Low Light Image Enhancement Based on a Pyramid Diffusion Model

In the process of low light image enhancement, there are problems such as information loss and unclear image details after enhancement. An improved method for low light image enhancement based on the pyramid diffusion model is proposed. The method improves the U-net network structure based on the pyramid diffusion model. A global and local perceptual network architecture is designed for further processing of the downsampled images. The global module captures the global context information through the channel attention mechanism. The local module constructs multi-scale feature block to extract image features at different scales. For provide fine-grained details, the local module uses contrast information to process and enhance the characterization of local features. Finally, an interactive fusion module is designed to fuse different features from the global branch and local branch. The mothod also improves the residual network and uses skip connection with convolution to connect the downsampling and upsampling. Experimental results on the LOL and LOLv2 datasets show that the improved method has better results, which demonstrates the effectiveness of this design.