Mobile Twin Recognition

Abstract

This study focused on one of the most advanced problems in facial recognition - twin differentiation. In recent years, an increasing number of mobile phones have been hacked using the face of the phone owner's sibling/twin, and there are hundreds of videos about this available on the internet. Our main goal is to improve mobile security and protect user data from outside interventions, and therefore we propose a technique which helps to recognize twins to the same extent as humans are able to do so. The main idea involves combining a modern convolutional neural network (CNN) approach with classical handcrafted features, which describe particular characteristics of the human face, such as an asymmetry. Our method was optimized for low performance mobile platforms and it can be simply used by any system with limited resources.