Painting using genetic algorithm with aesthetic evaluation of visual quality

Abstract

Creating art using artificial intelligence technologies is an emerging research topic. In particular, evolutionary computation has achieved several promising results in generating visual art and music. Evaluation of the items generated by evolutionary algorithms is a key issue at computational creativity. Interactive evolutionary algorithms are widely used to address this issue by incorporating human feedback in the fitness evaluation. However, this manner suffers from fatigue and decreasing sensitivity after long-term evaluation, which is commonly required in evolutionary algorithms. This paper proposes using an aesthetic evaluation of visual quality in the fitness evaluation for genetic algorithm (GA) to create paintings. Specifically, the fitness function considers two features for aesthetics. The generative ecosystemic art system, EvoEco, is applied as a test bench for the proposed method. Experimental results show that the proposed GA can generate satisfactory paintings by using aesthetic evaluation.