Fast Online Character Recognition Using a Novel Local-Global Feature Extraction Method

Online handwriting recognition is one of the most active subjects of research in the field of pattern recognition. Feature extraction is one of the main steps in handwritten recognition which has a significant impact on accuracy and speed of recognition. Low time complexity of feature extraction met...

Full description

Saved in:
Bibliographic Details
Published in2021 12th International Conference on Information and Knowledge Technology (IKT) pp. 183 - 187
Main Authors Parvizi, Ayoub, Kazemifard, Mohammad, Imani, Ziba
Format Conference Proceeding
LanguageEnglish
Published IEEE 14.12.2021
Subjects
Angle Based Feature
Character recognition
Computational efficiency
Data mining
Deep learning
Feature extraction
Handwriting recognition
Local-Global Feature
Memory management
Online Character Recognition
Pattern Recognition
Online AccessGet full text

Cover

More Information
Summary:Online handwriting recognition is one of the most active subjects of research in the field of pattern recognition. Feature extraction is one of the main steps in handwritten recognition which has a significant impact on accuracy and speed of recognition. Low time complexity of feature extraction method and short feature vector length reduce the computational cost and memory consumption. In this paper, a new feature extraction method is presented for fast online Persian/Arabic character recognition. In this method, in order to recognize the character pattern, the character signal sequence is first segmented into several equal parts in terms of number of points. Then the angle between horizon line and transient line of first and last points of each segment is considered as local feature. The global features are the angles between horizon line and transient line of first point of the character and the end point of segments. This method has been tested on two standard datasets including numbers and characters of Persian/Arabic languages. The experimental results show that the proposed method, in addition to high recognition accuracy, has less computational complexity and shorter vector length compared to other new methods of feature extraction and handwriting recognition.
DOI:10.1109/IKT54664.2021.9685875