Research On Privacy Decision Of Sharing Images In Social Networks Based On Questionnaire And Deep Learning

With the development of social networks,social intercourse with images has gradually become a feature of OSN(Online Social Network),and people are tending to share their images in OSN.However,these images often contain user's personal privacy information.Once the information is used by criminals or illegal organizations,it will affect the user's daily life,and worst of all,it will threaten the personal security of the users and their family.As social network image privacy becoming such an important problem,there have been some studies on privacy issues when sharing image in ONSs.However,these work have deficiencies:(1)There are already some works abroad conducted investigations and studies on privacy issues for users in foreign OSNs,and have obtained some conclusions on image privacy issues.However,there is a lack of similar work in our country.(2)Digital image privacy decision(in OSNs)is the method that provides sharing or privacy protection strategies for users when sharing their images in ONSs.At present,there are some work that research on image privacy decision models of ONSs.However,the difficulty of understanding the image privacy semantics,and the difficult to defining the image privacy,leading to these decision models cannot provide a good image privacy policy.In response to the above issues,the main research content of this dissertation includes:1)A questionnaire survey was conducted on the privacy decision-making and access control issues for OSNs'users in China when sharing image in social network.We designed a detailed questionnaire for domestic social network image sharing issues,divided it into four major parts,and published it on a questionnaire survey platform.Finally,we statistically analyzed the results,and obtained conclusions,such as there exist partial ordering among semantic tags and security levels,for the sharing and privacy protection of images in ONSs.The conclusions provide a theoretical basis for the further work of ONSs.2)A social network image privacy decision model was proposed based on deep learning.Firstly,we defined the privacy relationship of images by borrowing the idea form the technique of image visual relationship detection.Then we gave the procedure of the model for image privacy decision.What's more,we solve the key technique involved in the privacy decision model,which is the detection of image privacy relationship.We retrain the Faster RCNN model through Finetuning technology and privacy predicate image database to obtain the predicate detection model which is a part of the image privacy relationship detection.The user of this framework has few manual operations and is relatively simple compared to the state-of-the-art social network privacy policies,which can alleviate the serious problem of leakage of image privacy OSNs.