FDMLNet: A Frequency-Division and Multiscale Learning Network for Enhancing Low-Light Image.

Area of Science:

• Computer Vision
• Image Processing
• Deep Learning

Background:

• Low-illumination images suffer from poor brightness, blurriness, and color casts, negatively impacting visual quality and downstream applications.
• Data-driven methods show promise for image enhancement but often introduce artifacts like noise, color deviation, and unnatural enhancements.

Purpose of the Study:

• To develop an advanced deep learning network for effective low-illumination image enhancement.
• To address limitations of existing methods by mitigating noise, color deviation, and over/under enhancement.

Main Methods:

• A frequency division and multiscale learning network (FDMLNet) was proposed, comprising DetNet and StruNet subnets.
• Guided filter separates image frequencies; DetNet and StruNet process high and low frequencies, respectively.
• StruNet incorporates a feasible feature extraction module (FFEM) with multiscale learning (MSL) and dual-branch channel attention (DCAM) for enhanced multiscale representation.

Main Results:

• FDMLNet demonstrated superior performance on public benchmarks compared to state-of-the-art approaches.
• The network effectively enhanced image brightness and detail while preserving visual naturalness.
• Experiments validated the effectiveness of the multiscale feature expression and extraction capabilities.

Conclusions:

• FDMLNet offers a robust solution for low-illumination image enhancement.
• The proposed architecture, with its frequency division and multiscale learning, significantly improves image quality.
• The method provides a strong foundation for future research in low-light image restoration.