A Haze Removal Method Based on Additional Depth Information and Image Fusion

Abstract

To address the problem of image degradation in foggy days, we propose a haze removal method based on additional depth information and image fusion. With recent advances in depth-sensing technology, it has been realized that sensing devices can produce depth images in which the depth value are quite accurate. We adopt the depth estimation dataset of Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) which contains images collected from different real-world environments. The additional information includes the LiDaR scanning points and original depth images which can be used to estimate the optical depth of each point in the scene. In this paper, we investigate how to use additional depth information to remove haze for a single image. Our method focuses on LiDaR depth imaging, image fusion, and the atmospheric scattering model. We use LiDaR scanning points as input and then deduce a rough depth image with prominent features. The rough depth image is then combined with original depth image to improve reliability of depth estimation by image fusion. Using the atmospheric scattering model, we can remove haze for a single image. Experimental results show that our proposed approach provides better performance of dehazing under different fog conditions and holding the details of remote sensing images than current research methods.