Design of Novel Feature Vector for Recognition of Online Handwritten Bangla Basic Characters
In the present work, a new feature vector has been designed towards recognition of handwritten online Bangla basic characters. At first, Center of Gravity (CG) of a particular character sample is determined. After that a circle enclosing the character sam
- PDF / 218,295 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 17 Downloads / 159 Views
Abstract In the present work, a new feature vector has been designed towards recognition of handwritten online Bangla basic characters. At first, Center of Gravity (CG) of a particular character sample is determined. After that a circle enclosing the character sample is drawn whose radius is estimated as the distance of farthest data pixel from that CG. From this circular region, a 136-element feature vector is generated considering both the global as well as local information of the character sample. The feature set has been tested with several well-known classifiers on 10,000 isolated Bangla basic characters. Finally, Support Vector Machine (SVM) has produced 98.26 % recognition accuracy. Keywords Online character recognition Local feature
⋅
CG-based circle
⋅
Global feature
⋅
1 Introduction Online Handwriting Recognition (OHR) is now becoming an upcoming area of research due to exponentially increasing popularity of devices like Take Note, iPad, Smartphones etc. Also individuals from major part of the society are becoming Shibaprasad Sen (✉) ⋅ Ankan Bhattacharyya ⋅ Avik Das Future Institute of Engineering and Management, Kolkata, India e-mail: [email protected] Ankan Bhattacharyya e-mail: [email protected] Avik Das e-mail: [email protected] Ram Sarkar Jadavpur University, Kolkata, India e-mail: [email protected] Kaushik Roy West Bengal State University, Barasat, India e-mail: [email protected] © Springer Science+Business Media Singapore 2017 J.K. Mandal et al. (eds.), Proceedings of the First International Conference on Intelligent Computing and Communication, Advances in Intelligent Systems and Computing 458, DOI 10.1007/978-981-10-2035-3_50
485
486
Shibaprasad Sen et al.
habituated to write information freely on those devices in their natural handwriting style. In these devices, written data are saved as online information. Not only this saves extra time but also it reduces the chances of mistyping that may happen while writing with a keyboard. Hence, researchers are showing interest about OHR. Though some good research works are available for Devanagari [1–4], and English [5–8] scripts but while, talking about the Bangla script, researchers has paid a little attention which is evident from the limited research materials available in the literature. Authors in [9] have prompted an approach for estimating the features in an unsupervised way, based on disparity space that embeds the local neighborhoods surrounded by the pen positions in the trajectory. In [10], Bhattacharya, N. et al. have reported segmentation as well as recognition techniques for cursive online Bangla texts. In this work, after passing through segmentation module, texts are broken into set of primitives. Such primitives may represent the character or parts from basic or compound character set. A method for recognition of those primitives has been devised then. A different technique based on combination of online and offline feature extraction procedure, have been applied for segmentation of handwritten online Ba
Data Loading...
