Mask-based lensless face recognition system with dual-prior face restoration

Abstract

Face recognition, a biometric technology that analyzes facial features to authenticate individuals’ identities, has various applications and implications across different fields. However, the advancement of technologies such as the Internet of Things has posed challenges for face recognition systems in terms of size, weight, cost, and privacy issues. In response to these challenges, some scholars have suggested a mask-based lensless face recognition system that captures facial images through a mask, eliminating the need for lenses. Nevertheless, the performance of lensless face recognition systems is limited by mask-based imaging, resulting in suboptimal results. To address this limitation, we propose a novel mask-based lensless face recognition system based on the Dual-Prior Face Restoration (DPFR) model. This model utilizes a dual-prior generator to create distinct facial priors that aid the Generative Adversarial Network (GAN) blocks in reconstructing both the global face structure and local face details. Extensive experiments have been carried out on the FlatCam Face Dataset (FCFD) captured using a lens camera and Flatcam lensless camera. The enhanced accuracy, precision, and True Accept Rate (TAR) performance metrics validate the effectiveness of the proposed mask-based lensless face recognition system.