Information Technology Reference
In-Depth Information
Fig. 1.6 Example portraits using styles to heighten (L to R) sadness; happiness; disgust; anger;
fear and surprise
The Painting Fool's pieces is increased after people see the videos of it at work. Of
course, this approach was pioneered by Harold Cohen, as it has always been possible
to view AARON at work, and AARON was an inspiration for us in this respect.
Moreover, in discussions with Palle Dahlstedt about how software can better frame
and promote their own work, he suggested that the artefacts produced by music and
visual art systems should contain at least a trace of the construction process (see
Chap. 8). In simulating paint strokes and showing construction videos, we achieve
this with The Painting Fool.
One criticism of most image manipulation software is that it has no appreciation
of the images it is manipulating. Hence a Photoshop filter will apply the same tech-
niques to an image of kitten as it would to an image of a skyscraper, which clearly
has room for improvement. To address this, and following on from the Amélie
project, we addressed the question of whether The Painting Fool can detect emotion
in the people it is painting and use this information to produce more appropriate
portraits. Detecting emotion in images and videos is a well researched area, and we
worked with Maja Pantic and Michel Valstar in order to use their emotion detection
software (Valstar and Pantic 2006 ), in conjunction with The Painting Fool. The com-
bined system worked as follows: starting with the sitter for a portrait, we asked them
to express one of six emotions, namely happiness, sadness, fear, surprise, anger or
disgust, which was captured in a video of roughly 10 seconds duration. The emo-
tion detection software then identified three things: (i) the apex image, i.e. the still
image in the video where the emotion was most expressed, (ii) the locations of the
facial features in the apex image, and (iii) the emotion expressed by the sitter—with
around 80 % accuracy, achieved through methods described by Valstar and Pantic
( 2006 ). It was a fairly simple matter to enable The Painting Fool to use this in-
formation to choose a painting style from its database of mappings from styles to
emotions and then paint the apex images, using more detailed strokes on the facial
features to produce an acceptable likeness. We found subjectively that the styles for
surprise, disgust, sadness and happiness worked fairly well in terms of heighten-
ing the emotional content of the portraits, but that the styles for anger and fear did
not work particularly well, and better styles for these emotions need to be found.
Sample results for portraits in the six styles are given in Fig. 1.6 .
The combined system was entered for the British Computer Society's annual
Machine Intelligence Competition in 2007, where software has to be demonstrated
during a 15 minute slot. The audience voted for the Emotionally Aware Painting
Fool as demonstrating the biggest advancement towards machine intelligence, and
we won the competition. More importantly for The Painting Fool project, we can
Search WWH ::




Custom Search