zkSABER: Zero-knowledge Succinct Authentication using Biometric Embedding Representation

Abstract

Executing biometric matching between two embedding vectors on the blockchain remains a challenging problem due to inherent privacy concerns and the computational constraints imposed by block gas limits. To address these challenges, we propose zk-SABER, a succinct blockchain-based biometric authentication scheme that allows constant proof size and verification cost with respect to the embedding vector length. Our design combines a Merkle Tree and a biometric matching algorithm within a zkSNARK circuit to prove that a user’s biometric trait matches one of the registered templates in an anonymous manner. To ensure compatibility with state-of-the-art Deep Neural Network (DNN) models, we introduce a complete quantization pipeline that converts floating-point embeddings into zkSNARK-friendly representations. Our experiment results show constant transaction gas cost and proof size, regardless of the embedding vector length, thereby demonstrating the practicality of zk-SABER for real-world blockchain environments.