Research On Encrypted Image Retrieval Based On Multi-feature Fusion And LSH

With the continuous application and development of mobile devices,the storage demand for visual data such as images is increasing,and it has become a mainstream trend for outsourcing image data to the cloud server.In the cloud server,the image owner does not need to store the image resources locally,while the user only needs to obtain the authorization and access to the required image resources quickly,which brings great convenience to the user and the image owner.Due to the frequent occurrence of irmage resource privacy leaks,the security issues also pose challenges to the cloud server.In order to ensure the storage security of images,image resources cannot be directly outsourced to the cloud server,and images need to be encrypted and then upload to the cloud server.However,the retrieval of encrypted images usually is a difficult problem.Therefore,how to search images efficiently is a very challenging task in cloud computing.For existing problems of encrypted image retrieval in the cloud environment,we mainly studies the encrypted image retrieval technology based on multi-feature fusion and locally sensitive hash.The main contributions of this paper are as follows:(1)An encrypted image retrieval method based on deep learning and adaptive weighted fusion is proposed.Firstly,combined with deep learning,DenseNet is used to extract the depth feature of images.At the same time CLD(Color Layout Descriptor),EHD(Edge Histogram Descriptor)and BOW(Bag of words)features are also extracted and binarizted.Then we obtain the fusion features by the adaptive weighted fusion of the four kinds of features and reduce the dimension of fusion features by PCA.Finally,the pre-filter tables for the fusion features are constructed with the locality-sensitive hashing to improve the search efficiency.The images and indexes are protected by the secure k-nearest neighbor(kNN)algorithm and logistic encryption method.The experiments show that the mehod can effectively improve the retrieval accuracy and time of encrypted images while ensuring image security.(2)An encrypted image retrieval method searchable encryption method in cloud computing environment is proposed.First we encrypt the image of R,G,B three channel by the encryption operator,then design a secure adaptive fusion feature extractor,finally upload the encrypted image and the feature extractor to the cloud server directly,the cloud server can obtain the encrypted image's RGB,HSV and YUV histogram feature weights,and the fusion feature vector by feature extractor.The cloud server can directly compare the Euclidean distance between two feature vectors to calculate the similarity between images.Experimental results show that compared with the existing encryption retrieval scheme,the proposed method not only improves the security intensity of outsourcing image,but also improves the retrieval performance.