Score Based Biometric Template Selection Yong Li, Jianping Yin, En Zhu, Chunfeng Hu, Hui Chen School of Computer, National University of Defense Technology, Changsha, 410073, China
[email protected],
[email protected],
[email protected],
[email protected],
[email protected] Abstract A biometric identification procedure usually contains two stages: registration and authentication. Most biometric systems capture multiple samples of the same biometric trait (e.g., eight impressions of a person’s left index finger) at the stage of registration. As a result, it is essential to select several samples as templates. This paper proposes two algorithms maximum match scores (MMS) and greedy maximum match scores (GMMS) based on match scores for template selection. The proposed algorithms need not involve the specific details about the biometric data. Therefore, they are more flexible and can be used in various biometric systems. The two algorithms are compared with Random and sMDIST on the database of FVC2006DB1A, and the experimental results show that the proposed approaches can improve the accuracy of biometric system efficiently.
1. Introduction Biometric recognition refers to the use of distinctive physiological or behavioral characteristics for automatically confirming the identity of a person. Biometric systems usually include two stages: registration and authentication. In practice, most biometric systems will capture multiple samples of the same biometric trait at the stage of registration. For example, some fingerprint systems require the user to enroll eight impressions into the system. There are three techniques to deal with the multiple samples literally. The first method [2,3] is template selection which chooses several samples as templates from the multiple samples and extracts feature from each template during Enrollment. At the authentication stage, the inquiry fingerprint is matched with these templates and then score level fusion or decision level
978-1-4244-2175-6/08/$25.00 ©2008 IEEE
fusion is used to give the final decision. The second method [5,7] is to combine all the samples as a super template and extract features from it. The third method [1,4,6] is to extract features from each sample first and then combine the features together as one feature template. With the development of computer hardware and software especially for storage and parallel processing technique, storing multiple templates is feasible. Techniques on combining image files or features mostly used in fingerprint recognition may not suitable for other biometric recognition systems. Also, such techniques need deep comprehension of feature extraction and match details. In addition, it is difficult to combine too many raw biometric samples or feature templates together. So it is helpful to select template based on match scores after Enrollment. The rest of the paper is organized as follows. In Section 2 the model of templates selection has been studied. In Section 3 the relation between sample number and template number has been analyzed. Based on match scores, two algorithms have been described for template selection in section 4. To study the effectiveness of the proposed technique, section 5 gives the experimental results. The last section summarizes the results of this work and provides future directions for research.
2. Multiple Templates Selection Analyze 2.1 Templates selection and multibiometrics Biometric recognition systems usually keep several templates and execute multiple matches in authentication for an individual. The system validates a person’s identity by comparing the captured biometric data with his own biometric templates stored in the database. Systems which keep more than one template for each individual may be classified as a kind of multisample system. Based on the theory of
multibiometrics, multiple templates can gain high matching accuracy. Consider a two-class classification problem and a multiclassifier system consisting of N classifiers (assume N is odd); the majority vote rule classifies an input pattern as belonging to the class that obtains at least K=(N+1)/2 votes. If p is the probability that a single classifier performs correctly, then the probability of the multiclassifier is as follows[8]:
P( N ) =
N
∑(
N m
) p m (1 − p ) N − m
(1)
m= K
The formula (1) assumes that the classifiers themselves are statistically independent. However, multiple templates come from the same biometric trait, which can not be independent in practice. Anyway, multiple templates can improve the accuracy of biometric system, but how many templates are suitable and how do multiple templates achieve better performance can be a problem. In this paper, we studied the two problems and got some useful conclusions.
2.2 A model of template selection The problem of template selection may be posed as follows: Choose K(K