Comparison of Various Algorithms for Handwritten Character Recognition of Indian Languages

Abstract

In this paper, we present a comparison of various pre-processor, feature extraction methods and algorithms for handwritten character recognition of various Indian languages. Comparison of classifier, feature set and accuracy of offline handwritten character recognition of Gujarati, Devanagari, Gurmukhi, Kannada, Malayalam, Bangla and Hindi Indian languages. Comparison of classifier, feature set and accuracy of online handwritten character recognition of Assamese, Tamil, Devanagari, Malayalam, Gurmukhi, and Bangla Indian languages. Indian language wise best performance of each language is compared for both offline and online handwritten character recognition systems.

Introduction

I. INTRODUCTION

Optical character recognition system can be used to identify the handwritten characters. Handwritten character recognition (HCR) can be classified into two types - offline and online.

Offline handwritten character recognition system recognizes character written on paper or other such material using a pen or any device. In offline handwritten character recognition system, a character written on paper is converted into an image using a scanner or other imaging devices. The scanned image is further processed using different algorithms to remove noise, size variation etc. Preprocessed image is used to extract meaningful information from the written character. Extracted features are provided to the classifier as an input.

Indian languages have large and complex character set compare to English and other Latin scripts. Indian scripts include constants, vowels and composite characters representing a combination of constants and vowels. There is a similarity between characters of different Indian languages, mainly based on a geographic location of languages used.

Many characters in different Indian languages requires multiple strokes to write. Such a complex character set makes traditional keyboard not practical for Indian languages. Most of the Indian languages have major differences among each other and due to that, there cannot be a single handwritten character recognition system for all the Indian languages. We need to develop separate systems for every Indian language.

II. LITERATURE REVIEW

The character recognition system has training data of all character classes of a particular language. Offline handwritten character recognition is more complex compared to printed character recognition due to variation in writing style. In this system, handwritten documents are scanned and converted into digital image.

The scanned image is further processed using pre-processing methods, segmentation methods, text extraction methods, feature extraction methods and classification methods.

The character recognition system can be categorized into Offline and Online. We have compiled and compared different work of researchers of major Indian languages.

The following tables 1 to 7 shows and compare classifier, feature set and accuracy of offline handwritten character recognition of Gujarati, Devanagari, Gurmukhi, Kannada, Malayalam, Bangla and Hindi Indian languages.

Table 1. Comparison of character recognition system for Gujarati

Table 2. Comparison of character recognition system for Devanagari

Table 3. Comparison of character recognition system for Gurmukhi

Table 4. Comparison of character recognition system for Kannada

Table 5. Comparison of character recognition system for Malayalam

Table 6. Comparison of character recognition system for Bangla

Table 7. Comparison of character recognition system for Hindi

The following tables 8 to 14 shows and compare classifier, feature set and accuracy of online handwritten character recognition of Assamese, Tamil, Devanagari, Malayalam, Gurmukhi, and Bangla Indian languages.

Table 8. Comparison of character recognition system for Assamese

Table 9. Comparison of character recognition system for Tamil

Table 10. Comparison of character recognition system for Devanagari

Table 11. Comparison of character recognition system for Malayalam

Table 12. Comparison of character recognition system for Gurmukhi

Table 13. Comparison of character recognition system for Bangla

Table 14. Comparison of character recognition system for Gujarati       

III. RESULTS AND DISCUSSION

The comparison of various classifier, feature set and accuracy of offline and online handwritten character recognition for various Indian languages.  For offline handwritten character recognition, following are the best result achieved by the researcher for various Indian languages.

1. Gujarati: The classification was performed using a Support Vector Machine and naïve based classifiers. The feature set included chain code, zoning, and projection profile-based features and their possible combinations as a fusion feature set. The result showed the highest accuracy of 99.80% using SVM with the polynomial kernel classifier and chain code & zone-based features. [1]
2. Devanagari: In the pre-processing stage, smoothing, enhancing, and filtering methods were used. In the pre-classification stage, local and global features were used. The features were extracted using Zernike and Legendre moment. Classification was performed using multiclass SVM. The result showed a maximum accuracy of 98.30% for a handwritten character. [6]
3. Gurmukhi: The feature set had 117 feature values which included the local binary pattern with directional and regional features. Classification was performed using deep learning classifier. The result showed an accuracy of 99.3%. [11]
4. Kannada: In pre-processing skew detection and correction, binarization, noise removal, normalization, and thinning methods were used. The features set had 149 feature values which included structural and wavelet transform features. Classification was performed using the Artificial Neural Network classifier. The result showed an accuracy of 91%. [13]
5. Malayalam: Characters were separated based on foreground and background colours. The HLH patterns featured with horizontal, vertical and cross-sectional sequence check was used to identify a character. The authors obtained an accuracy of 88%. [16]
6. Bangla: Deep Belief Network (DBN) was used for classification. DBN learning process has two steps, unsupervised feature learning and supervised parameter tuning. The result showed an accuracy of 91.30%. [18]
7. Hindi: They used augmented data set with CNN. The proposed method achieved accuracy of 94.19% with average training time/epoch 8.43s. [22]

