A graph theoretic approach for user identification using palmprint biometrics

In this paper a novel method for user identification using spectral properties of a graph of the palmprint is proposed. The orientation of the principal lines of palmprint is used to categorize the palmprint image into either right hand or left hand. The region of interest containing the principal lines of palmprint is divided into four zones and ´Hu´ moments are extracted from these zones. The ´Hu´ moments are represented as weights of undirected graph and spectral properties of the graph are extracted and used as features vectors. This topological characterization to the image moments is a novel feature of this work. The principal eigenvectors of the cost adjacency matrix of the graph are employed to represent the characteristic features of the palmprint. User identification is performed using multiclass support vector machine. The proposed method is evaluated on a database of 193 (386 palms) users, with 20 palmprint images of each user from PolyU palmprint database. The experimental results with cumulative match characteristic (CMC) curve demonstrate the performance for a rank-one of 97.28%.