Usable OCR: what are the minimum performance requirements?
暂无分享,去创建一个
Forty-two subjects used a microcomputer and word processing software to type and proofread a 450-word document and then to correct errors in a number of other documents (of the same length) that had been created by OCR simulation [i.e., the documents looked like those typically obtained when using an optical character recognition (OCR) device for text entry]. The “OCR documents” contained both recognition errors (substitution errors, insertion errors, and deletion errors) and unrecognized characters. The percentage of characters requiring correction was varied from document to document. Text entry by OCR was found to be faster than manual entry (i.e., typing) if the OCR device can correctly recognize at least 94% of the individual alphanumeric characters. However, 98% correct recognition and computer-assisted proofreading were required in order to consistently obtain finished documents that had no more residual errors than typed documents.
[1] Örjan Ekeberg. Robust Dictionary Lookup Using Associative Networks , 1988, Int. J. Man Mach. Stud..
[2] Walter S. Rosenbaum,et al. Multifont OCR Postprocessing System , 1975, IBM J. Res. Dev..
[3] Stephen E. Newstead,et al. Efforts of display format on proof-reading with VDUs , 1988 .
[4] James L. Peterson,et al. Computer programs for detecting and correcting spelling errors , 1980, CACM.