Softcomputing Art Style Identification System

Abstract

The paper discuss the possibility to use the softcomputing methods to create effective and useful system for art style identification. The system should operate on the samples of paintings. The assumption is to use only the small parts of pictures with no high resolution. Different types of preprocessing methods are tested to create the input vectors for Convolutional Neural Network (CNN) which is an identification tool. The experiments are done for the significant dataset covering ten most classic art styles of paintings. Different types of CNN topology is discussed. The promising results could be an interesting subject for custodians, art historians or scientists. This may help them not only recognize the style with some certainty but also compare and mark the similarities and differences between styles or artists. The paper can be extended to help them in authenticating and determining the timeline of paintings.