3dhead / paint2022

A quick tool to transform any picture in a paint-by-number canvas

paint-by-numbers

A quick Python tool to transform any picture in a paint-by-number canvas, using opencv

How it works

1. Picture is resized, denoised and cleaned using morphomat
2. K-mean algorithm to detect the k-colors summarizing the best the picture
3. Transformation of the picture to match these colors
4. Contour detection and drawing of contours on new canvas
5. Add label of each shape on the canvas
6. Creation of the colormap for the user

This project includes

A set of PNG pictures in ./inputs folder, associated canvas and colormaps in the ./outputs folder The definition of the class Canvas in the python script canvas.py

Requirements:

• Numpy
• Matplotlib
• OpenCV

Execution:

run python run.py [path] [nb_color] [plot] [save] [pixel_size] with:

path: path of the source picture
nb_color : number of colors you want in the canvas (10 to 20)
plot: optional, boolean to set to True if you want to see some plots. Default False
save: optional, boolean to set to True if you want to save results in the ./outputs folder. Default True
pixel_size: optional, interger, size in pixel of the largest dimension of the output canvas (default 4000)

Result:

De Chirico, Place d'Italie, 1922

Expected result (20 colors)

Canvas

Colormap

Next Steps:

• Better positioning of the label in each region
• Better contour filtering to remove small regions