Quantifying Aesthetics of Visual Design Applied to Automatic Design

In this thesis, the author makes several contributions to the
study of design of graphical materials. The thesis begins with a review of the
relationship between design and aesthetics, and the use of mathematical models
to capture this relationship. Then, a novel method for linking linguistic concepts to colors
using the Latent Dirichlet Allocation Dual Topic Model is proposed.  Next, the thesis studies the relationship between aesthetics and spatial layout by formalizing the notion of visual balance.  Applying principles of salience and Gaussian mixture models over a body of about 120,000 aesthetically rated professional photographs, the author provides confirmation of Arnhem's theory about spatial layout.  The thesis concludes with a description of
tools to support automatically generating personalized design.