TWC: Small: Collaborative: Towards Privacy Preserving Online Image Sharing

图像的在线共享已成为用户连接的关键使能器。 通过社交媒体分享各种类型的图像，以代表用户的兴趣和体验。 虽然非常方便和社会价值，这种程度的普及引入了严重的隐私问题。首先，一旦共享图像可以去任何地方，因为复制/重新共享图像是简单的。其次，通过图像披露的信息揭示了用户的私人生活，影响到图像的所有者和其他主体。 恶意攻击者可以利用这些不必要的泄漏来发起上下文感知攻击（例如，钓鱼），甚至冒充攻击。 该项目正在开发帮助用户适当控制对其共享图像的访问的方法。研究人员正在开发新技术，以解决基于图像内容以及图像和用户元数据的图像隐私问题，方法是：（a）根据图像的视觉特性和用户的图像共享模式推断给定图像的敏感度，然后自动地对该图像应用适当的隐私设置，以及（B）通过使用所发现的用户的共享模式来根据对感兴趣域的本地可执行控制来定义访问策略。 除了技术成果，该项目正在开发一个框架，用于理解用户对隐私的理解方面的常见类别和图像类别。 该框架将使用户能够理解需要保护的图像，并帮助提高用户对经常无意或意外保护不足的图像类别的认识。

On-line sharing of images has become a key enabler of users' connectivity. Various types of images are shared through social media to represent users' interests and experiences. While extremely convenient and socially valuable, this level of pervasiveness introduces acute privacy concerns. First, once shared images may go anywhere, as copying / resharing images is straightforward. Second, the information disclosed through an image reveals aspects of users' private lives, affecting both the owner and other subjects in the image. Malicious attackers can take advantage of these unnecessary leaks to launch context-aware attacks (e.g., spearfishing) or even impersonation attacks. This project is developing methods to help users appropriately control access to their shared images.The investigators are developing new techniques to tackle image privacy based on the image content as well as images and users' meta-data, by (a) inferring the sensitivity of a given image based on the visual properties of the images and the users' image sharing patterns, and then automatically applying the appropriate privacy settings for that image, and (b) by using discovered users' sharing patters to define access policies according to the locally enforceable controls on the domain of interest. Beyond the technical results, this project is developing a framework for understanding the common classes and categories of images with respect to the users' understanding of privacy. This framework will enable an understanding of the images that need protection and help improve user awareness of classes of images that are often unintentionally or accidentally under-protected.