Research On Multi-table Permutation Encryption Based Image Retrieval Scheme

With the widespread use of multimedia devices,the number of multimedia data based on digital images is growing rapidly.To retrieve desired image from huge database,Content-Based Image Retrieval(CBIR)scheme has great development.Owing to the limit storage and computing ability of local device,many users choose to outsource CBIR work to the cloud server.However,it is not safe to store plain images on cloud servers,so secure CBIR has extensively research.In most of secure CBIR schemes,image users undertake most of work to extract features and establish index.This paper presents the image retrieval scheme based on multi-table permutation.Image features are extracted from encrypted images to reduce the workload of image users.The main idea of this paper is as follows.To address the problem that permutation encryption can't resist statistical attack,a scheme based on multi-table permutation encryption and global histogram is proposed.In this scheme,discrete cosine transform(DCT)is firstly performed on the color component to obtain DCT coefficients.Then,stream encryption technology,multi-table permutation and scrambling encryption are used to protect image content.Finally,AC-coefficients and color histogram are extracted from the encrypted image as global features,and the index between encrypted image and responding feature vector is established.The similarity between images is measured by calculating the Manhattan distance of feature vectors.Compared with previous scheme,the scheme improves image security and ensures retrieval accuracy.To address the problem of low retrieval accuracy in encrypted image retrieval,a scheme based on multi-table permutation encryption and local color histogram is proposed.On the basis of Bag-Of-Features(BOF)model,we propose the Bag-of-Encryption-Features(BOEF)model to extract image feature in encrypted domain.In this scheme,multi-table permutation and block scrambling encryption are used to protect image.Then color histograms are extracted from image block as local features,and all local features are clustered to form encrypted vocabulary book.The word frequency histogram based on the encrypted vocabulary book is calculated as image feature.The similarity between images is measured by calculating the Manhattan distance of feature vectors.This scheme proposes a BOEF model to extract local features from encrypted images,which further improves the retrieval accuracy.This paper proves that multi-table permutation is more secure than previous value permutation,and BOEF scheme has better retrieval performance.