Please use this identifier to cite or link to this item: http://hdl.handle.net/10603/124420
Title: quotOnline HandwrittenWord Recognition for Indic Scripts using Hidden Markov Models and Data driven Modeling of Writing Stylesquot
Researcher: Bharath A.
Guide(s): Sriganesh Madhvanath
Keywords: quotHidden Markov Models, Indic Scriptsquot
University: Birla Institute of Technology and Science
Completed Date: 
Abstract: Online handwriting recognition (OHWR) refers to the problem of machine recognition of handwriting captured in the form of pen trajectories, via a digitizing tablet and stylus. Over the years,a number of algorithms have been proposed for recognizing online handwriting, and today there are several commercial systems for recognizing European and Oriental scripts. OHWR technology holds significant promise for the Indic family of scripts, given that the Indic languages are used by a sixth of the world s population, and the greater ease of use of handwriting-based text input compared to keyboard-based methods for these scripts. The structure of the scripts newlineand the variety of shapes and writing styles pose some unique challenges that make adoption of existing word recognition systems developed for English or other languages, difficult. While there has been considerable research on recognition of isolated symbols and characters in Indic newlinescripts, research for recognizing larger writing units such as words or phrases is in its early stages. The systems developed to date have assumed various constraints on writing, or have been script-specific, or both.In this thesis, we address the problem of online handwritten word recognition for Indic scripts. In contrast to prior approaches, we propose techniques that are script-independent and data-driven, involving minimal manual intervention during training, and validate our approach newlinefor two important Indic scripts - Devanagari and Tamil. Our approach employs Hidden Markov newlineModels (HMM) to model strokes, symbols and words in the script. The HMMs are trained newlineusing a large number of word samples collected from over a hundred writers for each script,cleaned and annotated at the symbol level. newlineFrom the perspective of recognition, this thesis addresses two central issues that arise in the context of online Indic scripts: (i) variations in writing style of individual symbols, (ii) symbol order variations within and across characters. Variations in writing style at the symbol level result
Pagination: 3.10 MB
URI: http://hdl.handle.net/10603/124420
Appears in Departments:Computer Science & Information Systems

Files in This Item:
File Description SizeFormat 
thesis.pdfAttached File3.18 MBAdobe PDFView/Open


Items in Shodhganga are protected by copyright, with all rights reserved, unless otherwise indicated.