Research On The Key Technologies For The Release Of Private Image Data

With the development of the information age,the importance of data has gradually emerged in the modern era.Data-based services are widely used in all fields of society,one of which is based on image data.However,image publication in a direct way may lead to privacy leakage,because images are inherently sensitive.When user provides image data to a third-party application for data analysis,the user's sensitive information may be leaked.To solve the above problem,a series of solutions have been proposed.While,recent work in differential privacy has shown that it is possible to release image data while ensuring strong privacy guarantees.Differential privacy is a new way to protect data privacy,it has strong privacy protection capability and rigorous statistical model which make it widely researched and applied.In the recent years,k-anonymity,access control,and privacy encryption are employed to protect the privacy caused by releasing facial image.For these related technologies,if the attacker has certain background knowledge and passes the set access restrictions,the user's image and corresponding private information can be obtained.In response to these shortcomings,this paper adopts differential privacy to protect the privacy of image data.This paper,therefore,focuses on releasing facial image with differential privacy.For this objective,two differentially private releasing solutions are proposed,one of which is based on Fourier transform combined with differential privacy,and the other relies on the skill of Matrix Decomposition.The first solution employs the Fourier transform to compress the image data,which solves the low data utility caused by the large image size,and effectively controls the amount of noise.The second method uses the Matrix Decomposition,which can reduce the sensitivity of releasing image,and boost the accuracy of the publication.Both theoretical analysis and extensive experimental results show that the two methods outperform their competitors on data utility.