Effectiveness of Deep Neural Networks in Handwritten Character Recognition Performance Analysis of Limited and Extensive Class Structures
Turkish El Yazısı Karakterlerin Tanınmasında Derin Sinir Ağlarının Etkinliği Kısıtlı ve Geniş Sınıf Yapılarının Performans Analizi
Erdin Fidan1*, Mahir Kaya2
1Bilgisayar Mühendisliği, Tokat Gaziosmanpaşa Üniversitesi, Tokat, Türkiye
2Bilgisayar Mühendisliği, Tokat Gaziosmanpaşa Üniversitesi, Tokat, Türkiye
* Corresponding author: erdin.fidan6425@gop.edu.tr
Presented at the International Symposium on AI-Driven Engineering Systems (ISADES2025), Tokat, Turkiye, Jun 19, 2025
SETSCI Conference Proceedings, 2025, 22, Page (s): 119-123 , https://doi.org/10.36287/setsci.22.65.001
Published Date: 10 July 2025
In this study, a deep learning-based model was developed for handwritten character recognition. The model was tested with varying numbers of output classes (characters); while an accuracy rate of 96% was achieved in the scenario with only 10 character classes, an accuracy of 86% was obtained when the number of classes was increased to 62. The results clearly demonstrate the impact of the number of classes on model performance, revealing that the classification task becomes more complex as character variety increases. In this context, the proposed model was shown to perform with high accuracy in multi-class character recognition tasks, and its performance was analyzed in detail based on the number of classes. Additionally, the model training was planned in two stages: initial learning was conducted using the SGD optimization algorithm, followed by a fine-tuning process with the Adam algorithm.
Keywords - Handwriting Recognition, Deep Learning, Convolutional Neural Network, Batch Normalization, Dropout, Classification, Character Recognition
Bu çalışmada, el yazısı karakterlerin tanınmasına yönelik derin öğrenme tabanlı bir model geliştirilmiştir. Model, farklı sayıda çıkış sınıfı (karakter) için test edilmiş; yalnızca 10 karakterin sınıflandırıldığı senaryoda %96 doğruluk oranı elde edilirken, sınıf sayısı 62’ye çıkarıldığında %86 doğruluk oranı yakalanmıştır. Elde edilen sonuçlar, sınıf sayısının artmasının model performansı üzerindeki etkisini açıkça göstermekte ve karakter çeşitliliği arttıkça sınıflandırma probleminin daha karmaşık hâle geldiğini ortaya koymaktadır. Bu bağlamda, önerilen modelin çok sınıflı karakter tanıma görevlerinde yüksek doğrulukla çalışabildiği görülmüş ve modelin performansı sınıf sayısına bağlı olarak detaylı şekilde analiz edilmiştir. Ayrıca, modelin eğitimi iki aşamalı olarak planlanmış olup, ilk olarak SGD optimizasyon algoritması ile temel öğrenme sağlanmış, ardından Adam algoritması ile model ince ayar (fine-tuning) sürecine tabi tutulmuştur.
KeywordsTurkish - El Yazısı Tanıma,Derin Öğrenme, Konvolüsyonel Sinir Ağı, Batch Normalization, Dropout, Sınıflandırma, Karakter Tanıma
[1] Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE, vol. 86, no. 11, pp. 2278–2324, Nov. 1998.
[2] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet classification with deep convolutional neural networks,” in Advances in Neural Information Processing Systems, vol. 25, 2012, pp. 1097–1105.
[3] M. Kaya, “Feature fusion-based ensemble CNN learning optimization for automated detection of pediatric pneumonia,” Biomed. Signal Process. Control, vol. 87, p. 105472, 2024.
[4] M. Kaya and Y. Çetın-Kaya, “A novel deep learning architecture optimization for multiclass classification of Alzheimer’s disease level,” IEEE Access, vol. 12, pp. 46562–46581, 2024.
[5] Gonzalez, R. C., & Woods, R. E. (2018). Digital Image Processing. Pearson.
[6] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature.
[7] Barbedo, J. G. A. (2013). Digital image processing techniques for detecting, quantifying and classifying plant diseases. SpringerPlus.
[8] M. Kaya, S. Ulutürk, Y. Çetin Kaya, O. Altıntaş ve B. Turan "Optimization of Several Deep CNN Models for Waste Classification," Sakarya Üniversitesi Bilgisayar Bilimleri ve Mühendisliği Dergisi, http://saucis.sakarya.edu.tr/en/pub/issue/79575/1257100
[9] Cohen, G., Afshar, S., Tapson, J., & van Schaik, A. (2017). EMNIST: Extending MNIST to handwritten letters.
[10] de Campos, T. E., Babu, B. R., & Varma, M. (2009). Character recognition in natural images.
[11] Clanuwat, T., Bober-Irizar, M., Kitamoto, A., Lamb, A., Yamamoto, K., & Ha, D. (2018). Deep Learning for Classical Japanese Literature.
[12] Jindal, A., Dua, M., & Kumar, M. (2018). Handwritten Arabic character recognition using convolutional neural network.
[13] Lin, T. Y., Goyal, P., Girshick, R., He, K., & Dollár, P. (2017). Focal Loss for Dense Object Detection. ICCV.
[14] Buda, M., Maki, A., & Mazurowski, M. A. (2018). A systematic study of the class imbalance problem in convolutional neural networks. Neural Networks.
[15] Afzal, M. Z., Kölsch, A., Ahmed, S., & Liwicki, M. (2019). Deep learning based baseline detection in historical documents.
[16] Jaderberg, M., Simonyan, K., Vedaldi, A., & Zisserman, A. (2014). Synthetic data and artificial neural networks for natural scene text recognition.
[17] Miyazaki, K., & Kitamoto, A. (2019). Improving character recognition in Kuzushiji using ResNets.
[18] Alginahi, Y. M., Alabrah, A., & Alotaibi, M. (2020). Arabic handwritten character recognition using hybrid deep neural networks
[19] S. Mann, “Handwritten English Characters and Digits,” Kaggle, 2023. [Online]. Available: https://www.kaggle.com/datasets/sujaymann/handwritten-english-characters-and-digits
|
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. |