Stakeholder Groups in Computational Creativity Research and Practice - Computational Creativity Research: Towards Creative Machines - page 15

1.4.4 A Case Study in Automated Portraiture

As part of an exhibition with The Painting Fool 2 system [ 43 ] in 2013, we enabled
the software to produce portraits for people live in a gallery, as described in [ 19 ].
Managing the expectations and perceptions of the observers was a key aspect of this
project. To this end, we hung posters describing the behaviour of the software as
exhibiting aspects of intentionality, imagination, skill, appreciation, reflection and
learning (six of the essential behaviours described above). Moreover, the software's
actions and output were tailored to support the perception of these behaviours and
an impression of creativity in the software by observers present in the exhibition,
especially those sitting for a portrait.
Portraits were painted with people sitting in front of a laptop. It was immediately
made clear that (i) the software was modelling a 'mood' to direct its painting, and
(ii) the sitter was very much a tool for the software, not the other way around.
This was achieved by opening remarks from the software such as: “Thank you for
being my model. I'm in a negative mood right now, so I would like you to express
a sad emotion.” This was followed by The Painting Fool explicitly directing the
sitter, while video recording them. A still image was then extracted where the sitter
was expressing an emotion. Machine vision techniques were applied to remove the
background, into which was substituted one of 1,000 abstract art images, to which
one of 1,000 image filters was applied. The filter was chosen to increase the chances
that the resulting image might reflect a changing simulated mood gained through
reading newspaper articles, as described in [ 19 ]. The same filter was applied to the
face of the sitter placed in the foreground, producing in a few seconds an image
conception, or sketch for the portrait, such as the first image of Fig. 1.2 .
Following this, a canvas appeared on screen, and a hand holding either a pen-
cil, paint brush or pastel stick made virtual marks on the canvas leading to a non-
photorealistic rendering of the background and foreground of the portrait, taking
between 2 and 10 minutes, depending on the style. An example portrait is given
at the bottom of Fig. 1.2 , which was printed and given to the sitter, along with the
commentary (the whole of Fig. 1.2 ). The most important aspect of the commentary
is the expression of intention, by first showing a conception of the type of portrait the
software aimed to produced, then showing what it produced and finally analysing
and criticising—using machine vision techniques described in [ 44 ]—its results with
respect to its aims.
The purpose of the exhibition was cultural, not scientific, and no experimentation
was undertaken. From our experience, however, we contend that the behaviours
exhibited by the software and explained in poster form enabled people to be surprised
by the resulting portrait (and many of the 100 or so sitters in the exhibition were very
surprised), while still projecting creativity onto the software. This upheld the aim
of the You Can't Know my Mind exhibition: as it used some intelligence, and could
explain its actions, it was somewhat appropriate to employ the word 'mind' with
reference to The Painting Fool. However, as the process was unpredictable due to

2 Online presence: www.thepaintingfool.com .