Voronoi-ify Any Image

Take an image and use ML to turn it into a Voronoi map.

Uses Python 3.6.5.

How It Works

The initial code for turning the map into a Voronoi diagram comes from orlp on Stack Exchange. This link has a good explanation of the approach taken, I have just cleaned up the code and made small modifications for my own needs.

Inside the maps folder, put any jpg images, and use the filename format:

• 00_FILENAME_N.jpg

where:

• FILENAME is the name of the file. Please do not include spaces or any special characters.
• N is the number of Voronoi centroids you want to create. If you're not sure, try 2000.

Once you have all your desired files, run the Python script and the outputs will be inserted into the same folder. There will be multiple files, namely:

• 01_FILENAME_N_denoised.png will contain a denoised version of the file if this option has been selected. Only really required for noisy images such as high ISO photographs.
• 02_FILENAME_N_grayscale.png will contain a grayscale version of the image, which is used for edge detection purposes.
• 03_FILENAME_N_entropyweights.png contains the entropy weights of the image. This makes up 30% of the total weighting.
• 04_FILENAME_N_edgeweights.png contains the edge weights of the image. This makes up 70% of the total weighting.
• 05_FILENAME_N_allpoissinsamples.png contains every sample returned from the Poisson disc sampling, prior to this being trimmed down to the number specified by N.
• 06_FILENAME_N_filteredpoissinsamples.png contains only the N most relevant samples.
• 07_FILENAME_N_voronoi.png is a full colour image of the original image but using the N Voronoi samples.
• 08_FILENAME_N_voronoicentroids.png is the output required to use the D3.js visualisation described in the section Interactive Visualisation below.

The Effect Of N On The Output

Obviously, the higher number of samples that are used to construct the output the more faithful it will be to the original image. The GIF below shows a run through of the affect of N on the output:

Interactive Visualisation

In conjunction with the Python code above, there is a D3.js visualisation which uses the 08 file outputted above. See below for a preview: