Low-Illumination Image Enhancement Algorithm Based on Improved Multi-Scale Retinex and ABC Algorithm Optimization

Ying Sun, Zichen Zhao, Du Jiang, Xiliang Tong, Bo Tao, Guozhang Jiang, Jianyi Kong, Juntong Yun, Ying Liu, Xin Liu, Guojun Zhao, Zifan FangFrontiers in Bioengineering and Biotechnology 2022 · DOI

Abstract

In order to solve the problems of poor image quality, loss of detail information and excessive brightness enhancement during image enhancement in low light environment, we propose a low-light image enhancement algorithm based on improved multi-scale Retinex and Artificial Bee Colony (ABC) algorithm optimization in this paper. First of all, the algorithm makes two copies of the original image, afterwards, the irradiation component of the original image is obtained by used the structure extraction from texture via relative total variation for the first image, and combines it with the multi-scale Retinex algorithm to obtain the reflection component of the original image, which are simultaneously enhanced using histogram equalization, bilateral gamma function correction and bilateral filtering. In the next part, the second image is enhanced by histogram equalization and edge-preserving with Weighted Guided Image Filtering (WGIF). Finally, the weight-optimized image fusion is performed by ABC algorithm. The mean values of Information Entropy (IE), Average Gradient (AG) and Standard Deviation (SD) of the enhanced images are respectively 7.7878, 7.5560 and 67.0154, and the improvement compared to original image is respectively 2.4916, 5.8599 and 52.7553. The results of experiment show that the algorithm proposed in this paper improves the light loss problem in the image enhancement process, enhances the image sharpness, highlights the image details, restores the color of the image, and also reduces image noise with good edge preservation which enables a better visual perception of the image.

The Challenge of Low-Light Imaging

Capturing high-quality images in low-illumination environments remains a persistent challenge in computer vision. Standard low-light photos suffer from low contrast, heavy noise, and a loss of structural detail, which hinders both human perception and downstream tasks like target detection or medical imaging.

Existing solutions often fall short. Histogram Equalization (HE) frequently over-amplifies noise and loses fine details. Retinex-based methods, while good at dynamic range compression, often produce "halo" artifacts at high-contrast edges and suffer from color bias. Furthermore, while deep learning methods are powerful, they require massive datasets and significant computational resources for training. This research proposes a traditional-mathematical approach that is more robust and adaptive than simple filters but less data-hungry than deep neural networks.

Methodology

The proposed algorithm follows a "divide and conquer" strategy, splitting the original image into two distinct processing paths based on the Retinex theory, which decomposes an image into illumination and reflection components.

Retinex schematic illustrating the relationship between illumination, reflection, and the observer.
  1. The Compensation Layer: This path focuses on restoring color and illumination. Instead of using standard Gaussian blurring to estimate the illumination map (which causes halos), the authors use Relative Total Variation (RTV) for structure extraction. This is followed by Multi-Scale Retinex with Color Restoration (MSRCR), Histogram Equalization, and a bilateral gamma function to smoothly correct light-to-dark transitions.
  2. The Main Feature Layer: This path focuses on sharpness and noise reduction. It utilizes Weighted Guided Image Filtering (WGIF). Unlike standard Guided Image Filtering, WGIF uses an adaptive weighting factor based on local variance, allowing it to preserve edges more effectively while suppressing noise in flat regions.

Swarm Intelligence for Optimal Fusion

The core innovation of this framework is the fusion step. Rather than using fixed weights to combine the compensation and feature layers, the algorithm employs the Artificial Bee Colony (ABC) algorithm. ABC is a metaheuristic optimization tool that mimics the foraging behavior of honeybees. It iteratively searches for the optimal fusion weights by maximizing a "fitness function" composed of three image quality metrics: Information Entropy (IE), Average Gradient (AG), and Standard Deviation (SD). This allows the algorithm to be self-tuning, finding the best balance of detail and brightness for any given image.

Experimental Results and Analysis

The authors evaluated their algorithm using the LOLdataset (Low-Light dataset), a widely recognized benchmark. The quantitative results show a significant leap over traditional Retinex methods. Compared to the original low-light images, the proposed method improved Information Entropy (IE) by 2.4916, Average Gradient (AG) by 5.8599, and Standard Deviation (SD) by 52.7553.

Qualitative comparison showing detail restoration in a bedroom scene.

Visual results demonstrate that the algorithm restores details in deep shadows—such as the texture of bed linens or the patterns on a vase—that remain "annihilated" or noisy in other state-of-the-art methods. The algorithm effectively eliminates "halo" artifacts and noise common in traditional methods while maintaining superior brightness restoration.

Conclusion and Future Work

While the ABC algorithm is noted for its fast convergence (completing optimization in roughly 14 iterations), the overall pipeline is computationally complex. The combination of RTV structure extraction and iterative swarm optimization might be too slow for real-time video applications on low-power mobile hardware. Future work will involve further refining the fitness function and exploring ways to accelerate the structural extraction process to move closer to real-time performance.

Summary updated

Abstract

In order to solve the problems of poor image quality, loss of detail information and excessive brightness enhancement during image enhancement in low light environment, we propose a low-light image enhancement algorithm based on improved multi-scale Retinex and Artificial Bee Colony (ABC) algorithm optimization in this paper. First of all, the algorithm makes two copies of the original image, afterwards, the irradiation component of the original image is obtained by used the structure extraction from texture via relative total variation for the first image, and combines it with the multi-scale Retinex algorithm to obtain the reflection component of the original image, which are simultaneously enhanced using histogram equalization, bilateral gamma function correction and bilateral filtering. In the next part, the second image is enhanced by histogram equalization and edge-preserving with Weighted Guided Image Filtering (WGIF). Finally, the weight-optimized image fusion is performed by ABC algorithm. The mean values of Information Entropy (IE), Average Gradient (AG) and Standard Deviation (SD) of the enhanced images are respectively 7.7878, 7.5560 and 67.0154, and the improvement compared to original image is respectively 2.4916, 5.8599 and 52.7553. The results of experiment show that the algorithm proposed in this paper improves the light loss problem in the image enhancement process, enhances the image sharpness, highlights the image details, restores the color of the image, and also reduces image noise with good edge preservation which enables a better visual perception of the image.
BibTeX
@inproceedings{sun2022low,
  author = {Ying Sun and Zichen Zhao and Du Jiang and Xiliang Tong and Bo Tao and Guozhang Jiang and Jianyi Kong and Juntong Yun and Ying Liu and Xin Liu and Guojun Zhao and Zifan Fang},
  title = {{Low-Illumination Image Enhancement Algorithm Based on Improved Multi-Scale Retinex and ABC Algorithm Optimization}},
  booktitle = {Frontiers in Bioengineering and Biotechnology},
  year = {2022},
  doi = {10.3389/fbioe.2022.865820},
}
Figure