Quantum Fisher Information-Guided Diffusion Refinement for Fuzzy and Uncertain Surveillance Regions

Restoring surveillance images containing both structurally reliable and information-sparse fuzzy regions remains challenging because conventional restoration and diffusion-based methods may either suppress authentic textures or introduce visually plausible but unsupported details. To address this problem, we propose quantum Fisher information-guided diffusion refinement (QFI-Diff), a hybrid framework that incorporates quantum information-theoretic uncertainty into latent diffusion-based image restoration. Specifically, quantum Fisher information (QFI) is employed as a local sensitivity measure to construct a reliability map that distinguishes information-rich regions from uncertain and degraded regions. This map guides the reverse diffusion process by constraining reliable regions while permitting adaptive stochastic refinement in uncertain areas. We establish theoretical guarantees for the proposed framework, including asymptotic convergence of the associated stochastic differential equation, information-theoretic bounds based on the quantum Cramér-Rao bound, computational complexity bounds, and discrete restoration error bounds. Experiments on the DarkFace and RESIDE datasets demonstrate that QFI-Diff achieves competitive structural and perceptual restoration performance compared with the evaluated baselines, as reflected in the SSIM, LPIPS, and FID metrics. These results highlight the potential of quantum information-theoretic uncertainty measures for reliability-aware diffusion-based image restoration.