Deep learning for IR lensless face recognition in under-display system

We propose an under-display lensless face recognition system for privacy-preserving user authentication. Unlike conventional lens-based systems that capture raw facial images, our lensless approach optically encrypts facial information at the sensor level through a binary amplitude-coded mask positioned between the transparent display and the CMOS sensor. To improve robustness in indoor environments, we employ active infrared (IR) illumination and an IR-pass filter to suppress visible light interference. Our model uses a modified ResNet-50 network to extract features for identity verification. The model is trained on synthetically generated data from the ColorFERET dataset, while performance is validated using real IR-captured data. The system achieved an accuracy of up to 97%, demonstrating its effectiveness for secure authentication.