Finger Vein Segmentation from Infrared Images Using Spectral Clustering: An Approach for User Indentification

Abstract

Among biometric systems for user identification, finger vein patterns captured in the infrared spectrum have shown to be relevant for identifying users; and, in this way to provide a high level and low-cost security system. Unfortunately, the extraction of these vascular patterns is affected by many factors such as the capture device, light variations, force exerted on the finger, tissues, and bones with different morphology, finger position, etc. Therefore in this paper, we propose Spectral Clustering for the vein pattern extraction task from infrared images. To do so, the Spectral Clustering memory requirements for a large number of samples are attacked considering small disjoint partitions of the image and comparing resulting clusters in order to joint them avoiding the need for further expensive post-processing steps. Results are presented in terms of user classification error rates, showing that a good performance can be obtained by means of the proposed method.