Conference paper

Handwriting Recognition

Recognition of handwritten text by computer can significantly reduce the cost of manual key entry in applications such as bank check processing, postal mail routing, census and poll form capture, medical documents, and many others. In this review, I will describe the structure and performance of a handwriting recognition system. This system has demonstrated competitive performance in an evaluation by the {US} {C}ensus {B}ureau and is currently being used in the 1995 {US} {T}est {C}ensus for transcribing handwritten names, telephone numbers, and other information into {ASCII} text. The system operates on unsegmented handwritten input strings (i.e., there are no preprinted boxes into which respondents write characters). Like more general machine vision tasks, handwriting recognition forces us to address questions of feature extraction, figure/ground, image segmentation, shape variation, and the integration of top-down knowledge.

    Keywords: vision


    • EPFL-CONF-82338

    Record created on 2006-03-10, modified on 2017-05-10


  • There is no available fulltext. Please contact the lab or the authors.

Related material