Score Level Fusion of Multimodal Biometrics based on Entropy Function

Abstract

This paper presents the score level fusion of multimodal biometrics using Hanman-Anirban entropy function. Entropy function captures the uncertainty in the scores. The experimental results ascertain that Entropy based score level fusion outperforms over existing methods of score level fusion such as t-norms, sum and max. We have validated our claim on finger-knuckle-print (FKP) dataset consisting of left index, left middle, right index and right middle FKP. The features of FKPs are extracted using the Gabor Wavelet. The implementation is done using MATLAB and the performance of the proposed technique is evaluated using Receiver Operating characteristics (ROC) curve. The proposed score level fusion approach achieves significant improvement in the performance over the individual FKP. We obtain Genuine acceptance rate of 99% with FAR of 0.001 %.

Symmetric sum‐based biometric score fusion

“…In the future, we will evaluate our method at a large-scale including its robustness against spoofing attacks. Table 4 Comparison of fusion using different approaches on NIST-fingerprint database Score-level fusion method for FAR = 0.01 % GAR (%) S-sum using probabilistic t-norm 89 S-sum using Hamacher t-norm 75.5 S-sum using Yager t-norm with p = 10.3 90 S-sum using Schweizer & Sklar t-norm with p = 0.9 89 S-sum using max rule 90.75 S-sum using min rule 82.5 Max rule [11] 90.3 Min rule [11] 79.6 SVM [9] 91.4 Likelihood ratio [9] 91.4 Entropy-with-Frank p = 0.01 [14] 87.77 Entropy-with-Hamacher p = 0.01 [14] 85.42 Table 5 Classifier fusion using (S-sum using Yager t-norm) on NIST-multimodal database Classifier S-sum using Yager tnorm with p = 1.…”

“…Also, t‐norms were previously employed in score level fusion of multimodal biometrics using Hanman–Anirban entropy function. For the record, this approach delivered a good performance [14]. The proposed method has been evaluated on publicly available two benchmark multibiometric databases, i.e., NIST‐multimodal and NIST‐fingerprints.…”

An Approach for Full Reinforcement-Based Biometric Score Fusion

“…This operator is created by utilizing the product T-norm and its corresponding dual for T-conorm (refer to Eqn. 11,12).…”

“…Moeen et al [12] introduced a score-level fusion technique for person recognition, utilizing the entropy function (Hanman-Anirban). For experimentation, they used the fingerknuckle-print images obtained from (FKP) dataset consisting of left index, left middle, right index and right middle FKP.…”

“…In their evaluation of multimodal, multi-unit, and multi-algorithm systems, they demonstrated that the WQAM combining rule they proposed surpasses the earlier suggested score combining method, which relied on fixed techniques like T-norms, trained methods such as support vector machines, and density assessment approaches like likelihood ratio. [11] 2013 Sum, max, min rule iris and fingerprint [34] 2013 Mean, max, min, median rule Palm-print-vein [13] 2014 Mean rule face and fingerprint [14] 2015 Max rule palm-dorsal vein [12] 2016 Hanman-Anirban entropy function finger-knuckle-print (left index, left middle, right index and right middle) [33] 2017 Adaptive weighted Left and right palmprint [15] 2017 Dubois and Parad T-norm left and right wrist vein patterns [7] 2018 Symmetric sum rule face and fingerprint [16] 2019 T-norm wrist, palm vein images [8] 2020 Weighted quasiarithmetic mean face and fingerprint [9] 2020 Asym-AO palmprint, fingerprint, face and ocular left eye…”