Large-Small-Scale Structure Blended U-Net for Brightening Low-Light Images

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Existing low-light image enhancement methods often struggle with detail restoration and accurate color correction.
• Improving visual quality in low-illumination conditions remains a significant challenge in image processing.

Purpose of the Study:

• To develop a novel dual-branch network for superior low-light image enhancement.
• To address limitations in detail enhancement and color correction present in current methods.

Main Methods:

• A dual-branch network comprising a color correction network (CC-Net) and a light-boosting network (LB-Net).
• Utilizing the CIELAB color space for luminosity and color component extraction.
• Implementing a U-shaped network for CC-Net and a large-small-scale structure for LB-Net to explore multiscale features.
• Incorporating an efficient feature interaction module for cross-branch information exchange.

Main Results:

• The proposed method significantly enhances image brightness, detail, and color fidelity in low-illumination scenarios.
• Experimental results on public benchmarks show superior performance compared to state-of-the-art low-light enhancement techniques.
• Demonstrated improvement in object detection performance under low-light conditions.

Conclusions:

• The dual-branch network effectively tackles the challenges of detail enhancement and color correction in low-light images.
• The proposed approach offers a robust solution for improving the quality of low-light imagery and its applications.