Bidirectional Biometric Authentication Using Transciphering and (T)FHE

Joon Soo Yoo, Tae Min Ahn, Ji Won Yoon, [0.5em] School of Cybersecurity, Korea University, Seoul 02841, Republic of Korea · 2025

Read Bidirectional Biometric Authentication Using Transciphering and (T)FHE by Joon Soo Yoo, Tae Min Ahn, Ji Won Yoon, [0.5em] School of Cybersecurity, Korea University, Seoul 02841, Republic of Korea (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-12802.txt.

Introduction

Joon Soo Yoo, Tae Min Ahn, and Ji Won Yoon designed a private biometric login that works both ways in a 2025 paper. Leaked fingerprint or iris templates cannot be changed like passwords, so checking them under strong encryption is attractive. Full encryption schemes, though, ship huge data and strain key handling.

Their framework mixes encryption with transciphering and adds a trusted helper that never colludes. Double encryption guards the trust model, so client data stays private. The design shrinks returned ciphertexts, stops clients from faking a pass, and centers key management for scale. Built with a fast scheme and a stream cipher, it cut transmission size up to 121-fold on iris data against standard setups.

Catalog record

CREATORJoon Soo Yoo, Tae Min Ahn, Ji Won Yoon, [0.5em] School of Cybersecurity, Korea University, Seoul 02841, Republic of Korea
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE75 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-12802

