Forgery Detection in Paintings

• Activity Number: 415
• Type: Topic Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Yi Wang*+ and Ingrid Daubechies and Gungor Polatkan and Sina Jafarpour
• Companies: SAMSI/Duke University and Duke University and Princeton University and Yahoo! Research
• Keywords: image processing ; wavelet ; machine learning ; forgery detection ; hidden-markov-tree ; supervised learning

Abstract

This work studies how machine learning and image analysis tools can be used to assist art experts in the authentication of unknown or doubtful origin. Previous work has shown that variation in image clarity in the experimental data sets was correlated with authenticity, and may have acted as a confounding factor, artificially improving the results. Therefore, a data set with ground truth and uniform acquisition conditions is provided to determine the extent of this factor's influence. While many previously successful methods turn out to be ineffective, supervised machine learning on features extracted from Hidden-Markov-Tree modeling of the paintings' wavelet coefficients demonstrates its potential to distinguish copies from originals in this data set. In order to further study and improve this approach, a larger data set is created under similar conditions. In addition, more careful analysis and experiments are conducted to provide new insights.