This paper presents a deep learning method for human authentication based on hand dorsal characteristics. The proposed method uses the fingernail (FN) and the finger knuckle print (FKP) extracted from the ring, middle and index fingers. The proposed method was evaluated using a dataset of 1090 hand dorsal images (10 each from 109 persons) which are processed by the hand skin detection, the denoising method, and the procedure adopted for extraction of both finger knuckle and fingernail. A multimodal biometric scheme is used to improve the authentication performance of the proposed system and make it more resistant to spoofing attacks. A Deep learning-based approach using a convolutional neural network (CNN) with AlexNet as a pre-trained model is employed. Different features, extracted from hand images, were combined at different levels using normalization and fusion methods proposed by the authors. Experimental results demonstrate efficiency, robustness, and reliability of the proposed biometric system compared to existing alternatives. Consequently, it can be developed in many real-world applications
