Recognizing Art Style Automatically in Painting Using Convolutional Neural Network

Abstract

AbstractRecognition of art style in painting is an active and important research area in the computer visions field as art style recognition is not depending on the definite feature; however, extracting important in-definitive features from a painting image is a crucial fact. Moreover, using deep convolutional neural network (CNN) makes the model overfit and small CNN model makes the model underfit. To overcome these problems, in this work, a model is developed by us for art style recognition in painting based on the shallow convolutional neural network (SCNN). The proposed model is developed based on the two consecutive convolutional layers and single fully connected layer to extract painting features and recognize the art styles accordingly. For training and testing purposes in the art style recognition, we have used Wikipaintings dataset. We also used our created real-time dataset in this work. Our proposed model shows 60.37% accuracy which is the significant improvement over some current state of the arts.KeywordsFeature extractionOverfitUnderfitArt styles in paintingsShallow convolutional networkWikipaintings dataset