Secure Index Construction for Privacy-Preserving Large-Scale Image Retrieval

How to efficiently retrieve the images while preserving the user's privacy has gradually become a key problem in some applications such as Cloud storage, social networks. In this paper, a secure index used for image retrieval is constructed to protect the retrieval results being leaked to the malicious attackers. At first, inverted index is generated using visual words of images and then encrypted dually by randomized binary encoding and a key-based Gaussian random matrix respectively, generating a secure index. The similarity of the encrypted indexes is measured directly in the encrypted domain without decryption and the most similar k images will be returned as the retrieval results to the users. Experimental results and security analysis show that the secure index proposed in this paper can achieve comparable retrieval performance to the conventional large scale image retrieval method, while protecting user's privacy.