IEEE TRANSACTIONS ON CYBERNETICS, VOL. 43, NO. 3, JUNE 2013 843 Dynamic Detection-Rate-Based Bit Allocation With Genuine Interval Concealment for Binary Biometric Representation Meng-Hui Lim, Andrew Beng Jin Teoh, Senior Member, IEEE, and Kar-Ann Toh, Senior Member, IEEE Abstract—Biometric discretization is a key component in bio- metric cryptographic key generation. It converts an extracted biometric feature vector into a binary string via typical steps such as segmentation of each feature element into a number of labeled intervals, mapping of each interval-captured feature element onto a binary space, and concatenation of the resulted binary output of all feature elements into a binary string. Currently, the detection rate optimized bit allocation (DROBA) scheme is one of the most effective biometric discretization schemes in terms of its capability to assign binary bits dynamically to user-specific features with respect to their discriminability. However, we learn that DROBA suffers from potential discriminative feature misdetection and un- derdiscretization in its bit allocation process. This paper highlights such drawbacks and improves upon DROBA based on a novel two-stage algorithm: 1) a dynamic search method to efficiently recapture such misdetected features and to optimize the bit al- location of underdiscretized features and 2) a genuine interval concealment technique to alleviate crucial information leakage resulted from the dynamic search. Improvements in classification accuracy on two popular face data sets vindicate the feasibility of our approach compared with DROBA. Index Terms—Biometric discretization, detection rate, genuine interval concealment (GIC), quantization. I. I NTRODUCTION C RYPTOGRAPHIC keys serve as a crucial constituent in cryptographic applications. Each cryptographic key not only functions as a verification mechanism for the associated identity but also enables further cryptographic usage within the application itself (i.e., encryption/decryption). Biometric has been regarded as a potential source for crypto- graphic key generation due to its uniqueness and convenience. A biometric-generated cryptographic key, known as Bio-Crypto key, has the advantage of being representative. Moreover, it can be generated without the need of remembrance (password) or possession (token) which can be forgotten or lost, respectively. Manuscript received March 23, 2011; revised March 4, 2012 and August 2, 2012; accepted August 28, 2012. Date of publication October 2, 2012; date of current version May 10, 2013. This work was supported by the Ministry of Education, Science and Technology of the Korean Government under Korea Science and Engineering Foundation Grant 2011-8-1095. This paper was recommended by Associate Editor G.-B. Huang. M.-H. Lim is with the Department of Computer Science, Hong Kong Baptist University, Kowloon, Hong Kong (e-mail: menghui.lim@gmail.com). A. B. J. Teoh and K.-A. Toh are with the School of Electrical and Electronic Engineering, College of Engineering, Yonsei University, Seoul 120-749, South Korea (e-mail: bjteoh@yonsei.ac.kr; katoh@yonsei.ac.kr). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TSMCB.2012.2217127 However, due to severe consequence of biometric compromise, much effort has been contributed to developing changeable biometrics for key generation [1], [21]. A typical way of generating a Bio-Crypto key would be via a binary biometric discretization (biometric quantization with binary encoding) based on the statistical distribution of training data. Given a collection of extracted feature vectors of all users, a feature space is initially quantized into a number of intervals. Each feature element is then mapped into a short binary string according to the label of the interval where the feature element is enclosed within. Note that the mappings involved are usually on a per-dimension basis, although mappings based on two [5] or more dimensions [18] have just started to attract attention. Eventually, every individual binary output is concatenated to form a final bit string. Apart from Bio-Crypto key generation, biometric discretiza- tion plays another important role in many template protection schemes where a binary representation of biometric features is needed. Examples of template protection schemes include fuzzy commitment [12] and fuzzy extractor [7]. Moreover, the binary representation is efficient in terms of matching and storage requirements and is crucial when the processing time and storage space are of high concern. Fig. 1 shows a pictorial demonstration of the 1-D 3-bit bio- metric discretization scheme based on a binary reflected Gray code (BRGC) [9] encoding scheme which is able to confine the Hamming distance of adjacent intervals to unity. Generally, for the dth feature dimension, the feature compo- nent of user j can be modeled by a genuine user probability density function (pdf) p d j (f ) due to potential intraclass varia- tions, while the feature component of the entire population can be modeled by a background pdf p d bg (f ). Based on these two pdfs, the theoretical false acceptance rate (FAR) α d (n d ) and the false rejection rate (FRR) β d (n d ) of an n d -bit discretization scheme can respectively be quantified as α d j (n d )=  I d G (n d ) p d bg (f )df (1) β d j (n d )=1 − δ d j (n d ) (2) where δ d j (n d )=  I d G (n d ) p d j (f )df (3) 2168-2267/$31.00 © 2012 IEEE