KONVOLYUTSION NEYRON TARMOQ YORDAMIDA IMZO XUSUSIYATLARINI AVTOMATIK AJRATISH USULI
Keywords:
Konvolyutsion neyron tarmoq, CNN, imzo verifikatsiyasi, xususiyat ajratish, chuqur o'rganish, OC-SVM, biometrik autentifikatsiya.Abstract
Ushbu maqolada konvolyutsion neyron tarmoq (CNN) arxitekturasi yordamida shaxs imzosidan avtomatik ravishda xususiyatlarni ajratish usuli taqdim etiladi. Taklif etilgan yondashuv an'anaviy qo'lda xususiyat ajratish usullaridan farqli o'laroq, chuqur o'rganish tamoyillariga asoslanib, imzo tasvirlarining yuqori darajali semantik belgilarini avtomatik tarzda o'rganadi. Eksperimental natijalar CEDAR ma'lumotlar to'plami ustida olib borilgan bo'lib, taklif etilgan CNN modeli 96.8% aniqlik va 0.965 F1-Score ko'rsatkichiga erishdi. CNN va bir sinfli tayanch vektor mashinasi (OC-SVM) kombinatsiyasi esa 97.3% aniqlikni ta'minladi.
References
1. LeCun Y., Bengio Y., Hinton G. Deep learning // Nature. – 2015. – Vol. 521. – P. 436–444.
2. Hafemann L.G., Sabourin R., Oliveira L.S. Learning features for offline handwritten signature verification using deep convolutional neural networks // Pattern Recognition. – 2017. – Vol. 70. – P. 163–176.
3. Krizhevsky A., Sutskever I., Hinton G.E. ImageNet classification with deep convolutional neural networks // Advances in Neural Information Processing Systems. – 2012. – P. 1097–1105.
4. Simonyan K., Zisserman A. Very deep convolutional networks for large-scale image recognition // ICLR. – 2015.
5. Dey S., Dutta A., Toledo J.I. SigNet: Convolutional Siamese network for writer independent offline signature verification // arXiv. – 2017. – 1707.02131.
6. Schölkopf B., Platt J.C., Shawe-Taylor J. Estimating the support of a high-dimensional distribution // Neural Computation. – 2001. – Vol. 13. – P. 1443–1471.
7. Zheng L., Sun Z., Tan T. Transfer learning for signature verification // IEEE TIFS. – 2018.
8. CEDAR Signature Dataset URL: http://cedar.buffalo.edu/NIJ/data/.
