Finger Features Fusion Based On Tolerance Granular Space

Nowadays, the increasing demand for information security leads to a rapid development of multimodal biometric recognition system. Feature-level Fusion plays an important role in multimodal biometric recognition systems. However, the current theories on feature-level fusion are very imperfect in implementing feature combination organically.In this thesis, a Granular Computing(GrC) based scheme is proposed to realize the fusion of finger-print(FP), finger-knuckle-print(FKP), and finger-vein(FV) in feature level. The main work of the thesis is as follows:1. Traditional fusion method testing. A histogram-based feature-level fusion method and a score-level fusion method based on matching scores are implemented, respectively.Experiments show that multimodal methods are superior to unimodal methods in biometric recognition.2. Granular method selection for finger feature fusion. Firstly, the first layer granules are constructed, whose intensions are represented by their orient histograms, describing image textures, and extensions are formed by pixels, satisfying tolerance relationship. By concatenating and parallelizing the three biometric modalities, the second-layer granules are constructed with respect to intension-similarity and extension-similarity.3. A multimodal finger authentication and identification system is built based on an intension-fusion model. A multi-template poll scheme, based on the first granules is adopted in multimodal finger authentication system. Using first layer granules that belong to a sub-set roughly selected in second layer granules, an accurate matching scheme is introduced in multimodal finger identification system. Experiment shows that the proposed algorithm can achieve satisfactory identification results.