Thankfully Samson has found something more interesting than digging up our yard…he likes to chase critters that hop and slither around our yard. In fact he was fixated and barking like mad the other day – this is what he was barking at!
The Computer that Detects Forged Art
One of the things that continue to separate human intelligence from that of machines is the ability to recognize art. (Well, sometimes it’s hard for us humans to tell too…!)
A step towards remedying that has been taken by Milan Rajkovic at the University of Belgrade and Milos Milovanovic at the Mathematical Institute of the Serbian Academy of Sciences and Arts. They have developed a “machine-vision analysis technique” that detects in the pattern of brushstrokes in a painting the underlying cognitive processes of the artist, which indicates which of a pair of identical paintings is the original, and which the copy — even if painted by the same person.
This was undertaken to solve a legendary conundrum in the art world, involving two versions of the painting La saveur des larmes (The Flavour of Tears) by the Belgian surrealist Magritte. It has long been known that Magritte himself painted both versions, likely to make two sales to two different collectors, in the painter’s financially challenged, immediately-post-WWII period. Since Magritte took especial care to make each appear to be the same painting, it has proven impossible for art experts to determine which was painted first: even the inscriptions of the back match!
Enter Rajkovic and Milovanovic. The pair hypothesize that creating an original painting is a task more cognitively complex than copying one, resulting in more layered colours and differently organized brushstrokes, which can be detected by computer. The pair tested this hypothesis by engaging Dutch painter Charlotte Caspers to execute seven artworks using different media, and then create copies of them as closely as she could. Then Rajkovic and Milovanovic unleashed their analysis.
“[It] transforms a two-dimensional image into a time-frequency representation which captures information about the painting at various scales. These scales can be thought of as looking at progressively more blurred images of the paintings.
Rajkovic and Milovanovic perform this analysis using the red, green, and blue channels of a conventional RGB image of each painting, and they repeat the analysis for patches of each painting.
Sure enough, they say a difference in complexity is clearly visible between Caspers’s originals and copies. ‘For all patches and all the paintings, the mean global complexity of an original painting is larger than the corresponding value of a copy,’ they say.”
With the success of their trial run under their belts, Rajkovic and Milovanovic turned to La saveur des larmes 1.0 and 2.0. Using their techniques, they have definitively identified the original and the copy in relation to each other, but, in a move Gizmodo is calling “either a fantastic display of academic trolling or paranoid ass-covering,” they have omitted any other real-world details (home gallery, for example) that would indicate which is actually which.
Rajkovic and Milovanovic are currently being challenged to fully identify the paintings. While that means that someday this tantalizing mystery will sadly be solved, it means the next step for machine intelligence (and the forgery-detecting business!) may be in the horizon.