Enhanced batched character image processing

• US 5,852,685 A
• Filed: 11/15/1995
• Issued: 12/22/1998
• Est. Priority Date: 07/26/1993
• Status: Expired due to Fees

• Current Assignee: David H Shepard
• Original Assignee: Cognitronics Imaging Systems Inc.
• Inventors: Shepard, David H.
• Primary Examiner(s): Johns, Andrew W.
• Assistant Examiner(s): Davis, Monica S.
• Application Number:
  US08/558,830
• Time in Patent Office: 1,133 Days
• Field of Search: 382/209, 382/219, 382/221, 382/310, 382/213, 382/218, 382/309, 382/311, 382/187
• US Class Current: 382/311
• CPC Class Codes: G06V 10/987 with the intervention of an operator

Abstract

Character recognition processing wherein each of a batch of documents is scanned to produce corresponding scan data signals forming a rectilinear data array of binary bits at the intersections of a rectangular coordinate grid. These signals are stored and processed by a recognition algorithm to produce identity signals for recognized characters. Groups of non-recognized characters are presented simultaneously to permit rapid identification by inspection. The identification of recognized characters is verified at high speed by simultaneously presenting the character images as respective groups sorted to have the same recognized identities. High accuracy recognition is assured by including with the stored characters to be verified a number of images of purposely-incorrectly-identified characters, i.e., bogus errors. At the end of predetermined processing segments, such as one batch of documents, the results of verification are examined to determine how many bogus errors were caught by the operator. If not all of the bogus errors present were caught, the operator may review the segment until all are caught. Statistical analysis of the data will provide assurance of high accuracy recognition.

Claims

1. In a process for verifying the tentatively determined identity of recognition-processed characters, the steps of:

storing data representing the images of a number of characters which have been tentatively identified in a predetermined computer-controlled algorithm providing a recognition-processing segment;

storing data representing the images of a number of characters which are purposely-incorrectly-identified (PII);

sorting said stored tentatively-identified and PII image data into groups of characters of the same identification;

merging by combining the tentatively-identified characters of one group with a number of said PII characters from a corresponding group having the same (but incorrect) identification as said one group;

simultaneously displaying in a single uniform composite array a number of stored character images from each of said groups of the same identification;

determining by inspection the presence of characters in the displayed groups failing to have the characterization common to those groups of the same identification;

determining those characters failing to have said common characterization which were purposely-incorrectly-identified (PII);

developing statistics reflecting the number of PII characters found in said inspection;

determining from said PII statistics whether a desired sufficient number of incorrectly-tentatively-identified characters has been found; and

when an insufficient number of such characters has been found, inspecting said displayed groups again in an effort to find additional characters failing to have the characterization common to those groups.

2. The method of claim 1, wherein the operator is at intervals informed of the number of PII characters that were inserted.

3. The method of claim 1, wherein the operator is at intervals informed of the number of PII characters corrected since the last report to the operator.

4. The method of claim 1, including the step of utilizing said statistics to specify the probable accuracy to be achieved in verifying data.

5. The method of claim 1, including the stop of predetermining the probable accuracy with expectation of achievement by setting the rate of PII character insertion.

6. The method of claim 1, wherein the groups of characters to be verified are displayed in arrays of approximately ten by ten character images.

7. The method of claim 1, wherein the groups of characters to be verified are displayed in two-dimensional arrays of no more than ten character images in either dimension.

8. The method of claim 1, including the steps of:

establishing a desired completed accuracy rate for the recognition process;

calculating the rate of PII character insertion required to effect said established completed accuracy rate; and

merging the calculated number of PII characters with a group of sorted and tentatively identified characters to effect said desired accuracy rate.