A Comprehesive Multi-Level Low-Light Video Dataset for Benchmarking of Low-Light Image Enhancement Techniques

Images taken indoor and/or outdoor can suffer from adverse effects when there is inadequate lighting in the environment. The overall efficacy of computer vision systems may be compromised due to the resultant low dynamic range and elevated noise levels in the images. To overcome this problem, many traditional and deep learning-based enhancement methods have been employed. In terms of deep learning models, to achieve better results, the deep models need to be trained with good quality and rich datasets. Although there are many low-light static image datasets in the literature, there is a lack of multi-level low light image/video dataset. For a live video captured in a low-light environment, the level of the light naturally can change with respect to time. Therefore, the model should also adopt to those different level of low-light conditions. In this study, a new multi-level low-light video dataset is presented. This dataset includes 3 different scenes captured with a color camera under different levels of light. By using this dataset, in this study several state-of-the-art low-light image enhancement methods are tested and compared.