MFDNet is a lightweight multi-frequency deflare network designed for efficient nighttime flare removal. The network uses a Laplacian Pyramid to decompose images into low and high-frequency bands, separating illumination and content information. It includes two main modules: the Low-Frequency Flare Perception Module (LFFPM) for removing flare in the low-frequency part and the Hierarchical Fusion Reconstruction Module (HFRM) for reconstructing the flare-free image. LFFPM uses a Transformer for global feature extraction and a convolutional neural network for local feature capture, while HFRM gradually fuses outputs with high-frequency components. MFDNet reduces computational cost by processing in multiple frequency bands instead of directly removing flare on the input image. Experimental results show that MFDNet outperforms state-of-the-art methods on the Flare7K dataset, achieving high performance with low computational complexity. The network effectively removes various flare artifacts while preserving image details, and is efficient for high-resolution images. It is designed to handle both real-world and synthetic nighttime flare images, demonstrating superior performance in terms of PSNR, SSIM, and LPIPS. The method is also efficient, with lower GMACs and inference time compared to other state-of-the-art methods. Ablation studies confirm the effectiveness of each component, and the network is capable of handling a wide range of flare patterns. However, it has some limitations, such as difficulty in removing very bright and extensive flare shimmers. Overall, MFDNet provides an efficient and effective solution for nighttime flare removal.MFDNet is a lightweight multi-frequency deflare network designed for efficient nighttime flare removal. The network uses a Laplacian Pyramid to decompose images into low and high-frequency bands, separating illumination and content information. It includes two main modules: the Low-Frequency Flare Perception Module (LFFPM) for removing flare in the low-frequency part and the Hierarchical Fusion Reconstruction Module (HFRM) for reconstructing the flare-free image. LFFPM uses a Transformer for global feature extraction and a convolutional neural network for local feature capture, while HFRM gradually fuses outputs with high-frequency components. MFDNet reduces computational cost by processing in multiple frequency bands instead of directly removing flare on the input image. Experimental results show that MFDNet outperforms state-of-the-art methods on the Flare7K dataset, achieving high performance with low computational complexity. The network effectively removes various flare artifacts while preserving image details, and is efficient for high-resolution images. It is designed to handle both real-world and synthetic nighttime flare images, demonstrating superior performance in terms of PSNR, SSIM, and LPIPS. The method is also efficient, with lower GMACs and inference time compared to other state-of-the-art methods. Ablation studies confirm the effectiveness of each component, and the network is capable of handling a wide range of flare patterns. However, it has some limitations, such as difficulty in removing very bright and extensive flare shimmers. Overall, MFDNet provides an efficient and effective solution for nighttime flare removal.