"We're essentially training computers to be art historians," says James Wang. An assistant professor of information sciences and technology at Penn State, Wang is cataloguing works of Asian art with a newly developed computer program.
To teach the computer how to analyze images, he and his colleague, Jia Li, assistant professor of statistics, feed it conceptually grouped digital photographs of works of art along with descriptive text. "What we call the concept could be something you see in the image itself," explains Wang. "For example, we could enter a series of photographs of terracotta warriors with horses. The concept could also be a historical period or an individual artist. It could be several images painted during a certain dynasty."
The computer creates a statistical analysis of colors, textures, and shapes present in the given work and compares that work to the automatically learned knowledge, which is stored in the computer as a collection of statistical models. When faced with images it hasn't seen before, the program uses statistical models it developed during its "training" to identify an artist or period.
One application, says Wang, is a searchable database that allows experts and enthusiasts to explore specific details that might not be indexed or illustrated in print books. "The program is unique because it can literally take advantage of information from multiple sources while a human art historian can't," says Wang. "It can analyze an infrared or ultrasoundof the image and see every detail of a painting in a spectrum that a human expert wouldn't be able to see."
James Wang, Ph.D., is assistant professor of information science and technology,313C Information Science and Technology Building, University Park, PA 16802; 814-865-7889; jzw@psu.edu. His research was funded by the National Science Foundation.