IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL. 3, NO. 2, JUNE 2008 183 Template-Free Biometric-Key Generation by Means of Fuzzy Genetic Clustering Weiguo Sheng, Gareth Howells, Michael Fairhurst, and Farzin Deravi, Member, IEEE Abstract—Biometric authentication is increasingly gaining pop- ularity in a wide range of applications. However, the storage of the biometric templates and/or encryption keys that are necessary for such applications is a matter of serious concern, as the compro- mise of templates or keys necessarily compromises the information secured by those keys. In this paper, we propose a novel method, which requires storage of neither biometric templates nor encryp- tion keys, by directly generating the keys from statistical features of biometric data. An outline of the process is as follows: given bio- metric samples, a set of statistical features is first extracted from each sample. On each feature subset or single feature, we model the intra and interuser variation by clustering the data into nat- ural clusters using a fuzzy genetic clustering algorithm. Based on the modelling results, we subsequently quantify the consistency of each feature subset or single feature for each user. By selecting the most consistent feature subsets and/or single features for each user individually, we generate the key reliably without compromising its relative security. The proposed method is evaluated on hand- written signature data and compared with related methods, and the results are very promising. Index Terms—Biometric authentication, clustering, feature eval- uation, handwritten signatures, security. I. INTRODUCTION A UTHENTICATION methods to verify the identity of a person are a key ingredient of many of today’s se- cure systems. Secure and reliable authentication methods are increasingly important, particularly in view of the growing importance of electronic commerce. Traditional authentication schemes primarily utilize tokens or depend on some secret knowledge possessed by the user for verifying his or her identity [23]. While these techniques are widely used, they have several limitations. For example, neither token-based nor knowledge-based approaches can differentiate between an authorized user and a person having unauthorized access to the tokens or secret knowledge. Recently, biometrics-based authentication schemes that can overcome these limitations while offering usability advantages have significantly enhanced traditional authentication strategies. The most common approach to biometric authentication is to capture the biometric template (i.e., physiological and/or be- Manuscript received April 14, 2007; revised January 29, 2008. This work was supported by the UK Engineering and Physical Sciences Research Council (EPSRC EP/C00793X/1). The associate editor coordinating the review of this manuscript and approving it for publication was Dr. Bart Preneel. The authors are with the Department of Electronics, University of Kent, Kent CT2 7NT, U.K. (e-mail: W.Sheng@kent.ac.uk; W.G.J.Howells@kent.ac.uk; M.C.Fairhurst@kent.ac.uk; F.Deravi@kent.ac.uk). Digital Object Identifier 10.1109/TIFS.2008.922056 havioral characteristics [15]), of each user during an enrollment phase. The captured biometric template is then stored in a ref- erence database, together with the user name, encryption key, and access privileges, etc. During the authentication phase, a new biometric sample is matched against the database infor- mation and the encryption key is released upon successful bio- metric matching. By storing the templates, however, this ap- proach introduces a number of security and privacy risks. First, an attacker may steal templates from a database and construct artificial biometrics which satisfy a subsequent authentication process. Second, once compromised, biometrics cannot be up- dated, reissued, or destroyed. Third, the stored templates may expose sensitive personal information (for example, retina scans reflect information about diseases such as diabetes and strokes [2]). It is also important to note that since the biometric matching is usually completely decoupled from the key release and out- puts only an accept/reject decision, the match decision can be overridden by the hacker. To protect against threats such as these, various methods have been proposed in the literature [27]. One potentially viable ap- proach is to generate encryption keys from the statistical fea- tures of biometric data [3], [8], [14], [24], [25], [33], [34], [37], [38]. In this approach, a set of statistical features is first extracted from the given biometric sample. Each feature is then compared with a threshold to generate one or more key bit(s). By cascading the key bits from all of the features, an encryption key can be generated or computed and subsequently used for security ap- plications. We refer to encryption keys generated or computed in this case as bio keys. The advantage of this approach is that the bio keys can be dynamically generated, and storage of nei- ther templates nor keys is required. Generally, a bio-key gen- eration method should effectively quantize the feature space by modelling both the intra and interuser variation in the biometric representations, thereby separating users in the sense that bio keys produced by the same user are the same, but simultane- ously ensuring that those produced by different users are indeed different. Known implementations of this approach, however, either have limited capability to model the intra and interuser variation or usually do not take into account the interuser varia- tion at all and, therefore, cannot effectively quantize the feature space. Further, most of them rely on the quantization solutions of single features to generate the bio key by concatenating key bit(s) from each of them. This may introduce another significant drawback since generally not all single features are consistent and, indeed, may be consistent in relatively few cases. In this work, we are also interested in developing a biometric authentication method which dynamically generates bio keys from the statistical features of biometric data. In our method, 1556-6013/$25.00 © 2008 IEEE