System and method for enhanced character recogngition accuracy by adaptive probability weighting

• US 5,455,872 A
• Filed: 04/26/1993
• Issued: 10/03/1995
• Est. Priority Date: 04/26/1993
• Status: Expired due to Term

• Current Assignee: Lockheed Martin Corporation (Martin Marietta Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Bradley, Martin P. T.
• Primary Examiner(s): Mancuso, Joseph
• Assistant Examiner(s): NOT, DEFINED
• Application Number:
  US08/051,972
• Time in Patent Office: 890 Days
• Field of Search: 382/36, 382/38, 382/39, 382/57, 382/61, 382/10, 382/27, 382/37, 382/30, 395/149
• US Class Current: 382/228
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data

Abstract

A data processing system and method are disclosed for selecting which one of several character recognition programs should be used to optimize the accuracy in recognizing characters in a field in an image of a document form. Consideration is taken of the character form and field type for particular characters and an optimized selection is performed on a realtime basis among the several candidate character recognition programs which could be applied. The resulting character recognition operation has its accuracy maximized for reading a wide variety of character forms and field types in recognition of preprinted forms.

Claims

1. In a data processing system, including a scanner, a plurality of character recognition means, a recognition station processor and a monitoring and correction station processor, a method for selecting which one of said plurality of character recognition means to use for recognizing a character in a field in a document image, comprising the steps of:

generating a character form confidence factor in the recognition station processor, for each of said plurality of character recognition means;

generating a field type confidence factor in the recognition station processor, for each of said plurality of character recognition means;

inputting an adaptive probability weighting factor from said monitoring and correction station processor to said recognition station processor, for each of said plurality of character recognition means;

generating a first guess character and first confidence value and a second guess character and second confidence value, using each of said plurality of character recognition means;

computing a recognition means choice confidence factor in said recognition station processor, as a product of said character form confidence factor, said field type confidence factor, and said adaptive probability weighting factor, times a difference between said first confidence value and said second confidence value; and

selecting one of said plurality of character recognition means in said data processing system, having a maximum value for said recognition means choice confidence factor.

2. The method of claim 1, which further comprises:

accumulating an error count in said monitoring and correction station processor for one of said plurality of character recognition means in said data processing system;

computing a new value for said adaptive probability weighting factor in said monitoring and correction station processor, for at least one of said plurality of character recognition means, by modifying said adaptive probability weighting factor with a value derived from said error count.

3. The method of claim 2, which further comprises:

computing a recognition means choice confidence factor in said recognition station processor, as the product of said character form confidence factor, said field type confidence factor, and said new value for said adaptive probability weighting factor, times a difference between said first confidence value and said second confidence value.

4. The method of claim 1, which further comprises:

monitoring and correcting said first guess character in a manual monitoring and correction station processor in said data processing system;

accumulating an error count for one of said plurality of character recognition means in said monitoring and correction station processor; and

computing a new value for said adaptive probability weighting factor in said monitoring and correction station processor, for at least one of said plurality of character recognition means, by modifying said adaptive probability weighting factor with a value derived from said error count.

5. The method of claim 1, which further comprises:

monitoring and correcting said first guess character in an automatic monitoring and correction station processor in said data processing system;

accumulating an error count for one of said plurality of character recognition means in said data processing system; and

computing a new value for said adaptive probability weighting factor in said recognition station processor, for at least one of said plurality of character recognition means, by modifying said adaptive probability weighting derived from said error count.

6. In a data processing system, including a plurality of character recognition means, a recognition station processor and a monitoring and correction station processor, a method for selecting which one of said plurality of character recognition means to use for recognizing a character in a field in a document image, comprising the steps of:

inputting an adaptive probability weighting factor from said monitoring and correction station processor to said station recognition processor for each of said plurality of character recognition means;

generating a first guess character and first confidence value and a second guess character and second confidence value in said character recognition means, using each of said plurality of character recognition means;

computing a recognition means choice confidence factor in said recognition station processor, as a function of a product comprising an adaptive probability weighting factor times a difference between said first confidence value and said second confidence value; and

selecting in said recognition station processor one of said plurality of character recognition means, having a maximum value for said recognition means choice confidence factor.

