Information Technology Reference
In-Depth Information
11.3 The Evolution of an AJS
This section describes the evolution of an AJS over the course of the past decade. It
started as a heuristic based system, it was tested using the DJT, and it subsequently
became part of an evolutionary art tool. Prompted by the results obtained, an AJS
with learning abilities was developed and tested in a wide variety of experiments,
which are also described briefly.
11.3.1 A Heuristic AJS
Machado and Cardoso (
1998
) took inspiration from the works of Arnheim (
1956
;
1966
;
1969
), as well as from the research indicating a preference for simple repre-
sentations of the world, and a trend to perceive it in terms of regular, symmetric and
constant shapes (Wertheimer
1939
, Arnheim
1966
, Tyler
2002
, Field et al.
2000
).
They explored the working hypothesis that the aesthetic value was linked with the
sensorial and intellectual pleasure experienced when finding a compact percept (i.e.
internal representation) of a complex visual stimulus (cf. Chap. 12). The identifi-
cation of symmetry, repetition, rhythm, balance, etc. can be a way of reducing the
complexity of the percept, which would explain the universal nature of these aes-
thetic principles and the ability of the brain to recognise them “effortlessly”.
The approach rewards images that are simultaneously visually complex and easy
to perceive, employing estimates for the
Complexity of the Percept
(CP) and for the
Complexity of the Visual Stimulus
(CV). An estimate for CV should assess the pre-
dictability of the image pixels. JPEG image compression mainly affects the high fre-
quencies, which can normally be discarded without significant loss in image quality.
The amount, and quality (i.e. the error involved) of the compression achieved by this
method depends on the predictability of the pixels in the image being compressed.
Unlike JPEG compression, which only takes into account local information, fractal
image compression can take advantage of the self-similarities present in the im-
age. Machado and Cardoso (
1998
) assume that JPEG compression is less like the
way humans perceive images than fractal image compression, and hence use fractal
compression as a rough estimate of the CP. CP and CV are estimated through the di-
vision of the root mean square error by the compression ratio resulting, respectively,
from the fractal (quadratic tree based) and JPEG encoding of the image.
A time component is also considered (Machado and Cardoso
1998
;
2002
). As
time elapses, there is a variation in the detail level of image perception. Therefore,
it is necessary to estimate CP for specific points in time, in this case
t
0
and
t
1
,
which is achieved by carrying out a fractal image compression with increasing detail
levels. The proposed approach values images where CP is stable for different detail
levels. The idea being that as time goes by one should be able to acquire additional
information about the image, for example: the increase in size of the percept should
be balanced out by the increase in its level of detail. It is important to notice that
Machado and Cardoso neither suggested that the employed JPEG complexity was
