Information Technology Reference
In-Depth Information
both simulate and detect human emotions. For this reason, we chose to implement
some simple but foundational emotional modelling in The Painting Fool.
We first asked the question of whether we can train the software to paint in differ-
ent styles, so that it can choose a particular style in order to heighten the emotional
content of a painting. Note that this corresponds with part four of the meta-mountain
described previously, i.e. choosing styles in a meaningful way. We worked on por-
traits of the actress Audrey Tatou as she portrayed Amélie Poulain in the film
Le
Fabuleux Destin d'Amélie Poulain
. This source material seemed appropriate, as the
film is largely about the emotional rollercoaster that Amélie finds herself on, and
the actress portrays a full range of emotions during the film. Working with 22 stills
from the film, we first annotated the images to specify where the facial features
were, and then we repeatedly suggested painting styles to The Painting Fool. The
descriptions of styles specified the level of abstraction to obtain through the image
segmenting; the colour palette to map the regions of colour to; the natural media
to simulate while filling/outlining the regions and the brush stroke style to employ
while doing so. Largely through trial and error, we derived many of the styles by
hand, by experimenting until the pictures produced were subjectively interesting.
In addition to these hand-derived styles, we also enabled the software to randomly
generate painting styles. Each time a style—whether randomly generated or derived
by us—subjectively heightened the emotion portrayed by the actress in one of the
stills, we recorded this fact. In this way, we built up a knowledge base of around
100 mappings of painting styles to emotions, with roughly half the styles provided
by us, and the other half randomly generated (but evaluated by us).
Naturally, this tagging exercise was a very subjective endeavour, as all the emo-
tional assessment was undertaken by us. Therefore, in order to gain some feed-
back about the knowledge base, we built an online gallery of 222 portraits produced
from the 22 stills. The gallery is called Amélie's Progress and can be viewed at
www.thepaintingfool.com
.
The portraits are arranged from left to right to portray
emotions ranging from melancholy on the left to mild euphoria on the right. Some-
times, the emotion portrayed is largely due to the actress, but at other times, the
painting style has heightened the emotional content of the piece. Hence, on a num-
ber of occasions, we find the same still image painted in different ways on both the
left and the right hand sides of the gallery. An image of the entire gallery and some
individual portraits are presented in Fig.
1.5
.
The Amélie's Progress project raises some issues. In particular, we decided to
make the web site for The Painting Fool read as if The Painting Fool is a painter
discussing their work. This has been mildly divisive, with some people expressing
annoyance at the deceit, and others pointing out—as we believe—that if the software
is to be taken seriously as an artist in its own right, it cannot be portrayed merely
as a tool which we have used to produce pictures. In addition, we chose to enable
people working with The Painting Fool to see it paint its pictures stroke by stroke,
and we put videos of the construction of 24 of the Amélie portraits onto a video
wall, as part of the online gallery. This construction involves the sequential placing
of thousands of paint strokes on a canvas. In another area of the web site, there are
live demonstrations of paintings being constructed (as a Java Applet, rather than as
