You’ve heard about counterfeit paintings, you’ve heard about digital art and you probably have some idea about how experts authenticate paintings. What if a computer could learn everything a Rembrandt expert knows – and much more – but would not use it to identify the author of newly discovered paintings? Instead, imagine that data gathered by the machine could be used to create a painting that both looks like a Rembrandt and… does not look like any existing Rembrandt?
In a project called The Next Rembrandt, an extensive database of Rembrandt paintings, made of high definition pictures and 3D scans, was poured into a computer so it could analyse it and use colours, shapes and brushstrokes to create an artwork that would have all the characteristics an original Rembrandt has.
As a second step, it was decided that only data from portraits of middle-aged Caucasian males in dark clothing with a white collar, wearing a hat and facing to the right would be kept. Of course, this choice was not made randomly, but it representative of what Rembrandt painted the most is terms of gender, head direction, age and clothing.
Thanks to statistical analysis and algorithms, the tech teams then extracted “the features that make Rembrandt Rembrandt”, by comparing eyes, mouths, ears, face proportions from all the paintings to figure out what would make a fake Rembrandt come as close as possible to the real deal.
Finally, texture what added through a height map to recreate brushstrokes and surface of a Rembrandt painting.
And that is how, thanks to a lot of data, an amazing team and a sophisticated 3D printer, The Next Rembrandt Painting was created 347 years after the master’s death.
It is believable? Definitely. Would it fool an expert? At least for a second. Would Rembrandt have loved it? Probably, considering that the project is all about understanding him and his work and trying to create something new out of it.
And it isn’t just a new painting, it also is a new use of data, a further step into the understanding of a body of work by going beyond the appreciation of pattern to actually reproduce them.