Open Access
Balakavi, Rahul Mv
Graduate Program:
Computer Science and Engineering
Master of Science
Document Type:
Master Thesis
Date of Defense:
April 21, 2014
Committee Members:
  • George Kesidis, Thesis Advisor
  • Sencun Zhu, Thesis Advisor
  • Anna Cinzia Squicciarini, Thesis Advisor
  • privacy images learning models
Images are now one of the most common forms of digital content shared online in social network and other web 2.0 applications. With the availability of cheap digital cameras and huge amounts of storage space, photo publishing on social network application has become very popular. A portion of the images content that is uploaded onto the web is highly sensitive in nature, disclosing details of users’ personal life. In this thesis, we present an extensive study exploring the privacy needs and perspectives of users, when uploading their images. Our study investigates the importance of visual and textual features of images for privacy classification. One of the image features that we use in this research is a sentiment based visual representation, which was never used in the context of image classification till date. We also develop learning models to estimate privacy settings for newly uploaded images, based on carefully selected image-specific features. We identify a smallest set of features that by themselves or combined together with other features can perform well in predicting degree of sensitivity of images. In terms of privacy policies assigned for an image, we consider the case of binary classification, where an image is classified as either public or private and also a case where privacy is characterized by multiple sharing options.