Dehazing

This repository is used to publish the code, that usd in my image dehazing paper.

The C++ algorithm can set the parameters below:

block_size: the kernel size of twice minimum filter, before get the dark channel.
morphology_transoform_kernel_size: the kernel size of Morphological Transformations
is_cirle_wrong_point: circle the wrong pixel of airlight  

Also, there is a python source code. To calculate the dark channel, I used a erode calculation, same as the twice minimum filter.

Dependency

This algorithm is basesd on C++ and OpenCV. Install OpenCV from source code, more informatino can be found here.

Sample dataset

Download some sample datasets to test the algorithm.

Benchmark images

Run the program

C++

g++ dehaze.cpp dehaze.h `pkg-config opencv --cflags --libs` -o dehaze
./dehaze input.bmp

Python

python3 dehaze.py input_file file_type

Paper

@article{2019TJSKE-D-19-00004,
  title={Research on Haze Removal for Autonomous Car},
  author={Zhi, Wang and Daishi, Watabe},
  journal={Transactions of Japan Society of Kansei Engineering},
  volume={18},
  number={6},
  pages={417-421},
  year={2019},
  doi={10.5057/jjske.TJSKE-D-19-00004}
}