Research On Content-based Privacy-preserving Multi-source Outsourced Image Retrieval Technology

With the rapid development of Internet of Things(IoT)equipment,multimedia technology and computer vision,more and more images appear in our daily life.A large amount of image data provides abundant resources for our lives.In the meantime the emergence of a large number of image data increases the burden of local devices.Traditional storage devices and management methods are difficult to meet the new requirements.In the face of numerous and complex massive images,cloud technology has gradually penetrated into our lives,and cloud storage technology is utilized to store and manage a mass of data.As the increasing application of image processing technology in various fields,the demand for image retrieval is increasing.The existing text-based image retrieval system has gradually become impractical.First of all,the previous scheme only considers single source scene.However,in real life,image retrieval is more likely to involve multiple image sources.Secondly,with the increasing demand for search speed,existing work performing search operations in massive images is no longer able to meet practical needs.Therefore,with a view to solving the above problems,this thesis carries out a research on content-based multi-source image retrieval.First,the image owners extract the feature vector set locally,then encrypt and upload it to the cloud server.When the cloud server receives the search request from the authorized user,it performs image retrieval and returns the ID of the search result to the corresponding users.Finally,the users encrypt the corresponding images and sends them to the query user,So as to complete the image retrieval work.The specific research contents of this thesis are as follows:(1)A multi-source image retrieval scheme based on Local Sensitive Hashing(LSH)algorithm.In this study,first of all,each image owner extracts the image feature vector set locally,and then encrypts the image features through the improved secure multi-party summation protocol in this chapter,which can effectively support the similarity calculation of encrypted features collected from multiple sources.In addition,each image owner uses LSH algorithm to create a search index and upload it to the cloud server.The cloud server first summarizes the index tables uploaded by the owners to generate a total search index table,and then perform search operation.The use of LSH algorithm reduces the search scope and improves the search efficiency.At the same time,security analysis shows that the privacy of images and matching results is also well protected.(2)A multi-source image retrieval scheme based on k-means clustering algorithm.In this study,a dual-cloud server system is built.Each data owner first uploads the feature vector set that extracted locally to the two cloud servers through secret sharing.Then,the two cloud servers interact to cluster the encrypted image features through k-means clustering algorithm.In the search process,the cloud servers first calculate the distances between the query image and the cluster centers to select the nearest cluster,and then the distances between the feature vector in the nearest cluster and the query feature vector are calculated in order to return similar images.Finally,the experimental evaluation results show the effectiveness of our scheme.