In the past few years, collections of digitized fine-art paintings have been expanded rapidly, both in the size and variety of artworks. This brings up an extensive list of traditional challenges from data storage to visual classification. Additionally the opportunity of having access to large-scale collections of artworks opens the door for novel knowledge discovery research problems. My colleagues and I at the Art and AI lab at Rutgers University have been working on all the aforementioned tasks in the recent years.
In this talk, I will first cover our published results on the well-known task of painting tagging based on Style, Genre, and Artists. I will explain the applicability of visual features for the analysis of paintings, and how we tuned them to work on a large-collection of paintings. Next, I will explain our proposed algorithms for finding influence paths between artists based on visual analysis of the content of the their artworks. Additionally, I will elaborate how we built a large network of paintings and quantified the creativity of these artworks based on their novelty and influence. Finally, I will show results of validating our computational models on modern art data where we show that the model coincides with the aggregate of the opinions of art historian experts.