Multimodal Biometrics Using Fingerprint, Palmprint, and Iris With a Combined Fusion Approach

1. Introduction

As compare to traditional personal authentication methods like token-based authentication (key or id) or knowledge based authentication (password), biometrics provides more security (Chin, 2013). Biometrics verifies or identifies the persons based on measurement of their physiological or behavioral characteristics (Luque-baena, 2013). With biometrics, user has to physically interact with the system which makes biometrics stronger from security point of view. Unimodal biometrics employs a single biometric feature to authenticate the person but use of just single biometrics feature has to face with some problems like trouble with noisy data, lack of universality and lack of distinctiveness etc. By considering these bottlenecks of unimodal, biometric research community is now shifting to multimodal biometrics which uses multiple biometric features to authenticate the person (Conti, 2010, Sanjekar, 2013). The multibiometric system can be categorized as multi-sensors systems, multi-algorithm systems, multi-instance systems, multi-sample systems or multi-modal systems (Unar, 2014). The information gained from multiple biometric resources can be merged at various levels like sensor level, feature level, matching score level or decision level. The sensor level fusion and feature level fusion are the types of pre-classification fusion that means the modals are combined prior to classification phase. The match score level fusion, rank level fusion and decision level fusion are post-classification types of fusion where modals are integrated after classification phase. The fusion at feature level involves the consolidation of feature vectors from different biometric resources and forming the new feature vector for matching process. The integration of these feature vectors may lead to the very large feature vector also it may be very difficult to combine the feature vectors that are incompatible with each other (Ross, 2005, Dinca, 2017, Joshi, 2009). The feature level fusion can be categorized as serial feature fusion and parallel feature fusion. In former case simply concatenation of feature vectors one after another is carried out while later combines feature vectors into complex vectors (Haghighat, 2016). In feature level fusion there is a need of normalization of the features to convert them into same range which allows use of same threshold during matching process. The feature level fusion consists of rich information as compared with match score level fusion and decision level fusion.