Human forehead recognition: A novel biometric modality based on near-infrared laser backscattering feature image using deep transfer learning

Abstract

Human recognition systems are an essential tool for identity verification. Though various parts of the human body have been widely used as input data for decades, developing new biometric technology is still necessary to enhance the security system complexity. This article presents a novel biometric modality based on forehead feature images acquired from a specially designed near-infrared laser scanning system. The authors selected state-of-the-art deep convolutional neural networks (CNN), including VGGNet, ResNet, and Inception-v3, to demonstrate the human forehead recognition task. Though large-scale training data is generally required for learning a promising CNN model, they showed the feasibility to transfer the feature representation knowledge of the networks that were pre-trained on the data from a different domain and fine-tuned the target network on the limited dataset of forehead feature images. This transfer learning approach establishes the usability of human forehead recognition and allows us to implement this biometric modality for real-world application.