Robotics Reference
In-Depth Information
police guard, he would paint another “Vermeer”. He was duly provided
with the necessary paints and brushes, and he proved his case well before
the new painting was completed.
There is an ongoing debate within the art world on the authenticity of
van Gogh's
Sunflowers
, bought for just under $40 million by the Japanese
insurance company Yasuda Fire and Marine Insurance at Christie's in
London in 1987. Van Gogh created several paintings of sunflowers and
some art experts have challenged the authenticity of the Yasuda version
as a genuine van Gogh, although it was generally accepted as an original
van Gogh for several decades. The current view is that the painting is,
most likely, genuine, but the lack of absolute certainty over a painting
of this value is ample justification for employing Artificial Intelligence
techniques to identify the painters of works of art.
In 1998 a team at the University of Bremen trained a computer to
identify the drawings of the nineteenth-century French artist Delacroix,
which it managed to do with 87 percent accuracy. Instead of employing
techniques that had previously been used on drawings, such as analyzing
the length, thickness and curvature of lines, the computer scanned the
images of the drawings and simply designated each pixel as either black
or white. The system then analyzed the ratio of black-to-white within
each of the drawings, and across the entire set of drawings. This method
was cheap and simple, and the ratio proved to be a sufficient measure to
achieve an accuracy of 87 percent in identifying Delacroix drawings. But
while this is relatively high for a new technology, it is not good enough
for someone thinking of buying a work of art for hundreds of thousands
of dollars or more.
Since 2003 a project called Authentic has been under way at the Uni-
versity of Maastricht with the support of the Van Gogh Museum in Am-
sterdam. One of the system's first crude tests was to discriminate be-
tween paintings by van Gogh, Cezanne, and Gauguin, which it achieved
with 95 percent accuracy. The Authentic system detects certain elemen-
tary features in paintings and employs neural networks to identify their
painters. The choice of the features employed by these neural networks
was guided by discussions with an art expert, Frank Boom, and by some
of the extensive literature on art. The main general characteristics known
to be indicative of an artist's style are colour, brushwork, and texture.
21
21
Because a painter's brushwork yields a specific texture, brushwork and texture are treated within
the system as one characteristic.
