DQ is required in order to perform XOR operation. Table 4.9 shows the FAR/FRR of this implementation as well as comparison with Nandakumar (2010). Although, the results are poorer than Nandakumar’s work (2010), it can be justified that the proposed method is solely based on minutiae information while Nandakumar (2010) requires minutiae alignment (e.g. high curvature points), which is not ISO compliant. Thus the performance of the proposed method is comparable to state-of-the-art.
Table 4.9: FAR/FRR of fuzzy commitment implementation using proposed TKLSH-DQ as well as comparison with Nandakumar (2010).
Databases
FMR/FNMR Nandakumar
(2010)
FAR/FRR (%) BCH (n, k)
(511, 40)
BCH (n, k) (511, 58)
BCH (n, k) (511, 67) FVC2002
DB1 (0.1)/(12.5) (9.41)/(4) (4.18)/(6) (1.61)/(8.5) FVC2002
DB2 (0.1)/(8.9) (7.71)/(5.75) (3.05)/(9.5) (1.07)/(13.75) 4.7 Discussion and Summary
Several points regarding the usability of the proposed framework are further highlighted: (1) other than the modified PGTQ-based minutiae descriptor used in this implementation, various binary or real-valued minutiae descriptors, e.g. Minutiae Cylinder Code (MCC) (Cappelli et al., 2010) can be
flexibly replaced and integrated into the proposed framework. This is because the proposed framework only converts the matching score of the minutiae descriptors from a variable-size descriptor to a fixed-length representation; (2) the matching algorithm between minutiae descriptors has to be carefully designed because the matching scores are used to form the kernel matrix, which is sensitive to the performance; (3) besides the stability-dependence dynamic quantization methods, other feature binarization methods such as DROBA (Chen et al., 2009) can also be applied.
Finally, it is concluded that, in this chapter, a generic framework is proposed to convert variable-size minutiae descriptor into a fixed-length representation. The framework is comprised of four main components:
minutiae descriptor extraction, fixed-length feature generation by kernel methods, feature binarization and matching. The experiment shows that the performance of the proposed TKPCA-DQ and TKLSH-DQ is comparable to the-state-of-the-arts for the fixed-length representations. Besides the feasible accuracy performance, high matching efficiency is also achieved.
Furthermore, the framework provides good adaptability: The minutiae descriptor, kernel, and feature binarization components used in this implementation can easily be replaced with better state-of-the-art components.
Additionally, the randomness and the correlation between the binary templates of different identities are extensively examined using entropy estimation with second order dependency tree and statistical independence test. In conclusion, all these advantages justify the feasibility of the proposed framework in
CHAPTER 5
BIOMETRIC CRYPTOSYSTEM: A NEW BIOMETRIC KEY BINDING AND ITS IMPLEMENTATION FOR FINGERPRINT
MINUTIAE-BASED REPRESENTATION
Despite Fuzzy Commitment (FC) is a theoretically sound biometric- key binding scheme, it relies on error correction code (ECC) completely to mitigate biometric intra-user variations. Accordingly, FC suffers from the security–performance trade-off. That is, the larger key size/higher security always trades with poor key release success rate and vice versa. Additionally, the FC is highly susceptible to a number of security and privacy attacks.
Furthermore, FC is limited to simple distance metrics such as Hamming distance to measure the dissimilarity of biometric features. This implies many efficient matching algorithms are to be abandoned. In this chapter, an ECC- free key binding scheme along with cancellable transforms is proposed for minutiae-based fingerprint biometrics. Apart from that, the minutiae information is well protected with a strong non-invertible cancellable transform, which is crucial to prevent a number of security and privacy attacks. The scheme is not limited to binary biometrics as in FC but instead can be applied to various types of biometric features and hence a more effective matcher can be applied. Experiments conducted on FVC2002 and FVC2004 show that the accuracy performance is comparable to state-of-the-
arts. It is further demonstrated that the proposed scheme is robust against several major security and privacy attacks.