IPSODAC: AUTOMATED PROTECTION OF IMAGES IN SOCIAL NETWORKS USING SENSITIVE OBJECT DETECTION AND ACCESS CONTROL

ABSTRACT

Despite the introduction of various privacy tools by online social networks to safeguard users' privacy and data, ensuring a high degree of privacy remains challenging because of the complexity and ambiguity of setting these privacy policies and tools. To solve this problem, a new model is proposed that automates the privacy settings process for image sharing in online social networks, eliminating the burden and the necessity for any technical effort on the user's part. This work proposes a hierarchical deep learning approach to automatically recognize sensitive objects in images and classify them according to the classes defined by the online social network. Based on the relationship of a prospective image viewer to that of the image owner, a set of privacy rules and an access control method are employed to determine the level of access the prospective image viewer has to those images. Extensive training and experimentation were conducted on the Microsoft Azure Computer Vision platform utilizing real-world images by employing IPOSODAC. The results of the sensitive object detection effectively protect sensitive images in social networks.

Keywords: Privacy, Object Detection, Access Control, Social Network, Deep Learning, Images