Attention-Guided Multi-Scale Feature Fusion Network for Low-Light Image Enhancement

Area of Science:

• Computer Vision
• Image Processing

Background:

• Low-light image enhancement is crucial for improving visibility in computer vision applications.
• Existing methods often struggle with poor visibility, high noise, and low contrast in low-light conditions.
• Computational cost can be a significant barrier in complex enhancement networks.

Purpose of the Study:

• To propose an efficient and effective Attention-Guided Multi-scale feature fusion network (MSFFNet) for low-light image enhancement.
• To enhance contrast and brightness in images captured under low-light or night conditions.
• To reduce computational complexity while improving feature representation and detail recovery.

Main Methods:

• A single encoder-decoder architecture is utilized for multi-scale input and output, reducing computational overhead.
• Incorporation of Convolutional Block Attention Module (CBAM) in the encoder to suppress noise and refine color features.
• Integration of Feature Calibration Module (FCM), Attention Fusion Module (AFM), and Cascade Fusion Module (CFM) for enhanced feature mapping, contextual information capture, and perceptual field fusion.

Main Results:

• The proposed MSFFNet demonstrates superior performance in enhancing low-light images compared to existing methods.
• Experimental results show significant improvements in both visual quality and quantitative metrics.
• The network effectively suppresses noise and color differences while preserving and recovering image details.

Conclusions:

• MSFFNet offers an effective solution for low-light image enhancement with reduced computational cost.
• The attention-guided multi-scale fusion approach significantly improves image contrast, brightness, and detail recovery.
• The method shows strong potential for real-world applications requiring high-quality image enhancement in challenging lighting conditions.