Major improvements in art-identifying algorithm
DOI: 10.1063/PT.5.028871
MIT Technology Review : Babak Saleh and Ahmed Elgammal of Rutgers University in New Jersey have developed an algorithm that successfully identifies a painting’s artist 60% of the time and the artwork’s style 45% of the time. Saleh and Elgammal trained the algorithm using a subset of a collection of some 80 000 paintings. The collection spans 15 centuries, covers 27 styles, and represents 1000 artists. The algorithm is keyed to recognize a set of more than 400 characteristics ranging from setting (indoor, outdoor, portrait) to the presence of particular objects, such as horses. After the training, the algorithm was tested on paintings it had not yet seen. Beyond its success in identification, it was able to draw connections between artists and styles in a very short time, a skill that art historians often spend years acquiring.
