Finger vein identification using deeply-fused Convolutional Neural Network

Finger vein identification is a recently developed biometric technology and has become an essential field in biometrics, garnering increasing attention in recent years. As a biometric trait, using vein patterns allows for personal recognition with high security. In this paper, we have employed an improved deep network, named Merge Convolutional Neural Network (Merge CNN), which uses several CNNs with short paths. The scheme is based on the use of multiple identical CNNs with different input images qualities, and the unification of their outputs into a single layer. To achieve this, we designed different networks and trained them with the FV-USM dataset. The most optimal CNN architecture was used to build our final merged CNN labeled A, which is a combination of original image and image enhanced with Contrast Limited Adaptive Histogram (CLAH) method. Using six images for training, satisfactory performances were obtained from the FV-USM database with a recognition rate of 96.75%. Our proposed approach showed better performance than other methods exist in the literature, for the SDUMLA-HMT database with a recognition rate of 99.48%, when using five images for learning. Our proposed scheme can compete with state-of-the-art methods with recognition rate of 99.56% for the THU-FVFDT2 database.(c) 2020 The Authors. Production and hosting by Elsevier B.V. on behalf of King Saud University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).