Multi-granular Finger Feature Recognition Based On Quotient Space

Accurate personal identification technology is closely related with social security, so the identification techniques always play an important role in social stability. Traditionally, the reliability of token-based identification methods often suffer being lost, forgotten and illegally duplicated. Therefore, biometric methods have been regarded as the best alternatives for personal identification. In this field, many works show that multimodal biometrics is better than unimodal biometrics in stability and discrimination. In this thesis, fingerprint, finger-knuckle-print and finger vein are used as multimodal biometric traits for exploring proper fusion methods with the support of quotient space model in Granular-computing. The follows are the main works of this thesis.First, the quotient space with structural properties is used as a problem-solution model for recognition by reviewing four kinds of common Granular-computing models. In quotient space, recognition problem is hierachically performed by multi-granular computing. Secondly, a pixel-based quotient space is built for evaluating its performance in solving a multimodal biometrics task. Here, the used fingerprint, finger-knuckle-print and finger vein images are acquired by a home-made multimodal finger imaging device. To further extend the above quotient space model in feature level fusion, multi-granular recognition schemes are implemented respectively using ordinal features, direction of energy features and Hu invariant moment features. Finally, a new granular descriptor is proposed based on multiscale multimodal features.The experimental results show that the quotient space model is effective for solving a multi-granular recognition problem. And the recognition accuracy can be improved reliably as well as computing cost reduction.