Identification of Subjects and Bystanders in Photos with Feature-Based Machine Learning

With the proliferation of smartphones among everyday users, digital photography in public places has exploded. While this trend brings the hobby of photography to millions of consumers, it also reduces the privacy of individuals in public settings as people are inadvertently included in strangers’ photographs. These photos can then be uploaded to publicly accessible forums such as social media platforms without any knowledge of the bystanders in the image who might not want their location or surroundings disclosed. To combat these scenarios, we propose an automated system for detecting and distinguishing subjects from bystanders in digital images which will later be used with visual obfuscating techniques to protect bystander privacy. Techniques such as gaze direction tracking, facial size detection, and head pose estimation are all explored as methods of extracting useful features for training various classifiers. Specifically, Random Forest, Gradient Boosted Decision Tree, Multilayer Perceptron, and Support Vector Machine classifiers are trained and evaluated for accuracy.