EnsArtNet: Ensemble neural network architecture for identifying art styles from paintings

EnsArtNet: Ensemble neural network architecture for identifying art styles from paintings

Článek WoS

The digitization of paintings offers many benefits and opportunities for artists, collectors, and the public. It opens possibilities for researchers to investigate new hidden patterns that were not obvious to experts before. This work aims to develop a methodology that can identify and compare painting styles from various famous painters, such as Vincent van Gogh, Pablo Picasso, Claude Monet, and others, using an ensemble convolutional neural network (CNN). Our approach, named EnsArtNet, can distinguish between the styles of the artists' paintings with high accuracy and objectively measure the similarity with the other artists' styles. The proposed model was compared to several other state-of-the-art neural network architectures, and we show that EnsArtNet performs better than the compared one. Our model gives promising accuracy on two large-scale datasets: 84.93% on the WikiArt dataset and 86.65% on the Best Artworks of All Time dataset, which is better by more than 6% compared to other evaluated architectures. In this work, we also showed that a complex neural network architecture is efficient in this field of research, and an explanation using the GradCAM method supported it. Our methodology can help art researchers and enthusiasts analyze paintings' stylistic features and similarities and appreciate the creativity and diversity of visual arts. (c) 2025 Elsevier Masson SAS. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

The digitization of paintings offers many benefits and opportunities for artists, collectors, and the public. It opens possibilities for researchers to investigate new hidden patterns that were not obvious to experts before. This work aims to develop a methodology that can identify and compare painting styles from various famous painters, such as Vincent van Gogh, Pablo Picasso, Claude Monet, and others, using an ensemble convolutional neural network (CNN). Our approach, named EnsArtNet, can distinguish between the styles of the artists' paintings with high accuracy and objectively measure the similarity with the other artists' styles. The proposed model was compared to several other state-of-the-art neural network architectures, and we show that EnsArtNet performs better than the compared one. Our model gives promising accuracy on two large-scale datasets: 84.93% on the WikiArt dataset and 86.65% on the Best Artworks of All Time dataset, which is better by more than 6% compared to other evaluated architectures. In this work, we also showed that a complex neural network architecture is efficient in this field of research, and an explanation using the GradCAM method supported it. Our methodology can help art researchers and enthusiasts analyze paintings' stylistic features and similarities and appreciate the creativity and diversity of visual arts. (c) 2025 Elsevier Masson SAS. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

Deep learning; Artist identification; Painting classification; Convolutional neural network; Cultural heritage

Deep learning; Artist identification; Painting classification; Convolutional neural network; Cultural heritage

MEZINA, A.; BURGET, R.

2026

30.01.2025

ELSEVIER FRANCE-EDITIONS SCIENTIFIQUES MEDICALES ELSEVIER

FRANCE

1778-3674

JOURNAL OF CULTURAL HERITAGE

72

2025

Francouzská republika

71

80

10

https://www.sciencedirect.com/science/article/pii/S1296207425000056