Evaluation of Print Reader Output Correction Study

摘要

This study has tested and evaluated a basic computer software model developed for assisting an optical character recognition unit in deciding character identity by context dependent factors. An operational environment textual data base was used for the study. It had errors introduced as a result of output from an optical character recognition device. Correction techniques used are based, not on full dictionary lookup, but on n-gram occurrence lists, common word dictionaries, environmental dictionaries, and character confusion tables. A conclusion of this evaluation is that the basic programmed model provides a reliable approach for correcting identified errors. It is flexible and accommodates adaptive or learning techniques and is most effective when information about the character in error is supplied by the device. (Author)