[Paper Review] Secured Outsourced Content Based Image Retrieval Based on Encrypted Signatures Extracted From Homomorphically Encrypted Images

Reda Bellafqira, Gouenou Coatrieux+3 more|arXiv (Cornell University)|Apr 3, 2017

Advanced Steganography and Watermarking Techniques21 references3 citations

TL;DR

This paper proposes a novel Secure Outsourced Content-Based Image Retrieval (SOCBIR) scheme that enables cloud-based image search on homomorphically encrypted images without requiring additional communication between the user and server. By extracting encrypted wavelet-based global signatures directly from Paillier-encrypted images and using a novel fast comparison method for encrypted data, the approach achieves retrieval performance identical to unencrypted CBIR systems while ensuring full data confidentiality and zero interaction during search computation.

ABSTRACT

In this paper, we present a novel Secured Outsourced Content Based Image Retrieval solution, which allows looking for similar images stored into the cloud in a homomorphically encrypted form. Its originality is fourfold. In a first time, it extracts from a Paillier encrypted image a wavelet based global image signature. In second, this signature is extracted in an encrypted form and gives no clues about the image content. In a third time, its calculation does not require the cloud to communicate with a trusted third party as usually proposed by other existing schemes. More clearly, all computations required in order to look for similar images are conducted by the cloud only with no extra-communications. To make possible such a computation, we propose a new fast way to compare encrypted data, these ones being encrypted by the same public key or not, and using a recursive relationship in-between Paillier random values when computing the different resolution levels of the image wavelet transform. Experiments conducted in two distinct frameworks: medical image retrieval with as purpose diagnosis aid support, and face recognition for user authentication; indicate that the proposed SOCBIR does not change image retrieval performance.

Motivation & Objective

• To address the growing need for privacy-preserving image retrieval in cloud-based medical and biometric applications.
• To eliminate the need for trusted third parties or interactive communication during image search in encrypted databases.
• To enable secure extraction and comparison of image signatures directly from homomorphically encrypted images.
• To maintain retrieval accuracy equivalent to unencrypted CBIR systems despite full encryption of data and signatures.
• To reduce communication overhead in secure image retrieval by eliminating round-trip interactions between client and server.

Proposed method

• The method extracts a wavelet-based global image signature from a Paillier-encrypted image using a recursive relationship between Paillier random values across wavelet decomposition levels.
• It introduces a novel fast comparison technique for Paillier-encrypted data, even when encrypted with different public keys, enabling secure histogram comparison in the encrypted domain.
• The system constructs an encrypted histogram of wavelet coefficients using a noisy histogram procedure to preserve privacy while maintaining precision.
• All computations—signature extraction and similarity comparison—are performed solely by the cloud server on encrypted data, with no need for decryption or client-server interaction.
• The approach leverages the homomorphic properties of the Paillier cryptosystem to perform arithmetic operations on encrypted coefficients, ensuring correctness upon decryption.
• The method ensures that the encrypted signature reveals no information about the original image content, preserving confidentiality.

Experimental results

Research questions

• RQ1 Can a secure image retrieval system be built that allows the cloud to perform similarity search on encrypted images without any interaction with the user or a trusted third party?
• RQ2 Is it possible to extract and compare global image signatures directly from homomorphically encrypted images while preserving retrieval accuracy?
• RQ3 How does the performance of the proposed SOCBIR scheme compare to traditional CBIR systems operating on unencrypted data?
• RQ4 What is the impact of wavelet decomposition level and histogram quantization step on retrieval accuracy in the encrypted domain?
• RQ5 Can a fast and secure comparison mechanism for Paillier-encrypted data be designed to support efficient histogram-based similarity computation?

Key findings

• The proposed SOCBIR scheme achieves the same mean precision at five (over 90%) as the equivalent CBIR system in the clear domain across both medical and face recognition datasets.
• Retrieval performance remains stable across different wavelet decomposition levels (d = 0, 1, 2) and quantization step values (Δ ∈ [1, 2, 4, 8, 16, 32, 64]), with no significant drop in accuracy even at Δ = 64.
• The scheme maintains equivalent performance to the clear-domain system because the filter coefficients of the Haar wavelet transform are preserved in the encrypted domain, ensuring no precision loss from coefficient expansion.
• The method enables full computation of image signature extraction and comparison in the cloud with zero communication overhead after the initial encrypted image upload.
• The system is secure against privacy leakage, as the encrypted signature reveals no information about the original image content.
• Despite strong privacy and performance guarantees, the scheme requires transferring gigabits of encrypted data (e.g., for 1024-bit keys), limiting practical deployment until bandwidth is reduced.