GeneticLeonardo is a program that "evolves" drawings. The chromosomes encode the position, shape and color of 50 polygons. The fittest chromosomes are those who produce images most like a given image.
The goal is to approximate any image with only 50 polygons.
Here's how it works:
-
Generate a random population of chromosomes
-
Produce images by interpreting these chromosomes
-
Select the chromosomes that produce the best images
-
Mate the best chromosomes by exchanging information
-
Mutate the new chromosomes slightly
-
Kill the worst ranked chromosomes
-
Repeat from step 2 until satisfied
An executable version can be downloaded from:
http://github.com/ynd/genetic-drawing/raw/master/genetic_drawing-0.1.zip
You are more than welcome to contribute to this project! There is room for a lot of improvement.
The code is released as Open Source under the BSD license.
A zip of the source code is available at:
http://github.com/ynd/genetic-drawing/raw/master/genetic_drawing-src-0.1.zip
A zip of an earlier version which uses Genetic Programming is available too:
http://github.com/ynd/genetic-drawing/raw/gp/genetic_drawing-gp-src-0.1.zip
Note that the GP version doesn't converge very well yet.
The most important classes are:
-
gd.gui.EvolutionRunnable: Class in charge of actually running the evolution process.
-
gd.core.GAConfiguration: Encapsulates the settings of the genetic algorithm.
-
gd.core.GAInitialChromosomeFactory: Creates a suitable initial chromosome.
-
gd.core.GAPhenotypeExpresser: Class that can interpret the information encoded in the chromosomes.
-
gd.core.LMSFitnessFunction: Computes the fitness of a program as the Least-Mean-Sqare distance between the image it generates and the target image.
