Practical Privacy-Preserving Content-Based Retrieval

Abstract

Storage requirements for visual data have been increasing in recent years, following the emergence of many highly counteractive multimedia services and applications for mobile devices in both personal and corporate scenarios. In this paper we propose a secure framework for outsourced privacy-preserving storage and retrieval in large shared image repositories. Our proposal is based on IES-CBIR, a novel Image Encryption Scheme that exhibits Content-Based Image Retrieval properties. The framework enables both encrypted storage and searching using Content-Based Image Retrieval queries while preserving privacy against honest-but-curious cloud administrators. In this projects allows more efficient operations than existing proposals, both in terms of time and space complexity, and paves the way for new practical application scenarios.

Existing System

The existing system focuses on content-based multimedia retrieval over encrypted databases, where both the query and database documents are encrypted and their privacy is protected. The techniques proposed in the existing system enable ancient retrieval directly in the encrypted domain, without multiple rounds of communications between the user and the server. We demonstrate the proposed techniques using images, although these techniques are applicable to other multimedia modalities such as video. By analysing the requirements of secure retrieval scenarios, we propose two secure indexing schemes built upon visual words representation of images. In this scheme makes use of inverted indexes of visual words and the second scheme exploits randomized hash functions. Both indexing schemes achieve image retrieval and are scalable for large databases. We jointly exploit cryptography, image processing, and information retrieval techniques to ensure that the encrypted search indexes can preserve the search capability.

Proposed System

Our proposal is based on IES-CBIR, a novel Image Encryption Scheme that exhibits Content-Based Image Retrieval properties. The framework enables both encrypted storage and searching using Content-Based Image Retrieval queries. Images are outsourced to repositories that reside in the cloud. Each repository is used by multiples Users, where they can both add their own images and/or search using a query image. Each repository is created by a single user. Upon the creation of a repository, a new repository key is generated by that user and then shared with other trusted users, allowing them to search in the repository and add/update images. In this work, we use the Bag-Of-Visual-Words (BOVW) representation to build a vocabulary tree and an inverted list index for each repository. We choose this approach for indexing as it shows good search performance and scalability properties. In the BOVW model, feature-vectors are hierarchically clustered into a vocabulary tree (also known as codebook), where each node denotes a representative feature-vector in the collection and leaf nodes are selected as the most representative nodes (called visual

Architecture

Architecture Diagram