Automated analysis of composition and style of photographs and paintings

Open Access
Author:
Yao, Lei
Graduate Program:
Information Sciences and Technology
Degree:
Doctor of Philosophy
Document Type:
Dissertation
Date of Defense:
None
Committee Members:
  • James Z Wang, Dissertation Advisor
  • Jia Li, Dissertation Advisor
  • David J Hall, Committee Member
  • Xiaolong Zhang, Committee Member
  • Patrick John Mc Grady, Committee Member
Keywords:
  • painting analysis
  • photography composition
  • photo enhancement
  • image processing
  • computer vision
  • aesthetics inference
Abstract:
Computational aesthetics is a newly emerging cross-disciplinary field with its core situated in traditional research areas such as image processing and computer vision. Using a computer to interpret aesthetic terms for images is very challenging. In this dissertation, I focus on solving specific problems about analyzing the composition and style of photographs and paintings. First, I studied the problem of distinguishing van Gogh’s paintings from his contemporaries. The application of rhythmic and spontaneous brushstrokes is a prominent trait for van Gogh’s paintings. His unique brushstroke style is characterized by features calculated from automatically extracted brushstrokes. Statistical analysis on the extracted brushstroke features shows success in tackling real-world painting analysis tasks designed by art historians. Second, I explore the possibility of characterizing styles of paintings without visible brushstrokes, specifically artworks by Norval Morrisseau. Curve elegance measurements are used to differentiate authentic works from forgeries. Then, I present my studies on the topic of photography composition. Composition is closely related to the aesthetic qualities of images and a key factor that distinguishes professional photographs from snapshots. I design a spatial composition classifier to analyze the compositional properties for general photographs. A new integrated system is presented to render on-site photography feedback for users by retrieving high-quality exemplar photographs with similar compositions. User studies substantiate the system’s performance. Finally, I propose a dark-light re-composition algorithm to emulate the dodging and burning techniques used in darkroom photography. The algorithm performs region-wise intensity adjustments by utilizing the intensity distribution and the Notan structure of an exemplar. Overall, this dissertation studies computational approaches to the interpretation of artistic terms and explores potential applications in digital painting analysis, automatic photography feedback and photo enhancement.