Biometric systems are considered as human pattern recognition systems that can be used for individual identification and verification. The decision on the authenticity is done with the help of some specific measurable physiological or behavioral characteristics possessed by the individuals. Robust architecture of any biometric system provides very good performance of the system against rotation, translation, scaling effect and deformation of the image on the image plane. Further, there is a need of development of real-time biometric system. There exist many graph matching techniques used to design robust and real-time biometrics systems. This chapter discusses different types of graph matching techniques that have been successfully used in different biometric traits.
TopIntroduction
Biometric systems (Jain, et. al., 2004; Jain, et. al., 2006) are considered as human pattern recognition systems. They can be used for individual identification and verification which is determined by some specific measurable physiological or behavioral characteristics (Jain, et. al., 2004; Jain, et. al. 2006; Jain, et. al., 2007). These characteristics can be obtained from fingerprint, face, iris, retina, hand geometry and palmprint, signature, ear, gait and voice, etc. which satisfy the properties like universality, invariance, measurability, singularity, acceptance, reducibility, tamper resistance, comparable and inimitable. There exist many computational intelligence techniques (Jain, et. al., 2007) applied to biometric systems for feature extraction (Jain, et. al., 2007), template updating (Jain, et. al., 2007), matching and classification (Jain, et. al., 2007). However, this type of systems seeks efficient and robust performance in real time environments. These robust systems often degrade their performance because of uncontrolled environment and poor feature extraction, feature representation and pattern classification techniques.
There exist several graph matching techniques (Wiskott, et. al., 1997; Conte, et. al., 2003; Tarjoman, & Zarei, 2008; Fan, et. al., 1998; Mehrabian, & Heshemi-Tari, 2007; Abuhaiba, 2007) for identity verification of biometric samples which can solve problems like orientation, noise, non-invariant, etc that often occurred in fingerprint (Maltoni, et. al., 2003), face (Li, et. al., 2005), iris (Daugman, 1993), signature recognitions (Kisku, et. al., in press). Different graph topologies are successfully used for feature representations of these biometric cues (Jain, et. al., 2007). Graph algorithms (Conte, et. al., 2003; Gross, & Yellen, 2005) can be considered as a tool for matching two graphs obtained from feature sets extracted from two biometric cues (Jain, et. al., 2007). To describe the topological structure of biometric pattern, the locations at which the features are originated or extracted are used to define a graph. The small degree of distortions of features can easily be computed during matching of two graphs based on the position and distances between two nodes of the graph and also with the adjacency information of neighbor’s features.
This chapter makes an attempt and explain the way a graph can be used in the designing an efficient biometric system. Next section discusses the use of graphs in fingerprint, face and iris recognition. In Section 3, a complete graph topology has been used in a SIFT-based face recognition system. Section 4 describes the method of using probabilistic graphs and fuse invariant SIFT features of a face. Next section deals with the problem of using wavelet decomposition and monotonic decreasing graph to fuse biometric characteristics. Experimental results are given in Section 6 which concluding remarks are in the last section.