This paper proposes a heterogeneous window Transformer (HWformer) for image denoising. HWformer is designed to address the limitations of short-distance modeling by incorporating global context information and enhancing the interaction between long- and short-distance modeling. The key contributions of this work include the design of heterogeneous global windows to capture richer global context information, a shift mechanism in global windows to facilitate diversified information without increasing denoising time, and the introduction of a sparse technique in a feed-forward network to extract more local information from neighboring patches. HWformer achieves a denoising speed that is 30% faster than the popular Restormer. The proposed HWformer consists of a head, two global-window Transformer enhancement blocks (GTEBlocks), and a Transformer direction enhancement block (TDEBlock). The GTEBlocks use larger global windows to extract more global information, while the TDEBlock employs different directional shifts (horizontal, vertical, and common) to enhance the denoising performance. The sparse technique is integrated into the feed-forward network to capture more local information from neighboring patches. The HWformer is evaluated on both synthetic and real noisy image datasets, demonstrating superior performance in terms of PSNR, SSIM, FSIM, LPIPS, PSBR, and color difference metrics. The results show that HWformer outperforms several state-of-the-art methods in image denoising tasks, achieving higher PSNR values and better visual quality. The proposed method is also efficient, with a denoising time that is 30% faster than Restormer, making it suitable for real-time applications such as smartphones and cameras. The experiments confirm that HWformer is effective for both gray and color image denoising, and it is capable of restoring more details and texture information in real noisy images. The proposed method is suitable for deployment in real-world applications due to its efficiency and effectiveness in image denoising.This paper proposes a heterogeneous window Transformer (HWformer) for image denoising. HWformer is designed to address the limitations of short-distance modeling by incorporating global context information and enhancing the interaction between long- and short-distance modeling. The key contributions of this work include the design of heterogeneous global windows to capture richer global context information, a shift mechanism in global windows to facilitate diversified information without increasing denoising time, and the introduction of a sparse technique in a feed-forward network to extract more local information from neighboring patches. HWformer achieves a denoising speed that is 30% faster than the popular Restormer. The proposed HWformer consists of a head, two global-window Transformer enhancement blocks (GTEBlocks), and a Transformer direction enhancement block (TDEBlock). The GTEBlocks use larger global windows to extract more global information, while the TDEBlock employs different directional shifts (horizontal, vertical, and common) to enhance the denoising performance. The sparse technique is integrated into the feed-forward network to capture more local information from neighboring patches. The HWformer is evaluated on both synthetic and real noisy image datasets, demonstrating superior performance in terms of PSNR, SSIM, FSIM, LPIPS, PSBR, and color difference metrics. The results show that HWformer outperforms several state-of-the-art methods in image denoising tasks, achieving higher PSNR values and better visual quality. The proposed method is also efficient, with a denoising time that is 30% faster than Restormer, making it suitable for real-time applications such as smartphones and cameras. The experiments confirm that HWformer is effective for both gray and color image denoising, and it is capable of restoring more details and texture information in real noisy images. The proposed method is suitable for deployment in real-world applications due to its efficiency and effectiveness in image denoising.