A Comparative Analysis of FaceNet, VGGFace, and GhostFaceNets Face Recognition Algorithms For Potential Criminal Suspect Identification

https://doi.org/10.48185/jaai.v5i2.1237

Authors

Keywords:

Artificial Intelligence, Artificial Intelligence (AI), Deep Learning, CNN, FaceNets, GhostFaceNets, VarGFaceNets, Facial Recognition, Machine Learning

Abstract

The escalating concerns surrounding criminal activities underscore the imperative for bolstered security measures to safeguard public welfare. Despite concerted efforts, the identification of suspects remains fraught with limitations, hindering the attainment of comprehensive individual profiles. Leveraging advancements in facial detection and identification technologies, this study assesses the efficacy of three prominent deep learning models—FaceNet, VGGFace, and GhostFaceNets—in the domain of facial recognition for suspect identification. Drawing upon data collected in 2023, the investigation scrutinizes FaceNet's intricate methodologies, including triplet loss optimization and Euclidean space mappings, yielding exceptional accuracy rates of 97.05% during validation and 97.4% during testing. Conversely, VGGFace, while displaying commendable accuracies, registers marginally lower accuracy metrics, standing at 97.05% and 96.1% during validation and testing, respectively. GhostFaceNets, integrating novel architectural components, exhibit diminished accuracy rates, signaling avenues for refinement. These empirical insights underscore FaceNet's prowess in furnishing robust and reliable facial recognition outcomes, while delineating the imperative for iterative enhancements in GhostFaceNets to foster their pragmatic applicability in security domains.

Downloads

Download data is not yet available.

Published

2024-09-09

How to Cite

Muhammad Indra Ardiawan, & Gede Putra Kusuma Negarara. (2024). A Comparative Analysis of FaceNet, VGGFace, and GhostFaceNets Face Recognition Algorithms For Potential Criminal Suspect Identification. Journal of Applied Artificial Intelligence, 5(2), 34–49. https://doi.org/10.48185/jaai.v5i2.1237