Face Recognition on Mobile Devices Based on Frames Selection

The easy image capture process on a phone, the ability to move the camera, the non-intrusive characteristic that allows the authentication without interaction of the user, make the face a suitable biometric trait to be used on mobile devices. During the continuous use of the mobile, it is possible to analyze the expression, to determine the gender and ethnic, or to recognize the user. In this paper we propose a robust algorithm for face recognition on mobile devices. First, the best frames of a face sequence are selected based on the face pose, blurness, eyes and mouth expression. Then, a unique feature vector for the best selected frames is obtained using a deep learning model. Finally, a SoftMax function is used for authenticate the user. The experimental evaluation conducted on the UMD-AA dataset shows the robustness of the proposal, that outperforms state-of-the-art methods.