Computers enlisted in the detection of art copies and forgeries
Museums around the world had begun to digitize their collections to aid in conservation and research， but the notion of crunching* those reams of data was in its infancy.
So Johnson approached the Van Gogh Museum and offered to organize a conference. In exchange for the use of high-resolution scans from dozens of paintings， the three university teams would present their research at the event in Amsterdam.
Each team got 101 images， including 82 that had always been identified as van Goghs， six non-van Goshs that had a similar style and 13 for which the attributions* had been questioned at some point. The technique involves the use of "wavelets"-mathematical templates that identify characteristic patterns in the painting at a range of scales， from coarse to very fine. Van Gogh's style changed over the years， so Wang and Li used a rang of 23 representative paintings to "train" their computer program in what to look for.
Tiny， telltale characteristics
One finding was that when an artist had tried to copy van Gogh's style， the painting displayed telltale characteristics at a very small scale. It wasn't something you could see with the naked eye， says Princeton's Shannon M. Hughes， a Ph.D. students in electrical engineering.
All three teams did better than average at picking the real thing. Using several variations of its approach， for example， Princeton correctly classified as many as 55 out of 65 van Goshs.
The research lends itself to more than just telling apart real van Goghs from others. The teams are now pursuing additional challenges， such as telling when certain works were parinted.
It's all still in the rough stages， but as long as museums are amenable，* Johnsonand his colleagures vow that they will continue.
Museums around the world had begun to digitize their collections to aid in conservation and research， but the notion of crunching* those reams of data was in its infancy.
So Johnson approached the Van Gogh Museum and offered to organize a conference. In exchange for the use of high-resolution scans from dozens of paintings， the three university teams would present their research at the event in Amsterdam.
Each team got 101 images， including 82 that had always been identified as van Goghs， six non-van Goshs that had a similar style and 13 for which the attributions* had been questioned at some point. The technique involves the use of "wavelets"-mathematical templates that identify characteristic patterns in the painting at a range of scales， from coarse to very fine. Van Gogh's style changed over the years， so Wang and Li used a rang of 23 representative paintings to "train" their computer program in what to look for.
Tiny， telltale characteristics
One finding was that when an artist had tried to copy van Gogh's style， the painting displayed telltale characteristics at a very small scale. It wasn't something you could see with the naked eye， says Princeton's Shannon M. Hughes， a Ph.D. students in electrical engineering.
All three teams did better than average at picking the real thing. Using several variations of its approach， for example， Princeton correctly classified as many as 55 out of 65 van Goshs.
The research lends itself to more than just telling apart real van Goghs from others. The teams are now pursuing additional challenges， such as telling when certain works were parinted.
It's all still in the rough stages， but as long as museums are amenable，* Johnsonand his colleagures vow that they will continue.