Extracting biometric binary strings with minimal area under the FRR curve for the hamming distance classifier

C Chen,R Veldhuis

doi:10.1016/j.sigpro.2010.09.008

Abstract

Extracting binary strings from real-valued biometric templates is a fundamental step in template compression and protection systems, such as fuzzy commitment, fuzzy extractor, secure sketch and helper data systems. Quantization and coding are the straightforward way to extract binary representations from arbitrary real-valued biometric modalities. Afterwards, the binary strings can be compared by means of a Hamming distance classifier (HDC). One of the problems of the binary biometric representations is the allocation of quantization bits to the features. In this paper, we first give a theoretical model of the HDC, based on the features’ bit error probabilities after the quantization. This model predicts the false acceptance rate (FAR) and the false rejection rate (FRR) as a function of the Hamming distance threshold. Additionally, we propose the area under the FRR curve optimized bit allocation (AUF-OBA) principle. Given the features’ bit error probabilities, AUF-OBA assigns variable numbers of quantization bits to features, in such way that the analytical area under the FRR curve for the HDC is minimized. Experiments of AUF-OBA on the FVC2000 fingerprint database and the FRGC face database yield good verification performances. AUF-OBA is applicable to arbitrary biometric modalities, such as fingerprint texture, iris, signature and face.

Highlights

Binary biometric representations are used in data compression and template protection [1]
We propose an area under the false rejection rate (FRR) curve optimized bit allocation (AUF-OBA) principle that minimizes the area under the FRR curve for the Hamming distance classifier (HDC)
The false acceptance rate (FAR)/FRR performances for FRGCH and FRGCL are shown in Fig. 2 and Fig. 3, where the FAR is plotted in log scale

Summary

Introduction

Binary biometric representations are used in data compression and template protection [1]. In order to maximize the attacker’s efforts in guessing the target template, the bits should be statistically independent and identically distributed (i.i.d.). The straightforward way to extract binary strings is by quantizing and coding the real-valued biometric templates: Firstly, independent features are extracted from the raw measurements. The final binary string is the concatenation of the bits from every feature. Independent of the quantizer design, a detection rate optimized bit allocation (DROBA) principle [5] was proposed to assign the number of quantization bits, based on the density distribution of every feature, so that the analytical overall detection rate of the binary string is maximized

Methods

Results

Conclusion