RIPA: Real-time image privacy alert system

The problem of privacy and security threats arising from images uploaded onto popular social media and content sharing websites is prevalent now more than ever. As our digital footprints grow exponentially, the need to find a solution to these problems has become that much more significant. In order to address these problems, a lot of research work has been carried out for image privacy protection through privacy policy recommendations and configurations. Due to the recent advancement in the field of computer vision and deep learning we can now gain more detailed insights about the context of an image and about the relationships between objects within it, this makes it possible to better address these problems. The privacy and security threats arising from an image uploaded on-line are not only limited to the data owners. Unlike previous works that are mostly focused on individual privacy policies, we take into account privacy concerns of multiple objects depicted on the same photo (even people, animals or other objects in the background of a scenery photo) whereby these privacy concerns may not be those from the user who uploads the photo. Specifically, we first build a general knowledge base by leveraging convolution neural networks to classify sensitive and non-sensitive image content and then use our proposed metadata analysis module to analyze metadata embedded within the image. Next, we extract objects present in the photo and validate if there is any privacy violation of the objects' privacy concerns. If any sensitive object is found, we toggle the object and issue a privacy violation alert to the user who is uploading the image as well as the service provider.