Classification of Style in Fine-Art Paintings Using Transfer Learning and Weighted Image Patches

With the ongoing expansion of digitized artworks, the automated analysis and categorization of fine art paintings have become a rapidly growing research field. However, due to the implicit subjectivity and nuances separating different artistic movements, numerical art analysis implies significant challenges. This paper describes a new efficient method that improves the classification accuracy of fine-art paintings compared to the existing baseline methods. The proposed approach is based on transfer learning and classification of sub-regions or patches of the painting. A weighted sum of the individual-patch classification outcomes gives the final stylistic label of the analyzed painting. The patch weights are optimized to maximize the overall style recognition accuracy. Experimental validation based on two standard art classification datasets and six different pre-trained convolutional neural network (CNN) models (AlexNet, VGG-16, VGG-19, GoogLeNet, ResNet-50 and Inceptionv3) shows that the proposed approach outperforms the baseline techniques and offers low computational and data costs.