Unveiling the hand of a 19th century artist with binary image classification and bag-of-features

The shape, orientation and distribution of brushstrokes are distinctive markers left by the artist on the surface of a painting. Consequently, they are useful as additional evidence to decide on issues of attribution and authenticity. This paper proposes an image classification algorithm for the paintings of Amadeo de Souza-Cardoso, a modern Portuguese artist who is at the top of the Portuguese police's list of forged artists. The proposed system is based on a binary classification using bag-of-features (SIFT descriptors and a Gabor filter) for image representation. Experimental results using the algorithm show that the classification of the paintings by extracting brushstroke information is very promising.