How does reducing the dimension of feature vectors impact Biometric Systems that use Homomorphic Encryption?
Year: 2025
Type: conference-paper
Abstract: Homomorphic Encryption enables biometric systems to perform matching directly on encrypted feature vectors, preserving user privacy throughout the process. However, the high computational cost of encrypted-domain operations, especially on high-dimensional inputs, remain a major barrier to real-world use. This study presents a concrete example of how reducing the dimensionality of biometric feature vectors affects both matching accuracy and runtime, and discusses how system designers could estimate the trade-off between time savings and accuracy loss when choosing a target dimension. The code is available on Github.
Author: Andreis G. M. Purim
Institution: Universidade Estadual de Campinas (UNICAMP)
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4414127130