SD-PSFNet: Sequential and Dynamic Point Spread Function Network for Image Deraining

2025-11-22

AI Summary

Image deraining faces the challenge of modeling multiscale rain streaks strongly coupled with scene content. This paper proposes a physics-aware cascaded sequential restoration network. First, a learnable point spread function (PSF) module is designed to dynamically model the optical degradation induced by rain streaks. Second, an adaptive gated fusion mechanism is introduced to enable cross-stage feature optimization across the cascade—from coarse-grained rain removal to fine-grained structural reconstruction. By embedding physical priors into the deep learning architecture, the framework jointly ensures interpretability of the degradation process and expressive power of learned features. The method achieves state-of-the-art performance (PSNR/SSIM) on three benchmarks—Rain100H, RealRain-1k-L, and RealRain-1k-H—particularly excelling in dense rainfall conditions and complex background scenes, where it significantly improves rain-streak separation accuracy and structural fidelity.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Image deraining is crucial for vision applications but is challenged by the complex multi-scale physics of rain and its coupling with scenes. To address this challenge, a novel approach inspired by multi-stage image restoration is proposed, incorporating Point Spread Function (PSF) mechanisms to reveal the image degradation process while combining dynamic physical modeling with sequential feature fusion transfer, named SD-PSFNet. Specifically, SD-PSFNet employs a sequential restoration architecture with three cascaded stages, allowing multiple dynamic evaluations and refinements of the degradation process estimation. The network utilizes components with learned PSF mechanisms to dynamically simulate rain streak optics, enabling effective rain-background separation while progressively enhancing outputs through novel PSF components at each stage. Additionally, SD-PSFNet incorporates adaptive gated fusion for optimal cross-stage feature integration, enabling sequential refinement from coarse rain removal to fine detail restoration. Our model achieves state-of-the-art PSNR/SSIM metrics on Rain100H (33.12dB/0.9371), RealRain-1k-L (42.28dB/0.9872), and RealRain-1k-H (41.08dB/0.9838). In summary, SD-PSFNet demonstrates excellent capability in complex scenes and dense rainfall conditions, providing a new physics-aware approach to image deraining.

Problem

• Addressing complex multi-scale rain physics in image deraining
• Separating rain streaks from background scenes effectively
• Restoring image details in dense rainfall conditions

Innovation

• Sequential restoration architecture with three cascaded stages
• Dynamic PSF mechanisms simulating rain streak optics
• Adaptive gated fusion for cross-stage feature integration