1. Bézier curve-based saturation-aided optimal brightness adjustment for dark image clearness enhancement with image fusion
- Author
-
Chunxiao Liu, Xiaojie Zhong, and Zhichun Lu
- Subjects
Brightness ,Image fusion ,Exposure ,business.industry ,Computer science ,ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION ,020206 networking & telecommunications ,Bézier curve ,02 engineering and technology ,Evaluation function ,Image (mathematics) ,Signal Processing ,0202 electrical engineering, electronic engineering, information engineering ,020201 artificial intelligence & image processing ,Computer vision ,Artificial intelligence ,Electrical and Electronic Engineering ,business ,Saturation (chemistry) ,ComputingMethodologies_COMPUTERGRAPHICS ,Hue - Abstract
To solve such problems as insufficient illumination improvement and unnatural visual effects in the existing dark image enhancement approaches, we propose a Bezier curve-based and saturation-aided dark image enhancement method with image fusion. Firstly, a saturation-aided preliminary brightness map estimation approach is put forward. Both saturation and brightness information are utilized to compensate the illumination information, which makes for more colorful results. Then, an evaluation function based on the preliminary brightness map is devised to get an optimal Bezier curve for the illumination adjustment purpose. Bezier curve is more flexible than traditional curves, so it can be used to improve the illumination of dark images in various types. Secondly, we handle the potential overexposure and detail loss in the optimized brightness map by fusing it with the original as well as the CLAHE-processed brightness maps. And, a mask fusion strategy is designed to avoid over-enhancement for the area that is supposed to be smooth. Finally, the enhanced brightness map is combined with the original hue and saturation channels to get our enhanced results, which present richer details and look more vivid. Our method is quantitatively evaluated with NIQE and gets a high score 3.70 on a 101 dark image dataset, which is the best among the state-of-the-art approaches compared.
- Published
- 2020
- Full Text
- View/download PDF