NDMamba: Dual-Prior State-Space Model for Nighttime Deraining.

Area of Science:

• Computer Vision
• Deep Learning
• Image Processing

Background:

• Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) have advanced nighttime image deraining.
• Current methods lack optimal balance between computational efficiency and restoration performance.
• Existing models inadequately address low-light characteristics and rain-illumination interactions.

Purpose of the Study:

• To propose NDMamba, a dual-prior-guided state-space model for improved nighttime image deraining.
• To jointly model lighting conditions and rain degradation using a Prior Extraction Module (PEM).
• To enhance local features and refine deraining through a Prior-Guided Mamba Block (PGMB).

Main Methods:

• Developed NDMamba, a state-space model incorporating dual priors for nighttime deraining.
• Introduced a Prior Extraction Module (PEM) inspired by Retinex theory.
• Designed a Prior-Guided Mamba Block (PGMB) with Lighting-Adaptive Vision State-Space Module (LVSSM) and Rain Distribution Guidance Module (RDGM).

Main Results:

• NDMamba demonstrates superior performance compared to state-of-the-art methods.
• The model achieves significant improvements on both synthetic and real-world nighttime deraining datasets.
• Experimental results validate the effectiveness of the proposed dual-prior guidance approach.

Conclusions:

• NDMamba effectively addresses limitations in current nighttime image deraining techniques.
• The proposed model offers a promising solution for high-quality nighttime image restoration.
• Publicly available code facilitates further research and application in the field.