DeepTeeth: A Teeth-photo Based Human Authentication System for Mobile and Hand-held Devices

TL;DR

DeepTeeth introduces a novel teeth-photo biometric modality for mobile device authentication, utilizing deep learning with a custom loss function to achieve perfect accuracy on a new dataset.

Contribution

The paper presents the first teeth-photo based authentication system for mobile devices, with a specialized deep learning architecture and regularization scheme for small RoI classification.

Findings

Achieved perfect classification accuracy on the teeth-photo dataset.

Developed a highly discriminative embedding for small RoI images.

System is efficient, end-to-end, and suitable for real-time authentication.

Abstract

This paper proposes teeth-photo, a new biometric modality for human authentication on mobile and hand held devices. Biometrics samples are acquired using the camera mounted on mobile device with the help of a mobile application having specific markers to register the teeth area. Region of interest (RoI) is then extracted using the markers and the obtained sample is enhanced using contrast limited adaptive histogram equalization (CLAHE) for better visual clarity. We propose a deep learning architecture and novel regularization scheme to obtain highly discriminative embedding for small size RoI. Proposed custom loss function was able to achieve perfect classification for the tiny RoI of $75\times 75$ size. The model is end-to-end and few-shot and therefore is very efficient in terms of time and energy requirements. The system can be used in many ways including device unlocking and secure authentication. To the best of our understanding, this is the first work on teeth-photo based authentication for mobile device. Experiments have been conducted on an in-house teeth-photo database collected using our application. The database is made publicly available. Results have shown that the proposed system has perfect accuracy.