Page 86 - 2024-Vol20-Issue2

82 | AlKarem, Khalid, & Ali

TABLE II.
SUMMARY OF MODEL TRAINING, FITTING, AND TESTING RESULTS.

Data Obtained accuracy Obtained loss Error
Training set 0.995 0.0273
Validation-set 0.9536 0.2322
Test-set 1.000 0.0488

TABLE III.
COMPARISON EVALUATION WITH RECENT STUDIES

Study Accuracy
Wei et al.,2019 [22] 90.17%
Xiao & Ding (2022) [23] 95.66%
Ren et al., 2022 [24] 93.25%,
Lopes et al., 2022 [25] 85.0 %
The proposed method 95.36 %

mance of existing methods. Therefore, this work proposed to
develop deep learning model-based CNN to perform WI of-
fline signature verification. The proposed model was trained,
validated and tested on collected English signature dataset.
The developed model is utilized to make prediction and veri-
fication on Arabic signature samples which was created for
this study. The experimental outcomes have shown a good
performance in term of accuracy. However, this work needs
to enhance the accuracy via increasing the training dataset
samples. Fore future work, to improve the accuracy, this
work aims at exploring different techniques for features ex-
traction, selection, and classification. Exploring other datasets
challenges.

CONFLICT OF INTEREST

The authors have no conflict of relevant interest to this article.

Fig. 6. The outcomes of signature recognition using images REFERENCES
from Arabic dataset.
[1] Z. Hashim, H. M. Ahmed, A. H. Alkhayyat, et al., “A
comparative study among handwritten signature veri-
fication methods using machine learning techniques,”
Scientific Programming, vol. 2022, 2022.

[2] R. Tolosana, R. Vera-Rodriguez, J. Fierrez, and J. Ortega-
Garcia, “Exploring recurrent neural networks for on-line
handwritten signature biometrics,” Ieee Access, vol. 6,
pp. 5128–5138, 2018.

[3] X. Zhang, Z. Wu, L. Xie, Y. Li, F. Li, and J. Zhang,
“Multi-path siamese convolution network for offline
handwritten signature verification,” in 2022 The 8th In-
ternational Conference on Computing and Data Engi-
neering, pp. 51–58, 2022.