Abstract

Handwriting recognition is a challenging problem in image processing and machine learning, and research in this area has expanded significantly in recent years. The Devanagari script, with its rich linguistic and cultural heritage, forms the foundation of several major Indian languages. Digitizing handwritten Devanagari scripts enhances their visibility, accessibility, and global acceptance. Convolutional neural networks (CNNs) have played a crucial role in improving image processing, pattern recognition, and character recognition accuracy. In this study, a deep learning–based system using a residual neural network (ResNet197) was developed, trained, and evaluated for recognizing handwritten Devanagari characters and digits. The model was tested on three benchmark datasets and one combined dataset to ensure robustness and generalization. On the Devanagari handwritten character dataset (DHCD) consisting of 92,000 samples from the University of California, Irvine (UCI) machine learning repository, the model achieved 99.82% training accuracy, 98.47% validation accuracy, 0.18% training error, 1.53% validation loss, and 98.47% ± 0.2% test accuracy. On the Sampoorna Hindi Akshar Barakhadi Digital (SHABD) Devanagari dataset containing approximately 304,150 characters across 384 classes, it achieved 99.78% training accuracy, 98.27% validation accuracy, 0.22% training error, 1.73% validation loss, and 99.91% ± 0.5% test accuracy. On the handwritten Devanagari characters – Vowels and numerals dataset (~52,000 samples across 59 classes) from Mendeley Data, the model achieved 99.46% training accuracy, 98.27% validation accuracy, 0.54% training error, 1.73% validation loss, and 97.46% ± 0.4% test accuracy. Finally, when trained and evaluated on a combined dataset merging the DHCD and handwritten Devanagari characters – Vowels and numerals datasets (approximately 143,721 samples after merging overlapping classes), the ResNet197 model achieved 98.76% training accuracy, 97.53% validation accuracy, 1.24% training error, 2.47% validation loss, and 97.53% ± 0.3% test accuracy. These results demonstrate the strong generalization capability and robustness of the proposed ResNet197 model across diverse handwritten Devanagari datasets, confirming its potential for practical applications in multilingual handwritten text recognition.

Keywords

Handwriting recognition, Devanagari script, Deep learning, Convolutional neural network (CNN), Residual neural network (ResNet), Character recognition.

Cite this article

Sharma MK. Devanagari handwritten characters and digits recognition using ResNet197.International Journal of Advanced Technology and Engineering Exploration.2025;12(131):1-3

References

[1] Yawalkar PM, Kharat MU. Automatic handwritten character recognition of Devanagari language: a hybrid training algorithm for neural network. Evolutionary Intelligence. 2022; 15(2):1499-516.

[2] Xiao X, Jin L, Yang Y, Yang W, Sun J, Chang T. Building fast and compact convolutional neural networks for offline handwritten Chinese character recognition. Pattern Recognition. 2017; 72:72-81.

[3] Jayanthi J, Maheswari PU. Comparative study: enhancing legibility of ancient Indian script images from diverse stone background structures using 34 different pre-processing methods. Heritage Science. 2024; 12(1):1-17.

[4] Gaikwad S, Nalbalwar S, Nandgaonkar A. Recognition of offline handwritten Devanagari characters using new mask-based approach, histogram of oriented gradients and AdaBoost. Multimedia Tools and Applications. 2023; 82(28):43883-902.

[5] Zhang Z, Eli E, Mamat H, Aysa A, Ubul K. EA-ConvNeXt: an approach to script identification in natural scenes based on edge flow and coordinate attention. Electronics. 2023; 12(13):11-8.

[6] Deore SP, Pravin A. Devanagari handwritten character recognition using fine-tuned deep convolutional neural network on trivial dataset. Sādhanā. 2020; 45(1):243.

[7] Chatterjee S, Mukherjee H, Sen S, Obaidullah SM, Roy K. City name recognition for Indian postal automation: exploring script dependent and independent approach. Multimedia Tools and Applications. 2024; 83(8):22371-94.

[8] Pham TA, Le HH, Do NT. Offline handwritten signature verification using local and global features. Annals of Mathematics and Artificial Intelligence. 2015; 75(1):231-47.

[9] Suganthi K, Shenoy DN. Cheque information processing and validation using deep learning. In global conference on communications and information technologies (GCCIT) 2024 (pp. 1-6). IEEE.

[10] Sharma C, Sharma S, Sakshi, Chen HY. Advancements in handwritten Devanagari character recognition: a study on transfer learning and VGG16 algorithm. Discover Applied Sciences. 2024; 6(12):1-15.

[11] Bisht M, Gupta R. Offline handwritten Devanagari modified character recognition using convolutional neural network. Sādhanā. 2021; 46(1):20.

[12] Yadav B, Indian A, Gupta CP. Offline handwritten Devanagari character recognition using modified capsule neural network. Procedia Computer Science. 2025; 258:2878-87.

