Research On Efficient Image Retrieval Method For Encryption Domain Based On Deep Hashing

With the rapid development of the Internet,the rapid popularization of smart digital devices,and the sharp increase in the amount of image data,more and more data owners choose to outsource the storage and retrieval of local data to cloud servers.Since the image contains a large amount of user privacy data,there is a serious risk of privacy leakage.Therefore,exploring safe and efficient content-based image retrieval technology has become an important research topic in the field of multimedia information retrieval.The thesis mainly aims at the problem that the existing content-based encrypted domain image retrieval methods are difficult to achieve balance in terms of security,retrieval efficiency,and retrieval accuracy.Using key technologies such as deep hashing,image encryption,security index,and similarity measurement,the content-based encrypted domain image retrieval method is studied.The main research work is as follows:1.To improve the expressiveness of image features and realize safe and efficient image retrieval,a lightweight hyperchaotic image encryption method is designed,and an encrypted image retrieval method based on Feature Fusion Deeply Supervised Hashing(FFDSH)was proposed.Firstly,the encryption key is generated by using the key sequence generated by 4-dimensional hyperchaos and PRESENT algorithm,and the image is encrypted.Secondly,the convolutional neural network CNN and the visual geometry group VGG16 are used to extract the deep features of the image,and the two features are fused using feature fusion technology.Finally,the fusion feature is used to construct the deep hash sequence,and it is used as the FFDSH sequence.When searching,perform a double-layer similarity matching search according to the extracted FFDSH sequence to obtain the required image information.Experimental results show that compared with existing methods,the proposed method can achieve efficient image retrieval,and has obvious advantages in terms of security and retrieval performance.2.To achieve safe and efficient retrieval of image data in the cloud,extract more expressive image features and construct a secure searchable encryption method,a secure image retrieval method based on deep hashing and searchable encryption was proposed.Firstly,a deep learning framework based on transfer learning and residual network is designed to extract image features,and complete the construction of deep supervised hashing.Second,a central similarity measure and Paillier homomorphic encryption are used to construct a secure searchable index.Finally,according to the homomorphic characteristics of Paillier homomorphic encryption,a similarity measure method suitable for computing in the ciphertext is designed.Experimental results show that the proposed method has high security and retrieval accuracy,and can realize efficient image retrieval without revealing user privacy.Compared with the traditional hashing method,the retrieval accuracy is increased by at least 37%,and the retrieval time is saved by at least9.7% compared with the latest deep hashing method.3.To achieve safe and efficient large-scale image data retrieval on cloud servers and solve the problems of poor feature representation and low security in content-based encrypted image retrieval systems,an encrypted image retrieval method based on deep hashing and secure KNN was proposed.This method first introduces the instance normalization operation to improve the residual block;then uses the convolutional attention mechanism of the first channel and then the space to enhance the objects and regions that need attention,to obtain a feature index with stronger expressive ability;Finally,secure KNN is used to encrypt the index sequence to realize secure retrieval.Experimental results show that the security of the proposed method is guaranteed,and it has higher retrieval accuracy compared with existing retrieval methods.