Research On Cryptography Techniques For Encrypted Image Retrieval

With the rapid development of digital imaging technology and Internet technology,a large number of digital images are created and used in various fields.As resources limit many user terminals,more and more individuals or businesses tend to store image data on a cloud environment to save local storage and computing costs.Through the cloud environment,the majority of users can upload images to the cloud server anytime,anywhere.But users do not know the specific storage location of uploaded images in the cloud server,which raises users' concerns about image privacy leaks.For untrusted cloud servers,an effective method is to encrypt images before they are uploaded,but this will significantly reduce the operability of images.Besides,to meet actual efficiency requirements,many solutions will leak some information to the cloud server,and attackers can use this leaked information to destroy the confidentiality of data sets or queries.Therefore,how to search for images safely,efficiently,and accurately is an urgent problem that current research institutions and scholars need to solve.The main research work and contributions of this thesis are as follows:1.At present,there are some secure and efficient image retrieval schemes,but few of them can hide both k-NN response leakage and response length leakage.For data recovery attack against k-NN response leakage,this thesis uses the multi-level homomorphic encryption protocol to construct a secure multi-party computing model to realize the calculation of security similarity between encrypted feature vectors.Besides,the existing hidden response length leakage schemes are mostly based on simple padding or truncation.This thesis proposes a database padding algorithm that combines optimized padding and centralized multi-map to reduce the redundant storage of data while achieving response length hiding.2.Because of the expansion of image features after encryption,this thesis uses the compressed sensing measurement matrix to conduct compressed sensing sampling,compression,and encryption of image features.To further improve the security of the image retrieval scheme,this thesis organically combined the compressed sensing theory and Paillier encryption algorithm,and a secure binary embedding based on quantized random projection is used to achieve safe and efficient image retrieval.Furthermore,a similar image set of the query is returned by executing an n-out-of-k oblivious transfer protocol,which effectively protects the privacy information of image users and image databases.Finally,the security of this thesis proposed schemes is analyzed and proved,and the online public image data set is used for experimental verification.Theoretical analysis and experimental results show that the two secure and efficient image retrieval schemes proposed in this thesis have excellent retrieval performance and high security.