Skip to main navigation Skip to search Skip to main content

Fast Handwriting Recognition for Indexing Historical Documents

  • IBM

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

18 Scopus citations

Abstract

Handwriting Recognition (HR) has been successfully used in several applications such as postal address interpretation [1], bank check reading [2], and forms reading[3]. These applications are all characterized by small or fixed lexicons afforded by contextual knowledge. Machine recognition of handwriting in historical documents presents two primary challenges: (i) large lexicons (over 10,000 words) leading to low recognition accuracy (less than 50%) and (ii) a need for high speed HR given the millions of handwritten manuscripts in Digital Library repositories and that the speed is usually inversely proportional to lexicon size. This paper addresses the issue of speed when dealing with large lexicons. We present several techniques to improve the processing speed for a gain of up to 7 times in matching time and describe a method whereby the large lexicon is divided into smaller sets and processed in parallel. With 4 processors 18 times speedup for the matching phase is achieved.

Original languageEnglish
Title of host publicationProceedings First International Workshop on Document Image Analysis for Libraries - DIAL 2004
Pages314-320
Number of pages7
DOIs
StatePublished - 2004
EventProceedings First International Workshop on Document Image Analysis for Libraries DIAL 2004 - Palo Alto, CA, United States
Duration: Jan 23 2004Jan 24 2004

Publication series

NameProceedings - First International Workshop on Document Image Analysis for Libraries - DIAL 2004

Conference

ConferenceProceedings First International Workshop on Document Image Analysis for Libraries DIAL 2004
Country/TerritoryUnited States
CityPalo Alto, CA
Period01/23/0401/24/04

Fingerprint

Dive into the research topics of 'Fast Handwriting Recognition for Indexing Historical Documents'. Together they form a unique fingerprint.

Cite this