F for fake

An international team of researchers are developing a digital system to help tell original works of art from counterfeits.


As museums continue to digitise their art collections, it becomes increasingly easier for paintings to be copied by forgers.


But not if James Wang, associate professor of Information Sciences and Technology (IST), and Jia Li, associate professor of statistics at Penn State have anything to do with it.


The two researchers are part of an international team working on developing a digital system to help tell original works from counterfeits.


The system they have developed works by first digitising known authentic masterpieces. It then analyses them in some detail, creating a statistic model of the artist’s technique. Then, any other work believed to be the artist in question can be compared with the computer model to verify if it is unique or forged.


To test out the validity of their system, the researchers conducted 101 high-resolution grayscale scans of van Gogh paintings provided by the Van Gogh and Kröller-Müller Museums in the Netherlands.


Wang and Li then broke each scan down into sections measuring 512 pixels by 512 pixels, or about 2.5 inches by 2.5 inches in canvas size, and analysed them based on patterns and geometric characteristics of the brush strokes.


From the 101 scans they received from the museums, art historians identified 23 as unquestionably authentic van Gogh works. These were used by their computer system as a training database for van Gogh’s brushstroke styles.


Statistical models were created to capture the unique style, or ‘handwriting,’ that became the artist’s signature in 23 of the scans.


The other 78 – either works of van Gogh, works of van Gogh’s peers or paintings that had at one time been attributed to him but later found to be unauthentic – were compared against the generated models to test the algorithms.


The researchers system now allows any of the artist’s painting to be compared against existing data to help determine its authenticity.


For more information on the digital painting analysis project, please visit: digitalpaintinganalysis.org.