Written by Seoyoon Chang
Drawings, constructed with different lines ranging in various sizes and thickness, can be a difficult work to analyze. While we have art historians determining the value of a painting or detecting art forgeries, this can be a problem. For example, when Leonardo da Vinci's painting Salvator Mundi sold for 450 million dollars, suspicions arose which concerned that the painting was a fake. In addition, we have to perform a complicated process of tests such as radiometric dating, infrared spectroscopy, and other series of tests to analyze an artwork; this can waste time and costs. On the other hand, AI can analyze an artwork without doing any of these tests. AI, a newer and a more convenient method towards detecting art forgeries, not only can classify strokes but also check the authenticity of a painting. In a paper published in 2017 which researchers analyze strokes in line drawings by various artists, it is stated that they analyzed the "characteristics of individual strokes in drawings and comparing these characteristics to a large number of strokes by different artists using statistical inference and machine learning techniques". This process, which is said to be inspired by Maurits Michel van Dantzig, an art historian who suggested that the strokes of an artist should be compared to those of a forger, relies on uniqueness. The researchers designed an algorithm which experimented with a large dataset with over "70 thousand strokes". Furthermore, an RNN (recurrent neural network) system "learned" what features in the strokes were significant to identify the artist through tone variations and forms of each stroke. With these features in placement, the system was able to classify strokes with accuracy in between 70 to 90 percent, and achieved an accuracy of 100 percent for detecting fakes "in most settings". Art historians, who detect art forgeries usually based on the surface of a painting such as characteristics which are not common in an artist's work, is often "a gut feeling", according to Milko Den Leeuw, a painting conservator. As a result of the experiment conducted by the researchers from the same paper, it can be determined that AI can distinguish strokes between a painting of a real artist's and a forger's with high accuracy. Despite this, limitations are also present in the system. While the AI system can accurately identify art forgeries in line drawings just by looking at one stroke, paintings can be a big challenge. With thousands of strokes of different structures, shapes, and even intended mistakes, this can be tough for an AI system to handle. Another challenge for detecting art forgeries can be the condition of an artwork. In an article published by MIT Technology Review, it is explained that forgeries can only be identified when brushstrokes are conspicuously visible. In addition, the background of the drawing can cause a difference in tone and interfere with the strokes of the painting. In response to this, Elgammal, one of the authors of the paper, states in the same article that the researchers plan to test this method on Impressionist works and 19th-century art, as the research focused primarily on drawings by Picasso, Henry Matisse, Egon Schiele, Amedeo Modigliani, and other artists. AI, with a wide range of possibilities, could become one of the best approaches to detecting art forgeries. The AI system, which achieved a high accuracy, can be used to apply the same method with paintings and other forms of art. While art historians and traditional methods can undermine the authenticity of an art piece, AI can prove through more advancements its benefits in the art world. SOURCES:
0 Comments
Leave a Reply. |
Writers & Editors- Seoyoon Chang Archives
January 2019
Categories
All
|