Analysis of Dutch Master Paintings with Convolutional Neural Networks

TL;DR

This paper explores how convolutional neural networks can analyze Dutch master paintings to detect forgeries, attribute authorship, and identify regions of mixed or multiple authorship within artworks.

Contribution

It demonstrates the application of CNNs for detailed analysis of paintings, including forgery detection and attribution, revealing complex authorship details.

Findings

CNNs can accurately identify forgeries.

They can attribute paintings to specific artists.

The method reveals mixed authorship regions.

Abstract

Trained on the works of an artist under study and visually comparable works of other artists, convolutional neural networks can identify forgeries and provide attributions. They can also assign classification probabilities within a painting, revealing mixed authorship and identifying regions painted by different hands.