A Stroke of Annotation
Bren Professor Ramesh Jain and his colleagues at the National University of Singapore hope the new framework they created will ease the job of art experts by performing a wide range of annotation for the purposes of efficient search, retrieval and navigation.
By creating a framework that analyzes a painting by assessing both its visual characteristics, such as color, brushwork, and composition and higher-level concepts such as the artist names, painting style and art periods, the researchers hope to create a full-proof way to autonomously annotate paintings.
To test the system, the researchers selected over seven hundred paintings from the Medieval and Modern artistic periods and analyzed them according to their framework.
Each period has a distinctive visual element; Medieval paintings often exhibit a primary palette of colors such as red, blue, light-dark color contrasts and certain brushstrokes like mezzapasta, glazing and shading.
In contrast, paintings from the Modern period often exhibit complimentary colors, temperature contrasts and a variety of brush strokes such as scumbling, impasto, pointillism, divisionism and grattage.
To preserve color and brushwork information, a series of fixed-size 32x32 blocks were taken from each painting and these blocks were then analyzed according to the framework.
The result was an over 90 percent success rate in correctly annotating the paintings.
But the framework isn’t full proof, it did misidentify some Modernist paintings which had similar color and brushstrokes to Medieval style paintings.
