Low-Light Image Enhancement via Gradient Prior-Aided Network

Low-light images have low brightness and contrast, which bring huge obstacles to the intelligent video surveillance system. The enhancement of low-light must simultaneously consider interference factors such as brightness, artifacts, noise. To this end, in study, we propose a gradient prior-aided network (GPANet). main idea is improve network's ability extract edge features remove unwanted noise by introducing first-order (i.e., Sobel Filter) second-order Laplacian features. Unlike previous methods, proposed first information concatenate them with for multi-view feature analysis fusion encoder (MFE). Then, suggest multi-branch topology module (MTM) fuse decompose Finally, reconstruct through decomposition decoders (MDDs, including three sub-decoders) generate potentially normal-light images. first- will provide decoder multi-scale prior Furthermore, residual speed up convergence while ensuring stable performance.We conduct experiments on widely adopted datasets. results demonstrate advantages our method compared other methods from both qualitative quantitative perspectives. source code available at https://github.com/LouisYuxuLu/GPANet.