Statistical Analysis of Fine Art

This course presents the application of rigorous statistical analysis, machine learning, and data analysis to problems in the history and interpretation of fine art paintings, drawings, and other two-dimensional artworks. The course focuses on the aspects of these problems that are unlike those addressed widely elsewhere in statistical image analysis, such as applied to photographs, videos, and medical images. These novel problems include statistical analysis of brushstrokes and marks, medium, inferring artists' working methods, compositional principles, stylometry (quantification of style), the tracing of artistic influence, and art attribution and authentication. The course revisits classic problems, such as image-based object recognition and scene description, but in the environment of highly non-realistic, stylized artworks. Prerequisites: a course in machine learning, pattern recognition, or introductory data science; expertise in a high-level programming language of your choice (Matlab, Mathematica, R, Python, C/C++, ...); implementation knowledge of deep neural networks in a framework of your choice (PyTorch, TensorFlow, Keras, ...). Recommended: a course in Art and Art History; a course in image processing.