BroadStrokes

Classifying an Artist’s Impression Using Machine Learning

Developed by Ariyana Miri, Savanna Moss, Hannah Wilberding, and Brock Wilson.

Developed a convolutional neural network, k-nearest-neighbors classifier, and a support vector machine to predict an artist based on images of artworks.

How to Use the Classifiers

Generate Data Files

Run image_conversion.py to generate the train_data.gz and train_data.gz compressed text files based on the images in image_data/train_resized and image_data/test_resized. When loading the files into another program, the train_data ndarray should be of shape (2184, 50176) and the test_data ndarray should be of shape (546, 50176).

Run Algorithms

KNN.py is the k-nearest-neighbors classifier.

• Displays training accuracy, testing accuracy, number of mislabeled and correctly labeled images, a 3D scatter plot of the first three components created by the PCA of the testing data, and a confusion matrix graph.

SVM.py is the support vector machine.

• Displays training accuracy, testing accuracy, number of mislabeled and correctly labeled images, and a confusion matrix graph.

CNN.py is the convolutional neural network.

• Displays training loss and accuracy, testing loss and accuracy, and a line graph of the training and testing accuracy over time.