LIIA -Net: A lightweight illumination iterative adjustment network for low-light image enhancement

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Low-light image enhancement is crucial for visual perception but current methods struggle with illumination accuracy, noise amplification, and detail loss.
• Existing techniques often fail to effectively balance brightness, contrast, and structural integrity in degraded images.

Purpose of the Study:

• To develop a novel, lightweight network for effective low-light image enhancement.
• To address limitations of existing methods by improving illumination adjustment, reducing noise, and preserving structural details.

Main Methods:

• Proposed Lightweight Illumination Iterative Adjustment Network (LIIA-Net) processing images in both frequency and spatial domains.
• Utilized a linear cross attention module for fusing illumination and content features.
• Implemented an amplitude adaptive iterative adjustment module for frequency domain brightness regulation.
• Employed a mamba-based structure refinement module for spatial texture restoration.

Main Results:

• LIIA-Net achieves performance comparable to or exceeding state-of-the-art methods with only 0.48M parameters.
• Demonstrated significant improvements in image quality, including brightness, contrast, and structural details.
• Enhanced images from LIIA-Net led to a notable boost in downstream object detection accuracy.

Conclusions:

• LIIA-Net offers an efficient and effective solution for low-light image enhancement.
• The joint frequency and spatial domain processing effectively tackles noise and structural loss.
• The proposed method shows promise for improving performance in various computer vision applications, including object detection.