Mobile devices had their popularity and affordability greatly increased in recent years. As a consequence of their ubiquity, these devices now carry all sorts of personal data that should be accessed only by their owner. Even though knowledge-based procedures are still the main methods to secure the owner’s identity, recently biometric traits have been employed for a more secure and effortless authentication. In this work, we propose a facial verification method optimized to the mobile environment. It consists of a two-tiered procedure that combines hand-crafted features and a new convolutional neural network – HF-CNN –, an architecture tweaked for mobile devices that processes encoded information of a pair of face images. We also propose a technique to adapt our method’s acceptance cutoff to images with different characteristics than those present during training, by using the device owner’s enrollment gallery. The proposed solution outperforms state-of-the-art face verification methods, while having a model 16 times smaller and 4 times faster when processing an image in recent smartphone models. Finally, we present a new dataset of selfie pictures – RCD selfie dataset – that hopefully will support future research in this scenario.
@inproceedings{padilha18sibgrapi, author = “Rafael Padilha and Fernanda A. Andal{'o} and Ricardo da S. Torres and Anderson Rocha and Jacques Wainer”, title = “Two-tiered facial verification for mobile devices”, booktitle = “Workshop of Theses and Dissertations, Conference on Graphics, Patterns and Images ({WTD/SIBGRAPI})", year = 2018, address = “Foz do Igua{\c c}u, PR, Brazil”, }