FusionNet: A KNN-MLP Hybrid Model for Bengali Handwritten Digit Recognition using HOG and LBP Features
Abstract
Recent years have seen a surge of interest in research related to Bengali handwritten digit recognition, largely driven by its significant practical relevance and the pervasive utilization of the Bengali language. Convolutional Neural Networks (CNNs) have demonstrated notable success in this domain; however, hybrid approaches that integrate handcrafted feature extraction with conventional machine learning classifiers are emerging as effective alternatives. This study proposes and evaluates FusionNet, a hybrid model that combines the strengths of feature-based and learning-based methods through a two-stage classification pipeline. First, an optimized K-Nearest Neighbors (KNN) classifier generates a coarse label prediction based on handcrafted features. This prediction is then incorporated with origainal feature then fed into a Multi-Layer Perceptron (MLP), which performs the final classification. To enhance the system's robustness and generalization, few preprocessing techniques such as, binarization, Otsu’s threshold, and data augmentation were implemented. Then, two complementary feature extraction techniques were applied. Firstly, Histogram of Oriented Gradients (HOG) is utilized; and secondly, Local Binary Patterns (LBP). These features were computed in parallel to mitigate runtime overhead, thereby enabling reduced runtime. FusionNet's performance was benchmarked against EfficientNet-B0, a state-of-the-art pre-trained CNN model, using two datasets: a custom dataset reflecting diverse handwriting styles and the publicly available NumtaDb dataset. FusionNet attained an accuracy of 87% on the custom dataset and 96% on NumtaDb. In comparison, EfficientNet-B0 achieved 91% and 97%, respectively. Although EfficientNet-B0 exhibited marginally superior accuracy, FusionNet exhibited superior efficiency and lower computational demands, thus rendering it a compelling candidate for deployment in resource-constrained environments.References
A. Dalui, R. Sarkar, S. Sharma, A. Ghosh, S. Brahnam and S. R. Dash, ”A Deep Convolutional Neural Network Approach to Recognize Bangla
Handwritten Digits,” 2024 International Conference on Advancements in Smart, Secure and Intelligent Computing (ASSIC), Bhubaneswar, India, 2024, pp. 1-5, doi: 10.1109/ASSIC60049.2024.10507895.
Azgar, A., Imran Nazir, M., Akter, A., Saddam Hossain, M., Anwar
Hussen Wadud, M., Reazul Islam, M. (2024). MNIST Handwritten
Digit Recognition Using a Deep Learning-Based Modified Dual Input Convolutional Neural Network (DICNN) Model. In: Yang, XS., Sherratt, S., Dey, N., Joshi, A. (eds) Proceedings of Ninth International Congress on Information and Communication Technology. ICICT 2024
Lecture Notes in Networks and Systems, vol 1014. Springer, Singapore. https://doi.org/10.1007/978-981-97-3562-4 44
Ali, A., Senan, N., Murli, N. (2024). Convolutional Neural Network
Using Regularized Conditional Entropy Loss (CNNRCoE) for MNIST
Handwritten Digits Classification. In: Ghazali, R., Nawi, N.M., Deris,
M.M., Abawajy, J.H., Arbaiy, N. (eds) Recent Advances on Soft Computing and Data Mining. SCDM 2024. Lecture Notes in Networks and Systems, vol 1078. pringer, Cham. https://doi.org/10.1007/978-3031-66965-1 33
Pandey, Harikesh, and Arun Prakash Agrawal. ”Multi-lingual handwritten character recognition using Deep Learning.” In SCI, pp. 19-28. 2024
Amin, R.; Reza, M.S.; Okuyama, Y.; Tomioka, Y.; Shin,
J. A Fine-Tuned Hybrid Stacked CNN to Improve Bengali Handwritten Digit Recognition. Electronics 2023, 12, 3337.
https://doi.org/10.3390/electronics12153337
Chatterjee, S., Dutta, R.K., Ganguly, D., Chatterjee, K. and Roy, S., 2019. Bengali handwritten character classification using transfer learning on deep convolutional neural network. arXiv preprint arXiv:1902.11133.
Akhand, M.A.H., Ahmed, M. and Rahman, M.M., 2016. Convolutional Neural Network based Handwritten Bengali and Bengali English Mixed Numeral Recognition. International Journal of Image, Graphics & Signal Processing, 8(9).
