1 Binary Biometric Template Generation towards Security and Class Separability Eslam Hamouda ⋆† , Xiaohui Yuan † , Osama Ouda ⋆ Taher Hamza ⋆ , and Lei Chen ‡ ⋆ Faculty of Computers and Information, Mansoura University, Mansoura, Daqahlia, Egypt † Department of Computer Science and Engineering, University of North Texas, Denton, Texas, U.S.A. ‡ Department of Computer Science, Sam Houston State University, Huntsville, Texas, USA magic eslam2@yahoo.com, xiaohui.yuan@unt.edu, {oudaosama, taher hamza}@yahoo.com, chen@shsu.edu Abstract—Due to the wide usage of biometrics, its security issues deserve more attention. Many of biometric protection sys- tems require the biometric templates to be presented in a binary form. Therefore, extracting binary templates from real-valued biometric data is a key step in biometric data protection systems. In addition to meeting the security and privacy requirements, binary biometric templates allow fast matching and reduced storage. The main challenge of these approaches is how to convert the real-valued templates into corresponding binary representa- tion which retains the original information. In this paper, we present a novel method that employs Genetic Algorithms(GA)to generate a binarization scheme which used to transform the real-valued templates into robust binary ones. The main role of GA is to search for the optimal quantization and encoding parameters to generate the binarization scheme. Experiments were conducted with ORL face database for recognition. Our results demonstrated that binary templates achieved promising performance in terms of equal error rate for face recognition using a simple hamming distance classifier. Index Terms—Biometrics, Biometric Template, Binarization, Quantization, Encoding, Security, Privacy, Genetic Algorithm I. I NTRODUCTION Biometrics has been widely adopted in various applications that require authentication or verification. However, the digital presence of biometric data also invites attacks on a biometric system. The original biometric data could be recovered from its digital presence and the compromised biometric data could be cross-matched among biometric databases [1]. Therefore, securing biometric data is a crucial need in this network era. Many biometric template protection systems have been developed to ensure biometrics privacy and security using the transformed biometric templates rather than the original data [2]–[4]. Many of the existing biometric template protection sys- tems, such as fuzzy commitment scheme [5] and fuzzy vault scheme [6], are based on coding theory. A discrete form is needed for encrypting the biometric template. Therefore, extracting discrete templates from the original biometric data is a fundamental and crucial step. Moreover, it leads to fast matching and compressed storage for the biometric templates compared to the real-valued biometric systems [7] . Many methods have been proposed to transform real-valued biometric templates into binary templates to improve security and protect against system intrusions [8]–[13]. These methods focus on maximizing the security and privacy property yet with little considerations on the discrimination power of the generated templates. As a consequence, the recognition accu- racy of biometric system could be degraded. Two fundamental steps in a binarization process are quantization and encoding. To determine the quantization bins and the encoding method, heuristic techniques are usually used [14], [15]. When user- specific parameters are used, encryption for these parameters is needed because they must be saved as helper data in the biometric system. Quantization and encoding are crucial to the security and discrimination power of the derived biometric templates. Map- ping from the continuous real-value domain to the discrete binary domain is non-trivial, especially when optimal perfor- mance and balance between security and discrimination are demanded. In addition, the sheer volume of possible ways of conversion is daunting. To address these challenges, we propose a binary biometric template generation method using genetic algorithm (GA) [16] to search for the optimal config- uration for the quantization intervals and encoding schemes. In our method, a GA individual represents the binarization scheme. The length of an individual is the same as the number of dimensions for the input real-valued biometric data. The fitness function is designed to maximize the inter-class dis- tance while minimizing the intra-class distance to enhance the discrimination power for the new generated binary templates. The proposed method uses GA to search for the optimal quan- tization parameters (the number of discrete intervals and the intervals boundaries) for each real-valued feature dimension. Moreover, different discrete-to-binary encoding methods are used to encode each interval into binary representation in order to ensure the security requirement. To evaluate our proposed method, comparison studies were conducted using the AT&T ORL face database. The rest of this paper is organized as follows: a review of related works in extracting binary templates from the real- valued biometric templates is given in section II. The proposed method is introduced in section III. Experimental results are presented in section IV. Finally, section V concludes this paper with a summary of our results. II. RELATED WORKS The existing approaches of binary templates extraction can be categorized into two groups: local methods and global IEEE - 33044 5th ICCCNT - 2014 July 11-13, 2014, Hefei, China