Research On Privacy Preserving Content-based Image Retrieval

With the rapid development of imaging devices,number of images grows explosively.How to find desired images quickly and accurately from massive image database through image contents has become a research focus.Cloud computing provides huge storage capacities and powerful computing abilities for resource constrained users.By saving large number of images in the cloud,storage costs decrease and convenience increases,such method has been already a mainstream scheme.However,some problems exist,for example,when the data owner outsources CBIR services to the cloud server,the data owner will lose control of data which contains his private information.Once the data is attacked,the private information faces risks to be leaked.Detailed research which aims at above-mentioned problems has been given.In chapter II,required theory basis is demonstrated form different perspectives,such as CBIR features,SIFT algorithm extract characteristics.In chapter III,a kind of SIFT extract method with function of privacy protection is proposed based on two-side cloud server cooperation.In chapter IV,a kind of secure index construction plan is proposed based on Secure K-Dimension Tree(SKDTree),meanwhile,a series of security protocols in process of feature extraction and index construction are proposed.In chapter V,a kind of image retrieval scheme with privacy protection is proposed based on security thread model.Main research results are summarized as follows:1.A secure image retrieval scheme based on SKDTree is proposed.This scheme supports secure image feature extraction and secure index structure SKDTree construction.Image retrieval can be completed under circumstance of privacy protection on image data.Based on the scheme,data owner encrypts image before uploading,user retrieves the image in the encrypted images which effectively protects the privacy of the image.High speed and efficiency of image retrieval are also guaranteed.2.Several secure protocols for feature extraction and index construction are proposed,such as homomorphic encryption-based secure multiplication protocol,Garbled Circuit-based secure comparison protocol,secure Euclidean distance computing protocol,etc.The results are calculated under the intermediate results calculated on the server through security interaction between cloud servers,which effectively protects the data security and the secure sharing of data.Detailed security analysis of those protocols is given,which proves data privacy security.The above-mentioned research results preliminarily solve problems of image retrieval security in the case of data outsourcing applications.Update and maintenance of index structure are issues which need to be further discussed.