ABOUT US
The idea of developing an algorithm that recognizes counterfeit works of art came to Wolfgang Reuter in autumn 2016 during a conversation with Robert Ketterer, the owner of the art auction house of the same name in Munich. The auctioneer told Reuter, who was then deputy editor-in-chief of the news magazine Focus, about the typical forger's approach. "He doesn't come with a picture and say: 'I have a Nolde, Pechstein or some other well-known artist here,'" says Ketterer, "but he submits a picture and says: 'I found that in the belongings of my deceased mother. Is that worth anything?'”
Most counterfeits are not copies of actual originals, the auctioneer continues, but rather works in which the counterfeiter imitated the style of the artist in question, sometimes very well. It is therefore sometimes very difficult to prove a forgery.
Reuter, who had been programming and developing various apps – also in the field of image processing – for four years at the time, then asked: "Is there nothing digital yet that helps when examining paintings?” Ketterer replied that he didn’t believe so – and Reuter decided to try the use case.
Initial tests with 100 pictures each by four painters based on statistical methods such as Bayesian Bayesian inference were promising, but Reuter quickly realized that significantly better results would probably be possible with machine learning. The journalist, who had already studied physics for two semesters during his time at “Spiegel” in 2007 and 2008, taught himself the necessary knowledge as a student at the online university Coursera – in particular with the courses and specializations of Andrew Ng .
https://www.coursera.org/learn/machine-learning
https://www.coursera.org/specializations/deep-learning
With the help of so-called convolutional neural networks, Reuters application – meanwhile with 20 painters – achieved significantly better results. The journalist then presented the project to René Allonge , head of the art crime department at LKA Berlin. Between 2017 and 2020, the LKA supported Reuter, in particular with secured counterfeits, in order to test and further improve the algorithms.
"I enjoyed the whole thing so much," says Reuter, "that in 2017 I decided to change industries and work full-time with artificial intelligence." In the spring of 2018, Reuter first switched to the appliedAI initiative from UnternehmerTUM , the start-up up incubator by Susanne Klatten . He also worked as a freelance AI developer and further developed the painter and forgery detection model. In the spring and summer of 2019, the project was exhibited as a playful application in the Museum Buchheim ("Museum of Fantasy") in Bernried. Back then, visitors could compete against the algorithm. The respective visitor in the museum and the AI model were fed randomly selected images from the test set of the data set. The algorithm achieved a recognition accuracy of 93 percent (with 20 painters), the visitors averaged 60 percent.
In November 2019, Reuter joined Alexander Thamm GmbH as Leading Data Scientist. During this time, he continued to develop the algorithm – with astonishing results. In the meantime, the model, trained with 53 artists, assigned more than 90 percent of the pictures to the right painter. In spring 2020, Reuter founded Art Intelligence GmbH, which has so far carried out several dozen analyzes for customers.
Wolfgang Reuter, founder and managing director of Art Intelligence GmbH.
Playful application: Computer terminal exhibited in the Buchheim Museum, with which visitors could compete against the AI model.
Reuter, who had already been programming for four years and had developed various apps – including in the field of image processing – then asked: “Isn’t there anything digital that helps with the examination of paintings?” Ketterer replied that he was unaware of anything like that – and Reuter decided to try out the application.
Initial experiments with 100 paintings each from four artists, based on statistical methods such as Bayesian inference, were promising, but Reuter quickly realized that significantly better results would likely be possible with machine learning. The journalist, who had already studied physics for two semesters alongside his work at "Der Spiegel" in 2007 and 2008, acquired the necessary skills as a student at the online university Coursera – particularly through the courses and specializations offered by Andrew Ng .
https://www.coursera.org/learn/machine-learning
https://www.coursera.org/specializations/deep-learning
Using so-called convolutional neural networks, Reuter's application – now involving 20 painters – achieved significantly better results. The journalist then presented the project to René Allonge , head of the art crimes department at the Berlin State Criminal Police Office (LKA). Between 2017 and 2020, the LKA supported Reuter, particularly with verified forgeries, to test and further improve the algorithms.
“I enjoyed the whole thing so much,” says Reuter, “that in 2017 I decided to change industries and work full-time in artificial intelligence.” In the spring of 2018, Reuter initially joined the appliedAI initiative of UnternehmerTUM , Susanne Klatten ’s startup incubator. Alongside this, he worked freelance as an AI developer, further developing the painter and forgery detection model. In the spring and summer of 2019, the project was exhibited as a playful application at the Buchheim Museum (“Museum of Imagination”) in Bernried. Visitors could compete against the algorithm. Both the respective player in the museum and the AI model were randomly presented with selected images from the test set of the dataset. The algorithm achieved a recognition accuracy of 93 percent (with 20 painters), while the visitors averaged 60 percent.
In November 2019, Reuter joined Alexander Thamm GmbH as Lead Data Scientist. During this time, he further developed the algorithm – with remarkable results. Now, the model, trained with 53 artists, correctly attributes over 90 percent of images to the correct painter. In spring 2020, Reuter founded Art Intelligence GmbH, which has since conducted several dozen analyses for clients.