Low-Light Image Enhancement Using Photometric Alignment with Hierarchy Pyramid Network.

Area of Science:

• Computer Vision
• Image Processing

Background:

• Low-light conditions hinder high-level vision tasks.
• Existing data-driven methods often produce undesirable results like noise and color distortion.

Purpose of the Study:

• To propose an end-to-end pipeline for low-light image enhancement.
• To improve detail, reduce noise, and correct color distortion in degraded images.

Main Methods:

• A coarse-to-fine strategy using adaptive global photometric alignment.
• A hierarchy pyramid enhancement sub-network for quality optimization.
• A multi-residual cascade attention block (MRCAB) for high-resolution reconstruction.

Main Results:

• The proposed method effectively reduces style differences and improves illumination.
• It enhances local details and removes amplified noise.
• Achieved superior performance in detail and color reproduction compared to state-of-the-art methods.

Conclusions:

• The end-to-end pipeline significantly enhances low-light images.
• The method demonstrates effectiveness across various datasets.
• It offers a robust solution for improving image quality in poor illumination.