For online handwritten character recognition, following are the best result achieved by the researcher for various Indian languages.

a. Assamese: The Support Vector Machine was used for classification. The feature set included class-conditional probabilities features derived from a Gaussian mixture model. The result showed an accuracy of 97.67% for upper letters & processing time of 162.34 milliseconds, and 96.05% for lower letters & processing time of 335.56 milliseconds. The extended work of English word recognition showed an accuracy of 94.66%. [23]

b. Tamil: The proposed algorithm was a semi-supervised method which learns from labeled and unlabeled samples. The classification was performed using the naïve Bayes classifiers and the expectation-maximization algorithm. The result showed an accuracy of 91.81%. [28]

c. Devanagari: Normalization and smoothing pre-processing methods were used here. Single Engine Approach using SVM, Multiple SVM Engines and HMM were used for classification. The feature set included curve and coordinate points features. The result showed an accuracy of 97.27% using SVM and 83.08% using HMM. [31]

d. Malayalam: They used dot detection, dehooking, smoothing, thinning, loop detection, normalization, orientation normalization and equidistant resampling for pre-processing. They used normalized x-y co-ordinates, pen up/down, aspect ratio, curvature and writing direction for feature extraction. They used k-NN for classification. They achieved an accuracy of 98.12%. [36]

e. Gurmukhi: The Support Vector Machine with the RBF kernel was used for classification. The zone identification was performed using the x & y projection method. The feature set included x & y points, discrete Fourier transform features, and directional features. The result showed an accuracy of 94.8% for character and 99.75% for zone identification. [40]

f. Bangla: Classification with different strategies. Comparison between max pooling and average pooling schemes was done. The softmax and sigmoid activation functions were also compared. The result showed an accuracy of 99.40% using max pooling and softmax function. [45]

g. Gujarati: Classification with multi-layer classification using SVM at first layer and k-NN at second layer. The feature set is consist of derivative of pixel values, zoning and normalized chain code. The result showed an accuracy of 94.65%. [50]

Conclusion

Comparison of classifier, feature set and accuracy of offline handwritten character recognition of Gujarati, Devanagari, Gurmukhi, Kannada, Malayalam, Bangla and Hindi Indian languages and online handwritten character recognition of Assamese, Tamil, Devanagari, Malayalam, Gurmukhi, and Bangla Indian languages. For offline handwritten character recognition, Gujarati, Devanagari, Gurmukhi and Hindi language performed best among all Indian languages with SVM and deep learning algorithms. For online handwritten character recognition, Assamese, Devanagari, Gurmukhi, Bangla and Gujarati language performed best among all Indian languages with SVM and CNN learning algorithms.

