Artwork Classification and Recognition System based on Convolutional Neural Network

Abstract

From ancient ages, artworks have been the object of research in artist identification. Expert art historians primarily handle this issue manually. But an automatic artist recognition system using artwork is compulsory to lower the error percentage, and only a few progressive efforts are undertaken in this field, especially on Bangladeshi Artists. Our Convolutional Neural Network (CNN) model aims to determine the painter of a painting with a satisfactory accuracy standard. There are 450 paintings from 6 well-known Bangladeshi artists comprised in our novel dataset. Two different convolution kernels are used in model design, Model-1 has 3 X 3 convolutional kernels, and Model-2 has 5 X 5 kernel size. Our models achieve significantly higher classification accuracy as 87% for Model-1 and 89% for Model-2. Our result evaluation demonstrates that CNNs is not merely a robust learning tool for artist identification but also effective in predicting unique styles of an artisan.