Studying Digital Imagery of Ancient Paintings by Mixtures of Stochastic Models

Abstract

Abstract-This paper addresses learning based characterization of fine art painting styles. The research has the potential to provide a powerful tool to art historians for studying connections among artists or periods in the history of art. Depending on specific applications, paintings can be categorized in different ways. In this paper, we focus on comparing the painting styles of artists. To profile the style of an artist, a mixture of stochastic models is estimated using training images.

Graph-based methods for the automatic annotation and retrieval of art prints

“…Most of the methods above can be regarded as adaptations from the content-based image retrieval systems [14], where the emphasis is placed on the characterization of brush strokes using texture or color. The ancient Chinese painting classification studied by Li and Wang [29] is more similar to ours in the sense that they deal with the multi-class classification of painting styles. Finally, the work on the automatic brushwork annotation by Yelizaveta et al [48] is also similar to ours given that the authors are dealing with multi-class classification of brush strokes.…”

“…Art image analysis methodologies can also be used for art prints [42,27], but the great majority of these techniques have been developed for the analysis of digitized images of paintings, which contain richer visual information than prints. However, note that Li and Wang [29] have developed a system that analyzes ancient Chinese paintings, which are similar to art prints.…”

Painter classification using genetic algorithms

“…Kroner et al [18] applied automatic pattern recognition methods using histograms to classify a drawing to a certain artist. Li and Wang [20] applied a two-dimensional multi-resolution hidden Markov model to classify Chinese ink paintings and form a distinct digital signature. Kammerer et al [14] presented a strategy for the analysis of under-drawing strokes in infra-red reflectograms, in order to determine the drawing tool used to draft a painting.…”

Detection of forgery in paintings using supervised learning

“…Meanwhile, Berezhnoy and Postma [4,5] found methods for color and texture analysis by complementary colors of VG and specifying the spatial distribution of brushstrokes at different directions by circular filtering. Li and Wang [6] have used texture features obtained by training a 2D Hidden Markov Model on local wavelet coefficients combined with features obtained from detecting and segmenting individual brushstrokes from the images, such as length and average curvature of individual strokes.…”

“…This assumes that an artist's brushwork is characterized by signature features (caused, e.g., by the artist's habitual physical movements) which might be found by machine learning methods and used as an additional piece of evidence to rule upon authenticity. Indeed, early attempts in this area have already found considerable success [2,3,4,5,6]. …”