Finger knuckle image based personal authentication using DeepMatching

Recently under the lights of mega Indian Aadhaar [2] project, Indian government has been trying to initiate a robust linkage between several financial inclusion schemes (such as NREGA and MNREGA) and biometric based personal authentication mechanism. This will ensure a corruption free realization and management of such policies. Among commonly used biometric traits, fingerprint is the most accepted mode of identification. However, fingerprints are found monotonous in case of laborers and workers due to rough usages of hand in variety of day-to-day works. This will seriously affect any such initiation adversely, especially in India, which is primarily a rural country. On the contrary, quality of finger dorsal patterns remain better under such situations. In this work, we are proposing a novel finger-knuckle based biometric system to check the escapes that are present in transfer of payments through the various levels of bureaucracy financial inclusion projects. The method is based on finger knuckle local features that has been preprocessed by using ROI extraction, enhancement and proposed feature transformation schemes. In the classification process, a novel Deep-matching technique has been used to match the non-rigid regions between finger knuckle images. The experimental evaluation of proposed system has been carried out using publicly available PolyU finger-knuckle-print database of 8000 images collected from 165 subjects. Our results demonstrate the discriminative ability of transformed knuckle features (CRR-99.10%, and EER-0.98%) in improving the performance of traditional FKI biometric system.