Image Content Location Privacy Preserving in Social Network Travel Image Sharing

Abstract

It is a well-known truth that online image sharing can lead to privacy leakage. Although the privacy information may vary in forms, image contents only with private property will be concerned, while other contents that seems public but could reveal personal information are always ignored. In online image sharing, some images may contain labels or landmarks which could reveal the geographic positions where these photos were taken, in which the public content could disclose private location information. To handle such problem, we proposed a travel image location privacy protection system that aims at protecting travel images location. The key issue in this problem is which part of image content is related to the scene and to what extent. To determine such relevance, in our proposed system, several image process methods are utilized to find out potential objects and their belonging classes, and two proposed privacy strategies that respectively focus on quantity and relative position are further implemented to define initial privacy level of each classes. Finally, a privacy predict model is trained in online learning way so that it can be updated with future user feedbacks. We conducted experiment on travel image dataset that related to specific keyword, and the results have demonstrated the effectiveness of our proposed system.