Low-Light Enhancement Effect on Classification and Detection: An Empirical Study

Xu Wu,Zhihui Lai,Zhou Jie,Can Gao,Xianxu Hou,Ya-nan Zhang,Linlin Shen

2024-09-22

Abstract:Low-light images are commonly encountered in real-world scenarios, and numerous low-light image enhancement (LLIE) methods have been proposed to improve the visibility of these images. The primary goal of LLIE is to generate clearer images that are more visually pleasing to humans. However, the impact of LLIE methods in high-level vision tasks, such as image classification and object detection, which rely on high-quality image datasets, is not well {explored}. To explore the impact, we comprehensively evaluate LLIE methods on these high-level vision tasks by utilizing an empirical investigation comprising image classification and object detection experiments. The evaluation reveals a dichotomy: {\textit{While Low-Light Image Enhancement (LLIE) methods enhance human visual interpretation, their effect on computer vision tasks is inconsistent and can sometimes be harmful. }} Our findings suggest a disconnect between image enhancement for human visual perception and for machine analysis, indicating a need for LLIE methods tailored to support high-level vision tasks effectively. This insight is crucial for the development of LLIE techniques that align with the needs of both human and machine vision.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: the impact of low - light image enhancement (LLIE) methods on high - level vision tasks (such as image classification and object detection) has not been fully explored. Although LLIE methods aim to improve the visibility and clarity of low - light images, while these methods improve human visual perception, their effects on computer vision tasks are inconsistent and sometimes even harmful.

Specifically, the paper empirically explores the following aspects:

1. **The impact of low - light image enhancement on human visual perception**:
- LLIE methods can significantly improve the visibility of low - light images, making them clearer and more in line with the aesthetic standards of human vision.

2. **The impact of low - light image enhancement on computer vision tasks**:
- Through the evaluation of high - level vision tasks such as image classification and object detection, it is found that the effects of LLIE methods are not always positive. Sometimes, the enhanced images will instead reduce the performance of these tasks.

3. **The differences between human vision and machine vision**:
- The paper points out that there is a disconnection between LLIE methods in optimizing human visual perception and improving the performance of machine vision tasks. This indicates the need to develop LLIE methods specifically for high - level vision tasks.

4. **Experimental verification**:
- The paper constructs a low - light image dataset containing 19 categories and conducts a large number of experiments to evaluate the impact of different LLIE methods on image classification and object detection tasks.

In summary, the main purpose of this paper is to reveal the actual impact of LLIE methods on high - level vision tasks under low - light conditions through systematic experiments and analysis, and to provide guidance for future research to develop LLIE techniques more suitable for high - level vision tasks.

### Formula representation

- **PSNR (Peak Signal - to - Noise Ratio)**:
\[
\text{PSNR} = 10\log_{10}\left(\frac{(2^n - 1)^2}{\text{MSE}(x, y)}\right)
\]
where \(x\) and \(y\) represent the reference image and the enhanced image respectively, \(n\) is the number of binary bits per pixel, and \(\text{MSE}(\cdot)\) represents the mean - squared error.

- **SSIM (Structural Similarity Index)**:
SSIM is an image quality evaluation index based on contrast, brightness and structure, taking human visual perception into account.

- **BRISQUE (Blind/No - Reference Image Spatial Quality Evaluator)**:
\[
D(\mu_1, \mu_2, \sigma_1, \sigma_2)=\sqrt{(\mu_1 - \mu_2)^T(\sigma_1 + \sigma_2)^{-1}(\mu_1 - \mu_2)}
\]
where \(\mu_1\) and \(\mu_2\) represent the mean vectors of the multivariate Gaussian distribution model and the input image respectively, and \(\sigma_1\) and \(\sigma_2\) represent their covariance matrices.

These formulas are used to evaluate the performance of LLIE methods on different datasets.