Evolving aesthetic images using multiobjective optimization

We consider the problem of using evolutionary multiobjective optimization to evolve visual imagery. In our method, images (phenomes) are generated from expressions (genomes), and then color segmented so that they can be evaluated under a number of different aesthetic criteria. Our principal task is to formulate fitness functions that make the best use of these elementary aesthetic components. We demonstrate the benefits obtained from using more than one objective function. We also discuss technical issues that arose as a consequence of treating our computational aesthetics problem as a "real-world" application of evolutionary multiobjective optimization.