Retinexmamba: Retinex-based Mamba for Low-light Image Enhancement

Jiesong Bai,Yuhao Yin,Qiyuan He,Yuanxian Li,Xiaofeng Zhang

2024-05-20

Abstract:In the field of low-light image enhancement, both traditional Retinex methods and advanced deep learning techniques such as Retinexformer have shown distinct advantages and limitations. Traditional Retinex methods, designed to mimic the human eye's perception of brightness and color, decompose images into illumination and reflection components but struggle with noise management and detail preservation under low light conditions. Retinexformer enhances illumination estimation through traditional self-attention mechanisms, but faces challenges with insufficient interpretability and suboptimal enhancement effects. To overcome these limitations, this paper introduces the RetinexMamba architecture. RetinexMamba not only captures the physical intuitiveness of traditional Retinex methods but also integrates the deep learning framework of Retinexformer, leveraging the computational efficiency of State Space Models (SSMs) to enhance processing speed. This architecture features innovative illumination estimators and damage restorer mechanisms that maintain image quality during enhancement. Moreover, RetinexMamba replaces the IG-MSA (Illumination-Guided Multi-Head Attention) in Retinexformer with a Fused-Attention mechanism, improving the model's interpretability. Experimental evaluations on the LOL dataset show that RetinexMamba outperforms existing deep learning approaches based on Retinex theory in both quantitative and qualitative metrics, confirming its effectiveness and superiority in enhancing low-light images.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The paper aims to address issues in the field of low-light image enhancement, particularly focusing on the limitations of traditional Retinex methods and deep learning-based methods (such as Retinexformer) when dealing with such images. Specifically, the goals of the paper can be summarized as follows:

1. **Overcome the limitations of traditional Retinex methods**: Although traditional Retinex methods can mimic human eye perception of brightness and color by decomposing images into illumination and reflectance components, they face challenges in noise management and detail preservation under low-light conditions.

2. **Improve the shortcomings of Retinexformer**: Retinexformer enhances illumination estimation through traditional self-attention mechanisms but faces issues with interpretability and suboptimal enhancement effects.

3. **Propose the RetinexMamba architecture**: This architecture combines the physical intuitiveness of traditional Retinex methods with the deep learning framework of Retinexformer and utilizes State Space Models (SSMs) to improve processing speed. It introduces innovative illumination estimators and damage repair mechanisms to maintain image quality.

4. **Enhance model interpretability**: By replacing the Illumination-Guided Multi-Head Attention (IG-MSA) in Retinexformer with Fused-Attention, the model's interpretability is improved.

5. **Empirical evaluation and surpassing existing technologies**: Through quantitative and qualitative experiments on the LOL dataset, RetinexMamba's effectiveness in low-light image enhancement and its advantages over existing deep learning methods based on Retinex theory are demonstrated.

In summary, this research is dedicated to developing a new deep learning framework to enhance images under low-light conditions more efficiently and with higher quality, while addressing some key issues in existing methods.