Structural synthesis of neural network models for biometric recognition in the feature space of multidimensional dynamic curves

Purpose of research. The study aims to improve the reliability of biometric user identification based on dynamic signatures by developing neural network models operating in the feature space of multidimensional dynamic curves. The focus is on the structural and parametric synthesis of a classification neural network architecture using statistical, harmonic, and wavelet-transformed features extracted from the dynamic signature.

Methods. The proposed identification model performs parallel recognition of multidimensional curve fragments using various methods, including statistical, metric, and neural classifiers. The analysis is based on a set of dynamic signature parameters, such as pen coordinates, pressure, velocity, acceleration, and their derived features. Statistical metrics – mean values, standard deviations, coefficients of variation, entropy, and equivocation – are combined with Discrete Fourier Transform (DFT), Discrete Cosine Transform (DCT), and Discrete Wavelet Transform (DWT) to form an informative feature space. These features are then used to synthesize an MLP classifier whose architecture is adapted to the input data.

Results. Experimental results confirm that using secondary features significantly increases identification accuracy compared to traditional methods. A set of 3–5 key parameters along with their spectral derivatives allows for accuracy levels of 0,8 to 0,95 with a limited number of users, maintaining around 0,7 when scaling. The average improvement in identification accuracy was 25–35% over statistical methods and 5–15% over metric-based algorithms.

Conclusion. To ensure the required level of identification reliability, it is recommended to apply a multi-level approach involving separate processing of dynamic signature parameters followed by result integration. The most effective configurations were based on neural network models combined with metric and correlation methods operating in the space of spectral and statistical features.

1. Lomov N., Petrova D., Ryazantseva Yu. Biometrics in the financial sector 2020: Benefits for the consumer. URL: https://www.fintechru.org/analytics/analiticheskaya-zapiska-pobiometrii/?ysclid=lvwb2xh8z610856013 (accessed 08.09.2025).

2. Report on the analysis of the size, share and trends of the global biometric systems market – an overview of the industry and forecast up to 2032 // Data Bridge.

3. Dancers A.Kh., Danilov E.A. Modern approaches to the recognition and authentication of signatures: methods of image processing. XXI vek: itogi proshlogo i pro-blemy nastoyashchego plyus = XXI Century: Results of the Past and Problems of the Present Plus. 2025;14(2):71–76. (In Russ.) EDN SCSZTT

4. Marcel S., Nixon M.S., Fierrez J., Evans N. Advances in Computer Vision and Pattern Recognition. Springer, 2023. 522 p.

5. Starovoitov V.V., Akhundzhanov U.Yu. Distribution of local curve values as a structural feature for off-line verification of a handwritten signature. Sistemnyi analiz i prikladnaya informatika = System Analysis and Applied Informatics. 2023;(2):49–58. (In Russ.)

6. Dancers A.Kh., Danilov E.A., Martyshkin A.I. Review and comparison of some methods of signature authentication by dynamic parameters. Vestnik Ryazanskogo gosudarstvennogo radiotekhnicheskogo universiteta = Bulletin of the Ryazan State Radio Engineering University. 2025;(92):213–224. (In Russ.) https://doi.org/10.21667/1995-4565-2025-92-213-224

7. Carmona P.L., Salvador S.A., Fred L.N. Mathematical Methodologies in Pattern Recognition and Machine Learning: Contributions from the International Conference on Pattern Recognition Applications and Methods. Springer; 2022. 204 p.

8. Ivanov A.I. Using the inverse chi-square Pearson criterion in the multi-plicative synthesis of new statistical criteria from already known criteria to test the hypothesis of the normal distribution of these small samples. Izvestiya vysshikh uchebnykh zavedenii. Povolzhskii region. Tekhnicheskie nauki = News of Higher Educational Institutions. Volga region. Technical Sciences. 2023;(2):15–23. (In Russ.) EDN LPXYSF

9. Ivanov A.I. Synthesis of new integro-differential statistical criteria and their equivalent artificial neurons for small samples. Samara: Izdatel'stvo Samarskogo nauchnogo tsentra RAN; 2022. P. 161–163. (In Russ.) EDN JHGXBI

10. Dancers A.Kh., Danilov E.A. Kinematic model of the formation of reference dynamic parameters of the user's signature. Vestnik Ryazanskogo gosudarstvennogo radiotekhnicheskogo universiteta = Bulletin of the Ryazan State Radio Engineering University. 2025;(92):170–178. (In Russ.) https://doi.org/10.21667/1995-4565-2025-92-170-178

11. Jain A. K., Griess F.D., Connell S.D. Online signature verification. Pattern Recognition. 2022;(35):2963–2972.

12. Wijdan Yassen A. AlKarem, Khawla H. Ali, Eman Thabet Khalid. Handwritten Signature Verification Method Using Convolutional Neural Network. Iraqi Journal for Electrical and Electronic Engineering. 2024;(20):77–84. https://doi.org/10.37917/ijeee.20.2.7

13. Tolosana R., Vera-Rodriguez R., Fierrez J., Ortega-Garcia J. DeepSign: Deep learning for signature analysis. IEEE Transactions on Information Forensics and Security. 2020;(15):229–240. https://doi.org/10.1109/TBIOM.2021.3054533

14. Vargas J.F., Ferrer M.A., Travieso C.M., Alonso J.B. Off-line handwritten signature GPDS-960 corpus. IEEE Transactions on Information Forensics and Security. 2017;(12):596– 605. https://doi.org/10.1109/ICDAR.2007.4377018

15. Farabet C., Couprie C., Najman L., LeCun Y. Learning hierarchical features for scene labeling. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2023;(8):1915– 1929.

16. Batista L., Granger E., Sabourin R. Dynamic selection of generative – discriminative en-sembles for off-line signature verification. Pattern Recognition. 2016;(52):182–197. https://doi.org/10.1016/j.patcog.2011.10.011

17. Akhundzhanov U.Yu., Starovoitov V.V. Preliminary processing of images of handwritten signatures for subsequent recognition. Sistemnyi analiz i prikladnaya informatika = System Analysis and Stock Informatics. 2022;(2):4–9. (In Russ.)

18. Ivanov A.I., Kupriyanov E.N. Synthesis of new more powerful statistical Cree-Terii through the multiplicative union of the classical Frozini and Murota – Takeuchi criteria with the Hirst criterion to test the hypothesis of the normality of small samples. Nadezhnost' = Reliability. 2022;22(1):52–55. (In Russ.) https://doi.org/10.21683/1729-2646-2022-22-1-52-55. EDN CWHSKS

19. Vdovkina S.A., Shibanov S.V. Library for online verification of handwritten signatures. Vestnik Penzenskogo gosudarstvennogo universiteta = Bulletin of Penza State University. 2022;(2):76–84. (In Russ.)