Can we use machines to analyze paintings? This paper presents a new technique--the prevailing orientation extraction technique (POET)--that is designed for the “automatic extraction of the orientation of brushstrokes from digital reproductions of paintings.” The work “is motivated by the difficulty of segmenting individual brushstrokes in Van Gogh’s works,” which is also difficult for human operators.
In a painting, shape, brushstroke distribution, and orientation are low-level, spatial characteristics. POET is based on the application of a circular filter and an ad hoc orientation extraction phase. In the filtering stage, a circular filter yields to a “convolved image in which oriented parallel contours are enhanced.” In the orientation-extraction stage, the convolved image is transformed “into a set of binary oriented objects which correspond to the bright brushstrokes.”
The paper is very well written, clear, and very detailed. The experimental phase is robust, having a deep validation step that is really needed in cases where ground truth data is absent. The paper would attract more readers if it had a longer and more general introduction about how automatic techniques could improve painting analyses. As it is, the paper is great for those who work in the field.