Information Technology Reference
In-Depth Information
Painting Using Genetic Algorithm
with Aesthetic Evaluation of Visual Quality
Sheng-Yu Feng and Chuan-Kang Ting
Department of Computer Science and Information Engineering
National Chung Cheng University, Taiwan
{
fsy102m,ckting
}
@cs.ccu.edu.tw
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 is-
sue 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 decreas-
ing sensitivity after long-term evaluation, which is commonly required in
evolutionary algorithms. This paper proposes using an aesthetic evalua-
tion of visual quality in the fitness evaluation for genetic algorithm (GA)
to create paintings. Specifically, the fitness function considers two fea-
tures 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.
1 Introduction
Computational creativity has received increasing attention owing to the advance
of artificial intelligence technologies. Creativity involves with the generation of
appropriate novelty and represents some ideas based on previous works [2]. Sev-
eral artificial intelligence technologies have been applied to achieve creativity by
computers. In particular, using evolutionary computation to generate artworks,
create visual art, and compose music has gained considerable promising results
[4,10,13]. Evolutionary algorithms (EAs) are nature-inspired global search ap-
proaches and have succeeded in solving a variety of complex optimization prob-
lems. EAs manipulate a set of chromosomes representing candidate solutions
and change them by mimicking the evolutionary operators in nature, such as
selection, crossover, and mutation. The candidate solutions are evaluated by the
fitness function. Following Darwin's theory “Survival of the Fittest,” the chro-
mosomes with relatively high fitness values are selected to survive into the next
generation.
Fitness evaluation plays an important role in EAs because it guides the search
direction. In computational creativity using EAs, it ordinarily lacks a specific fit-
ness function for evaluating chromosomes. Human feedback is commonly used as
Search WWH ::
Custom Search