A Dual-Domain Fourier Fusion Network for Low- Light Image Enhancement and Deblurring

To address the coexistence of low illumination and blur in real-world images, we introduce an enhanced framework incorporating a Dual-Domain Fourier Fusion Module (DFFM). Inspired by the fact that in the frequency domain, amplitude affects illumination while phase affects structural information, the DFFM integrates local feature extraction through convolutional neural networks with global context modeling in the frequency domain. To reduce redundancy and improve inference speed, we adopt a real-domain inverse transform in DFFM, eliminating the need for the separation of real and imaginary parts and thereby reducing model complexity. Additionally, we design a frequency-domain loss function that considers both amplitude and phase, facilitating better feature learning and faster convergence. Experiments on the LOL-Blur and Real-LOL-Blur datasets demonstrate that our method achieves competitive performance, obtaining a favorable trade-off between restoration accuracy and perceptual quality.