[Paper Review] Outsource Photo Sharing and Searching for Mobile Devices With Privacy Protection

Lan Zhang, Taeho Jung+4 more|arXiv (Cornell University)|Oct 24, 2014

Advanced Image and Video Retrieval Techniques12 references3 citations

TL;DR

This paper proposes SouTu, a privacy-preserving framework enabling mobile users to outsource photo sharing and searching to untrusted cloud servers without exposing image content, queries, or search results. It uses novel non-interactive outsourced vector distance protocols for real and binary feature vectors, achieving low communication and computation overhead with only 6% more storage and 27% less communication on some datasets compared to non-private systems.

ABSTRACT

With the proliferation of mobile devices, cloud-based photo sharing and searching services are becoming common due to the mobile devices' resource constrains. Meanwhile, there is also increasing concern about privacy in photos. In this work, we present a framework \ourprotocolNSP, which enables cloud servers to provide privacy-preserving photo sharing and search as a service to mobile device users. Privacy-seeking users can share their photos via our framework to allow only their authorized friends to browse and search their photos using resource-bounded mobile devices. This is achieved by our carefully designed architecture and novel outsourced privacy-preserving computation protocols, through which no information about the outsourced photos or even the search contents (including the results) would be revealed to the cloud servers. Our framework is compatible with most of the existing image search technologies, and it requires few changes to the existing cloud systems. The evaluation of our prototype system with 31,772 real-life images shows the communication and computation efficiency of our system.

Motivation & Objective

• Address the growing privacy concerns in cloud-based photo sharing and searching, especially regarding automated image recognition and unauthorized access.
• Enable mobile users to outsource heavy storage and search tasks to the cloud while preserving privacy of photos, queries, and search results.
• Design a system that supports efficient, secure content-based image search on encrypted feature vectors without requiring online interaction between owners and queriers.
• Ensure compatibility with existing image search technologies and cloud infrastructure to enable practical deployment.

Proposed method

• Design a modular framework, SouTu, that separates privacy protection from core image search functionality, allowing seamless integration with existing cloud services.
• Develop two non-interactive outsourced vector distance computation protocols—one for real-valued feature vectors and another for binary vectors—enabling secure distance evaluation without interaction between the owner and querier.
• Use secure encoding of image descriptors (e.g., SIFT, binary descriptors) such that only authorized users can compute similarity, while the cloud remains oblivious to content.
• Introduce an optimized variant, SouTu_bin, which uses binary feature vectors to halve the computation cost at the expense of minor accuracy loss.
• Implement k-n oblivious transfer to hide the actual search pattern, ensuring that the cloud cannot infer which images are being queried.
• Apply compression techniques to reduce communication overhead for uploading queries and downloading results, especially for large-scale image databases.

Experimental results

Research questions

• RQ1 How can photo sharing and searching be securely outsourced to untrusted cloud servers while preserving the privacy of image content and search queries?
• RQ2 Can efficient, non-interactive secure computation protocols be designed for vector distance evaluation in image search, avoiding the need for online interaction between users?
• RQ3 To what extent can privacy-preserving image search be made compatible with existing image search systems and feature extraction methods?
• RQ4 What is the performance overhead (communication, computation) of the proposed privacy-preserving framework on mobile and cloud platforms?

Key findings

• The communication cost of SouTu is only about 6% higher than a non-private system for the LFW dataset, and 27% lower for the Holiday dataset due to efficient compression.
• Query upload overhead is reduced to 350B (SouTu) and 90B (SouTu_bin) after compression, compared to 36KB and 9KB respectively before compression.
• The cloud-side computation for Euclidean distance takes 0.18 seconds for real vectors (SouTu) and only 0.018 seconds for binary vectors (SouTu_bin), enabling sub-minute response times on laptops.
• For typical photo collections (e.g., hundreds of face images per user), a single search request can be processed in under one minute on a laptop, with performance significantly better on powerful cloud servers.
• The framework achieves strong privacy guarantees: neither the cloud nor any untrusted party can learn the content of images, queries, or search results.
• The system is transparent to users—privacy protection is automatic and requires no changes to user behavior or existing image search workflows.