Double random phase-encoded image reconstruction based on denoising diffusion models

Optical cryptosystems based on double random phase encoding (DRPE) offers a robust method for image encryption, effectively safeguarding images against unauthorized access. However, the inherent randomness of DRPE introduces significant challenges for image processing tasks, including reconstruction and classification. To address these challenges, this study proposes a new approach utilizing diffusion models. Our framework utilizes diffusion models to learn and mitigate the complex noise patterns introduced by DRPE, aiming to reconstruct the original images with high fidelity. Additionally, we explore the efficacy of diffusion models in image reconstruction tasks by evaluating their performance on both encrypted and original datasets, providing insights into their capacity for learning and transferring knowledge across different image versions.