SAR-to-Optical Image Translation Using Conditional Denoising Diffusion Probabilistic Models
Jiang Qin, Bin Zou, Lamei Zhang, Yu Qiu · 2024
Due to the special imaging mechanism, SAR images exhibit speckle noise, top-bottom inversion, which make them hard to be visually interpreted. Compared with optical remote sensing images, visual interpretation of SAR images often requires a lot of expert experience and knowledge, which limits wider applications in multimodal tasks, like the multimodal registration and information fusion. In this paper, we propose an image translation method based on conditional denoising diffusion probabilistic models, aiming that translate SAR images to optical images to facilitate more intuitive visual interpretations and downstream applications. In particular, both structure information and content information are introduced into the diffusion model as conditional constraints, to ensure the content and structure consistency of the synthetic optical images through the iterative refinement strategy. SAR-EO dataset is adpoted for the performance evaluation. According to the quantitative and qualitative results, the proposed method preserves high content and structure consistency during the translation, which exhibits strong SAR-to-optical image translation capability.