Geleneksel Görüntü İşleme ve Derin Öğrenmenin Birleşimi Hibrid Model ile El Yazısı Rakam Sınıflandırma

Öz

Bu çalışma, rakam sınıflandırma problemlerinde yaygın olarak kullanılan Konvolüsyonel Sinir Ağları’nın (CNN) dönme ve ölçekleme gibi geometrik varyasyonlara karşı yaşadığı yetersizlikleri gidermek amacıyla hibrit bir model önermektedir. Önerilen yaklaşım, CNN mimarisini Hu ve Zernike Momentlerinin sağladığı geometrik değişmezlik özellikleriyle entegre ederek özellik birleştirme tekniğini kullanmaktadır. Modelin dayanıklılığı, standart veriler yerine gürültü ve bozulma içeren zorlu SVHN veri seti üzerinde test edilmiştir. Çalışma kapsamında modern ve hafif bir mimari olan MobileNetV3 ile yapılan kapsamlı kıyaslamalar, hibrit modelin %90.97 ile istatistiksel olarak rekabetçi bir doğruluk oranına ulaştığını göstermiştir. Ancak asıl fark işlem hızında ortaya çıkmış; hibrit model 760 ms işlem süresiyle, 3276 ms süren MobileNetV3 modeline kıyasla 4 kattan fazla hız avantajı sağlamıştır. Bu sonuçlar, önerilen yöntemin yüksek doğruluk ve hız gerektiren gerçek zamanlı uygulamalar için modern alternatiflerden çok daha verimli bir çözüm sunduğunu kanıtlamaktadır.

Anahtar Kelimeler

• Hibrit Model
• El Yazısı Rakam Sınıflandırması
• Görüntü İşleme Teknikleri
• Optik Karakter Tanıma (OCR)
• Öznitelik Çıkarımı

Handwritten Digit Classification with a Hybrid Model Combining Traditional Image Processing and Deep Learning

Öz

This study proposes a hybrid model to address the insufficiencies of Convolutional Neural Networks (CNNs), widely used in digit classification, against geometric variations such as rotation and scaling. The proposed approach integrates CNN architecture with the geometric invariance properties of Hu and Zernike Moments using a feature fusion technique. To test robustness, the model was evaluated on the challenging SVHN dataset, containing noise and real-world distortions, rather than standard datasets. Comprehensive comparisons with MobileNetV3, a modern lightweight architecture, showed that the hybrid model achieved a statistically competitive accuracy rate of 90.97%. However, the significant difference appeared in processing speed; the hybrid model demonstrated more than a 4-fold speed advantage (760 ms) compared to MobileNetV3 (3276 ms). These findings prove that the proposed method offers a significantly more efficient solution than modern alternatives for real-time applications requiring high accuracy and low computational cost.

Anahtar Kelimeler

• Hybrid Model
• Handwritten Digit Classification
• Image Processing Techniques
• Optical Character Recognition (OCR)
• Feature Extraction

Kaynakça

1. 1. Patidar, P.K., Tomar, D.S., Pateriya, R.K. & Sharma, Y.K. (2023). Precision agriculture: Crop image segmentation and loss evaluation through drone surveillance. 2023 Third International Conference on Secure Cyber Computing and Communication (ICSCCC), Jalandhar, India, 495-500.
2. 2. Wang, Y. (2025). Deep learning based image classification algorithm. In 2025 IEEE 5th International Conference on Power, Electronics and Computer Applications (ICPECA), 505-509. IEEE.
3. 3. He, K., Zhang, X., Ren, S. & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 770-778.
4. 4. LeCun, Y., Bottou, L., Bengio, Y. & Haffner, P. (2002). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
5. 5. Tomar, A. & Patidar, H. (2023). Optimizing CNN model performance for MNIST and CIFAR classification using rectified sigmoid and ReS activation functions. In 2023 7th International Conference On Computing, Communication, Control and Automation (ICCUBEA),1-6. IEEE.
6. 6. Singh, B.K., Rai, A., Kundu, K., Kalita, K. & Agrawal, R. (2024). An empirical analysis of invariance Hu's moment feature over a digital image. In 2024 11th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO), 1-5. IEEE.
7. 7. Lai, Z., Yao, Z., Lai, G., Wang, C. & Feng, R. (2024). A novel face swapping detection scheme using the Pseudo Zernike transform based robust watermarking. Electronics, 13(24), 4955.
8. 8. Lei, G., Lai, W., Meng, Q., Liu, H., Shi, D., Cui, W. & Han, K. (2023). Efficient and noise-resistant single-pixel imaging based on Pseudo-Zernike moments. Optics Express, 31(24), 39893-39905.

9. 9. Fu, D., Zhou, X., Xu, L., Hou, K. & Chen, X. (2023). Robust reversible watermarking by fractional order Zernike moments and pseudo-Zernike moments. IEEE Transactions on Circuits and Systems for Video Technology, 33(12), 7310-7326.
10. 10. Jusman, Y., Anam, M.K., Puspita, S. & Saleh, E. (2021). Machine learnings of dental caries images based on Hu moment invariants features. In 2021 International Seminar on Application for Technology of Information and Communication (iSemantic), 296-299. IEEE.
11. 11. Zheng, N., Zhang, G., Zhang, Y. & Sheykhahmad, F.R. (2023). Brain tumor diagnosis based on Zernike moments and support vector machine optimized by chaotic arithmetic optimization algorithm. Biomedical Signal Processing and Control, 82, 104543.
12. 12. Gül, E. & Bilgiç, S. (2025). Secret image hiding method based on multi-cover image with stego image sequence detection using perceptive hash. Çukurova Üniversitesi Mühendislik Fakültesi Dergisi, 40(4), 927-936.
13. 13. Wan, G. & Yao, L. (2023). LMFRNet: A lightweight convolutional neural network model for image analysis. Electronics, 13(1), 129.
14. 14. AbuRass, S., Huneiti, A. & Al-Zoubi, M.B. (2020). Enhancing convolutional neural network using Hu’s moments. International Journal of Advanced Computer Science and Applications, 11(12), 130-137.
15. 15. Ala, M., Şahin, M., Kılıç, M. & Dişken, G. (2025). An investigation on design criteria of heat exchangers by using tree models of machine learning methods. Çukurova Üniversitesi Mühendislik Fakültesi Dergisi, 40(2), 375-386.