A LIGHTWEIGHT FACE RECOGNITION SYSTEM FOR MEDIUM‑SCALE DATASETS: ALGORITHM SELECTION AND VECTOR DATABASE IMPLEMENTATION

Main Article Content

Abstract

Face recognition is widely used for access control, attendance, and surveillance, yet many real systems are medium‑scale (10k–50k images) rather than billion‑scale. This study examines algorithm choice for a 13,000‑image system (similar size to LFW), comparing classic methods (Eigenfaces, Fisherfaces, LBPH) with modern deep models (FaceNet, ArcFace, SFace). We evaluate accuracy, runtime, model footprint, and deployment simplicity. Results show SFace — a compact CNN that yields 128‑D embeddings — provides the best balance for medium deployments, exceeding 95% accuracy on LFW while occupying just 37 MB and running effectively on CPU. We also show a straightforward in‑memory vector store using cosine similarity can serve 13,000 embeddings with sub‑millisecond queries. A full reference implementation using OpenCV and NumPy is included. These findings give practical guidance for developers selecting face recognition solutions optimized for accuracy, efficiency, and ease of deployment in medium‑scale production environments.

Article Details

Section

Articles

References

[1] G. B. Huang, M. Ramesh, T. Berg, and E. Learned Miller, “Labeled Faces in the Wild: A database for studying face recognition in unconstrained environments,” University of Massachusetts, Amherst, Tech. Rep. 07 49, 2007.

[2] M. Turk and A. Pentland, “Eigenfaces for recognition,” Journal of Cognitive Neuroscience, vol. 3, no. 1, pp. 71 86, 1991.

[3] P. N. Belhumeur, J. P. Hespanha, and D. J. Kriegman, “Eigenfaces vs. Fisherfaces: Recognition using class specific linear projection,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 19, no. 7, pp. 711 720, 1997.

[4] W. Deng, “SFace: Lightweight face recognition model,” Beijing University of Posts and Telecommunications, 2021. [Online]. Available: OpenCV Zoo.

[5] L. Wolf, T. Hassner, and I. Maoz, “Face recognition in unconstrained videos with matched background similarity,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2011, pp. 529 534.

[6] T. Ahonen, A. Hadid, and M. Pietikainen, “Face description with local binary patterns: Application to face recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 28, no. 12, pp. 2037 2041, 2006.

[7] F. Schroff, D. Kalenichenko, and J. Philbin, “FaceNet: A unified embedding for face recognition and clustering,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2015, pp. 815 823.

[8] J. Deng, J. Guo, N. Xue, and S. Zafeiriou, “ArcFace: Additive angular margin loss for deep face recognition,” in Proc. IEEE/CVF Conf. Computer Vision and Pattern Recognition, 2019, pp. 4690 4699.

[9] Qodirov, A., “Comparative analysis of Eigenfaces, Fisherfaces, and LBP algorithms on the LFW dataset,” Al Farg’oniy avlodlari, vol. 1, no. 4, pp. 193 205, 2023.

[10] J. Johnson, M. Douze, and H. Jégou, “Billion scale similarity search with GPUs,” IEEE Transactions on Big Data, vol. 7, no. 3, pp. 535 547, 2019.

[11] W. Wu, H. Peng, and S. Yu, “Yunet: A tiny millisecond level face detector,” Machine Intelligence Research, vol. 20, no. 5, pp. 656 665, 2023.

[12] “A Comprehensive Survey of Lightweight Neural Networks for Face Recognition,” KCI, 2025.

[13] “FaceLiVT: Face recognition using linear vision transformer with structural reparameterization for mobile device,” in Proc. IEEE International Conference on Image Processing (ICIP), 2025, pp. 1720 1725.