Bilateral Image Inpainting

This project implements a new inpainting method that proposes convolution with a bilateral averaging kernel. Traditional convolution-based inpainting use image information from only a space neighborhood. A more robust approach (in structure and texture preserving) of convolution–based inpainting is using information from both space and range. This implementation produces some fascinating results when it comes to structural and textural restoration of damaged regions of the image.

Steps to follow:

• Clone the project repository

  • git clone https://github.com/raj-shubham/Bilateral_Image_Inpainting.git
  • Requirements: python, pip (preferred version: python3, pip3) compatible with your system

• Change path in the following files:

  • Bilateral_Image_Inpainting/code/config/settings.config

    "master_working_directory": "/Users/mymac/Desktop/Assignments/DIP-Assignments/Project/test/Bilateral_Image_Inpainting/code/"
    "log_directory": "/Users/mymac/Desktop/Assignments/DIP-Assignments/Project/test/Bilateral_Image_Inpainting/code/config"
    
    to 
    
    "master_working_directory": "Your_Directory/Bilateral_Image_Inpainting/code/"
    "log_directory": "Your_Directory/Bilateral_Image_Inpainting/code/config"

  • Bilateral_Image_Inpainting/code/GUI/mousecallback.py

    master_working_path = "/Users/mymac/Desktop/Assignments/DIP-Assignments/Project/test/Bilateral_Image_Inpainting/code/"
    
    to 
    
    master_working_path = "Your_Directory/Bilateral_Image_Inpainting/code/"

Install the required python modules used in the project (execute inside /Bilateral_Image_Inpainting):

$ pip3 install -r requirements.txt

Compile using the following command (execute inside /Bilateral_Image_Inpainting):

$ python3 -m compileall .

Path to relevant files:

Bilateral_Image_Inpainting/code/GUI/mousecallback.py
Bilateral_Image_Inpainting/code/GUI/comparative_analysis.py
Bilateral_Image_Inpainting/code/config/settings.config

Help

$ python3 mousecallback.py -h
or 
$ python3 mousecallback.py --help

usage: mousecallback.py [-h]
                        image_width kernel_width sigma_range sigma_domain
                        filter_itr

Performs inpainting eg. python3 mousecallback.py 20 5 10.0 0.5 2

positional arguments:
  image_width   neighbourhood range of image (type: int ; eg. 20)
  kernel_width  kernel size (odd number) (type: int ; eg. 5)
  sigma_range   variation across intensity (type: float ; eg. 10.0)
  sigma_domain  variation across distance (type: float ; eg. 0.5)
  filter_itr    iterations over damaged regions (type: int ; eg. 2)

optional arguments:
  -h, --help    show this help message and exit

$ python3 comparative_analysis.py -h
or 
$ python3 comparative_analysis.py --help

usage: comparative_analysis.py [-h] iterations sigma_range sigma_domain

Compares different inpainting solutions eg. python3 comparative_analysis.py 5 25.0 0.5

positional arguments:
  iterations    input number of iterations of inpainting algorithms (type: int
                ; eg. 5)
  sigma_range   variation across intensity (type: float ; eg. 25.0)
  sigma_domain  variation across distance (type: float ; eg. 0.5)

optional arguments:
  -h, --help    show this help message and exit

Execute

$ python3 mousecallback.py 20 5 10.0 0.5 2

Select the damaged regions by dragging your mouse pointer over those areas while pressing the left mouse key

Press 'y' when you are done with the selection of damaged regions

Wait until the inpainting is in progress...

An opencv window with inpainted image appears after some time

Press 'y' to see the changes made on the damaged regions along with the original damaged image

Result: