Vincent van Gogh's Self portrait with grey felt hat in the van Gogh Museum in Amsterdam is representative of this artist's self portraits and his distinctive open brush work. The following is based on:
Art scholars are often interested in the intermedite steps in the development of a work and the decisions an artist makes while executing a painting. For that reason, scholars often study the underdrawings or pentimenti of a painting, to see how the the composition develops. The De-pict algorithm addresses this problem in the very limited domain of brush strokes in "open" passages, where multiple layers are visible.
In brief, the De-pict algorithm determines the number of layers of brush strokes by performing k-means clustering on the colors. For instance, if there are three colors, there are likely three layers. Then the algorithm identifies the top layer as the one where the brush strokes have roughly the same shape. (Brush strokes of the lower layers have more variant shapes because they have been broken up or occluded by the upper layers.) Then the algorithm digitally removes the pixels corresponding to this top layer and then inpaints or fills in the missing strokes "behind" the removed strokes. Then the algorithm identifies the current top layer, removes it, inpaints, and so on, iterating through all the layers. Then, the layers can be built up, from deepest to nearest, to give an approximation of the appearance of the painting as the artist added layer after layer—almost like standing next to the artist, watching his canvas as he paints.
We applied De-pict to a passage in van Gogh's Self portrait with grey felt hat and it reveals plausible stages in the work. Of course, we cannot validate the algorithm thoroughly because we don't have access to the true intermediate stages. Nevertheless, we're now applying De-pict to our own paintings, where we photograph the intermediate stages so we can compare the intermediate stages produced by the algorithm to the actual intermediate stages.
The De-pict algorithm, suitably refined and tuned to the statistical properties of the brush strokes of a given painter, could help art scholars see the intermediate step in the development of paintings with open brush strokes.