[13] Truong TN, Nguyen CT, Zanibbi R, Mouchère H, Nakagawa M. A survey on handwritten mathematical expression recognition: the rise of encoder-decoder and GNN models. Pattern Recognition. 2024; 153:110531.

[14] Singh H, Sharma RK, Singh VP. Online handwriting recognition systems for Indic and non-Indic scripts: a review. Artificial Intelligence Review. 2021; 54(2):1525-79.

[15] Mishra M, Choudhury T, Sarkar T. Devanagari handwritten character recognition. In India council international subsections conference (INDISCON) 2021 (pp. 1-6). IEEE.

[16] Malhotra R, Addis MT. Ge’ez digit recognition model based on convolutional neural network. Applied Artificial Intelligence. 2024; 38(1):2400641.

[17] Choudhury H, Prasanna SM. Representation of online handwriting using multi-component sinusoidal model. Pattern Recognition. 2019; 91:200-15.

[18] Narang SR, Kumar M, Jindal MK. DeepNetDevanagari: a deep learning model for Devanagari ancient character recognition. Multimedia Tools and Applications. 2021; 80(13):20671-86.

[19] Lee H, Grosse R, Ranganath R, Ng AY. Convolutional deep belief networks for scalable unsupervised learning of hierarchical representations. In proceedings of the 26th annual international conference on machine learning 2009 (pp. 609-16). ACM.

[20] Shah KR, Badgujar DD. Devnagari handwritten character recognition (DHCR) for ancient documents: a review. In conference on information & communication technologies 2013 (pp. 656-60). IEEE.

[21] Shahzad NU, Wali A, Javaid A, Naeem M. A novel parallelnet model with domain-specific feature integration for handwritten pattern recognition. Signal, Image and Video Processing. 2025; 19(9):719.

[22] Boufenar C, Kerboua A, Batouche M. Investigation on deep learning for off-line handwritten Arabic character recognition. Cognitive Systems Research. 2018; 50:180-95.

[23] Lin D, Lin F, Lv Y, Cai F, Cao D. Chinese character CAPTCHA recognition and performance estimation via deep neural network. Neurocomputing. 2018; 288:11-9.

[24] Macwan JJ, Goswami MM, Vyas AN. A survey on offline handwritten North Indian script symbol recognition. In international conference on electrical, electronics, and optimization techniques (ICEEOT) 2016 (pp. 2747-52). IEEE.

[25] Alzubi J, Nayyar A, Kumar A. Machine learning from theory to algorithms: an overview. In journal of physics: conference series 2018 (pp. 1-16). IOP Publishing.

[26] Tawiah TA. A review of algorithms and techniques for image-based recognition and inference in mobile robotic systems. International Journal of Advanced Robotic Systems. 2020; 17(6):1-25.

[27] Mhapsekar M, Mhapsekar P, Mhatre A, Sawant V. Implementation of residual network (ResNet) for Devanagari handwritten character recognition. In advanced computing technologies and applications: proceedings of 2nd international conference on advanced computing technologies and applications 2020 (pp. 137-48). Singapore: Springer Singapore.

[28] Krizhevsky A, Sutskever I, Hinton GE. Imagenet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems. 2012; 60(6):84-90.

[29] Chakraborty S, Paul S. Bengali handwritten character transformation: basic to compound and compound to basic using convolutional neural network. In 2nd international conference on robotics, electrical and signal processing techniques 2021 (pp. 142-6). IEEE.

[30] Deore SP, Pravin A. On-line Devanagari handwritten character recognition using moments features. In international conference on recent trends in image processing and pattern recognition 2018 (pp. 37-48). Singapore: Springer Singapore.

[31] Ghosh K, Chakraborty A, Parui SK, Majumder P. Improving information retrieval performance on OCRed text in the absence of clean text ground truth. Information Processing & Management. 2016; 52(5):873-84.

[32] Singh R, Shukla AK, Mishra RK, Bedi SS. An improved approach for Devanagari handwritten characters recognition system. In applied information processing systems: proceedings of ICCET 2021 (pp. 217-26). Singapore: Springer Singapore.

[33] Agarwal B, Ramampiaro H, Langseth H, Ruocco M. A deep network model for paraphrase detection in short text messages. Information Processing & Management. 2018; 54(6):922-37.

[34] Jangid M, Srivastava S. Handwritten Devanagari character recognition using layer-wise training of deep convolutional neural networks and adaptive gradient methods. Journal of Imaging. 2018; 4(2):1-14.

[35] Dokare I, Gadge S, Kharde K, Bhere S, Jadhav R. Recognition of handwritten Devanagari character using convolutional neural network. In 3rd international conference on signal processing and communication (ICPSC) 2021 (pp. 353-9). IEEE.

[36] Manocha SK, Tewari P. Deep learning approaches for Devanagari handwriting recognition. In 9th international conference on reliability, infocom technologies and optimization (Trends and Future Directions) (ICRITO) 2021 (pp. 1-5). IEEE.

