The Painting Fool: Stories from Building an Automated Painter - Computers and Creativity - page 28

Fig. 1.10 Pictures from the “PresidENTS”, “Fish Fingers” and “After AARON” galleries in The
Painting Fool's “Ever So Slightly...” exhibition
their content, and we won't spoil the fun of working out the wordplay here
(see www.thepaintingfool.com ) . The fourth one reflects the influence that Cohen's
AARON system has had on the scene generation aspects of The Painting Fool—as
we see in the images shown, the pictures strongly reference the contents of the pic-
tures produced by AARON, although we did not try to capture its visual style. Note
that the human figures were produced by context free design grammars, and the ab-
stract images on the walls of the room were similarly produced. The gradient effect
of the ceiling and floor used a constraints approach to the generation of rectangles
which were filled in using simulated pencils.
1.4.4 Collage Generation
The pictures produced in the “Ever So Slightly...”seriesrepresentastepintheright
direction towards imaginative behaviour. However, looking at the meta-mountain
we have described for The Painting Fool, the software needs to construct scenes
for a purpose. Moreover, while the paintings in the series may be amusing and
mildly thought-provoking as simple word/art puzzles, they are certainly not the most
provocative of works. One aspect of the human painting process that is rarely simu-
lated in computer art programatically is the ability to construct paintings to convey
a particular message within a cultural context. We looked at using text and image
resources from the internet as source materials for the production of artwork that
might have a cultural impact (Krzeczkowska 2009 ). In particular, the software began
by downloading headline news stories from the websites of the Guardian newspaper
and other news sources. Using text extraction software (El-Hage 2009 ), based on
the TextRank algorithm (Mihalcea and Tarau 2004 ), the most important nouns were
extracted from the text of the news story. These were then used as keywords for
searches in image repositories, including Google images and Flickr. The resulting
images were juxtaposed in a collage which was turned into a segmentation, and the
non-photorealistic rendering software from The Painting Fool was used to produce a
painterly rendition of the subject material. With the exception of the text extraction
software, the process was largely devoid of AI techniques, and this is something we
plan to work on. However, the results were often remarkably salient. As an example,
one morning, the software downloaded the lead story from the Guardian, which was