Identifying the painter using texture features and machine learning algorithms

Every artist has their own unique style of painting. The quantitative analysis of artworks is therefore essential to better understand the statistical differences between the paintings of different artists. In this study, we test if we can distinguish the works of Juan Luna from other Filipino painters using features derived from different sections of their paintings - foreground, background, foreground and background. We extracted texture features from the Gray Level Co-occurrence Matrix (GLCM) of the patches obtained from these sections and apply neural networks and Support Vector Machine (SVM) on duplets and triplets of features. From k-fold validation, the SVM on duplet of features from the background section of the paintings gave the highest accuracy of 83% for 375 dpi and 82% for 100 dpi. We have therefore shown that our approach can distinguish the works of Juan Luna from other Filipino painters.