Generating and Validating DSA Private Keys from Online Face Images for Digital Signatures

Abstract

Signing digital documents is attracting more attention in recent years, according to the rapidly growing number of digital documents being exchanged online. The digital signature proves the authenticity of the document and the sender’s approval on the contents of the document. However, storing the private keys of users for digital signing imposes threats toward gaining unauthorized access, which can result in producing false signatures. Thus, in this paper, a novel approach is proposed to extract the private component of the key used to produce the digital signature from online face image. Hence, this private component is never stored in any database, so that, false signatures cannot be produced and the sender’s approval cannot be denied. The proposed method uses a convolutional neural network that is trained using a semi-supervised approach, so that, the values used for the training are extracted based on the predictions of the neural network. To avoid the need for training a complex neural network, the proposed neural network makes use of existing pretrained neural networks, that already have the knowledge about the distinctive features in the faces. The use of the MTCNN for face detection and Facenet for face recognition, in addition to the proposed neural network, to achieved the best performance. The performance of the proposed method is evaluated using the Colored FERET Faces Database Version 2 and has achieved robustness rate of 13.48% and uniqueness of 100%.