Experts from the United States and Thailand have developed a method of reliable reproduction of the world’s masterpieces of painting using 3D printing and artificial intelligence. Thanks to this combination, it is possible to reproduce the color palette of the image as close as possible to the original. The method is based on the printing of many layers of different color inks, so you can get as close to the original color for each piece of the picture. The technology is going to be presented at the SIGGRAPH Asia 2018 conference.
In the world practice is often used for the reproduction of masterpieces of world art. This is done not for the sake of some machinations, but in order to get acquainted with the works of more people. In addition, in this way the owners of works of art can protect the originals from destruction.
Typically, reproductions are created using high-precision scanners and printers. However, the possibilities of this technique are very limited and do not allow to convey the beauty of the original in the smallest details. This is due to several shortcomings. For example, one of them is that commonly used printers use a combination of four colors to convey the color of the original, which reduces color accuracy. In addition, printers usually produce colorimetric rather than spectral reproduction of the original color, which is why the printed picture is close to the original only at a certain reference lighting.
To get a more accurate copy allows the development of researchers from the Massachusetts Institute of technology — RePaint system, in which artificial intelligence controls the 3D printer and allows you to transfer the original colors regardless of lighting.
Creating a reproduction using this method occurs in several stages. First, a high-quality scan of the original is performed. After that, the system calculates the 3D printing parameters. The last step is, in fact, the very print copy on a 3D printer. To obtain the most accurate color reproduction, engineers use a special method of shooting. Placed on the substrate of the original picture takes a multispectral camera. When scanning, the liquid crystal filter in front of the camera changes its ten-nanometer bandwidth from 420 nanometers to 720 nanometers. At the same time, the camera takes monochrome images, and then combines them into a single image, in which each pixel corresponds to 31 spectral values.
To recreate the copy, a 3D printer is used, capable of printing a variety of different translucent inks, superimposed on each other layer by layer. For a smooth transition between colors, the classic method of creating a halftone image is used. The whole process is controlled by two neural networks, one of which predicts the spectrum of the array of layers of different materials. This neural network was trained by engineers on the basis of a printed plate with a set of squares the size of a millimeter consisting of different combinations of layers.
With the help of the first neural network taught the second. It is used for the inverse problem — predicts the optimal arrangement of layers from different materials for the image provided to it.
Unfortunately, at the moment the technology has a limitation. The system is capable of producing a reproduction no larger than a postcard. However, if the technology can be scaled, the museums will have another opportunity to save priceless originals, showing visitors the exact copies.