Kurdish Handwritten Text Recognition: A DenseNet121-Transformer Architecture with Constrained Synthetic Line Generation

Karez A. Hamad; Shareef M. Shareef

doi:10.14500/aro.12820

Authors

Karez A. Hamad Department of Software and Informatics Engineering, College of Engineering, Salahaddin University-Erbil, Erbil, Kurdistan Region, Iraq https://orcid.org/0000-0003-4638-6642
Shareef M. Shareef Department of Business Management, Mergasour Technical Institute, Erbil Polytechnic University, Erbil, Kurdistan Region, Iraq https://orcid.org/0000-0002-9112-2198

DOI:

https://doi.org/10.14500/aro.12820

Keywords:

Arabic-script recognition, Data augmentation, DenseNet-Transformer, Kurdish handwriting recognition, Synthetic line generation

Abstract

Based on the available peer-reviewed literature, Kurdish handwritten recognition remains at a nascent stage, with existing models limited to isolated character, digit, and word recognition. The demand for digitizing Kurdish handwritten documents has grown rapidly, particularly in regions undergoing government digitization, such as the Kurdistan Regional Government. The primary obstacle hindering progress beyond isolated characters or words is the absence of text-based handwritten datasets. To address this gap, a comprehensive Kurdish handwritten dataset encompassing paragraph-, line-, and word-level samples is first introduced. A recognition architecture combining DenseNet-121 as the CNN backbone with a Transformer-based encoder-decoder for sequence modeling is then proposed for Kurdish handwritten line recognition, representing one of the earliest reported efforts in this domain. To augment the limited line-level training data, a constrained recipe-based synthetic line generation framework is developed that concatenates real handwritten word images while enforcing text uniqueness, single-writer consistency, and leakage-free data partitioning. Additional training strategies are investigated, including cross-lingual transfer learning from Arabic handwritten data and fixed-content handwritten line integration for improving recognition accuracy. The best configuration achieved a character error rate (CER) of 0.0593 and a word error rate (WER) of 0.3083 without language model integration, further reduced to a CER of 0.0534 and a WER of 0.2746 with an 8-g language model. The source code and trained models are publicly available at: https://huggingface.co/Karez/ KHLR.

Downloads

Download data is not yet available.

References

Abdo, H.A., Abdu, A., Al-Antari, M.A., Manza, R.R., Talo, M., Gu, Y.H., and Bawiskar, S., 2024. End-to-end deep learning framework for Arabic handwritten legal amount recognition and digital courtesy conversion. Mathematics, 12(14), p.2256.

Abdullah, A.A., Abdulla, S.H., Toufiq, D.M., Maghdid, H.S., Rashid, T.A., Farho, P.F., Sabr, S.S., Taher, A,H., Hamad, D.S., Veisi, H., and Asaad, A.T., 2024. NER- RoBERTa: Fine-Tuning Roberta for Named Entity Recognition (Ner) Within Low-Resource Languages. Available from: https://arxiv.org/abs/2412.15252

Ahmed, R.M., Rashid, T.A., Fattah, P., Alsadoon, A., Bacanin, N., Mirjalili, S., Vimal, S., and Chhabra, A., 2022. Kurdish handwritten character recognition using deep learning techniques. Gene Expression Patterns, 46, p.119278.

Ali, S.H., and Abdulrazzaq, M.B., 2024a. KurdSet: A Kurdish Handwritten characters recognition dataset using convolutional neural network. Computers, Materials and Continua, 79(1), pp.429-448.

Ali, S.H., and Abdulrazzaq, M.B., 2024b. KurdSet handwritten digits recognition based on different convolutional neural networks models. TEM Journal, 13(1), pp.221-233.

Alsaqi, I.M., and Fattah, P., 2025. Kurdish Handwritten Word Recognition withCRNN Using Multi-Backbone Evaluation. Academic Journal of International University of Erbil, 2(4), pp.439-449.

Anjum, T., and Khan, N., 2020. An Attention Based Method for Offline Handwritten Urdu Text Recognition. In: 2020 17th International Conference on Frontiers in Handwriting Recognition (ICFHR). pp.169-174.

Anjum, T., and Khan, N., 2023. CALText: Contextual attention localization for offline handwritten text. Neural Processing Letters, 55(6), pp.7227-7257.

Bensouilah, M., Taffar, M., and Zennir, M.N., 2024. gMLP guided deep networks model for character-based handwritten text transcription. Multimedia Tools and Applications, 83(5), pp.13557-13575.

Cascianelli, S., Cornia, M., Baraldi, L., Piazzi, M.L., Schiuma, R., and Cucchiara, R., 2021. Learning to read L’Infinito: Handwritten text recognition with synthetic training data. In: Tsapatsoulis, N., Panayides, A., Theocharides, T., Lanitis, A., Pattichis, C., and Vento, M., Eds. Computer Analysis of Images and Patterns. International Publishing, Springer, Cham, pp.340-350.

