Low-Light Image Enhancement Network Based on Recursive Network.

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Low-light conditions severely degrade image quality, hindering computer vision applications.
• Existing low-light image enhancement algorithms face challenges in detail recovery and brightness enhancement.

Purpose of the Study:

• To propose a novel multi-scale feature fusion image enhancement network with a recursive structure.
• To improve the performance of computer vision tasks by enhancing low-light images.

Main Methods:

• A recursive network architecture is employed, processing images through multiple stages.
• The network integrates a Convolutional Block Attention Module (CBAM) for focused feature extraction.
• A Multi-scale Inception U-Net (MIU) module fuses features across different scales.

Main Results:

• The proposed network effectively recovers image details and increases brightness.
• Experimental results demonstrate superior performance compared to existing methods in reducing image degradation.
• Subjective and objective analyses confirm the algorithm's efficacy on public datasets.

Conclusions:

• The proposed recursive multi-scale feature fusion network offers a simple yet effective solution for low-light image enhancement.
• This approach significantly improves image quality, benefiting subsequent computer vision tasks.
• The method shows promise for practical applications requiring high-quality images in challenging lighting conditions.