FaceID: Verification of face in selfie and ID document

Abstract

Various activities in everyday life require us to verify our identity by demonstrating our ID document containing face images, for example, voter ID, passports, driver licence, to human administrators. However, this procedure is reluctant, unreliable and labor comprehensive. An automatic framework for verifying ID record photographs to live face pictures (selfies) progressively and with high precision is required. Cross-domain biometrics is another requirement, which represents se-veral additional challenges, including harsh illumination conditions, pose variations, noise, among others. In this paper, we propose an algorithm to meet this objective. We first extract faces from ID document and selfie using Multi-task Cascaded Convolutional Networks. To extract prominent features from the data, we apply a VGG face model which is a CNN-based transfer learning approach. Finally, we validate the methods using a novel FaceId-Selfie dataset comprising 600 individuals using cosine distance measure. Results show that 74% accuracy is achieved on FaceId-Selfie dataset.