Novel mobile palmprint databases for biometric authentication

Mobile palmprint biometric authentication has attracted a lot of attention as an interesting analytics tool for representing discriminative features. Despite the advances in this technology, there are some challenges including lack of enough data and invariant templates to the rotation, illumination, and translation. In this paper, we provide two mobile palmprint databases and we can address the aforementioned challenges via deep convolutional neural networks. In the best of our knowledge, this paper is the first study in which mobile palmprint images were acquired in some special views and then were evaluated via deep learning training algorithms. To evaluate our mobile palmprint images, some well-known convolutional neural networks are applied for verification task. By using these networks, the best performing results are achieved via GoogLeNet and CNN-F architectures in terms of cost of the training phase and classification accuracy of the test phase obtained in the 1-to-1 matching procedure.