Finger-Knuckle-Print verification by fusing invariant texture and structure scores

Finger-Knuckle-Print (FKP) biometric traits for person recognition have recently gained much attention from both the research community and industry, owing to their distinctive features and higher usability or user friendliness. In this paper, a reliable and robust personal identification approach using FKP is presented. The proposed framework merges two types of matching scores extracted from structure and texture images. The Region Covariances Algorithm (RCA) has been employed by the presented method to extract the structure and the texture images from each FKP captured image. Gabor Filter bank and Kernel Fisher Discriminant (KFD) methods have been utilised to obtain distinctive feature vectors. Finally, the Cosine Mahalanobis distance similarity metric is used for classification. Experimental analyses were performed on Hong Kong Polytechnic University (PolyU) FKP database. Experimental results show that our proposed system achieves better results than prior state-of-the-art systems. In addition, fused scores using weighted sum rule in the proposed framework render very good performance compared to min, max and simple sum rules.