References

[1] Sharma, P. Thakkar, D. M. Adhyaru, and T. H. Zaveri, “Features Fusion based Approach for Handwritten Gujarati Character Recognition”, NIRMA UNIVERISTY J. Eng. Technol., vol. 5, no. 2, pp. 13–19, 2016. [2] S. B. Sunilkumar, R. Y. Arvind, M. P. Mihir, and M. P. Mitul, “Gujarati handwritten character recognition using grey level co-occurrence matrix and dynamic time warping technique”, in International Conference On Smart Technologies For Smart Nation (SmartTechCon), 2017, pp. 340–346. [3] A. A. Desai, “Support vector machine for identification of handwritten Gujarati alphabets using hybrid feature space”, CSI Trans. ICT, vol. 2, no. January, pp. 235–241, 2015. [4] C. Patel and A. Desai, “Gujarati Handwritten Character Recognition Using Hybrid Method Based on Binary Tree-Classifier And K-Nearest Neighbour”, Int. J. Eng. Res. Technol., vol. 2, no. 6, pp. 2337–2345, 2013. [5] Dr. Mrs. Mamta Jagdish Baheti, “Principal Component Analysis and Support Vector Machine approach for Gujarati Handwritten Numeral Recognition”, International Journal of Scientific Research in Computer Science, Engineering and Information Technology, Volume 3, Issue 3 , 2018 [6] K.V.Kale, S.V.Chavan, M.M.Kazi, and Y.S.Rode, “Handwritten and Printed Devanagari Compound using Multiclass SVM Classifier with Orthogonal moment Feature”, Int. J. Comput. Appl., vol. 71, no. 24, pp. 31–37, 2013. [7] M. Bhalerao, S. Bonde, A. Nandedkar, and S. Pilawan, “Combined Classifier Approach for Offline Handwritten Devanagari Character Recognition Using Multiple Features”, Comput. Vis. Bio Inspired Comput., pp. 45–54, 2018. [8] D. Khanduja, N. Nain, and S. Panwar, “A Hybrid Feature Extraction Algorithm for Devanagari Script”, ACM Trans. Asian Low-Resource Lang. Inf. Process., vol. 15, no. 1, 2015. [9] R. Sharma and T. Mudgal, “Primitive Feature-Based Optical Character Recognition of the Devanagari Script”, Prog. Adv. Comput. Intell., vol. 714, pp. 249–259, 2018. [10] S. Ansari and U. Sutar, “Devanagari Handwritten Character Recognition using Hybrid Features Extraction and Feed Forward Neural Network Classifier ( FFNN )”, Int. J. Comput. Appl., vol. 129, no. 7, pp. 22–27, 2015. [11] N. Kumar and S. Gupta, “A Novel Handwritten Gurmukhi Character Recognition System Based on Deep Neural Networks”, Int. J. Pure Appl. Math., vol. 117, no. 21, pp. 663–678, 2017. [12] M. Kumar, M. K. Jindal, and R. K. Sharma, “A Novel Framework for Grading of Writers Using Offline Gurmukhi Characters”, Proc. Natl. Acad. Sci. India Sect. A - Phys. Sci., vol. 86, no. 3, pp. 405–415, 2016. [13] S. Pasha and M. C. Padma, “Handwritten Kannada character recognition using wavelet transform and structural features”, 2015 Int. Conf. Emerg. Res. Electron. Comput. Sci. Technol., pp. 346–351, 2015. [14] S. K. Niranjan, V. Kumar, G. Hemantha Kumar, and V. N. Manjunath Aradhya, “FLD based unconstrained handwritten kannada character recognition”, Proc. 2008 2nd Int. Conf. Futur. Gener. Commun. Networking, FGCN 2008, vol. 3, no. 3, pp. 7–10, 2008. [15] G. S. Veena, T. N. R. Kumar, and A. Sushma, “Handwritten Off-line Kannada Character / Word Recognition Using Hidden Markov Model”, in International Conference on Cognition and Recognition, 2018, pp. 357–369. [16] A. Abdul Rahiman, A. Shajan, A. Elizabeth, M. K. Divya, G. Manoj Kumar, and M. S. Rajasree, “Isolated handwritten Malayalam character recognition using HLH intensity patterns”, ICMLC 2010 - 2nd Int. Conf. Mach. Learn. Comput., pp. 147–151, 2010. [17] J. John, K. V. Pramod, and K. Balakrishnan, “Offline handwritten Malayalam Character Recognition based on chain code histogram”, 2011 Int. Conf. Emerg. Trends Electr. Comput. Technol. ICETECT 2011, pp. 736–741, 2011. [18] M. M. R. Sazal, S. K. Biswas, M. F. Amin, and K. Murase, “Bangla handwritten character recognition using deep belief network”, 2013 Int. Conf. Electr. Inf. Commun. Technol., pp. 1–5, 2014. [19] R. Pramanik and