Information Technology Reference
In-Depth Information
explosions and fields of graves with the kinds of neutral words that were extracted
from the newspaper article, such as “Afghanistan”, “troops” and “British”.
We also used the collage generation project to raise the issue of playfulness in
the software, as the collages would often contain strange additions, such as an image
of Frank Lloyd-Wright's
Falling Water building
being included in a collage arising
from a news story about the England cricket team (see Krzeczkowska et al. (
2010
)
for an explanation of how this happened). We don't claim that the word “playful”
should be used to describe the software, but it does show potential for this kind of
behaviour.
1.4.5 Paint Dances
In many respects, it is an odd choice to build an automated artist that simulates tradi-
tional media such as paints and brushes. It would seem to be missing an opportunity
for the software to invent its own medium of expression and exploit that. In future,
we would have no problem with the software performing such medium invention,
and we would see that as a very creative act. However, we argue that while the soft-
ware is in its early stages it is more important for its behaviour to be understood
in traditional artistic terms, so that its creativity can be more easily appreciated. In
particular, as described in Sect.
1.3.6
, we want the software to produce paintings
that look like they could have been physically produced by a human, but simultane-
ously look like they would not have been painted by a person because they are so
innovative in technique and in substance.
Notwithstanding this notion, we wanted to follow an opportunity for the soft-
ware to work in a new medium, albeit one which is not far removed from painting,
namely,
paint dances
. We define a paint dance as an animation of paint, pencil, or
pastel strokes moving around a canvas in such a way that the strokes occasionally
come together to produce recognisable subject material. We worked with portraits,
in particular our subject material was images of the attendees of the 2009 Dagstuhl
seminar on Computational Creativity. The technical difficulties involved are dis-
cussedinColton(
2010
). In summary, to achieve the paint dances, we first imple-
mented a way to tell which pairs of strokes from two different paintings were most
closely matched. Following this, from the 60,000 pencil strokes used in the 32 por-
traits of the Dagstuhl attendees, we used a K-means clustering method to extract just
enough generic strokes to paint each picture in such a way that the fidelity would re-
main high enough for a likeness to be maintained. The final technical hurdle was to
write software to perform the animations by moving and rotating the paint strokes in
such a way as to come together at the right time to achieve a portrait. We have so far
completed two paint dances: “meeting of minds”, where pairs of pencil portraits are
shown together, with the strokes meeting in the centre of the picture as they move to
form two new portraits, and “eye to eye”, where each painted portrait is formed in-
dividually, with spare paint strokes orbiting the scene. The images in Fig.
1.12
show
the stills from a transition in both pieces. These videos formed part of an group