Sufian, A., Ghosh, A., Naskar, A., Sultana, F., Sil, J. and Rahman, M.H., 2022. Bdnet: bengali handwritten numeral digit recognition based on densely connected convolutional neural networks. Journal of King Saud University-Computer and Information Sciences, 34(6), pp.2610-2620.
Maity, S., Dey, A., Chowdhury, A. and Banerjee, A., 2020. Handwritten Bengali character recognition using deep convolution neural network.
In Machine Learning, Image Processing, Network Security and Data Sciences: Second International Conference, MIND 2020, Silchar, India, July 30-31, 2020, Proceedings, Part II 2 (pp. 84-92). Springer Singapore.
Amin, R., Reza, M.S., Okuyama, Y., Tomioka, Y. and Shin, J., 2023. A Fine-Tuned Hybrid Stacked CNN to Improve Bengali Handwritten Digit Recognition. Electronics, 12(15), p.3337.
Azad, M.A., Singha, H.S. and Nahid, M.M.H., 2020, November. Bangla handwritten character recognition using deep convolutional autoencoder neural network. In 2020 2nd International Conference on Advanced Information and Communication Technology (ICAICT) (pp. 295-300). IEEE.
Mondal, S. and Mahfuz, N., 2020. Convolutional neural networks based bengali handwritten character recognition. In Cyber Security and Computer Science: Second EAI International Conference, ICONCS 2020, Dhaka, Bangladesh, February 15-16, 2020, Proceedings 2 (pp. 718-729). Springer International Publishing.
Datta, T., Purkaystha, B. and Islam, M.S., Bengali Handwritten Character Recognition Using Deep Convolutional Neural Network. In 2017 20th International Conference on Computer and Information Technology (ICCIT) (pp. 22-24).
Bappi, J.O., Rony, M.A.T. and Islam, M.S., 2024. BNVGLENET: Hypercomplex Bangla =handwriting character recognition with hierarchical class expansion using Convolutional Neural Networks. Natural Language Processing Journal, 7, p.100068.
Shawon, A., Rahman, M.J.U., Mahmud, F. and Zaman, M.A., 2018, September. Bangla handwritten digit recognition using deep cnn for large and unbiased dataset. In 2018 international conference on Bangla speech and language processing (ICBSLP) (pp. 1-6). IEEE.
https://doi.org/10.48550/arXiv.1806.02452
https://yann.lecun.com/exdb/mnist/
Ahamed, M., Kabir, R.B., Dipto, T.T., Al Mushabbir, M., Ahmed, S. and Kabir, M.H., Performance Analysis of Few-Shot Learning Approaches for Bangla Handwritten Character and Digit Recognition.
M. Tan and Q. Le, ”EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks,” International Conference on Machine Learning (ICML), pp. 6105-6114, 2019.
J. Deng, W. Dong, R. Socher, L. -J. Li, Kai Li and Li Fei-Fei, ”ImageNet: A large-scale hierarchical image database,” 2009 IEEE Conference on Computer Vision and Pattern Recognition, Miami, FL, USA, 2009, pp. 248-255, doi: 10.1109/CVPR.2009.5206848.
S. A. Chaudhari and R. M. Gulati, ”An OCR for separation and identification of mixed English — Gujarati digits using kNN classifier,” 2013 International Conference on Intelligent Systems and Signal Processing (ISSP), Vallabh Vidyanagar, India, 2013, pp. 190-193, doi: 10.1109/ISSP.2013.6526900
Matei, O., Pop, P.C. Valean, H. Optical character recognition in real˘ environments using neural networks and k-nearest neighbor. Appl Intell 39, 739–748 (2013). https://doi.org/10.1007/s10489-013-0456-2
Anukriti Rajput, Anish Kumar Singh. (2024). Handwritten Digit Recognition Accuracy Comparison Using Knn,Cnn And Svm. Educational Administration: Theory and Practice, 30(2), 638–643. https://doi.org/10.53555/kuey.v30i2.1676
DOI:
https://doi.org/10.31449/inf.v49i4.9117Downloads
Published
How to Cite
Issue
Section
License
Authors retain copyright in their work. By submitting to and publishing with Informatica, authors grant the publisher (Slovene Society Informatika) the non-exclusive right to publish, reproduce, and distribute the article and to identify itself as the original publisher.
All articles are published under the Creative Commons Attribution license CC BY 3.0. Under this license, others may share and adapt the work for any purpose, provided appropriate credit is given and changes (if any) are indicated.
Authors may deposit and share the submitted version, accepted manuscript, and published version, provided the original publication in Informatica is properly cited.
