Using Genetic Algorithms to Generate Evolutionary Art in C# and .NET

Admittedly the art in Figure 1 is a bit disturbing, but nonetheless, it could probably pass for art and it was not painted by a human hand, but by a computer algorithm. A number of people in the art world are experimenting with what is known as evolutionary art and even sold a few of their works. Genetic Art is a subset of evolutionary art and uses genetic algorithms to generate the art. To learn more about evolutionary and genetic art, I recommend visiting the Organic, Genetic, and Evolutionary Art Page.

Figure 1 - Genetic Art from Computer Artworks

In this article we will experiment a bit with genetic algorithms and creating fitness functions that produce art on a Windows Form. For the purpose of simplicity, we will generate art in 2 dimensional space using approximations to data points determined from a few general mathematical formulas. The genetic algorithm program we are using utilizes Multiple Expression Programming (MEP). Multiple Expression Programming produces gene sequences of mathematical formulas and in this example the fitness of the formulas are compared against a set of data. To learn more about MEP, check out my previous article, Using Multiple Expression Programming to Derive the Pythagorean Theorem in C# and .NET.

The basic idea behind creating art from mathematical equations, is to produce a color from a formula operating on an x,y coordinate. (In three dimensions you would produce a color from an (x,y,z) coordinate.). In our example we gear our formulas to produce 4 possible colors from results ranging from 0-3. The first dataset for comparing our fitness was derived from the simple formula shown below:

(a + b) mod 4

This formula forces all our simulated results to values between 0 and 3 The genetic algorithm will produce genes that try to converge on a formula that approximates the formula shown above. Its actually better that the algorithm not converge to the correct answer (a+b) mod 4, because the approximation to the formula is what produces the original art. In fact, you may want to choose functions in your MEP genes that are a bit more unusual (In this examples some of the genes include sin(a*b), a/b, and abs (a) + abs(b). Also, you can place a few inaccurate results in the fitness set, to add some interesting effects. Below are some of the resulting artworks produced from genes after 1000 generations of converging on the formula:

Figure 2 - Genetic Art Produced from the fitness formula (a+b) mod 4

Seeing as the module 4 convergence did some interesting things, we chose a similar formula for our next art gallery, ab mod 4. We also introduced some (ab) mod 5 data into the dataset to cause some strange convergences:

(ab) mod 4

Figure 3 - Textures generated from the fitness set of (ab) mod 4

Finally, keeping with our success of the modulus function, we created a set of numbers based on:

a^b mod 5

Changing the color scheme a bit, by producing a different set of colors from the results of (0-4) and running the algorithm on a 1000 generations against the a^b mod 5 data, we get the art textures shown below:

Figure 4 - Art Work based on running the genetic algorithm on the formula set a^b mod 5

Some may argue that this computer generated art is not exactly Van Gogh, but it does provoke the eye a bit, and there is no limit to the number of creations you can produce. By tweaking the set of measured results, the gene formulas utilized by the algorithm, the number of colors produced, and the variety of colors produced, you can really come out with a few cool abstract pieces (who knows, maybe even a masterpiece!)