RAW-CLIP Fusion: Unleashing Semantic-Aware Denoising for Sensor-Agnostic Low-Light Imaging.

Area of Science:

• Computational Photography
• Computer Vision
• Image Processing

Background:

• Denoising low-light images is difficult due to low signal-to-noise ratios and sensor-specific noise.
• Existing calibration-free methods struggle with extreme low-light and diverse sensors.
• Mismatches between synthetic and real noise hinder current approaches.

Purpose of the Study:

• To develop a calibration-free denoising framework for extreme low-light RAW images.
• To leverage large-scale vision models for cross-domain feature fusion in denoising.
• To improve generalization across different image sensors.

Main Methods:

• Introduced CLIP-Guided Denoising (CLD), utilizing CLIP embeddings from sRGB pre-trained models.
• Employed cross-domain feature fusion to guide RAW image denoising.
• Validated the approach on SID and ELD datasets.

Main Results:

• CLD achieves state-of-the-art performance in calibration-free denoising.
• Significantly outperforms prior methods in extreme low-light scenarios.
• Demonstrates robust generalization across unseen sensor domains.

Conclusions:

• CLIP-Guided Denoising offers a robust solution for low-light RAW image denoising.
• The framework effectively utilizes semantically rich, noise-invariant features from large vision models.
• CLD overcomes limitations of current methods, enabling high-quality denoising across diverse sensors without calibration.