7. In a data processing system, including a plurality of character recognition means and a recognition station processor, a method for selecting which one of said plurality of character recognition means to use for recognizing a character in a field in a document image, comprising the steps of:

generating a character form confidence factor in said recognition station processor, for each of said plurality of character recognition means;

generating a field type confidence factor in said recognition station processor, for each of said plurality of character recognition means;

generating a first guess character and first confidence value and a second guess character and second confidence value using each of said plurality of character recognition means;

computing a recognition means choice confidence factor in said recognition station processor, as a product of said character form confidence factor and said field type confidence factor, times a difference between said first confidence value and said second confidence value; and

selecting in said recognition station processor one of said plurality of character recognition means having a maximum value for said recognition means choice confidence factor.

8. In a data processing system, including a plurality of:

character recognition means, a recognition station processor and a monitoring and correction station processor, a method for selecting which one of said plurality of character recognition means to use for recognizing a character in a field in a document image, comprising the steps of;

inputting an adaptive probability weighting factor from said monitoring and correction station processor to said recognition station processor for each of said plurality of character recognition means;

generating a first guess character and first confidence value and a second guess character and second confidence value using each of said plurality of character recognition means;

computing a recognition means choice confidence factor in said recognition station processor as a product of said adaptive probability weighting factor, times a difference between said first confidence value and said second confidence value; and

selecting in said recognition station processor one of said plurality of character recognition means having a maximum value for said recognition means choice confidence factor.

9. The method of claim 8, which further comprises:

accumulating an error count for one of said plurality of character recognition means in said monitoring and correction station processor;

computing a new value for said adaptive probability weighting factor in said monitoring and correction station processor, for at least one of said plurality of character recognition means, by modifying said adaptive probability weighting factor with a value derived from said error count.

10. The method of claim 9, which further comprises:

computing a recognition means choice confidence factor in said recognition station processor, as a product of said new value for said adaptive probability weighting factor, times a difference between said first confidence value and said second confidence value.

11. The method of claim 8, which further comprises:

monitoring and correcting said first guess character in a manual monitoring and correction station processor in said data processing system;

accumulating an error count for one of said plurality of character recognition means in said manual monitoring and correction station processor;

computing a new value for said adaptive probability weighting factor in said recognition station processor, for at least one of said plurality of character recognition means, by modifying said adaptive probability weighting factor with a value derived from said error count.

12. The method of claim 8, which further comprises:

monitoring and correcting said first guess character in the monitoring and correction station processor operated in an automatic mode;

accumulating an error count for one of said plurality of character recognition means in said monitoring and correction station processor;

computing a new value for said adaptive probability weighting factor in said recognition station processor, for at least one of said plurality of character recognition means, by modifying said adaptive probability weighting factor with a value derived from said error count.

13. In a data processing system, including a document image input means, a forms recognition means, a field extraction means, a recognition station processor, a monitoring and correction station processor and a plurality of character recognition means, a method for selecting which one of said plurality of character recognition means to use for a character in a field in a document image, comprising the steps of:

generating a character form confidence factor in said recognition station processor, for each of said plurality of character recognition means;

generating a field type confidence factor in said recognition station processor for each of said plurality of character recognition means;

generating an adaptive probability weighting factor in said recognition station processor, for each of said plurality of character recognition means;

generating a first guess character and first confidence value and a second guess character and second confidence value using each of said plurality of character recognition means;

computing an OCR engine choice confidence factor in said recognition station processor, as a product of said character form confidence factor, said field type confidence factor, and said adaptive probability weighting factor, times a difference between said first confidence value and said second confidence value;

selecting in said recognition station processor one of said plurality of character recognition means having a maximum value for said OCR engine choice confidence factor;

accumulating an error count for one of said plurality of character recognition means in said monitoring and correction station processor;

computing a new value for said adaptive probability weighting factor in said monitoring and correction station processor, for each of said plurality of character recognition means, by modifying said adaptive probability weighting factor with a value derived from said error count.

14. In a data processing system, including a plurality of character recognition means, an apparatus for selecting which one of said plurality of character recognition means to use for a character in a field in a document image, comprising:

a memory for storing a character form confidence factor for each of said plurality of character recognition means;

said memory storing a field type confidence factor for each of said plurality of character recognition means;

said memory storing an adaptive probability weighting factor for each of said plurality of character recognition means;

each of said plurality of character recognition means generating a first guess character and first confidence value and a second guess character and second confidence value;

a first processor means coupled to said memory and to at least one of said plurality of character recognition means, for computing a recognition means choice confidence factor, as a product of said character form confidence factor, said field type confidence factor, and said adaptive probability weighting factor, times a difference between said first confidence value and said second confidence value; and

said first processor means including means for selecting one of said plurality of character recognition means in said data processing system, having a maximum value for said recognition means choice confidence factor.

