IEEE Systems, Man and Cybernetics Magazine - January 2020 - 21

painting works in the test set that were not created by
the computer. The experimental results are shown in Figure 4. From Figure 4(a), it is clear that, for testing the
artworks of the 10 test subjects, the fidelity rate of the
system's creation corresponding to nine test subjects
was higher than 50%. The fidelity rate corresponding to
test subject seven was 90%, which indicates that the creativity of the system we built is high. In addition, we
also tested the running-time delay of the system, with
the results displayed in Figure 4(b). The data transmission delay time was 0.6-0.8 s, the training time model
was 1-1.3 s, while the model test time was less than
50 ms. Accordingly, deploying the algorithm in practical
applications is feasible.
Conclusion
The AI rapidly being developed is not limited to use in computing and logical analysis. To transfer the aesthetic judgment and creativity of a human to a machine, the creative
game system CreativeBioMan was introduced in this article.

100
90
80
Verisimilitude Rate (%)

was recognized utilizing the wearable-clothing expressive
robot. After obtaining a user's emotion data, the hue of
the final works could be slightly adjusted according to the
user's emotion. If the user's emotion was positive, then the
hue of the work was set at "warm"; if his or her emotion
was to the contrary, then it would be set to "cold." The
duration of each monitoring session lasted 8-15 min.
After each test, users labeled the signal of their whole
creation process to mark the style of painting desired. The
data set was used to train the imagine classification model
for different painting styles to match the painting styles in
the users' historical works. The data sets built were used
to train the users' personalized creation process.
We first preprocessed the 22 EEG signals paths for
each user. A five-order, low-pass Butterworth filter with
fz = 50 Hz was used to filter the radio-frequency component. The EEG signal was divided into frames, and the
size of a window was 256 pixels. A short-time, discrete
Fourier transform was utilized to extract the rhythm
bands of the EEG signal, thereby obtaining the energy values of d, i, a, and b as the approximate entropy, largest Lyapunov index, and the Kolmogorov entropy as the
signal features of the EEG, respectively. These values
were then input into the LSTM network for classification
and to obtain the corresponding labeling result.
We collected between five to 10 works of users in different styles and marked the painting style of each work. The
labels were divided into four classes: oil paintings, traditional Chinese paintings, sketches, and cartoons. Histogram equalization was used to obtain the luminance image
in the whole luminance range. All users' historical works
were resized to be the same size and were then uploaded
to the server.

70
60
50
40
30
20

life_like =

i=1 j=1

n#m

6 7
Tester
(a)

9 10 Average

1.4
1.2
1
0.8
0.6
Transport Delay
Model Testing Delay
Model Training Delay

0.4
0.2

| | goal

Time Delay (s)

Experimental Results and Analysis
By establishing the aforementioned platform and data
set, the CreativeBioMan system could generate paintings
using the users' style. Our work assessed the system
according to the picture effect generated by the system
and the fidelity of artworks.
To define the artwork's creation fidelity, the works generated by the computer were mixed with those created by
real artists and other painters to distinguish and select
them. If the painters could not select the works created by
the computer, this was an indication that the computer had
a similar creativity to that of real painters, i.e., the fidelity
of the creation was very high. Concretely, it can be
expressed by the following mathematical formula:
i, j

# non_machine # 100%.

(9)

The formula goal_i, j equals 0 if the ith judge finds
the work created by computer in the jth test set; otherwise, it equals 1. In addition, m means m test sets, n
means n judges, and non_machine is the proportion of

Tester
(b)
Figure 4. The experiment results: (a) verisimilitude

rate and (b) time delay of the CreativeBioMan system.

Ja nu a r y 2020

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE

IEEE Systems, Man and Cybernetics Magazine - January 2020

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - January 2020