An implementation of the algorithm described in Single Image Haze Removal Using Dark Channel Prior [He et al. 09] (page), with the modifications proposed in Guided Filtering [He et al. 10] for faster transmission refinement. It was implemented with Python 2.7 and uses the SciPy and NumPy libraries.
In order to run the program one needs:
-
Python 2.7 installed;
-
SciPy installed;
-
NumPy installed.
Having those requirements, one should be able to run the program with the following command line (considering one is in the src folder):
$ python main.py -i ../images/cones.jpg -o ../results/cones_res.jpg
This programs calls the main module of the program to receive the arguments. The first argument -i is the path to the input image that will be dehazed. While the -o argument is the path to the output image, that is, the dehazed version of the input image. These are the only two required arguments.
For optional arguments, one can type:
$ python main.py -h
This will display the set of arguments available.
A set of benchmark images can be found under the folder images. Most were taken from the main base paper page, but some were taken from the page of Dehazing Using Color-Lines [Fattal 14].
Results of applying the program to some of the benchmark images can be found under the folder results.
There is a document under references listing all the papers used in the development of this project. However, the two main references for this project were:
- Single Image Haze Removal Using Dark Channel Prior [He et al. 09], CVRP;
- Guided Filtering [He et al. 10], ECCV.
This project was developed as a Final Project for the "INF01050 - Computational Photography" class, 2016, at UFRGS.