RULE-BASED ALGORITHMS FOR HANDWRITTEN CHARACTER RECOGNITION

Document Type

Dissertation

Publication Date

2007

Abstract

Machine simulation of human reading has been the subject of intensive research for the last three decades. The interest devoted to this field is not explained only by the exciting challenges involved, but also the huge benefits that a system,designed in the context of a commercial application, could bring.Handwriting is a skill that is personal to individuals. It consists of artificial graphical marks on a surface; its purpose is to communicate something. It has continued to persist as a means of communication and recording information in day-to-day life due to the convenience of paper and pen as compared to keyboards for numerous day-to-day situations. The Character Recognition (CR) is the task of transforming language represented in its spatial form of graphical marks into its symbolic representation.The recognition of handwritten characters is quite difficult due to the wide variability of hand printing and cursive script. Most of the efforts done in the field of character recognition were dedicated to recognize Latin, Japanese and Chinese characters. Arabic CR faces technical problems not encountered in any other language, which makes the problem more challenging.In CR we have four major research directions varying according to the problem nature. These major research categories are:1. Recognition of Isolated Characters.2. Explicit Word Segmentation Before Recognition.3. Simultaneous recognition and segmentation of cursive writing.4. Global Whole Word recognition.The proposed work in this thesis is dedicated to handwritten Arabic character recognition. The problem is to be investigated from different sides: the isolated character and cursive handwriting problems, the off-line and on-line point of view, the single writer and multi-writer variability problems, single character decision and multi-decisions outputs. The objective is to achieve the best possible recognition accuracy using the most logical rule-based algorithms.   The thesis is divided to two parts:In the first part, we proposed a rule-based off-line character recognition system for isolated Arabic alphabet written by a single writer. The system was a single-output system. The most common off-line features used by most of the researchers were used but we were able to achieve high results, comparable to those achieved by other researchers in literature, by using feature fusion and proposing the idea of a multiple-classifier system besides using a classification hierarchy based on the structural features of Arabic characters.In the second part, we proposed the basic stages towards a system that addresses the problem of recognizing on-line cursive handwriting. A database of handwriting was collected from multi-writers, and a new technique for text line separation was used. Rule-based algorithms were used to perform simultaneous segmentation and recognition of word portions in an unconstrained cursively handwritten document using dynamic programming. The output of these stages is in the form of a ranked list of the possible decisions. We were able to correctly segmentand recognize most of the test words. The correct segmentation-recognition results were located in the top choices of the ranked list. In the future, linguistics can be used to select the best decision from this list.

Link to Full Text

Share

COinS