ALEX: Automated Low-Light Enhancement eXpert for Intelligent Security Systems Using Vision Transformer

Abstract Closed-Circuit TeleVision (CCTV) performance in low-light conditions often results in poor image quality. This study introduces Automated Low-Light Enhancement eXpert (ALEX), a new architecture that combines ViTRA with SwinIR to improve image clarity. ALEX utilizes Relative Lighten Cross-Attention (RLCA) and Relative Position Encoding (RPE) in the HVI color space to enhance light intensity and color, followed by SwinIR for depth restoration and resolution enhancement. Evaluation on benchmark datasets like LOLv1, LOLv2, and SID shows that ALEX outperforms existing methods like HVI-CIDNet and ViTRA, yielding sharper, more natural results based on PSNR, SSIM, and other metrics. Real-world CCTV tests demonstrate that ALEX improves image quality, even with dimmed or downscaled images. While the integration of SwinIR increases complexity and inference time, ALEX proves to be an effective low-light enhancement solution, offering significant potential for intelligent surveillance systems and future real-time applications on resource-constrained devices.