Information Technology Reference
In-Depth Information
We posit that the ability to recognise at least some of these aesthetic properties
is common to all humans, acknowledging that the way different humans may react
to different aesthetic principles, to their relationships, and value aesthetic principles
may vary. Likewise, the degree of awareness to principles of aesthetical order and
the inclination to use aesthetic criteria when valuing artefacts also differs.
In Machado et al. (
2003
) we find the following definition: Artificial Art Critics
are “systems that are capable to see/listen to an artwork and perform some sort of
evaluation of the perceived piece”. Unfortunately, the term “art critic” can be easily
misunderstood, given that it may be perceived as the equivalent of a human mak-
ing an artistic critique or a written analysis of an artwork, rather than an aesthetic
judgement. For this reason, we abandon this nomenclature.
Taking all of the above into consideration, for the scope of this chapter, we de-
fine an aesthetic judgement system (AJS) as a system that performs an aesthetic
assessment of an image based on its aesthetic properties. For instance, a system
that: measures the degree of accordance of an artwork with a given aesthetic theory;
measures several aesthetic properties of an image; makes an assessment of an art-
work according to the aesthetic preferences of a given user, set of users, community,
etc.; identifies the aesthetic current of an artwork; assesses the aesthetic consistency
of a set of works; etc.
It is important to note that the system should make its judgement based on
aesthetic properties. A system that assesses the aesthetic value of an artwork by
analysing its aesthetic properties can be considered an AJS. A system that performs
the same task by using optical character recognition to identify the signed name of
the author and determines aesthetic value by the popularity of the author cannot be
considered an AJS.
An AJS may provide a quantitative judgement, e.g. a single numeric value, a
vector, or a classification in one or more dimensions. An AJS may also provide a
qualitative assessment or assessments. Ultimately, the adequacy of the output de-
pends on the task at hand. For instance, to guide an evolutionary algorithm using
roulette wheel selection, a quantitative judgement, or one that can be converted to
quantities, is required. However, to guide the same algorithm using tournament se-
lection, only a qualitative assessment is needed, i.e. knowing if a given individual is
better suited to the task at hand than another, we do not need to quantify how much
better it is.
The AJSs can be divided into two categories. The first category explores systems
that rely on a theory of visual aesthetics and use an AJS to explore this theory by
computing it, e.g. Rigau et al. (
2008
), Staudek (
2002
;
2003
), Taylor et al. (
1999
),
Machado and Cardoso (
1998
), Spehar et al. (
2003
), Schmidhuber (
1997
;
1998
;
2007
), see also the chapters by Galanter (Chap. 10) and Schmidhuber (Chap. 12)
in this volume.
The second category presents learning systems which include some kind of adap-
tive capacity that potentially allows them to learn user preferences, trends, aesthetic
theories, etc. Although there are different approaches, usually these systems extract
information from images (e.g. a set of metrics) using a machine learning system that
performs an aesthetics-based evaluation or classification. There are numerous exam-
ples of this architecture in the fields of content based image retrieval and computer