Coquenet, D., Chatelain, C., and Paquet, T., 2023. DAN: A segmentation-free document attention network for handwritten document recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence 45(7), pp.8227-8243.

Dhiaf, M., Rouhou, A.C., Kessentini, Y., and Salem, S.B., 2023. MSdocTr-Lite: A lite transformer for full page multi-script handwriting recognition. Pattern Recognition Letters, 169, pp.28-34.

Fang, S., Xie, H., Wang, Y., Mao, Z., and Zhang, Y., 2021. Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). pp.7098-7107.

Gader, T.B.A., and Echi, A.K., 2022. Attention-based deep learning model for Arabic handwritten text recognition. Machine Graphics and Vision, 31(1/4), pp.49-73.

Ganai, A.F., and Khursheed, F., 2023. Computationally efficient recognition of unconstrained handwritten Urdu script using BERT with vision transformers. Neural Computing and Applications, 35(34), pp.24161-24177.

Hamad, K.A., and Shareef, S.M., 2025. Towards the digitization of Kurdish Handwritten script using deep learning: A comprehensive analysis. International Journal on Document Analysis and Recognition (IJDAR), 29(2), pp.302-330.

He, K., Zhang, X., Ren, S., and Sun, J., 2016. Deep Residual Learning for Image Recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp. 770-778.

Howard, A., Sandler, M., Chu, G., Chen, L.C., Chen, B., Tan, M., Wang, W., Zhu, Y., Pang, R., Vasudevan, V., Le, Q.V., and Adam, H., 2019. Searching for MobileNetV3. In: Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV).

Huang, G., Liu, Z., Van Der Maaten, L., and Weinberger, K.Q., 2017. Densely Connected Convolutional Networks. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. pp.4700-4708.

Ibrahim, R.M., Saman Nariman, G., and Majeed, H.D., 2023. offline kurdish character handwritten recognition (okchr) using cnn with various preprocessing techniques. Journal of Engineering Science and Technology, 18, pp.3113-3127.

Ingle, R.R., Fujii, Y., Deselaers, T., Baccash, J., and Popat, A.C., 2019. A Scalable Handwritten Text Recognition System. In: 2019 International Conference on Document Analysis and Recognition (ICDAR). pp.17-24.

Hamad, K., and Kaya, M., 2016. A detailed analysis of optical character recognition technology. International Journal of Applied Mathematics Electronics and Computers, 4(1), pp.244-249.

Li, M., Lv, T., Chen, J., Cui, L., Lu, Y., Florencio, D., Zhang, C., Li, Z., and Wei, F., 2023. TrOCR: Transformer-based optical character recognition with pre trained models. Proceedings of the AAAI Conference on Artificial Intelligence, 37(11), pp.13094-13102.

Mahmoud, S.A., Ahmad, I., Al-Khatib, W.G., Alshayeb, M., Tanvir Parvez, M., Märgner, V., and Fink, G.A., 2014. KHATT: An open Arabic offline handwritten text database. Pattern Recognition, 47(3), pp.1096-1112.

Momeni, S., and BabaAli, B., 2024. A transformer-based approach for Arabic offline handwritten text recognition. Signal, Image and Video Processing, 18(4), pp.3053-3062.

Rabaev, I., and Litvak, M., 2025. Recent advances in text line segmentation and baseline detection in historical document images: A systematic review. International Journal on Document Analysis and Recognition (IJDAR), 29, pp.3-39.

Rustay, S., Mohammed, J., Rasool, R., Taha, B., and Khdir, H., 2012. فەرهەنگی .First. Available from: https://kurdipedia.org/default .هەرز ە aspx?q=2016022118152294085 [Last accessed on 2024 Aug 19].

Shareef, S.M., and Ali, A.M., 2024. Deep learning-based digitization of Kurdish text handwritten in the e-government system. Indonesian Journal of Electrical Engineering and Computer Science, 35(3), p.1865.

Shen, X., and Messina, R., 2016. A Method of Synthesizing Handwritten Chinese Images for Data Augmentation. In: 2016 15th International Conference on Frontiers in Handwriting Recognition (ICFHR). pp.114-119.

Shi, B., Bai, X., and Yao, C., 2016. An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence 39(11), pp.2298-2304.

Tan, M., and Le, Q., 2019. EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks. In: Chaudhuri, K., and Salakhutdinov, R., Eds. Proceedings of the 36th International Conference on Machine Learning. Proceedings of Machine Learning Research. PMLR, pp.6105-6114. Available from: https://proceedings.mlr.press/v97/tan19a.html. [Last accessed on 2025 Sep 10].

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A., Kaiser, L., and Polosukhin, I., 2017. Attention is All you Need. Advances in Neural Information Processing Systems 30. Neural Information Processing Systems Foundation, Long Beach.

Veisi, H., MohammadAmini, M., and Hosseini, H., 2020. Toward Kurdish language processing: Experiments in collecting and processing the AsoSoft text corpus. Digital Scholarship in the Humanities, 35(1), pp.176-193.