Article Preview
Top1. Introduction
Biometrics has emerged over the last decade as a major field that deals with part of security issues. It involves automatically identifying individuals based on their physiological (e.g., fingerprint, face, iris, palm print, etc.) and / or behavioral (e.g., voice, gait, signature, etc.) characteristics, using mathematical analysis. This leads to a more natural and reliable identification of individuals by focusing on "who they are", rather than being concerned with "what they know" (e.g., username and password) or "what they have" (e.g., access card). Practically speaking, biometric systems are currently available in different types and in large numbers, covering several functionalities. Depending on the number of modalities, biometric systems are divided into two categories: unimodal biometric systems and multimodal biometric systems. In the former category, only one biometric modality is considered while carrying out the identification process. The latter category deals generally with several biometric modalities that are combined within the identification process. Multimodal biometric systems include also the following subcategories (Taouche et al., 2014): multi-sensor systems, multi-algorithm systems, multi-instance systems and multi-sample systems.
Despite their simplicity of implementation, unimodal biometric systems have several limitations, among which we mention: the limited recognition rate, the non-universality and the lack of individuality of the chosen biometric modality in addition to the potential intrusion attempts. Hence, multimodal biometric systems are believed to be an effective way to address these limitations. The identification process within multimodal systems comprises the following phases. Firstly, the considered modalities are acquired through sensing devices. Then, features are extracted from the modalities using descriptors. After that, a matching between the features extracted and the template features stored in the database is performed. Finally, a decision is made for which an acceptance or a rejection is provided. A fusion can take place at different levels, depending on the modalities processed: acquisition level, feature level, score level and decision level.
In this work, we treat the case of multimodal biometric systems involving palmprint modality. Palmprint recognition is a relatively young authentication technology that has gained more attention from researchers (Kong, Zhang, & Kamel, 2009; Jain & Feng, 2009). Compared to other biometric modalities, the palmprint trait satisfies most of the critical properties of biometric characteristics such as universality, individuality, stability, and collectability in addition to its simplicity of use. In the literature, several approaches for palmprint recognition have already been proposed, in which the palmprint modality has been considered either alone or combined with other modalities such as fingerprints, face, ear and hand shape (Ghulam Mohi-ud-Din et al., 2011; Raghavendra et al., 2011; Xu et al., 2011, 2013; Saini and Sinha, 2015; Farmanbar and Toygar, 2016; Hezil and Boukrouche, 2017; Charfi et al., 2017; etc.). Although such proposals have solved many authentication issues, their analysis allowed us to identify some weakness that may slow down their general performances, related to: 1) possibility to make certain improvements so that the recognition rate would be enhanced, 2) negative effects on the running time and storage space, resulting from the use of more than one sample training image.