This project is a simple implementation of Poisson Blending algorithm whose code size is 5 lines. It uses scipy and contain only esssential logics. i.e. input validations are missing and single data type and channel are support.

The contants of poisson_blend.py are following. You can see that there are only 5 lines. They are all of out Poisson Blending algorithm. In fact, They are optimized for code size. It means that there are some redundant computations.

from scipy.ndimage import laplace
def poisson_blend(target_img, src_img, mask_img, iter: int = 1024):
    for _ in range(iter):
        target_img = target_img   0.25 * mask_img * laplace(target_img - src_img)
    return target_img.clip(0, 1)

Above codes generate the folloing right image with three left images as its input. These input images area source image, mask image and target image from left to right.

• Only importing gradients method support
• Mixigin gradients methods is not support
• Convergence checking of jacobi method is not suuport

$ git clone https://github.com/ar90n/poisson-blending-in-5lines.git
$ cd poisson-blending-in-5lines
$ poetry install
$ poetry run jupytext --to notebook ./notebook/example.py ./notebook/example.ipynb
$ poetry run jupyter notebook ./notebook/example.ipynb

• Poisson Blending
• Erkaman / poisson_blend
• Poisson Image Editing

This software is released under the Apache License, see LICENSE.