Image Restoration by Learning Morphological Opening-Closing Network

This repository contains the official Tensorflow implementation of the Opening-Closing network as described in the paper Image Restoration by Learning Morphological Opening-Closing Network (Medium Blog)

Overview

In this work, we revisit the concept of structuring elements (SE) for morphological operations and attempt to incorporate it with learning-based methods. We propose the Opening-Closing network, consisting solely of basic morphological operations such as Opening and Closing, for de-raining and de-hazing real-life images. Unlike convolutions, morphological operations have inherent non-linearity that helps extract interpretable features and produce complex decision boundaries. These properties allow our network to get results comparable to other state of the art methods, despite having a fraction of parameters.

Files

├── src_haze    # directory contains the code for color image dehazing 
├── src_rain    # directory contains the code for color/grayscale image de-raining  
├── src_shape   # directory contains the code for simulation 

Results

Real rainy images

Real hazy images

Please cite our paper if you find the code useful.

@inproceedings{mondal2020image,
  title={Image Restoration by Learning Morphological Opening-Closing Network },
  author={Mondal, Ranjan and Shankar Dey, Moni   and Chanda, Bhabatosh},
  journal={Mathematical Morphology-Theory and Applications},
  number={1},
  pages={87--107},
  year={2020},
  publisher={De Gruyter}
}

@inproceedings{mondal2019morphological,
  title={Morphological networks for image de-raining},
  author={Mondal, Ranjan and Purkait, Pulak and Santra, Sanchayan and Chanda, Bhabatosh},
  booktitle={International Conference on Discrete Geometry for Computer Imagery},
  pages={262--275},
  year={2019},
  organization={Springer}
}