S. Bag, “Shape decomposition-based handwritten compound character recognition for Bangla OCR”, J. Vis. Commun. Image Represent., vol. 50, no. August 2017, pp. 123–134, 2018. [20] F. I. Alam and B. Banik, “Offline Isolated Bangla Handwritten Character Recognition Using Spatial Relationships”, in International Conference on Informatics, Electronics and Vision (ICIEV), 2013. [21] S. Ramachandrula, S. Jain, and H. Ravishankar, “Offline Handwritten Word Recognition in Hindi”, in the workshop on Document Analysis and Recognition, 2012, no. ii, pp. 49–54. [22] Ajay Indian, Karamjit Bhatia, and Krishan Kumar, “Offline Handwritten Hindi Character Recognition Using Deep Learning with Augmented Dataset”, Cyber Security in Intelligent Computing and Communications, 2022 https://doi.org/10.1007/978-981-16-8012-0_11 [23] S. Mandal, S. R. M. Prasanna, and S. Sundaram, “GMM posterior features for improving online handwriting recognition”, Expert Syst. Appl., vol. 97, pp. 421–433, 2018. [24] H. Choudhury, S. Mandal, S. Devnath, S. R. M. Prasanna, and S. Sundaram, “Combining HMM and SVM based stroke classifiers for online Assamese handwritten character recognition”, in 12th IEEE International Conference Electronics, Energy, Environment, Communication, Computer, Control: (E3-C3), INDICON 2015, 2016, pp. 1–6. [25] S. Mandal, H. Choudhury, S. R. M. Prasanna, and S. Sundaram, “Frequency count based two stage classification for online handwritten character recognition”, 2016 Int. Conf. Signal Process. Commun. SPCOM 2016, 2016. [26] H. Choudhury, S. Mandal, S. Devnath, S. R. M. Prasanna, and S. Sundaram, “Comparison of assamese character recognizer using stroke level and character level engines”, 2015 21st Natl. Conf. Commun. NCC 2015, 2015. [27] S. Mandal, S. R. M. Prasanna, and S. Sundaram, “An improved discriminative region selection methodology for online handwriting recognition”, Int. J. Doc. Anal. Recognit., 2018. [28] R. Kunwar, U. Pal, and M. Blumenstein, “Semi-supervised Online Learning of Handwritten Characters Using a Bayesian Classifier”, 2013 2nd IAPR Asian Conf. Pattern Recognit., pp. 717–721, 2013. [29] A. Bharath and S. Madhvanath, “HMM-based lexicon-driven and lexicon-free word recognition for online handwritten indic scripts”, IEEE Trans. Pattern Anal. Mach. Intell., vol. 34, no. 4, pp. 670–682, 2012. [30] K. H. Aparna, V. Subramanian, M. Kasirajan, G. V. Prakash, and V. S. Chakravarthy, “Online Handwriting Recognition for Tamil”, Ninth Int. Work. Front. Handwrit. Recognit., pp. 438–443, 2004. [31] H. Swethalakshmi, “Online handwritten character recognition of Devanagari and Telugu Characters using support vector machines”, Tenth Int. Work. Front. Handwrit. Recognit., pp. 1–6, 2006. [32] R. Ghosh and P. P. Roy, “Study of Zone-Based Feature for Online Handwritten Signature Recognition and Verification in Devanagari Script”, in International Conference on Computer Vision and Image Processing, 2017, pp. 523–530. [33] K. C. Santosh and E. Iwata, “Stroke-Based Cursive Character Recognition”, Adv. Character Recognit., 2012. [34] R. Ghosh and P. P. Roy, “Study of two zone-based features for online Bengali and Devanagari character recognition”, Proc. Int. Conf. Doc. Anal. Recognition, ICDAR, vol. 2015–Novem, pp. 401–405, 2015. [35] A. Bharath and S. Madhvanath, “HMM-based lexicon-driven and lexicon-free word recognition for online handwritten indic scripts”, IEEE Trans. Pattern Anal. Mach. Intell., vol. 34, no. 4, pp. 670–682, 2012. [36] M. Sreeraj and S. Idicula, “k-NN Based On-Line Handwritten Character Recognition System”, Integr. Intell. Comput. (ICIIC), 2010 First Int. Conf., pp. 171–176, 2010. [37] K. P. Primekumar and S. M. Idiculla, “On-line Malayalam Handwritten Character Recognition using HMM and SVM”, Int. Conf. Signal Process., Image Process. Pattern Recognit. [ ICSIPR], pp. 1–5, 2013. [38] A. Arora and A. M. Namboodiri, “A hybrid model for recognition of online