Searchable Encryption And Secure Classification On Encrypted Image Data

In recent years,the development of cloud computing technology has improved rapidly.Due to the storage capability,powerful computational ability,and convenient services provided by cloud servers,there have been more and more users willing to carry out large-scale data processing on cloud servers,such as image classification and retrieval.The sharing and retrieval of medical data via cloud systems can facilitate timely medical/disease diagnosis.However,cloud servers or malicious adversaries may violate the privacy of users and the leakage of medical resources leads to serious consequences.Therefore it is of practical significance to propose a secure classification and retrieval scheme on encrypted medical data.Searchable Encryption(SE)technology is one of the effective solutions to this kind of problem.However,existing SE schemes have limitations on encrypted medical images.Therefore,a Privacy-Preserving Image Searching(PPIS)scheme is proposed in this paper.PPIS scheme designs several primitives using homomorphic encryption to construct a secure Convolutional Neural Network(CNN)model,which securely extracts the feature vector of encrypted query image and provides secure and accurate inference and similar searching.In practical applications,the query and the results usually appear with different representations(or modalities,such as image,audio,video,and text),and maintaining privacy is crucial in our privacy-aware and interconnected society.Therefore a Secure Cross-Modal Retrieval(SCMR)scheme is also proposed in this paper.SCMR scheme utilizes Collective Matrix Factorization(CMF)to construct a common subspace for heterogeneous data and obtain the mapping matrices,which are encrypted and outsourced for the cloud server to securely calculate the feature of encrypted query and find relavant multi-modal results.The above two schemes are evaluated by theoretical analyses and experiments,which demonstrate that these two schemes can not only protect data privacy,but also perform classification and retrieval tasks accurately and efficiently.