People search and activity mining in large-scale community-contributed photos

• Authors: Yan-Ying Chen
• Venue: Proceedings of the 20th ACM international conference on Multimedia
• Year: 2012

A growing number of users are contributing a huge amount of photos containing people (e.g., family, classmates, colleagues, etc.) to social media for the purpose of photo sharing and social communication. There arises a strong need for automatically analyzing the people shown in large-scale photos because these visual data comprise abundant consumer activities which greatly benefit demographic analysis and enhance marketing research. In this work, we aim at learning facial attributes (gender, race, age, etc.) by these publicly available photos and exploiting the detected facial attributes for locating designated persons, profiling user preferences and predicting social group types. In addition, community-contributed data possess rich contexts such as tags, geo-locations and time stamps, which strongly correlate with user intentions and preferences. The knowledge would be informative to actively refine the recognition models and promising towards improvement of photo management, personalized recommendation and social networking. Most importantly, this framework effectively relieves costly annotation efforts and ensures scalability for large-scale media.