Multimodal Finger Feature Recognition Based On Traditional Granulation

Today, the demand for accurate person identification using a security means is increasing dramatically. Traditionally, the token and knowledge-based identification methods can be counterfeited and cracked effortlessly, these means, therefore, are not effective for property protection in practice. In contrast, the biometric traits hold their own advantages in identification by using biological patterns of human body. However, unimodal biometrics is unfortunately imperfect in dealing with an actual identification task since a unimodal trait always behaves its own limitations in stability. Hence, multimodal biometrics has been a hot topic in biometric community. Unfortunately, the theory about the feature fusion of multimodal biometric patterns has not been well established yet. This greatly hinders the development of multimodal biometric technology in real applications. Hence, this dissertation proposes a Granular-computing(GrC) based strategy in order to exploit a new solution for multimodal biometric feature fusion in theory.In this dissertation, the fingerprint, finger-knuckle-print and finger vein from an identical finger are used for investigating the multimodal feature fusion problems using GrC theory. Here, the tolerance granular space is selected for modeling the process of feature granularity in a bottom-up manner considering the convenience of combining granular computing with multimodal feature analysis. First, based on minutia point extraction, the Delaunay triangulation and the circular granulation methods are respectively proposed for granulating multimodal features using a clustering scheme in granule-level construction. Second, another multilevel granular model based on the spatial neighborhood relationship is proposed due to the sensitivity of the above proposed methods to minutia point extraction. In this model, we are able to solve the feature deforming problem in granulation by selecting different dividing schemes to different modalities and weaken the affect of finger pose variations on the recognition performance by using sliding-window granulating operation.Based on the database built using a home-made multimodal finger image acquisition device, we implement the recognition process in top-down granular matching scheme. The experimental results show that the proposed methods are powerful in persona identification. This shows that it is effective using GrC to address the problems in multimodal fusion recognition.