Seeing Through the PRISM: Compound & Controllable Restoration of Scientific Images

Abstract

Scientific and environmental imagery often suffer from complex mixtures of noise related to the sensor and the environment. Existing restoration methods typically remove one degradation at a time, leading to cascading artifacts, overcorrection, or loss of meaningful signal. In scientific applications, restoration must be able to simultaneously handle compound degradations while allowing experts to selectively remove subsets of distortions without erasing important features. To address these challenges, we present PRISM (Precision Restoration with Interpretable Separation of Mixtures). PRISM is a prompted conditional diffusion framework which combines compound-aware supervision over mixed degradations with a weighted contrastive disentanglement objective that aligns primitives and their mixtures in the latent space. This compositional geometry enables high-fidelity joint removal of overlapping distortions while also allowing flexible, targeted fixes through natural language prompts. Across microscopy, wildlife monitoring, remote sensing, and urban weather datasets, PRISM outperforms state-of-the-art baselines on complex compound degradations, including zero-shot mixtures not seen during training. Importantly, we show that selective restoration significantly improves downstream scientific accuracy in several domains over standard "black-box" restoration. These results establish PRISM as a generalizable and controllable framework for high-fidelity restoration in domains where scientific utility is a priority.

Figures (23)

• Figure 1: Expert-in-the-Loop Restoration with PRISM. PRISM enables robust compound restoration and zero-shot handling of unseen mixtures. It supports both automatic restoration and prompt-driven, selective correction for scientific analysis.
• Figure 2: Overview of PRISM. We first fine-tune CLIP's image encoder to disentangle image embeddings by distortion. The degraded input and user prompt are then used to condition the latent diffusion backbone, the coarse outputs of which are refined with a Semantic Content Preservation Module (SCPM) to yield the final restored output. Appendix Figure 12 shows the SCPM's architecture.
• Figure 3: PRISM trained on composite examples scales best with the number of distortions. This outperforms our model trained on each degradation separately as well as comparable baselines, emphasized by the $\Delta$ PSNR of test images with 1 vs. 4 distortions) above each bar.
• Figure 4: Latent disentanglement enables both stepwise and single-shot restoration, closing the gap between prompting strategies.
• Figure 5: Structured, compositional latent geometry supports both automated (left) and expert-driven (right) generalization. With an expert-in-the-loop, prompts progressively target distortions.