Supervised Learning-Based Feature Selection for Mondrian Paintings Style Authentication

Abstract

This paper concerns feature selection for computational analysis in authenticating works of art. The various features designed and extracted from art work in art forgery detection or the identification of the characteristics of art work style are valuable only when they have a meaningful influence on a given task such as classification. This paper presents features applicable to authenticating the painting style of Piet Mondrian and demonstrates meaningful features by using two supervised learning algorithms, a decision tree induction algorithm C4.5 and the Feature Generating Machine (FGM), both of which are used to select important features in the course of learning.