Related Experiment Video
Updated: Sep 21, 2025
07:15
Transient Optical Clearing Using Absorbing Molecules for Ex Vivo and In Vivo Imaging
Published on: July 11, 2025
535
Super-Pixel Guided Low-Light Images Enhancement with Features Restoration
Xiaoming Liu1, Yan Yang1, Yuanhong Zhong1
1School of Microelectronics and Communications Engineering, Chongqing University, Chongqing 400044, China.
Sensors (Basel, Switzerland)
|May 28, 2022
Summary
This study introduces a novel low-light image enhancement method using Convolutional Neural Networks (CNNs) and Attentive Neural Processes (ANPs). The technique effectively improves visual quality and aids subsequent tasks like target detection.
Area of Science:
- Computer Vision
- Image Processing
- Artificial Intelligence
Background:
- Low-light image enhancement is crucial for visual perception and downstream tasks like object detection.
- Existing methods struggle to balance visual quality with feature preservation for subsequent analysis.
Purpose of the Study:
- To develop an effective low-light image enhancement technique that balances visual perception and utility for high-level tasks.
- To introduce a novel approach combining Convolutional Neural Networks (CNNs), super-pixel segmentation, and Attentive Neural Processes (ANPs).
Main Methods:
- Utilizing shallow CNNs for initial feature restoration in low-light images.
- Applying super-pixel segmentation to group similar image regions.
- Employing Attentive Neural Processes (ANPs) for localized enhancement within super-pixels.
Main Results:
- Achieved high scores in Peak Signal to Noise Ratio (PSNR), Structural Similarity (SSIM), and Natural Image Quality Evaluator (NIQE).
- Demonstrated superior performance in Scale-Invariant Feature Transform (SIFT) feature detection and target detection tasks.
- Validated effectiveness on both synthetic and real-world low-light images.
Conclusions:
- The proposed method significantly enhances low-light images, improving both visual quality and feature information.
- This approach provides a robust foundation for subsequent computer vision tasks, outperforming existing methods.
