HAIMNet: A Hierarchical Adaptive Interaction Modulation Network for Low-Light Image Enhancement

Area of Science:

• Computer Vision
• Image Processing
• Artificial Intelligence

Background:

• Low-light image enhancement (LLIE) is crucial for nighttime visual perception and tasks like object detection.
• Existing methods often fail to coherently restore luminance, texture, and color under complex lighting.
• Low-light conditions lead to insufficient luminance, lost details, and unstable colors.

Purpose of the Study:

• To propose a novel network, the Hierarchical Adaptive Interaction Modulation Network (HAIMNet), for effective low-light image enhancement.
• To address the limitations of current methods in simultaneously restoring luminance, texture, and color.
• To improve the naturalness and stability of enhanced images under extreme low-light conditions.

Main Methods:

• The HAIMNet network decouples luminance and chromaticity in the Horizontal/Vertical-Intensity (HVI) color space.
• An inter-branch attention-modulation block (IAMB) enhances luminance-texture consistency.
• A cross-branch gated affine fusion module (CGAF) calibrates features and reduces color deviations.

Main Results:

• HAIMNet demonstrated superior performance across 11 public datasets.
• The enhanced images exhibited high naturalness and stability, even in extremely dark conditions.
• The method showed effectiveness, robustness, and generalization capabilities.

Conclusions:

• HAIMNet offers a significant advancement in low-light image enhancement.
• The proposed network effectively balances luminance, texture, and color restoration.
• HAIMNet improves visual perception and performance in downstream computer vision tasks.