Painting Using Genetic Algorithm with Aesthetic Evaluation of Visual Quality - Technologies and Applications of Artificial Intelligence - page 132

Tabl e 2. Parameter setting

Parameter Value

Paint size 256 × 256

Representation Agents

Population size 10

Selection 5-tournament

Crossover Uniform

Crossover rate P c 1.0

Mutation Uniform

Mutation rate P m 5/chromosome length

Survivor μ + ʻ

Termination 10 generations

The objective of GA is to find the paintings that maximize the above fitness
value considering edge distribution and line intersection.

4 Simulation Results

This study conducts simulations to examine the performance of the proposed
GA-based painting system. Table 2 lists the parameter setting for the GA. Two
factors are taken into account in the simulations: The first is the use of smooth
rule; the second is the presence of uncertainty in the moving direction. The study
investigates the results from the four combinations of these two factors.

1) No Smoothing + Deterministic Direction
We first experimented with drawing a pixel in a deterministic direction without
performing the smoothing rule at each step. The color of a pixel is calculated
according to its neighbors' information. The simulation results in Fig. 6a shows
that the generated patterns lack diversity. This is caused by the use of deter-
ministic direction for drawing pixels, which results in many straight lines in
the patterns. To address the monotony issue, we adopt the smoothing rule and
enable uncertainty in the direction to move.

2) Smoothing + Deterministic Direction
Figure 6b presents the results of drawing a pixel in a deterministic direction with
smoothing at each step. The generated patterns are similar with Fig. 6a because
the moving directions are deterministic. However, the gaps between pixels are
moderated by smoothing among neighbors. It helps to generate smoother and
more harmonious patterns in Fig. 6b than the patterns in Fig. 6a.