Research on Low-Light Image Enhancement Algorithm Based on Color-Improved Zero-DCE

In the field of computer vision, low-light image enhancement is crucial for tasks such as target tracking, detection and entity segmentation. In recent years, although supervised learning-based low-light image enhancement methods have achieved promising results, they still have problems such as relying on a large amount of paired data and high computational cost. To solve these problems, unsupervised methods such as Zero-DCE have emerged, however, through practical observation, the model has bias for color processing. Therefore, in this paper, the color loss function of Zero-DCE model is improved, and experiments are carried out on LOL and VE-LOL-L datasets to compare the performance of the model before and after the improvement, as well as ablation experiments and hyperparameter selection experiments. The results show that the improved model significantly improves the image reconstruction quality and structural similarity, effectively improves the low-light image enhancement effect, and provides a new direction for subsequent related research.