Recognizing images from two-dimensional images is nothing new – fire up Google yourself, and it’s quite easy to search by image and find information about that image and many related images. But in the art world, such tools are often insufficient, especially when it comes to attributing a painting of unknown (or unconfirmed) origin. Now, researchers are taking three-dimensional paint recognition to bridge that gap, using the topography of paint application to develop a textural signature that can be used to identify the artist of a painting.
“Many notable artists, including El Greco, Rembrandt and Peter Paul Rubens, employed studios of varying sizes and structures to meet market demands for their art,” explain the authors in their paper published in Heritage Science. “In the case of the workshops, the different artists try to create a complete painting with a singular style, defying the methods [used to attribute paintings to their painters]. Moreover, the challenges of such attributions create conflicts when the attribution is closely tied to the apparent value of objects in the art market. Therefore, there is a need for unbiased and quantitative methods to provide insight into disputed attributions of studio paintings.
The researchers recruited a team of nine painting students from the Cleveland Institute of Art, tasking them each with creating triplicate paintings of a photograph of a water lily. Next, a team of art historians and a paintings conservator selected the four most stylistically similar artists. Surface height information from the paintings of these four artists was then captured at a spatial resolution of 50 microns, approximately 400 times thinner than a penny, and sufficient to capture the characteristics of fine brushstrokes that are summed up often at differences of hundreds of microns.
This high-resolution topography – captured over 12cm by 15cm areas on each board – was then divided into one-centimeter-square patches, allowing each board to produce 180 patches. An ensemble convolutional neural network model was then trained with most of those hundreds of patches, learning to assign the other patches based solely on stylistic differences in how the artists applied the paint.
The data preparation and analysis workflow. Image reproduced with the kind permission of the authors.
The researchers found this method to be between 60 and 90 percent accurate, and more than twice as accurate as models using image recognition under certain conditions. “Remarkably, scales of short length, even as small as a diameter of silk, were the key to reliably distinguishing performers,” the authors concluded. “These results are promising for real-world attribution, especially in the case of workshop practice.”
To learn more about this research, read the article here.
