Possibilistic Modeling Palmprint and Fingerprint based Multimodal Biometric Recognition System

Unimodal biometric recognition systems have random performances which can be efficient for some contexts and not for others. Fusing multiple modalities improves recognition performance and reduces some limitations of biometric systems based on a single modality, such as intentional fraud or universality problem (inability to acquire data from certain persons for some modes). Furthermore, in real- world applications, biometric data imperfections affect the performance of the recognition algorithms which may be caused by several factors such as the poor image quality, noise and adverse illumination and/or contrast. To handle data imperfections or redundancy and cover data variability, possibilistic modeling is a powerful tool. In this paper, we propose a robust multimodal biometric recognition system integrating fingerprint and palmprint based on possibilistic modeling approach. The proposed approach relies on possibility theory concepts for modeling biometric features. Accordingly, a set of relevant biometric features extracted from image samples is statistically analyzed and represented by a possibility distribution. The biometric templates from the palmprint and fingerprint are used for decision making by applying a score level data fusion process. Validation of the proposed approach is done on the public palmprint image database, CASIA (Chinese Academy of Sciences Institution of Automation) and the public fingerprint database, FVC (Fingerprint Verification Competition). As performance metric, we adopt the area under the ROC curve. Each both modalities give an AUC (0,8694) and (0,9531) respectively for palmprint and fingerprint when used alone. A typique multimodal system using both modalities gives an AUC of (0,9531). The proposed multimodal system has an AUC very close to unity (0,9997).