Diffusion-Based Image Super-Resolution with Full-Components Wavelet Convolution

Image super-resolution is an essential task in computer vision, aiming to restore high-resolution images from low-resolution ones. While CNN-based and GAN-based methods have improved performance over traditional interpolation and reconstruction techniques, they often suffer from limited high-frequency detail recovery and computational inefficiency. Recently, generative approaches based on diffusion models have demonstrated remarkable capabilities in reconstructing high-fidelity visual content. Nevertheless, these methods often require substantial computational resources and may fail to adequately retain high-frequency information. In response to these limitations, this paper proposes a novel diffusion-based framework incorporating a lightweight Full-Components Wavelet Convolution (FCWC). The FCWC effectively integrates multi-level high-frequency information while significantly reducing parameter count compared to standard convolutions. Integrated into the ResNet blocks of the residual network, FCWC strengthens feature representation and residual learning capabilities. The overall enhanced network is embedded within the residual denoising diffusion model, further improving reconstruction quality. Experimental evaluations show that our method exhibits superior performance and surpasses previous methods.