Date of Graduation


Document Type


Degree Name

Master of Science in Computer Science (MS)

Degree Level



Computer Science & Computer Engineering


Qinghua Li

Committee Member

Brajendra Panda

Second Committee Member

Dale Thompson


Browser extension, Mobile Systems, Privacy, Risk, Security, Visual eavesdroppers


As smartphones have gained popularity over recent years, they have provided usersconvenient access to services and integrated sensors that were previously only available through larger, stationary computing devices. This trend of ubiquitous, mobile devices provides unparalleled convenience and productivity for users who wish to perform everyday actions such as taking photos, participating in social media, reading emails, or checking online banking transactions. However, the increasing use of mobile devices in public spaces by users has negative implications for their own privacy and, in some cases, that of bystanders around them.

Specifically, digital photography trends in public have negative implications for bystanders who can be captured inadvertently in users’ photos. Those who are captured often have no knowledge of being photographed and have no control over how photos of them are distributed. To address this growing issue, a novel system is proposed for protecting the privacy of bystanders captured in public photos. A fully automated approach to accurately distinguish the intended subjects from strangers is explored. A feature-based classification scheme utilizing entire photos is presented. Additionally, the privacy-minded case of only utilizing local face images with no contextual information from the original image is explored with a convolutional neural network-based classifier. Three methods of face anonymization are implemented and compared: black boxing, Gaussian blurring, and pose-tolerant face swapping. To validate these methods, a comprehensive user survey is conducted to understand the difference in viability between them.

Beyond photographing, the privacy of mobile device users can sometimes be impacted in public spaces, as visual eavesdropping or “shoulder surfing” attacks on device screens become feasible. Malicious individuals can easily glean personal data from smartphone and mobile device screens while they are accessed visually. In order to protect displayed user content, anovel, sensor-based visual eavesdropping detection scheme using integrated device cameras is proposed. In order to selectively obfuscate private content while an attacker is nearby, a dynamic scheme for detecting and hiding private content is also developed utilizing User-Interface-as-an-Image (UIaaI). A deep, convolutional object detection network is trained and utilized to identify sensitive content under this scheme. To allow users to customize the types ofcontent to hide, dynamic training sample generation is introduced to retrain the content detection network with very few original UI samples. Web applications are also considered with a Chrome browser extension which automates the detection and obfuscation of sensitive web page fields through HTML parsing and CSS injection.