Enhancing image quality for single-pixel imaging with a score-based diffusion model

Reconstructing high-quality images at low measurement rates remains a problem in single-pixel imaging. Traditional approaches, including methods based on compressed sensing and end-to-end deep learning models, often fail to deliver satisfactory results, especially under low sampling rates. These methods generally exhibit inherent limitations when dealing with undersampling data, resulting in poor image reconstruction quality. To address this issue, we propose a novel approach using the score-based diffusion model. By leveraging the strong and stable generative capabilities of the diffusion model, we fit the underlying distribution of target images. During the reverse diffusion sampling process, we impose conditional constraints using the undersampling measurements, enabling high-quality image reconstruction. The simulation experiments show that our method outperforms previous methods in terms of PSNR or SSIM at different sampling rates, demonstrating superior reconstruction quality.