Multi-Algorithm Fusion with Template Protection E.J.C. Kelkboom, X. Zhou, J. Breebaart, R.N.J. Veldhuis, C. Busch Abstract— The popularity of biometrics and its widespread use introduces privacy risks. To mitigate these risks, solutions such as the helper-data system, fuzzy vault, fuzzy extractors, and cancelable biometrics were introduced, also known as the ﬁeld of template protection. In parallel to these developments, fusion of multiple sources of biometric information have shown to improve the veriﬁcation performance of the biometric system. In this work we analyze fusion of the protected template from two 3D recognition algorithms (multi-algorithm fusion) at feature-, score-, and decision-level. We show that fusion can be applied at the known fusion-levels with the template protection technique known as the Helper-Data System. We also illustrate the required changes of the Helper-Data System and its corresponding limitations. Furthermore, our experimental results, based on 3D face range images of the FRGC v2 dataset, show that indeed fusion improves the veriﬁcation performance. I. I NTRODUCTION There is a growing popularity of using biometrics in applications ranging from simple home or business applica- tions with a small and limited group of enrolled people (for example access control to buildings or rooms) to large-scale systems used by an entire nation or even the entire world (for example identity cards with biometrics or the electronic passport e-Passport). However, its widespread use increases the privacy risks such as identity theft or activity monitoring by cross-matching between biometric databases of different applications. The ﬁeld of template protection provides the technology that mitigates these privacy risks by transforming the biometric template with a one-way function in order to guarantee the irreversibility property and by randomizing the biometric template in order to guarantee that multiple protected templates from the same biometric sample cannot be linked with each other. In the literature, multiple solutions have been presented to solve these problems. Some examples are the Fuzzy Commitment Scheme [1], Helper-Data Systems (HDS) [2], [3], [4], Fuzzy Vaults [5], [6], Fuzzy Extractors [7], [8], and Cancelable Biometrics [9]. In parallel to these developments, fusion of multiple sources of biometric information has shown to improve the recognition performance of the biometric system. As stated in [10], the basic principle of fusion is the reconciliation of evidence presented by multiple sources of biometric infor- mation in order to enhance the classiﬁcation performance. As E.J.C. Kelkboom and J. Breebaart are with Philips Research, The Netherlands {Emile.Kelkboom, Jeroen.Breebaart}@philips.com R.N.J. Veldhuis is with the University of Twente, Fac. EEMCS, The Netherlands R.N.J.Veldhuis@utwente.nl X. Zhou and C. Busch are with the Fraunhofer Institute for Computer Graphics Research IGD, Germany {Xuebing.Zhou, Christoph.Busch}@igd.fraunhofer.de described in [10], multiple sources of biometric information can be extracted from the same biometric modality by (see Fig. 1 for the case of ﬁngerprints): (i) capturing a sample of multiple instances (left and right index ﬁngerprint or iris) with the same sensor, (ii) using different sensors to acquire a different type of biometric samples from the same instance, (iii) capturing multiple samples using the same sensor and instance, and (iv) extracting multiple feature representations of the same biometric sample using different algorithms. These cases are referred to as the multi-instance, multi- sensor, multi-sample, and multi-algorithm systems, respec- tively. Further more, the ﬁfth type is the multi-modal system, which is the fusion of sources of biometric information from multiple modalities, for example ﬁngerprint, face, iris, voice, palm or retina. To complete the summary from [10], the sixth type is referred to as the hybrid system, which consists of a combination of the aforementioned fusion types. Each multi- biometric fusion type can be implemented at feature-level, score-level, or decision-level of the biometric system. In [11], multi-sample, multi-instance, and multi-modal fusion has been applied using the Fuzzy Vault as the template protection system. For multi-sample fusion a single mosaiced template is obtained from multiple ﬁngerprint impressions from which the vault is constructed. For multi-instance fusion the union of the minutiae sets of the left and right index ﬁngers is used to construct the vault. For multi-modal fusion, a ﬁngerprint and an iris sample are combined by concatenating the unordered minutiae set with the transformed iriscode extracted from the ﬁngerprint and iris samples, respectively. The concatenated unordered set is used to construct the vault. The recognition performance improved for all three cases as well as the claimed security. Our Contribution: Our work consists of applying multi- algorithm fusion with the Helper-Data System. We show that fusion can be applied at feature-, score-, and decision- level and illustrate the required changes of the Helper-Data System and its corresponding limitations. We experimentally determine the performance of different fusion methods at each level. The experiments are based on 3D face range images of the FRGC v2 dataset [12], where we use two recognition algorithms from different vendors. The outline of this paper is as follows. In Section II we brieﬂy discuss the HDS system, while in Section III we discuss the application of multi-algorithm fusion at feature-, score-, and decision-level using the HDS system. The ex- perimental setup and results are provided in Section IV. We ﬁnish with the conclusions in Section V. 978-1-4244-5020-6/09/$25.00 ©2009 IEEE