Enhancing the Spatial Resolution of Galaxy Images Using an Advanced Conditional Denoising Diffusion Probabilistic Model

High-resolution images of galaxies can reveal more galaxy features. However, due to the optical performance limitations of telescopes, there are significant differences in the resolution and sharpness of galaxy images obtained by different telescopes. Therefore, high-resolution galaxy images are reconstructed by mapping low-resolution images into the high-resolution image domain through image processing techniques without changing the existing telescope equipment. In this paper, we propose an advanced conditional denoising diffusion probabilistic model, GalaxSR3, which is an optimized version of the SR3 model for reconstructing galaxy images. We constructed and trained low and high-resolution datasets using the Sloan Digital Sky Survey (SDSS) and Dark Energy Spectroscopic Sky Survey (DESI). Experimental results demonstrate that GalaxSR3 achieves a Peak Signal to Noise Ratio (PSNR) of 30.77 dB and a Learned Perceptual Image Patch Similarity (LPIPS) value of 0.05, outperforming mainstream models such as SRGAN, EDSR, and HAT. The reconstructed images using the GalaxSR3 model can significantly improve the resolution, enhance the boundary contrast between the cantilever and the nucleus of galaxies, and accurately reconstruct the details of galaxies.