Research On Content-based Image Security Retrieval Technology

With the popularity of the Internet,the amount of multimedia data is growing rapidly.Content-based image retrieval(CBIR)can search out similar images from large-scale images and realize the utilization of data.For data owners,outsourcing the management and maintenance of image data to cloud service providers can effectively reduce costs,but there is a problem of privacy disclosure.The research on image privacy protection methods in data outsourcing scenario is becoming more and more active.Many image security retrieval schemes have been proposed one after another.However,these schemes have many problems,such as large amount of client computing,large number of interaction rounds between image owner,cloud server and query user,and large communication cost,so it is difficult to apply them directly.In the actual environment,the same service can be supported by multiple ECs.In this thesis,content-based image security retrieval is the research topic,focusing on image feature extraction,index design and image retrieval methods under the dual server model.The main contents are as follows:1.An image security retrieval scheme based on BOVW feature is proposed.This scheme combines SIFT feature security extraction and local sensitive hash algorithm to achieve the security extraction of image BOVW features.In order to protect the BOVW feature of image,an inverted index based on word frequency division is designed to store the index on different server.Finally,the corresponding image retrieval method is designed.This scheme reduces the participation of image owners and query users in the retrieval process to the maximum extent.2.A secure image retrieval scheme based on CNN features is proposed.Based on the feature extraction of image CNN,an extensible hash index based on dimension partition is designed.Firstly,the image CNN features are indexed in different dimensions to reduce the risk of privacy disclosure of cloud target image data pattern.Then,the bucket splitting mechanism is introduced to avoid the impact of uneven data distribution on retrieval performance,which can support a large number of image updating scenarios.At the same time,the corresponding image retrieval algorithm is designed.3.The security and performance of the proposed scheme are analyzed theoretically and verified by experiments.Based on different image data sets,the influence of different parameters on the scheme performance is tested,and the optimized parameters are given.The experimental results show that compared with the existing schemes,the scheme proposed in this thesis can effectively reduce the participation of clients,and take less time in index building and image retrieval.