Features-Level Fusion of accumulated Edge Magnitudes Images in Finger-Knuckle-Print authentification

Abstract

These days, new technological solutions are gradually being implemented to fight against identity theft, document fraud, cybercrime, and threats from terrorism. Among these technologies, biometrics quickly emerged as the most relevant to identify and authenticate personnel reliably and quickly, based on unique biological characteristics. This paper presents a novel and efficient scheme to extract features for Finger-Knuckle-Print (FKP) recognition. First, the Accumulated Edge Magnitudes Images (AEMI) has been extracted from each ROI FKP image. Then, Gabor Filter bank is utilized to extract features from these AEMIs images. The extracted features from AEMIs images are concatenated to form a large feature vector. Finally, kernel fisher analysis (KFA) is used as a dimensionality reduction technique to build the best feature representant for each FKP image. For matching stage, the cosine distance is used to ascertain the identity of the person. Experimental results using publicly available PolyU FKP dataset show that the presented framework notably attained best performance and outperformed the state-of-the-art techniques.