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Bidirectional Biometric Authentication Using Transciphering and (T)FHE Joon Soo Yoo, Tae Min Ahn, Ji Won Yoon, [0.5em] School of Cybersecurity, Korea University, Seoul 02841, Republic of Korea Abstract Biometric authentication systems pose privacy risks, as leaked templates such as iris or fingerprints can lead to security breaches. Fully Homomorphic Encryption (FHE) enables secure encrypted evaluation, but its deployment is hindered by large ciphertexts, high key overhead, and limited trust models. We propose the Bidirectional Transciphering Framework (BTF), combining FHE, transciphering, and a non-colluding trusted party to enable efficient and privacy-preserving biometric authentication. The key architectural innovation is the introduction of a trusted party that assists in evaluation and key management, along with a double encryption mechanism to preserve the FHE trust model, where client data remains private. BTF addresses three core deployment challenges: reducing the size of returned FHE ciphertexts, preventing clients from falsely reporting successful authentication, and enabling scalable, centralized FHE key management. We implement BTF using TFHE and the Trivium cipher, and evaluate it on iris-based biometric data. Our results show up to a 121 × \times reduction in transmission size compared to standard FHE models, demonstrating practical scalability and deployment potential. 1 Introduction Authentication serves as the first line of defense, ensuring that only legitimate users are authorized to access resources, data, or systems. As cloud services expand rapidly, especially with the rise of online banking, e-commerce, and IoT devices, weak authentication mechanisms can lead to significant security issues. Therefore, the need for a robust authentication system is critical for verifying user identity and protecting sensitive information. The current biometric authentication system, particularly for online authentication without the use of a Trusted Execution Environment (TEE) [ 32 ] , requires the server to store either the raw or transformed biometric data of the user. Storing raw biometric data presents a significant privacy risk, as this sensitive information becomes a prime target for attackers if the server is compromised. To mitigate this risk, technologies such as cancelable biometrics [ 29 ] and biometric cryptosystems (BCS) [ 34 ] transform the biometric data into a non-reversible format. While these transformed data types do not expose the raw biometric information directly in the event of a breach, they are not entirely risk-free. If the transformed data or the associated helper data in BCS is not sufficiently secured, attackers may still exploit this information to impersonate the user. Although these systems are designed to minimize such risks through robust cryptographic techniques, vulnerabilities can still arise depending on the implementation. Fully Homomorphic Encryption (FHE) [ 35 , 23 ] is highly suited for privacy-sensitive scenarios because it allows servers to perform computations on encrypted data without needing to decrypt it, unlike traditional models. This ensures that biometric data, once encrypted on the client side, remains encrypted during transmission, storage, and evaluation (matching). Furthermore, FHE is primarily based on lattice-based cryptography, which is considered resistant to quantum attacks, making it post-quantum secure. Given these advantages, constructing biometric authentication systems using FHE is emerging as a promising trend, offering a higher level of security and resilience against server compromises. Despite FHE being one of the most secure cryptographic technologies, it still faces several challenges when it comes to practical deployment—particularly in terms of evaluation time and ciphertext size. For instance, in the TFHE [ 8 , 9 ] scheme, which is a logic-based FHE approach, a ciphertext can be more than 2,000 times larger than a single plaintext bit, depending on the security level 1 1 1 For example, when n=630 , for a \lambda=128 -bit security level . Similarly, the CKKS [ 7 , 6 ] encryption scheme, which is a popular arithmetic-based scheme, allows SIMD operations that pack multiple plaintexts into a single ciphertext, yet still struggles with large ciphertext sizes. Therefore, the significant ciphertext expansion, compared to symmetric ciphers that do not require such overhead, remains a major obstacle to the practical use of FHE. Table 1: Comparison of Approaches in Biometric Authentication. ( \mathsf{SC} represents stream cipher) Reference Single Server Transcipher Return Size \downarrow Biometric System Symmetric Cipher Encryption Scheme Balenbois et al. [ 1 ] ✓ ✓ ✗ ✗ Trivium & Kreyvium TFHE Cho et al. [ 10 ] ✓ ✓ ✗ ✗ HERA CKKS Zhou et al. [ 37 ] ✓ ✗ ✗ ✓ N/A Functional Encryption Barrero et al. [ 19 ] ✗ ✗ ✗ ✓ N/A Paillier Our Work ✗ ✓ ✓ ✓ FHE-friendly SC FHE Transciphering [ 25 , 1 , 10 , 21 , 27 ] is a technique for converting one encryption scheme to another to achieve a balance between security and efficiency. In the context of FHE, transciphering allows the system to leverage the powerful evaluation capabilities of FHE on the server side, while relying on lightweight symmetric encryption during transmission between the client and server. The key idea is to homomorphically decrypt a stream cipher ciphertext that has itself been encrypted under the FHE scheme. Specifically, the client first encrypts a message m using a symmetric stream cipher with key k , resulting in a ciphertext c=\mathsf{E}(m,k) . Rather than encrypting m directly under FHE, the client sends c to the server. The server then encrypts c under the FHE public key \mathsf{pk} , producing \mathsf{Enc}(c,\mathsf{pk}) . It then performs a homomorphic decryption of c to obtain \mathsf{Enc}(m) —the message now encrypted under the FHE scheme. This transformation enables the client to benefit from the low overhead of symmetric encryption while offloading the more costly homomorphic operations to the server. Importantly, because the client transmits only a small symmetric ciphertext, the network overhead remains comparable to traditional encrypted communication, avoiding the large transmission costs associated with native FHE ciphertexts. 1.1 Transciphering Challenges in Authentication Systems While it may seem straightforward to naively adapt existing transciphering frameworks to the domain of FHE-based biometric authentication, several practical challenges arise in this setting. Most notably, current transciphering approaches primarily focus on minimizing the transmission overhead from the client to the server. However, they often overlook the size of the ciphertext returned to the client after computation, which can also contribute significantly to the overall network overhead. Another challenge arises from the trust model typically assumed in FHE-based systems. After the server completes the homomorphic evaluation, the authentication result remains encrypted under the client’s key, meaning only the client can decrypt it. In this setting, the server must rely on the client to report whether authentication was successful. However, this creates a security vulnerability: a malicious client could falsely claim a successful authentication, thereby compromising the integrity of the system. As a result, the traditional FHE model—where only the client possesses the decryption key—can be fundamentally unsuitable for biometric authentication, where the correctness of the result must be verifiable by the server. Lastly, the server must handle multiple cryptographic keys from each client, which introduces substantial scalability concerns. In particular, the client is required to transmit both a public key for encrypting the stream ciphertext under the FHE scheme and an evaluation key for enabling homomorphic operations. Under our TFHE implementation, the size of the public key is approximately 4.93 MB, while the evaluation key reaches 41.6 MB. These values correspond to a single client.

Read the complete text