Page 33 - AC-2-2

Arts & Communication Identification of Pollock Art

spectroscopy, and surface analysis of the painting
materials. Computer vision analysis through machine
learning or signal processing is not yet recognized as a
tool for formal authentication of paintings. With the
state-of-the-art generative artificial intelligence, the style
of painters can be mimicked by a computer to create a
painting that might be challenging for computers to
analyze for forgery. Yet, in cases where the painting
materials are verified to be old, computer analysis can be
used as an aid for authentication of art.
5. Conclusion
Figure 6. The values of Chebyshev histogram bins computed from the The recent advances in machine learning and computer
authentic Pollock paintings and the faked Pollock paintings.
vision have enabled a large number of tasks that were
previously not considered possible by computers. Analysis
and authentication of visual art can benefit greatly from the
availability of such methods. While tools such as generative
AI can be used to generate art, the analysis of existing art can
be done by explainable methods that do not necessarily rely
on deep neural networks. The ability to understand the way
machine learning works is critical for the understanding of
the art and can be a useful tool in profiling and understanding
the art in a quantitative manner. Such analysis can be used
in addition to the traditional qualitative analysis and can
lead to new verifiable insights about art.

Acknowledgments
Figure 7. The t-values of the t-test comparison between the Chebyshev None.
histogram bins computed from the authentic Pollock paintings and the
faked Pollock paintings. Funding

The analysis shows that when the feature set was used This project was funded in part by NSF grant 2148878.
in combination with machine learning algorithms, it could Conflict of interest
identify between a Pollock painting and a non-Pollock
painting in accuracy far higher than mere chance. The The author declares no competing interests.
differences between the numerical image content descriptors
show a wide range of differences in multiple aspects of the Author contributions
visual content. While fractals show the strongest difference, This is single-authored article.
aspects such as the image entropy and polynomial
distribution of the pixel intensities also exhibit very strong Ethics approval and consent to participate
statistical differences between Pollock and non-Pollock drip
paintings. These findings demonstrate the uniqueness of the Not applicable.
work of Jackson Pollock compared to careful attempts to Consent for publication
mimic his work and create paintings that aim at making the
impression of authentic Jackson Pollock work. Not applicable.

Due to the high monetary value of some classic art, Availability of data
legal authentication of art has been a field of primary
concern. In addition to the history of a certain piece Images of Jackson Pollock paintings were obtained from
of art as reflected through the track record of trades online sources. Faked Pollock paintings cannot be shared
and transactions, art can also be authenticated through publicly. Code can be accessed at http://people.cs.ksu.
forensic tools such as analysis of the pigments, X-rays, edu/~lshamir/downloads/udat.

Volume 2 Issue 2 (2024) 7 doi: 10.36922/ac.1628