To enhance the capability of capturing both local and global information in image denoising, a Swin Transformer-based image denoising network (SwinFPSFNet) is proposed by integrating a feature pyramid network (FPN) and a spatial frequency block (SFB). The proposed network consists of three stages. In the shallow feature extraction stage, a feature pyramid network is employed to strengthen local feature representation. In the deep feature extraction stage, a spatial frequency block incorporating fast Fourier convolution (FFC) is introduced to simultaneously capture global and local features. Finally, the denoising performance is further enhanced by aggregating shallow and deep features. In addition, a Gaussian noise degradation model is constructed, and various data augmentation strategies are applied to enhance the generalization ability. Experimental results on the CBSD68, Kodak24, and Urban100 datasets demonstrate that, compared to main stream denoising methods such as BM3D, DnCNN, FFDNet, and SwinIR, SwinFPSFNet achieves a superior balance between noise suppression and detail preservation by effectively leveraging both local and global information.