On a Computation Model for Human and Machine Symmetry Perception

Restricted (Penn State Only)
Funk, Christopher A
Graduate Program:
Computer Science and Engineering
Doctor of Philosophy
Document Type:
Date of Defense:
October 12, 2018
Committee Members:
  • Yanxi Liu, Dissertation Advisor
  • Yanxi Liu, Committee Chair
  • Robert Collins, Committee Member
  • William Evan Higgins, Committee Member
  • David T Reitter, Outside Member
  • Symmetry
  • Deep Learning
  • Convolutional Neural Networks
  • Machine Learning
  • Wallpaper Groups
  • Topology Comparison
  • EEG
Symmetry is fundamental to human perception since it draws attention to specific objects and helps perceptual grouping of the messy world into distinct, well-defined categories. Symmetry and regularity have always been key ingredients that influence the visual beauty perceived by humans in art and architecture. Computational symmetry, formally defined by Professor Yanxi Liu in 2000, refers to the practice of representing, detection, and reasoning of symmetries from digital data through an algorithmic approach. This problem is challenging enough that, after decades of research, a fully automated and robust symmetry detection system remains elusive. This work lies at a nexus of computer science, cognitive science, and mathematics. The main focus of the work is to create the next generation computational symmetry algorithms simulating human intelligence. This work develops algorithms to detect real-world symmetries including reflection, rotation, and translation from photos and synthetic images containing periodic patterns represented by the 17 wallpaper groups. New datasets that are orders of magnitude larger than previous datasets have been created to train Convolutional Neural Networks to detect the imperfect symmetries and to investigate how the Convolutional Neural Networks are learning the wallpaper hierarchy. The novelty of this work is to explore human and machine perception of symmetries jointly and to develop novel ways for symmetry detection in computer vision.