Classifying Paintings into Movements using HOG and LBP Features

Abstract

With the increase in availability of paintings on the web, the importance of organizing art collections cannot be overstated. By classifying paintings based on art movements, information about paintings on the web can be well structured. This will also help us garner insights into more obscure paintings and the styles they embody. This paper discusses a method of classifying paintings into two art movements, namely Cubism and Romanticism, using two texture descriptors: Local Binary Pattern (LBP) and Histogram of Oriented Gradients (HOG). The paper also extends the applicability of the method to different classifiers. Some classifiers used for classification are KNN, NuSVC, LinearSVC, GaussNB, Decision Tree, Random Forest, AdaBoost and Gradient Boost. A subset of the Pandora database is used as a dataset. From the results, it can be inferred that using the Gradient Boost classifier gives the highest overall accuracy for Cubism and Romanticism when LBP and HOG are used as texture descriptors. Moreover, it can be seen that LBP emerges as the best feature for classifying paintings into the Cubism and Romanticism Art Movements.