15. The data processing system of claim 14, which further comprises:

a second processor including accumulating means coupled to said means for selection for accumulating an error count of said first guess character;

said second processor means computing a new value for said adaptive probability weighting factor for at least one of said plurality of character recognition means, by modifying said adaptive probability weighting factor with a value derived from said error count.

16. The data processing system of claim 15, which further comprises:

said first processor means coupled to said second processor means, for receiving said new value for said adaptive probability weighting factor and computing a recognition means choice confidence factor in said data processing system, as a product of said character form confidence factor, said field type confidence factor, and said new value for said adaptive probability weighting factor, times a difference between said first confidence value and said second confidence value.

17. The data processing system of claim 16, which further comprises:

a manual monitoring and correcting station in said data processing system, which houses said accumulator means and second processor means;

display means in said manual station, coupled to said selection means, for presenting said document image and said first guess character to an operator; and

input means in said manual station for receiving a corrected character from said operator, to substitute for said first guess character.

18. The data processing system of claim 16, which further comprises:

an automatic monitoring and correcting means in said data processing system, coupled to said selection means, for automatically detecting an error in said first guess character and generating a corrected character to substitute for said first guess character.

19. In a data processing system, including a plurality of character recognition means a recognition station processor, and a monitoring and station processor, an apparatus for selecting which one of said plurality of character recognition means to use for a character in a field in a document image, comprising:

each of said plurality of character recognition means generating a first guess character and first confidence value and a second guess character and second confidence value;

means for inputting an adaptive probability weighting factor from said monitoring and correction station processor to said recognition station processor for each of said plurality of character recognition means;

a first processor means coupled to said memory and to each of said plurality of character recognition means, for computing a recognition means choice confidence factor, as a function of a product comprising an adaptive probability weighting factor times a difference between said first confidence value and said second confidence value;

selection means included in said first processor means, for selecting one of said plurality of character recognition means in said data processing system, having a maximum value for said recognition means choice confidence factor.

20. In a data processing system, including a plurality of character recognition means, an apparatus for selecting which one of said plurality of character recognition means to use for a character in a field in a document image, comprising:

a memory for storing a character form confidence factor for each of said plurality of character recognition means;

said memory storing a field type confidence factor for each of said plurality of character recognition means;

each of said plurality of character recognition means generating a first guess character and first confidence value and a second guess character and second confidence value;

a first processor means included in said memory and to each of said plurality of character recognition means, for computing a recognition means choice confidence factor, as a product of said character form confidence factor and said field type confidence factor, times a difference between said first confidence value and said second confidence value; and

selection means coupled to said first processor means, for selecting one of said plurality of character recognition means in said data processing system, having a maximum value for said recognition means choice confidence factor.

21. In a data processing system, including a plurality of character recognition means, an apparatus for selecting which one of said plurality of character recognition means to use for a character in a field in a document image, comprising:

a memory for storing an adaptive probability weighting factor for each of said plurality of character recognition means;

each of said plurality of character recognition means generating a first guess character and first confidence value and a second guess character and second confidence value;

a first processor means coupled to said memory and to each of said plurality of character recognition means, for computing a recognition means choice confidence factor, as a product of said adaptive probability weighting factor, times a difference between said first confidence value and said second confidence value;

selection means coupled to said first processor means, for selecting one of said plurality of character recognition means in said data processing system, having a maximum value for said recognition means choice confidence factor.

22. The data processing system of claim 21, which further comprises:

a second processor means including an accumulating means coupled to said selection means, for accumulating an error count of said first guess character; and

second processor means computing a new value for said adaptive probability weighting factor for at least one of said plurality of character recognition means, by modifying said adaptive probability weighting factor with a value derived from said error count.

23. The data processing system of claim 22, which further comprises:

said first processor means coupled to said second processor means, for receiving said new value for said adaptive probability weighting factor and computing a recognition means choice confidence factor in said data processing system, as a product of said new value for said adaptive probability weighting factor, times a difference between said first confidence value and said second confidence value.

24. The data processing system of claim 23, which further comprises:

a manual monitoring and correcting station in said data processing system, which houses said accumulator means and second processor means;

display means in said manual station, coupled to said selection means, for presenting said document image and said first guess character to an operator;

input means in said manual station for receiving a corrected character from said operator, to substitute for said first guess character.

25. The data processing system of claim 23, which further comprises:

an automatic monitoring and correcting means in said data processing system, coupled to said selection means, for automatically detecting an error in said first guess character and generating a corrected character to substitute for said first guess character.