Analysis Machine Learning techniques to artistic movement recognition

The proposed methods for solving the problem of recognizing the style of the visual arts were tested on the Pandora dataset and included the construction of a classifier over the features obtained by classic methods of computer vision, as well as over the features obtained from the inner layers of state-of-the-art neural networks. Among all the proposed methods, the classifier based on logistic regression in combination with the internal features of the GoogLeNet v3 network provided best results. Also, the data classes were visualized in the context of the neural network output. The proposed classifier, using the features of classic computer vision, is competitive against the background of other classifiers using similar features. The use of pre-trained GoogLeNet and logistic regression gave the best result among the published works, and the proposed visualization provided new insights for the data.