Single Image Dehazing

In this work we propose HR-Dehazer, a novel and accurate method for image dehazing. An encoder-decoder neural network is trained to learn a direct mapping between a hazy image and its respective clear version. We designed a special loss that forces the network to keep into account the semantics of the input image and to promote consistency among local structures. In addition, this loss makes the system more invariant to scale changes. Quantitative results on the recently released Dense-Haze dataset introduced for the NTIRE2019-Dehazing Challenge demonstrates the effectiveness of the proposed method. Furthermore, qualitative results on real data show that the described solution generalizes well to different never-seen scenarios.

Publications

1.

Single Image Dehazing by Predicting Atmospheric Scattering Parameters
(Simone Bianco, Luigi Celona, Flavio Piccoli) In London Imaging Meeting, volume 2020, number 1, pp. 74-77, 2020.

BibTex Doi Project Page 

@inproceedings{bianco2020single,
 author = {Bianco, Simone and Celona, Luigi and Piccoli, Flavio},
 year = {2020},
 pages = {74-77},
 title = {Single Image Dehazing by Predicting Atmospheric Scattering Parameters},
 volume = {2020},
 number = {1},
 organization = {Society for Imaging Science and Technology},
 booktitle = {London Imaging Meeting},
 doi = {https://doi.org/10.2352/issn.2694-118X.2020.LIM-11},
 projectref = {http://www.ivl.disco.unimib.it/activities/single-image-dehazing/}}