[37] Deshmukh A, Dhulekar P, Dharmale S. Handwritten devanagari character recognf dition. International Journal of Engineering Research & Technology (IJERT). 2014; 3(4):842–6.

[38] Pande SD, Chetty MS. Position invariant spline curve based image retrieval using control points. International Journal of Intelligent Engineering and Systems. 2019; 12(4):177-91.

[39] Pande SD, Jadhav PP, Joshi R, Sawant AD, Muddebihalkar V, Rathod S, et al. Digitization of handwritten Devanagari text using CNN transfer learning–a better customer service support. Neuroscience Informatics. 2022; 2(3):1-9.

[40] Lonare S, Patil R, Kachoria R. Devanagari handwritten character recognition using dynamic routing algorithm. In AI, IoT, big data and cloud computing for industry 4.0 2023 (pp. 275-87). Cham: Springer International Publishing.

[41] Wang J, Sun K, Cheng T, Jiang B, Deng C, Zhao Y, et al. Deep high-resolution representation learning for visual recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2020; 43(10):3349-64.

[42] Gurav Y, Bhagat P, Jadhav R, Sinha S. Devanagari handwritten character recognition using convolutional neural networks. In international conference on electrical, communication, and computer engineering (ICECCE) 2020 (pp. 1-6). IEEE.

[43] Gaikwad SS, Nalbalwar SL, Nandgaonkar AB. Devanagari handwritten characters recognition using DCT, geometric and hue moments feature extraction techniques. Sādhanā. 2022; 47(3):112.

[44] Deore SP. A DHCR_ SmartNet: a smart Devanagari handwritten character recognition using level-wised CNN architecture. Computer Science. 2022; 23:301-20.

[45] Guha R, Das N, Kundu M, Nasipuri M, Santosh KC. Devnet: an efficient CNN architecture for handwritten Devanagari character recognition. International Journal of Pattern Recognition and Artificial Intelligence. 2020; 34(12):2052009.

[46] Singh S, Garg NK, Kumar M. VGG16: offline handwritten Devanagari word recognition using transfer learning. Multimedia Tools and Applications. 2024; 83(29):72561-94.

[47] Ghanimi HM. Linear discriminant analysis-based deep learning algorithms for numerical character handwriting recognition. Journal of Autonomous Intelligence. 2024; 7(5):1-12.

[48] Alhamad HA, Shehab M, Shambour MK, Abu-hashem MA, Abuthawabeh A, Al-aqrabi H, et al. Handwritten recognition techniques: a comprehensive review. Symmetry. 2024; 16(6):1-25.

[49] Sethi M, Kumar M, Jindal MK. Forensic handwriting analysis: a hybrid classification framework for writer identification in Devanagari script. Multimedia Tools and Applications. 2025; 84:33183-94.

[50] Jabde M, Patil CH, Vibhute AD, Saini JR. A systematic review of multilingual numeral recognition systems. Artificial Intelligence Review. 2025; 58(4):1-27.

[51] Madhav KV, Veeramaneni A, Sriya PA, Sumanth K, Singh T, Duraisamy P. Handwritten Devanagari numeral recognition using deep learning. In 3rd international conference for innovation in technology (INOCON) 2024 (pp. 1-7). IEEE.

[52] Kukade J, Pawar RS, Rai I, Patidar M, Jain K, Pandit D. Decoding the Devanagari-handwritten Hindi recognition using deep learning methodology. In international conference on advances in computing research on science engineering and technology (ACROSET) 2024 (pp. 1-6). IEEE.

[53] Zhang A, Lipton ZC, Li M, Smola AJ. Dive into deep learning. Cambridge University Press; 2023.

[54] Yang L. Theoretical analysis of adam optimizer in the presence of gradient skewness. International Journal of Applied Science. 2024; 7(2):27-42.

[55] Acharya S, Pant AK, Gyawali PK. Deep learning based large scale handwritten Devanagari character recognition. In 9th international conference on software, knowledge, information management and applications (SKIMA) 2015 (pp. 1-6). IEEE.

[56] Pant AK, Panday SP, Joshi SR. Off-line Nepali handwritten character recognition using multilayer perceptron and radial basis function neural networks. In third Asian Himalayas international conference on internet 2012 (pp. 1-5). IEEE.

[57] https://www.kaggle.com/datasets/sushantshetty/shabd-complete-hindi-characters-dataset. Accessed 20 September 2025.

[58] Duddela SP, Mehta RVK. Handwritten Devanagari characters - vowels and numerals (38,750 isolated images). Mendeley Data. 2021.

[59] Yu C, Dian-ren C, Yang L, Lei C. Otsu's thresholding method based on gray level-gradient two-dimensional histogram. In 2nd international Asia conference on informatics in control, automation and robotics (CAR 2010) 2010 (pp. 282-5). IEEE.