handwriting in Indian scripts”, in International Conference on Frontiers in Handwriting Recognition, ICFHR 2010, 2010, pp. 433–438. [39] K. Baiju and S. K, “Online Recognition of Malayalam Handwritten Scripts - A comparison using KNN, MLP and SVM”, Intl. Conf. Adv. Comput. Commun. Informatics, no. Sept. 21-24, 2016, pp. 1–5, 2016. [40] H. Singh, R. K. Sharma, and V. P. Singh, “Efficient zone identification approach for the recognition of online handwritten Gurmukhi script”, Neural Comput. Appl., vol. 3456789, 2018. [41] S. Singh, A. Sharma, and I. Chhabra, “A dominant points-based feature extraction approach to recognize online handwritten strokes”, Int. J. Doc. Anal. Recognit., vol. 20, no. 1, pp. 37–58, 2017. [42] A. Sharma and K. Dahiya, “Online Handwriting Recognition of Gurmukhi and Devanagari Characters in Mobile Phone Devices”, in International Journal of Computer Applications, 2012, pp. 37–41. [43] R. Kaur and M. Singh, “Stroke based online handwritten Gurmukhi character recognition”, 2016 Int. Conf. Adv. Comput. Commun. Informatics, ICACCI 2016, pp. 598–601, 2016. [44] K. Verma and R. K. Sharma, “An Efficient Writing-Zone Identification Technique for Online Handwritten Gurmukhi Character Recognition”, Proc. Natl. Acad. Sci. India Sect. A Phys. Sci., 2017. [45] S. Sen, D. Shaoo, S. Paul, R. Sarkar, and K. Roy, “Online Handwritten Bangla Character Recognition Using CNN?: A Deep Learning Approach”, Intell. Eng. Informatics. Adv. Intell. Syst. Comput., vol. 695, pp. 413–420, 2018. [46] S. Sen, M. Mitra, A. Bhattacharyya, R. Sarkar, F. Schwenker, and K. Roy, “Feature Selection for Recognition of Online Handwritten Bangla Characters”, Neural Process. Lett., 2019. [47] S. Sen, A. Bhattacharyya, A. Das, R. Sarkar, and K. Roy, “Design of Novel Feature Vector for Recognition of Online Handwritten Bangla Basic Characters”, in the First International Conference on Intelligent Computing and Communication. Advances in Intelligent Systems and Computing, 2016, vol. 458, pp. 485–494. [48] S. Sen, R. Sarkar, K. Roy, and N. Hori, “Recognize Online Handwritten Bangla Characters Using Hausdorff Distance-Based Feature”, in the 5th International Conference on Frontiers in Intelligent Computing: Theory and Applications. Advances in Intelligent Systems and Computing, 2017, vol. 515, pp. 541–549. [49] S. Sen, A. Bhattacharyya, P. K. Singh, R. A. M. Sarkar, K. Roy, and D. Doermann, “Application of Structural and Topological Features to Recognize Online Handwritten Bangla Characters”, ACM Trans. Asian Low-Resour. Lang. Inf. Process., vol. 17, no. 3, 2018. [50] Vishal Naik and Apurva Desai, “Online Handwritten Gujarati Character Recognition: Two-Layer Classification Approach”, International Journal of Research in Advent Technology, Vol.7, No.6, June 2019 DOI: 10.32622/ijrat.762019128 [51] Vishal Naik and Apurva Desai, “Multi-layer Classification Approach for Online Handwritten Gujarati Character Recognition”, Computational Intelligence: Theories, Applications and Future Directions—Volume II, Advances in Intelligent Systems and Computing, pp 595-606,2019 https://doi.org/10.1007/978-981-13-1135-2_45 [52] Vishal Naik and Apurva Desai, “Online handwritten Gujarati character recognition using SVM, MLP, and K-NN”, 8th International Conference on Computing, Communication and Networking Technologies (ICCCNT), 2017 DOI: 10.1109/ICCCNT.2017.8203926 [53] Vishal Naik and Apurva Desai, “Online Handwritten Gujarati Numeral Recognition Using Support Vector Machine”, International Journal of Computer Sciences and Engineering, Vol.-6, Issue-9, Sept. 2018 DOI: 10.26438/ijcse/v6i9.416421 [54] Vishal Naik and Apurva Desai, “Online Handwritten Gujarati Word Recognition”, International Journal of Computer Vision and Image Processing, Vol-9, Issue-1, 2019 DOI: 10.4018/IJCVIP.2019010103

Copyright

Copyright © 2023 Dr. Vishal Naik, Heli Mehta. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.