LightDeblur: A Lightweight Image Deblurring Network via CNN-Transformer Hybrid Architecture

Single image deblurring aims to restore a sharp image by removing blurred areas in the single image. Although existing deblurring methods achieve significant progress, these approaches tend to require lots of network parameters and huge computational costs, which limits their practical applications. This work proposes a lightweight image deblurring net-work that integrates Convolutional Neural Networks (CNNs) and Transformers, which is based on an encoder-decoder architecture. Specifically, we propose the inverted residual fast fourier transformation (IRFFT) to reduce model complex-ity. Furthermore, we developed a sparse global self-attention (SGSA) that leverages the most useful global features for image reconstruction while reducing computational complexity. Finally, we also proposed a lightweight feed-forward network (LFFN) to generate better features for image deblurring. Experimental results on benchmark tests indicate that our method achieves competitive performance while maintaining a lightweight structure. Additionally, it demonstrates superior generalization and robustness, producing deblurred images that are visually more realistic.