Still in its infancy, the ‘sparse coding’ system, as it is known, breaks down works of art into tiny, digitalised pieces to create a visual library of an artist’s work, as well as their style and technique. The larger the body of work available, the more consistent patterns that appear across an artist’s output can be identified.
The system, which is an extension of research into vision and neural networks, effectively reduces analysis of artworks to mathematical calculations, which can be assessed objectively.
Importantly, similar mathematical patterns between works identify them as being created by the same hand, and the research scientists are confident that fakes will not be able to recreate such similarities.
The system, unveiled early in January by a team led by Daniel Rockmore of Dartmouth College in the US, remains limited.
To be really effective, it needs to draw on a fair-sized body of consistent work with similar subject matter. They have also found that, for the time being, it works best with landscapes, but expect its effectiveness to spread.
The research team have said that they see the system as a useful extension of existing methods for verifying works art rather than as a replacement.
It is not clear how similar the system is to the one announced in May 2007 by Richard Johnson of Cornell University.
He said that he had developed a computer program that could identify fake paintings by developing a “visual signature” for artists.
He said he had already created a signature for Van Gogh’s style by analysing a database of 101 paintings by the artist and his known imitators. The signature was then distilled into numbers and used as a benchmark by which to judge other works.
By Ivan Macquisten
