Image Privacy Detection And Protection

The development of science and technology facilitates the generation of image and makes image widely used,which prompts more and more images to be shared and applied in multiple ways,such as online social networks like Facebook and Flickr,or Data trading platforms like Datatang.However,images contain a lot of privacy infor-mation.If proper privacy protection cannot be implemented,users will face the risk of privacy leakage.Although current social networks allow users to set their privacy options,whether it is setting privacy or filtering privacy images is a tedious task for most users,and when manual operations are not appropriate,image sharing may lead to unnecessary privacy disclosure or privacy violations.Therefore,it is very important to design an automated privacy image protection method.Based on the requirements of image privacy protection in the context of data shar-ing,we focus on the research of image privacy protection.First,we design a question-naire and analyze the characteristics of users' privacy of images based on 417 feedback studies.Then,we design a fine-grained,personalized and interpretable privacy image detection algorithm,considering the deficiencies of the existing work and the user's subjective definition of privacy.The user-understandable multi-layer graph structure is used as a feature representation of the image,and rules are used to describe the user's pri-vacy definition.The user's personalized image privacy definition is constructed through rule learning,and the granularity of privacy detection is refined from the file level to partial image.Finally,we design a privacy image protection algorithm that minimizes the loss of image utility,and converts the balance between image privacy protection and image utility into a combined optimization problem.We collected 8744 images from 20 users to form the personal dataset and crawled the PicAlert dataset on Flickr to jointly verify our algorithm.The results verify the effectiveness of our design to detect and protect personalized image privacy.