A Secure CBIR Method based on Bag-of-Visual-Words Model under Cloud Environment

Abstract

Cloud computing platform has powerful computing capacity and nearly infinite resource pool, which provides a strong guarantee for mass data storage and computing. Considering double security threats from malicious external attackers and "honest-but-curious" CSP (cloud service provider), users need to encrypt images to ensure the data security before outsourcing images to cloud. But encryption can have an impact on the necessary data services, such as content based image retrieval (CBIR). A secure CBIR method based on BoVW (Bag of Visual Words) model under cloud environment is proposed in the paper. Images are expressed as frequency histogram by BoVW model, orthogonal decomposition is utilized to divide it into two individual parts of component coefficients thus encryption operation and feature extraction operation can be executed separately, and orthogonal composition is used to fuse the encrypted operation results to construct secure image index. After encrypted image index and encrypted images are outsourced to CSP, distance comparison can be executed by CSP on feature extraction field without violating data privacy. Encrypted images with the closest distance to query trapdoor are returned to users to decrypt and obtain plain images. Any encryption algorithms can be used to encrypt images and search index by using orthogonal transformation, so that the proposed method is practicable. Retrieval precision is improved and better performance are achieved by using BoVW model. The security analysis and experimental results show our scheme has obvious advantages in security and retrieval performance.