LQE-Net: Hybrid Deep Network for Enhancing Low-Light and Low-Quality CCTV Footage

Accurate recognition of faces and vehicle license plates in CCTV footage is crucial for public safety and crime prevention, yet low-light and low-resolution conditions greatly hinder recognition performance. To address these challenges, we propose LQE-Net, a hybrid deep learning model that combines Zero-DCE for low-light enhancement, GFPGAN for facial detail restoration, and SRGAN for license plate resolution improvement. Experimental results show that LQE-Net significantly outperforms traditional methods across multiple metrics, achieving a 98.26% OCR recognition rate for license plates and higher PSNR, SSIM, and LPIPS scores. This improvement demonstrates LQE-Net’s effectiveness in enhancing visual quality and recognition accuracy, even in challenging lighting and resolution conditions. By optimizing these combined techniques, LQE-Net provides a practical and efficient solution for real-time CCTV surveillance systems, reinforcing reliability in diverse environments.