Polarization-enhanced super-resolution imaging reconstruction based on a denoising diffusion probabilistic model

Super-resolution imaging reconstruction from corresponding low-resolution inputs has garnered increasing interest due to its fundamental significance in physics and its potential applications in target detection, material recognition, and semantic segmentation. In this work, we propose a polarization-enhanced super-resolution diffusion model (PSRDM) that leverages the powerful generative capabilities of diffusion models and the high-frequency information contained in degree of linear polarization (DoLP) images. The PSRDM consists of a double-branch feature extraction module (DBFEM) and a denoising network. The DBFEM extracts and fuses low-frequency features from the low-resolution intensity image and high-frequency features from the DoLP image, providing enriched input for subsequent training and inference. The denoising network then reconstructs the corresponding high-resolution images through a diffusion and denoising process. Experiment results demonstrate that the proposed model outperforms state-of-the-art methods in super-resolution image reconstruction.