multi-dimensional-poisson-blending

A multidimensional generalization of the seamless cloning algorithm suggested in Perez et al. (2003) [1], AKA Poisson Blending.

Examples

2D

This repository includes three 2D example sets, demonstrated in both grayscale and RGB modes in main.py. RGB results are attached below:

3D

The main.py script also generates a 3D example, where an MRI scan of a brain is blended into an MRI scan of a spinal cord. A screenshot of the blended 3D image, presented with napari, is attached below:

Usage

To use the code, git clone this repository. To install dependencies, run the following in a terminal window:

pip install -r requirements.txt

To blend grayscale images, use PoissonBlending.Poisson2DBlender or PoissonBlending.Poisson3DBlender. Both classes expect numpy arrays as inputs. Loading images into acceptable formats is the user's responsibility, and see main.py for working examples.

To blend RGB (or any other multichannel format) images, create an instance of any PoissonBlender (e.g., Poisson2DBlender, Poisson3DBlender) with colored images as inputs. Then use the Colored decorator class to add multichannel support. For example:

from PoissonBlending import Poisson2DBlender, Colored

rgb_2d_blender = Colored(Poisson2DBlender(source, target, mask))
rgb_2d_blender.blend()
rgb_2d_blender.show_results()

Working examples can be found in main.py.

Sources

• 2D implementation inspired by poissonpy [2].
• The OpenNeuro [3] platform was used for 3D example images. Specifically, the brain image is from Greco et al. (2024) [4] and the spinal cord image is from Dabbagh et al. (2024) [5].

References

[1] Pérez, P., Gangnet, M., & Blake, A. (2003). Poisson image editing. In ACM Transactions on Graphics, Volume 22, Issue 3 (pp. 313-318). doi: doi:10.1145/882262.882269

[2] https://github.com/bchao1/poissonpy

[3] https://openneuro.org/

[4] Viviana Greco and Tamas A. Foldes and Neil A. Harrison and Kevin Murphy and Marta Wawrzuta and Mahmoud E. A. Abdellahi and Penelope A. Lewis (2024). Depotentiation of emotional reactivity using TMR during REM sleep. OpenNeuro. [Dataset] doi: doi:10.18112/openneuro.ds005530.v1.0.8

[5] Alice Dabbagh and Ulrike Horn and Merve Kaptan and Toralf Mildner and Roland Mueller and Joeran Lepsien and Nikolaus Weiskopf and Jonathan C.W. Brooks and Juergen Finsterbusch and Falk Eippert (2024). Reliability of taskbased fMRI in the dorsal horn of the human spinal cord. OpenNeuro. [Dataset] doi: doi:10.18112/openneuro.ds004926.v1.3.0