Research On Block Features Based Encrypted Image Retrieval Scheme

With the rapid development of technology,some content-based image retrieval(CBIR)schemes researchers put forward to use image information efficiently.However,because of the huge amount of images,it is difficult for users to store and search images by themselves.With the development of cloud computing,users can outsource their works to the cloud server.However,if the plaintext image is directly outsourced to the cloud server,it may disclose some private information.Therefore,the research on ciphertext CBIR came into being and got some achievements.However,there are still some problems,such as encrypt may leak information or the retrieval effect is not ideal.Based on the extraction of encrypted image features,we study encrypted image retrieval schemes based on random mapping and deep learning.It can reduce the burden of users,protect image security and support image retrieval.The research of this paper is as follows:An encrypted image retrieval scheme based on random mapping is proposed.Encryption methods used in existing schemes may not secure,for example,value substitution may disclose some image information.In order to enhance security,the plaintext image is encrypted by AES and block permutation.Random mapping is used to extract local features.The dimension of the features is determined by the number of templates.Considering that the difference of features may affect retrieval efficiency,the activation function is used to constrain the features.After that,the feature will be extracted by the bag-of-words(BOW)model to represent the image.Finally,the Manhattan distance between features is calculated to compare similarity.Experiments show that AES and block permutation enhances image security,and random mapping features from encrypted images can retrieve similar images.An encrypted image retrieval scheme based on deep learning is proposed.In most schemes,image encryption is carried out in the spatial domain,and the dependence between pixels may be destroyed,so the encrypted image cannot be compressed well.The JPEG image can be encrypted by stream encryption and scrambling encryption,which can protect the image content without expanding storage space.The intermediate data is parsed from the encrypted image,and the variable-length integer(VLI)code is extracted to construct a matrix.Through the powerful learning ability of the neural network,the framework of deep learning is constructed to analyze the image information and extract features.Finally,similar images are retrieved by Euclidean distance between features.The scheme ensures the security of the JPEG image and achieves good results in retrieval accuracy.