Online Handwritten Character and Word Recognition in Indic scripts
Loading...
Date
item.page.authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Online handwriting recognition problem has been well known in pattern recognition and machine learning community for long. But, it is still a challenging
newlinetask to recognize accurately the online handwritten texts written in various Indic
newlinescripts like Devanagari, Bengali, Telugu and Tamil. These are four most popular Indic scripts. The works available in the literature on these Indic scripts
newlinevary from one script to other. Regarding character level recognition, researchers
newlinehave worked on Devanagari and Tamil scripts with both simple and compound
newlinecharacter recognition but works in Bengali script are confined only within simple characters. Real time requirements are for texts having both characters as
newlinewell as numerals. None of these scripts have been analyzed with numerals in
newlinecombination with characters. The present research work proposes two different
newlineapproaches - one without combining the outcomes of Support Vector Machine
newline(SVM) and Hidden Markov Model (HMM) classifiers and the other by combining the outcomes of these two classifiers, to recognize both online handwritten
newlinesimple and compound characters as well as numerals in Devanagari, Bengali and
newlineTamil scripts. The present research work trains the system initially by generating
newlineseparate training datasets for numerals, simple characters and compound characters and then a single training dataset for all these symbols. The first approach
newlineproposes novel zone-based feature extraction approaches, one of which is used
newlinein the second approach as well. Regarding isolated word level recognition, little
newlinenumber of research works are available in Devanagari, Bengali and Tamil scripts,
newlinebut works in Telugu script are confined within only character level recognition.
newlineThe present work also proposes two different approaches to develop an online
newlinehandwritten script identification and isolated word recognition system in different Indic scripts.