What is it about?
This research explores how artificial intelligence can help experts assess whether a historical sketch is likely to be by a particular artist. Authenticating drawings is often difficult because there may be only a small number of trusted examples available for comparison, and sketches can vary greatly in style, condition, and purpose. Instead of relying on large image datasets, this study uses a compact set of visual measurements that describe features such as line texture, contrast, tonal variation, mark-making, and visual complexity. The system learns the typical visual pattern of an artist’s authenticated sketches and then checks whether a new sketch fits within that pattern or appears unusual. The method was tested on sketches by ten historical artists using images from major open-access museum and cultural heritage collections. The aim is not to replace art historians, conservators, or connoisseurs, but to provide an additional source of transparent, quantitative evidence that can support expert judgement in difficult attribution cases.
Featured Image
Photo by Art Institute of Chicago on Unsplash
Why is it important?
This work is important because many AI approaches to art attribution require thousands of training images, which are rarely available for historical sketches. By using a small number of meaningful and interpretable image features, this study shows that machine learning can still provide useful evidence even when reference material is limited. The approach is timely because museums, researchers, collectors, and the art market increasingly need reproducible and explainable tools for assessing authenticity. The method offers a way to quantify stylistic similarity while keeping the evidence understandable to human experts. It could help identify works that deserve closer examination, reduce the risk of accepting misattributed works, and strengthen the evidence base used in art authentication.
Perspectives
Historical sketch authentication has traditionally depended on expert visual judgement, provenance research, and material analysis. This research adds a complementary digital perspective by showing that aspects of an artist’s mark-making can be measured and modelled in a reproducible way. The findings suggest that AI can be most useful in this field when it is designed to assist expert interpretation rather than replace it. The study also points towards a broader future for computer-assisted connoisseurship, where human expertise and machine analysis work together. By focusing on transparency, small datasets, and interpretable visual features, the framework may be especially valuable for cultural heritage settings where the number of securely attributed works is limited, but the need for careful and defensible attribution remains high.
Professor Hassan Ugail
University of Bradford
Read the Original
This page is a summary of: Verification of historical sketches via one-class learning on compact feature representations, PLOS One, June 2026, PLOS,
DOI: 10.1371/journal.pone.0344796.
You can read the full text:
Contributors
The following have contributed to this page
