A digital technique for art authentication

doi:10.1073/pnas.0406398101

A digital technique for art authentication

Departments of ^*Computer Science and ^†Mathematics, Dartmouth College, Hanover, NH 03755

Communicated by David L. Donoho, Stanford University, Stanford, CA, September 1, 2004 (received for review May 13, 2004)

Abstract

We describe a computational technique for authenticating works of art, specifically paintings and drawings, from high-resolution digital scans of the original works. This approach builds a statistical model of an artist from the scans of a set of authenticated works against which new works then are compared. The statistical model consists of first- and higher-order wavelet statistics. We show preliminary results from our analysis of 13 drawings that at various times have been attributed to Pieter Bruegel the Elder; these results confirm expert authentications. We also apply these techniques to the problem of determining the number of artists that may have contributed to a painting attributed to Pietro Perugino and again achieve an analysis agreeing with expert opinion.

Footnotes

↵ ‡ To whom correspondence should be addressed at: 6211 Sudikoff Laboratory, Department of Computer Science, Dartmouth College, Hanover, NH 03755. E-mail: farid{at}cs.dartmouth.edu.
Author contributions: S.L., D.R., and H.F. designed research; S.L. and H.F. performed research; S.L., D.R., and H.F. analyzed data; and D.R. and H.F. wrote the paper.
Abbreviation: MDS, multidimensional scaling.
↵ § Henceforth we will give the benefit of the doubt to the imitator and use the term “imitation” rather than the more charged “forgery.”
↵ ¶ Although converting from color to grayscale results in a significant loss of information, we did so to make it more likely that the measured statistical features and subsequent classification were more likely to be based on the artist's strokes and not on simple color differences.
↵ ∥ We also have experimented with both Laplacian and steerable pyramid decompositions. Results from a steerable pyramid (with eight orientation subbands) were similar to the results included above (which use only three orientation subbands). Furthermore, the Laplacian pyramid generally gave poor results. So, although it seems that oriented subbands are necessary, it also seems that a finer tuning of orientation is not necessary for this particular task.
↵ ** Farid, H. & Lyu, S. IEEE Workshop on Statistical Analysis in Computer Vision (in conjunction with Computer Vision and Pattern Recognition), June 17–23, 2003, Madison, WI.
↵ †† Integer rounding is used when computing the spatial positions of a parent, e.g., x/2 or x/4.