Research On Efficient Large-scale Encrypted Image Retrieval Technology

With the rapid development of the Internet and the widespread popularity of multimedia devices,the number of images on the Internet are increasing explosively.To reduce storage and maintenance costs,more and more image owners are willing to outsource image datasets to public clouds.To protect the privacy of images,images should be encrypted before outsourcing.However,this makes it difficult for public clouds to provide content-based image retrieval services.How to achieve efficient encrypted image retrieval has become a research focus in recent years.Research scholars have proposed many encrypted image retrieval schemes.Although most existing schemes can achieve efficient encrypted image retrieval,their image owners consume a lot of time and computing resources when building secure indexes.This is not practical for image owners with limited resources.Other schemes allow public clouds to extract image features from encrypted images and build indexes,thus greatly reducing the resource consumption of image owners.However,these schemes can only use low-level image features to achieve encrypted image retrieval,which is not practical for large-scale encrypted image retrieval.Aiming at the above problems,this paper studies efficient large-scale encrypted image retrieval technologies.The main contributions of this paper are:1.This paper proposes an efficient large-scale encrypted image retrieval scheme which allows image owners to fully outsource the task of secure index construction to public clouds.In order to obtain good retrieval accuracy,the scheme employs fine-tuned convolutional neural networks to extract image features.In order to protect data privacy,the scheme employs AES algorithm to protect images,and uses asymmetric scalar-product preserving encryption algorithm to protect the image features.In order to improve retrieval efficiency,the scheme allows cloud servers to directly use the encrypted image features to build a secure hierarchical index graph.Participation from image owners is not needed during the construction and update of the secure index,so this scheme is suitable for image owners with limited resources.2.This paper not only theoretically analyzes the security and efficiency of the proposed scheme,but also gives the experimental results and comparative analysis of the proposed scheme and a recently proposed encrypted image retrieval scheme on the Cifar-10 dataset.The experimental results show that the proposed scheme is feasible,and it performs better than the existing similar scheme on the construction and update of secure indexes.