Image Dynamic Range Enhancement based on Fusion Pyramid

Learning techniques have been widely applied in image processing, including dynamic range enhancement. But they require a lot of computational resources to train and run. As an alternative, in this paper we propose an algorithm utilizing only traditional image processing techniques to achieve better results on test datasets. We adopt a contrast-based fusion approach. First, the original image is segmented into multiple regions. Then, several enhanced images are generated and evaluated for their quality. Finally, the best enhanced images for each region is selected and then fused together using an image pyramid. We achieve a 3.9% performance improvement over the state of the art, based on experimental results on datasets.