Single-Image-Dehazing-Python

python implementation of the paper: "Efficient Image Dehazing with Boundary Constraint and Contextual Regularization"

Installation and Running the tests

method 1

pip install image_dehazer

Usage:

import image_dehazer										# Load the library

HazeImg = cv2.imread('image_path')						# read input image -- (**must be a color image**)
HazeCorrectedImg, HazeTransmissionMap = image_dehazer.remove_haze(HazeImg)		# Remove Haze

cv2.imshow('input image', HazeImg);						# display the original hazy image
cv2.imshow('enhanced_image', HazeCorrectedImg);			# display the result
cv2.waitKey(0)											# hold the display window

user controllable parameters (with their default values):

airlightEstimation_windowSze=15
boundaryConstraint_windowSze=3
C0=20
C1=300
regularize_lambda=0.1
sigma=0.5
delta=0.85
showHazeTrasmissionMap=True

method 2

1. Go to the src folder
2. run the file "example.py"
3. sample images are stored in the "Images/" folder
4. Output images will be stored in the "outputImages/" folder

Libraries needed:

1.numpy==1.19.0

2.opencv-python

3.scipy

Theory

This code is an implementation of the paper "Efficient Image Dehazing with Boundary Constraint and Contextual Regularization" The algorithm can be divided into 4 parts:

• Airlight estimation
• Calculating boundary constraints
• Estimate and refine Transmission
• Perform Dehazing using the estimated Airlight and Transmission

License

• This project is licensed under the BSD 2 License - see the LICENSE.md file for details

Acknowledgements

• The author would like to thank "Gaofeng MENG" and his implementation of his algorithm: https://github.com/gfmeng/imagedehaze

• The author would like to thank Gaofeng MENG, Ying WANG, Jiangyong DUAN, Shiming XIANG, Chunhong PAN for their paper "Efficient Image Dehazing with Boundary Constraint and Contextual Regularization"

• The author would like to thank Alexandre Boucaud. The function psf2otf was obtained from his repository. (https://github.com/aboucaud/pypher/blob/master/pypher/pypher.py)

• The Author would like to thank Dr. Suresh Merugu for his matlab implementation of the codes. This repository is the python implementation of the matlab codes.

• The Author would like to thank Mayank Singal for his repository "PyTorch-Image-Dehazing" which gives a pytorch implementation of the AOD-Net architecture. Link to ICCV 2017 paper

Merugu, Suresh. (2014). Re: How to detect fog in an image and then enhance the image to remove fog?. Retrieved from: https://www.researchgate.net/post/How_to_detect_fog_in_an_image_and_then_enhance_the_image_to_remove_fog/53ae3f10d2fd64c3648b45a9/citation/download.

Citation

@INPROCEEDINGS{6751186, 
  author={G. Meng and Y. Wang and J. Duan and S. Xiang and C. Pan}, 
  booktitle={IEEE International Conference on Computer Vision}, 
  title={Efficient Image Dehazing with Boundary Constraint and Contextual Regularization}, 
  year={2013}, 
  volume={}, 
  number={}, 
  pages={617-624}, 
  month={Dec},}

Results

Performance Comparison:

In this section, I am comparing the dehazing output with that of AOD-Net. I am using this python implementation of AOD-Net to run a pretrained AOD-Net model

Here are some cases where AOD-Net is better: