World Families Forums

Special Forums - Note: You must be Logged In to Post. Anyone can browse. => Questions and News Articles of the Week => Topic started by: Marilyn Teaff Barton on April 08, 2005, 08:09:23 PM

Title: Wordspotter Searches Historical Documents
Post by: Marilyn Teaff Barton on April 08, 2005, 08:09:23 PM
NewsFactor Network (03/29/05); Martin, Mike

University of Massachusetts Amherst (UMass) computer science professor R. Manmatha has developed a computer interface that can search handwritten documents for information in a manner similar to the Google search engine. Toni Rath, a UMass grad student who helped create a demo of Manmatha's search tool, says the concept is comparable to searching text documents written in one language using queries in another language, as different handwriting is similar to different languages. He says, "Our system learns from a parallel body of transcribed scanned images. That is, the word images form a 'visual language.'" The UMass team's work was partially underwritten by the National Endowment for the Humanities and the National Science Foundation, and detailed at a recent ACM SIGIR conference. Xerox European Research Center director Chris Dance reports that even human readers can have trouble recognizing handwritten documents because of differences in the shapes of characters written by a single writer, while variability in pen-stroke thickness and writers' distinctive quirks are additional factors. Xerox Research Center information retrieval and machine learning expert Eric Gaussier says, "Most approaches in searching handwritten documents have taken the path of first solving the handwriting recognition problem, then solving the search problem with standard indexing and search techniques." The problem is that trying to address the enormously difficult challenge of handwriting recognition may be an even tougher proposition than handwritten document retrieval