Learning With Self-Calibrator for Fast and Robust Low-Light Image Enhancement.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Convolutional Neural Networks (CNNs) excel at low-light image enhancement.
• Existing methods struggle to balance image quality and computational efficiency.
• This inefficiency limits practical use in real-world scenarios and downstream tasks.

Purpose of the Study:

• To develop an efficient and high-quality low-light image enhancement method.
• To introduce a novel Self-Calibrated Illumination (SCI) learning scheme.
• To improve the practical applicability of image enhancement techniques.

Main Methods:

• Proposed a Self-Calibrated Illumination (SCI) learning scheme.
• Utilized a weight-sharing illumination estimation process with an embedded self-calibrator.
• Introduced an additivity condition for a reinforced SCI++ version, enhancing interpretability and stability.

Main Results:

• Achieved significant gains using only a single basic block for inference, drastically reducing computation cost.
• Demonstrated higher quality and efficiency in restoring clean images from diverse low-light scenes.
• Verified applicability across various low-light vision tasks, outperforming existing methods.

Conclusions:

• The SCI learning scheme offers a new perspective for boosting model capability in image enhancement.
• SCI and SCI++ provide interpretable, effective, and efficient solutions for low-light image enhancement.
• The proposed methods are highly applicable and performant for real-world computer vision challenges.