REAL VAQT REJIMIDA O‘ZGARUVCHAN TASHQI OMILLAR TA’SIRIDA YUZNI TANIB OLISH ALGORITMLARINI TAKOMILLASHTIRISH
Keywords:
yuzni tanib olish, real vaqt, ArcFace, domain moslashish, adaptiv augmentatsiya, diqqat mexanizmi, tashqi omillar, FAR, IJB-C benchmark, transfer learning.Abstract
Ushbu maqolada real vaqt rejimida o'zgaruvchan tashqi omillar — yorug'lik intensivligi, ob'havo sharoiti, yuz qoplamasi (niqob, ko'zoynak) va kamera sifati o'zgarishlari — ta'sirida yuzni tanib olish algoritmlarini takomillashtirish masalasi ko'rib chiqiladi. Tadqiqotning dolzarbligi shundaki, amaliy xavfsizlik, border-control va raqamli to'lov tizimlarida real vaqtdagi tanib olish aniqligi tashqi omillar ta'sirida 25–50% ga pasayishi kuzatiladi. Muammoni hal etish uchun moslashuvchan ma'lumotlarni kuchaytirish (adaptive augmentation), domain moslashish (domain adaptation) va ko'p omilli diqqat mexanizmi (multi-factor attention) asosidagi takomillashtirilgan ArcFace arxitekturasi taklif etiladi. IJB-C va CFP-FP benchmark to'plamlarida o'tkazilgan tajribalar 97.2% tanib olish aniqligini va 0.31% yolg'on qabul qilish darajasini (FAR) ko'rsatdi. Ilmiy yangilik — real vaqt rejimida tashqi omillarning dinamik tasnifi va unga mos adaptiv preprocessing konveyerini avtomatik tanlash mexanizmini yaratishdan iborat.
References
[1] Guo G., Zhang N. A survey on deep learning based face recognition // Computer Vision and Image Understanding. — 2019. — Vol. 189. — P. 102805. DOI: 10.1016/j.cviu.2019.102805
[2] Wang M., Deng W. Deep face recognition: A survey // Neurocomputing. — 2021. — Vol. 429. — P. 215–244. DOI: 10.1016/j.neucom.2020.10.081
[3] Li S., Xu X., Wan L., Zhu T. Measure and manage: An empirical study of face recognition performance degradation // Proc. IEEE FG. — 2021. — P. 1–8. DOI: 10.1109/FG52635.2021.9667074
[4] Neto P.C., Pinto J.R., Cardoso J.S. Beyond the visible: A survey on cross-spectral face recognition // Pattern Recognition. — 2022. — Vol. 132. — P. 108929. DOI: 10.1016/j.patcog.2022.108929
[5] Damer N. et al. The effect of wearing a mask on face recognition performance: An exploratory study // Proc. IEEE BIOSIG. — 2020. — P. 1–6. DOI: 10.1109/BIOSIG49314.2020.9262043
[6] Turk M., Pentland A. Eigenfaces for recognition // Journal of Cognitive Neuroscience. — 1991. — Vol. 3, No. 1. — P. 71–86. DOI: 10.1162/jocn.1991.3.1.71
[7] Belhumeur P.N., Hespanha J.P., Kriegman D.J. Eigenfaces vs. Fisherfaces: Recognition using class specific linear projection // IEEE Trans. Pattern Anal. Mach. Intell. — 1997. — Vol. 19, No. 7. — P. 711–720. DOI: 10.1109/34.598228
[8] Deng J., Guo J., Xue N., Zafeiriou S. ArcFace: Additive angular margin loss for deep face recognition // Proc. IEEE CVPR. — 2019. — P. 4690–4699. DOI: 10.1109/CVPR.2019.00482
[9] Kim M., Jain A.K., Liu X. AdaFace: Quality adaptive margin for face recognition // Proc. IEEE CVPR. — 2022. — P. 18729–18738. DOI: 10.1109/CVPR52688.2022.01819
[10] Li Y. et al. A cropping-based method for masked face recognition // Applied Sciences. — 2021. — Vol. 11, No. 10. — P. 4416. DOI: 10.3390/app11104416
[11] Ganin Y., Lempitsky V. Unsupervised domain adaptation by backpropagation // Proc. ICML. — 2015. — Vol. 37. — P. 1180–1189.
[12] Hu J., Shen L., Sun G. Squeeze-and-excitation networks // Proc. IEEE CVPR. — 2018. — P. 7132–7141. DOI: 10.1109/CVPR.2018.00745
