Low-light image enhancement network based on luminance prior and depth feature extraction

Low-light image enhancement has important background and significance in today’s digital image processing and computer vision fields Images captured in low light conditions are often affected by, among other things, noise, blur and low brightness, which poses challenges to image clarity, color and detail. In order to promote more accurate analysis and interpretation of low-light image enhancement. This paper introduces a very effective method for low-light image enhancement. Firstly, based on the luminance prior, the luminance components of low-light images are pre-enhanced and fused to significantly improve the brightness and contrast of the images; then, a dense block depth is introduced into the deep feature extraction network of the images to extract the multiscale features of the images, and an attention gating mechanism is added to ensure efficient information propagation and feature learning. Extensive experiments show that our approach makes significant progress on the LOL dataset, demonstrating its robustness in challenging scenarios. The proposed model performs well in comparison experiments, effectively mitigates the underexposure problem, and efficiently recovers image texture and color in low-light scenes.