Contactless Finger Knuckle Authentication under Severe Pose Deformations

Contactless biometrics identification using finger knuckle images has shown significant potential for the e-business and forensic applications. One of the key challenges in accurately matching the real-world contactless finger knuckle images is related to the knuckle pattern deformations that are involuntarily generated due to finger pose changes. Earlier work in this area therefore acquired fixed pose finger images for the authentication and therefore the performance achieved from such images cannot reflect the expected performance under the deployment scenarios. This paper adopts a new approach to accurately match such finger knuckle images and presents first attempt to authenticate finger-knuckle patterns under severe pose changes. This approach attempts to correct pose related deformations by identifying the correspondence between a fixed number of chosen points between two matched images. The match score is computed using local feature descriptors, at each of these correspondence points, and consolidated to generate average match score. The experimental results are presented in this paper, both using two-session and single-session index finger knuckle images from 221 different subjects, using publicly available database. These results are outperforming and indicate the merit of spatial-domain approach to match deformed finger knuckle images using a fixed number of correspondence points.