Study On The Construction And Hierarchical Analysis Of Finger Feature Granules

In recent years, multimodal biometric-based technology has been an attractive research direction in biometrics. Finger-based personal identification has been widely used because of its high user acceptance and convenience. How to reliably and effectively fuse the multimodal finger-features together, however, has still been a difficult problem in practice. To obtain a more accurate and robust identification system, a new multimodal finger-based recognition scheme is proposed based on granular computing(GrC). Taking fingerprint(FP), finger-vein(FV) and finger-knuckle-print(FKP) as the constitutions of a finger trait, the multimodal finger-feature fusion and recognition scheme is achieved by constructing hierarchical and structuralized feature granules.First, considering the line network characteristics of three modalities, it is necessary to study the feature analysis methods to express the stable finger features structure. To extract the ridge texture features of the finger images, three different feature extraction methods are adopted, namely Rotation Invariant Uniform Local Gabor Binary Patterns(RIU-LGBP), Gabor Orientation coding and Magnitude coding(OrientCode&MagCode) and Gabor Ordinal measures(GOM), respectively. After feature extraction, combining the three modalities feature maps in a color-based manner, the original feature domain of a finger is then constituted. To represent finger features effectively, it is granulated at multiple levels of information granularity in a bottom-up manner based on GrC. For three different feature extraction methods, three different feature granulation methods are proposed, namely the granulation method based on clustering theory, the rectangle granulation and the circular granulation based on the spatial location. Moreover, a top-down matching method is proposed to test the performance of the multi-level feature granules. Experimental results show that the proposed methods are able to effectively achieve multimodal finger-feature fusion and recognition, while the top-down identification method not only improves the matching accuracy rate but also improves the matching efficiency.