Weave analysis of paintings on canvas from radiographs

A spectral algorithm is described for measuring from radiographs the weaving densities of the horizontal and vertical threads that comprise a painting's canvas. A framework for relating spectra to canvas weave type is presented. The so-called thread density and angle maps obtained from the algorithm reveal the canvas's distinctive density variations and provide insights into how the canvas was prepared. Applying a two-stage correlation procedure to the density variations allowed determination of which paintings' support could have been cut from the same piece of canvas. The first stage uses a nonparametric test of the similarity of the probability distributions of two painting's thread counts. The second stage is a new correlation procedure more stringent than the usual cross-correlation function. Examples drawn from the paintings by Vincent van Gogh illustrate the algorithms.

► “Short-space” spectral analysis coupled with weave modeling leads to accurate canvas analysis.
► Our new correlation approach applies to data containing a large offset and unknown statistics.
► Results from this algorithm have led to re-dating paintings and insights into the artist's process.