Research On The Encrypted Image Search Of Mobile Terminal Users On The Cloud Server

In recent years,with the development of mobile terminal digital imaging technology and mobile Internet,a large number of digital images have been created and applied in various fields.And with the rise of cloud services,more and more users who are limited by the memory resources of mobile terminals tend to store digital images in the cloud server,but the problem of personal privacy comes with it.In view of the problem that the information transmitted in the public channel is not secure and the cloud server is not trusted,a more effective solution is to encrypt the image before uploading it.However,encrypted images cannot be retrieved in a conventional way.How to retrieve the target image from the encrypted image safely,efficiently and accurately is a hot topic in current academic research.The main research work and contributions of this paper are as follows:1.In the existing image encryption methods,most of the authors pay too much attention to the security of the algorithm,focusing on improving the statistical testing,differential attack analysis and resistance to known plaintext attacks.However,the complexity of the encryption scheme,the use of memory and algorithm throughput are not given enough attention.In this paper,a lightweight image encryption algorithm is proposed.Logistic chaotic iteration is used to generate chaotic sequence,which provides random number for subsequent image pixel scrambling,DNA encryption and pixel confusion.In addition,the logistic homogenization algorithm is used to overcome the inherent uneven distribution of Logistic chaotic sequence.In addition,this paper also gives the index of the algorithm in the aspect of security testing.Experiments show that the algorithm can well resist all kinds of security attacks,and the complexity of the algorithm is low.2.The similarity between images can be measured by the feature distance of images,so mobile terminal users can use their own feature values to compare with the image features on the cloud server to achieve image retrieval.However,image features are also sensitive data for a user,which can not be transmitted on the common channel,otherwise it is easy to be obtained by a third party.Therefore,this paper proposes a lightweight non-interactive image privacy audit protocol.Using this protocol,the mobile user and the cloud server can get the Hamming distance between the two features without transmitting the plaintext image feature values to each other.In addition,this paper demonstrates the correctness of the protocol on the mathematical level,and proves that the protocol has the characteristics of privacy protection,content privacy and audit.Finally,this paper proves the high efficiency of the protocol through experiments,especially the mobile party in the protocol interaction,only needs to run a small amount of mathematical operations to complete the protocol.3.In view of the shortcomings of existing deep hash methods in model volume and operation speed.This paper presents an image depth hash algorithm based on binary neural network.Inspired by deep pairwise supervised hashing(DPSH),the algorithm firstly replaces the original CNN-F feature extraction network with VGG-11 network in the original model,which improves the accuracy of the model.The improved DPSH network is quantized by binary technology,which can greatly reduce the volume of the model and accelerate the forward propagation speed of the model under the condition of slightly reducing the accuracy of the model.Finally,through the experiment,it is found that although the accuracy of the binarization model decreases,the model size and operation speed are improved,which is very suitable for mobile terminals and embedded devices.