2020-073 – A Machine Learning Method of Art Authentication

A researcher at Kansas State University has developed a novel software algorithm to authenticatepaintings quickly and with high accuracy. This method uses a machine learning algorithm with acomprehensive set of numerical image content descriptors, including the WND-CHARM featureset, which determines the authenticity of a painting. The platform software has been developed andtested for authenticating Jackson Pollock paintings with about 96% accuracy. The mechanism of this software algorithm is that each painting is separated to a grid of smallerparts of the painting. The mathematical features of all paintings are computed from each tile of thegrid by using the larger WND-CHARM feature set as well as other helpful features. A machinelearning system is then used to determine whether the patterns of the numerical content descriptorof the paintings are also typical to the patterns in other paintings of that painter, or are more typicalto other painters. If the patterns match, as determined by the machine learning system, it can bedetermined that the painting is the authentic work of the painter. Aarushi Gupta-Sheth aarushi@ksu.edu 785-532-3907