A novel approach to low-light image and video enhancement using adaptive dual super-resolution generative adversarial networks and top-hat filtering

Abstract

Image and video enhancement under low-light conditions is challenging, as the task involves more than just brightness adjustment. Without addressing issues such as artifacts, distortions, and noise in dark regions, brightness improvement alone can worsen the quality. This paper presents a novel approach to low-light image and video enhancement based on the adaptive fusion of Dual Super-Resolution Generative Adversarial Network (DSRGAN) models, followed by Top-Hat Gradient-Domain Filtering (THGDF). A soft thresholding mechanism is used to integrate the Memory Residual Super-Resolution Generative Adversarial Network (MRSRGAN) and the Weighted Perception Super-Resolution Generative Adversarial Network (WPSRGAN). MRSRGAN enhances fine details, improving the objective performance of the image, while WPSRGAN improves overall details, enhancing the subjective performance. Top-hat gradient-domain filtering is then applied to remove artifacts, distortions, and noise in both images and videos, resulting in outstanding perception scores. The proposed approach is validated using the quality assessment metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measurement (SSIM), and Information Fidelity Criterion (IFC). Extensive experiments conducted on publicly available source codes and databases demonstrate that the proposed method is more effective than the existing state-of-the-art techniques.