Fig. 1. Comparison of the output results of different architecture models and the proposed method. (a) Input image; (b) based on convolutional neural network; (c) based on Transformer; (d) proposed method

Fig. 2. The architecture of MFF-Net

Fig. 3. Details of long distance feature extraction module

Fig. 4. Details of patch feature extract module

Fig. 5. Details of global feature extraction module

Fig. 6. Details of image reconstruction module

Fig. 7. Low light input images

Fig. 8. Comparison of brightness enhancement effects. (a) SCI algorithm; (b) Zero-DCE algorithm; (c) MIRNet algorithm; (d) proposed algorithm; (e) normal light images

Fig. 9. Comparison of detail and color richness. (a) KinD algorithm; (b) KinD++ algorithm; (c) LIME algorithm; (d) Retinex-Net algorithm; (e) algorithm of this article; (f) normal light images

Fig. 10. Schematic diagrams of the ablation experiment structure. (a) Eliminate the GFE module; (b) eliminate the PFE module; (c) eliminate the GFE and PFE modules at the same time; (d) proposed method

Fig. 11. Comparison of boundary artifact elimination results. (a) Input low-light image; (b) Transformer method; (c) proposed method; (d) normal light images