Browse Prior Art Database

Sequential Micr Character Recognizer

IP.com Disclosure Number: IPCOM000062709D
Original Publication Date: 1986-Sep-01
Included in the Prior Art Database: 2005-Mar-09
Document File: 1 page(s) / 12K

Publishing Venue

IBM

Related People

Martin, WC: AUTHOR

Abstract

An unknown waveform representing a scanned character is analyzed by processing a signal through a sequence of different linear filters. The waveform corresponding to the first character is located by finding a well-defined peak. The character is then identified using sequential filters, and a typical waveform length value pre-assigned to the identified character is used to locate the general region in which the next character is likely to be found. A suitable character set is the set of 14 E13B MICR characters described in ANSI Standard X3.2-1970; however, it is applicable to a much less restrictive set.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 69% of the total text.

Page 1 of 1

Sequential Micr Character Recognizer

An unknown waveform representing a scanned character is analyzed by processing a signal through a sequence of different linear filters. The waveform corresponding to the first character is located by finding a well-defined peak. The character is then identified using sequential filters, and a typical waveform length value pre-assigned to the identified character is used to locate the general region in which the next character is likely to be found. A suitable character set is the set of 14 E13B MICR characters described in ANSI Standard X3.2-1970; however, it is applicable to a much less restrictive set.

The method assumes that a parameter common to all 14 characters can be estimated using different linear sequential (Kalman) filters, one for each possible character. An additi filter is used corresponding to the typical waveform of an isolated ink spot. The difference between the actual waveform measurement and the predicted value of the waveform measurement are examined to locate the filters producing the smallest difference or residual, i.e., a near-zero mean and small variance. Unlikely candidate characters can be dropped from consideration at early stages in the estimation process.

While any parameter associated with the character set can be estimated, the procedure selected here is to estimate the shift along the time axis of the waveform associated with each character relative to the peak of the first lobe of the wavefor...