Genetic is a Python 2 implementation of a genetic hill climber for image matching. It was inspired in part by the JavaScript implementation at http://alteredqualia.com/visualization/evolve/.
- BeautifulSoup
- numpy
- pyglet
Genetic expects to be passed a single image file to match:
python genetic.py IMAGE_FILEWhen running it will display a window which contains the target image, along with the current attempt at matching that image (starting from a random seed).
Closing the window causes the current attempt to be saved as an SVG file
alongside the target image (with an additional .svg extension).
Matching can be continued from an existing SVG using the --resume argument.
Alternatively the starting number of polygons can be changed with --polygons.
See the --help for more details.
Genetic iterates through a large number of possible images in search of one that most closely matches the target. Each candidate image is comprised of a number of polygons (currently triangles) of various colours and dimensions.
On each iteration, a number of mutations are applied to the current best candidate image in order to produce a new candidate. The mutations are one of:
- reshaping a triangle (i.e: moving some of the corners)
- recolouring a triangle
- changing the order of a pair of triangles (so that one is now above the other)
The relative difference, in terms of pixels, between the target image and the new candidate are then compared. If this is better than the current best candidate then the new candidate is used as the current best.