A semi-automatic annotation tool for Arabic online handwritten text

Document Type

Dissertation

Publication Date

2012

Abstract

Digital documents are now omnipresent. Techniques and algorithms to organise, process, and understand these documents are still developing.Evolving digitizers like Tablet PCs, Palmtops, and many other handled pen-based devices are wide spreading. Thousands of online documents are generated by them. These documents are non-textual documents of textual content. The digitizer device captures a writer's ink information, his speed, the pressure applied etc. Algorithms for accessing such documents at the content-level are still missing, especially for Arabic Language. This thesis addresses a fundamental problem in this area: Annotation.Annotation is the process of identifying an object with a textual description that precisely describes the object or entities that make up the object. The need for annotating documents is increasingly getting important. Annotated datasets of handwriting are a prerequisite for the design and training of handwriting recognition algorithms and information retrieval search engines.Manual annotation of handwriting datasets is laborious and error-proneprocess, especially if one were to annotate handwriting at the character or stroke level. The problem becomes compounded for handwritten documents,where the characters correspond to one or more strokes. There have been approaches that do automatic and semi-automatic annotation of handwriting data. However, the annotation is still highly manual. In this thesis we propose novel, automated algorithms which make this task easy and fast through a user-friendly annotation tool for online handwriting data at the character level, given a parallel dataset of online handwritten data and the corresponding typed text(ground truth).   In short, the contributions of this thesis are:1. Development of the first semi-automatic annotation tool for Arabic online handwriting data that provides transcriptions essentially needed by every handwriting recognition engine.2. The annotation tool saves more than 50% of the human time needed for manual annotation.3. Introduction of the first Arabic online handwritten sentence dataset forhandwriting text line recognition engines.4. Development of text line extraction system for Arabic online handwrittenscripts using dynamic programming based on a new cost function design. The system successfully extracts over 98% of the text lines.5. Development of novel word extraction system from Arabic online handwritten text lines using multiple classifiers' decision fusion; thatoutperforms single classifiers results.6. Development of a HMM-based segmentation-annotation system for annotating Arabic online handwritten words with the help of parallel text using a novel HMM design; that outperforms the common HMM design results.

Link to Full Text

Share

COinS