Home // International Journal On Advances in Intelligent Systems, volume 18, numbers 1 and 2, 2025 // View article

Exploring Deep Learning Techniques for Artist and Style Recognition in the Paintings-100 Dataset

Authors:
Erica M. Knizhnik
Brian Rivera
Atreyee Sinha
Sugata Banerji

Keywords: painting classification; image dataset; style classification; artist classification; CNN ensemble

Abstract:
Painting classification is a challenging interdisciplinary research problem in computer vision. With more fine-art paintings being available in the form of high-resolution digital scans, the development of effective classification algorithms has become vital. Such algorithms would have numerous applications, including but not limited to museum curation, several different industries, painting theft and forgery investigation, and art education. While some progress has been done in this field, accurately identifying the painter or the artistic style from the painting remains a complex task. Towards that end, we present an enhanced image dataset comprising high-resolution painting images from 100 diverse artists across 14 distinct styles. This dataset builds upon the Painting-91 dataset originally created by Khan et al. Our main contributions in this work are three-fold. First, we improve the older dataset by correcting errors, enhancing image resolution, and expanding it with more images, artists, and styles. Second, we perform an extensive evaluation of this newly constructed Paintings-100 dataset using several different convolutional neural network (CNN)-based classification techniques for both artist and style recognition tasks. Finally, we explore the different stylistic characteristics that the networks focus on to recognize the specific artists and styles of paintings, and demonstrate that our proposed and improved dataset is more suitable for patch-based models than the earlier published Painting-91 dataset due to larger image resolutions.

Pages: 11 to 21

Copyright: Copyright (c) to authors, 2025. Used with permission.

Publication date: June 30, 2025

Published in: journal

ISSN: 1942-2679