A Computational Approach For Understanding The Aesthetics Of Gray Scale Images

Open Access
Suryanarayan, Poonam
Graduate Program:
Electrical Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
December 07, 2010
Committee Members:
  • James Z Wang, Thesis Advisor
  • David Jonathan Miller, Thesis Advisor
  • Prof Jia Li, Thesis Advisor
  • aesthetics
  • modeling
  • gray scale
Image aesthetics is an important as well as a highly challenging field due to its subjective nature. There have been a number of computational studies on image aesthetics which try to predict the aesthetic quality of an image using its visual features. These studies are very generic in nature and have ignored the differences in aesthetic principles of monochromatic and color images. This thesis tries to understand the visual primitives which govern the aesthetics of black and white images through photography cues provided by experts. Newer visual features have been developed to learn a good mathematical model to predict the aesthetic quality of black and white images. A new set of ground truth images along with their aesthetic ratings have been collected from a digital photo contest website, dpchallenge.com. This study is a first of its kind to address the gray scale image aesthetics specifically. A regression model is used in a near real-time demo to provide aesthetic ratings to any image uploaded by the user.