Denoising & Inpainting using Loopy Belief Propagation

This package provides an implementation of the loopy belief propagation (LBP) algorithm in denoising and inpainting greyscale images.

Description

This package contains 2 modules:

image_damager : Add Gaussian noise to an image and destroy a portion of it.

denoising_inpainting : Denoise and inpaint an image using the LBP algorithm. The current implementation can take many hours for a high number of iterations or high-resolution images.

The algorithm is based on the following paper:

Szeliski R., Zabih R., Scharstein D., Veksler O., Kolmogorov V., Agarwala A., Tappen M., and Rother C. (2008). A comparative study of energy minimization methods for Markov random fields with smoothness-based priors. IEEE Transactions on Pattern Analysis and Machine Intelligence.

Getting Started

Prerequisites

The scripts ware implemented and tested with Python 3.8 but previous versions can also be used.

The following libraries need to be installed:

NumPy
OpenCV
Matplotlib

Installation

Install the package from the repository with the following commands:


        git clone https://github.com/NickGeramanis/denoising-inpainting-lbp

        cd denoising-inpainting-lbp 

        pip3 install -e .

Usage

To damage an image, execute the function damage_image()` of the image_damager module with the following parameters:

image_path, noise_mean_value, noise_variance

For example:


        from denoising_inpainting_lbp import image_damager 

        

        image_damager.damage_image('path/to/image.png', 0, 0.1)

Note that a mask image will also be produced that indicates which pixels have been damaged.

To perform denoising and inpainting on an image using the LBP algorithm, execute the function denoise_inpaint() of the denoising_inpainting module with the following parameters:

image_path, mask_image_path, n_iterations, lambda, energy_lower_bound, max_smoothness_penalty

If the smoothness cost function is not truncated do not provide max_smoothness_penalty.

If energy_lower_bound is not known provide 1.

For example:


        from denoising_inpainting_lbp import denoising_inpainting 

        

        denoising_inpainting.denoise_inpaint('path/to/image.png',
        'path/to/mask.png', 1, 5, 37580519.6)

Furthermore, some unit tests have been implemented in the folder tests to verify the proper functioning of the code.

About

Status

Under maintenance.

License

Distributed under the GPL-3.0 License. See LICENSE for more information.

Authors

Nick Geramanis