Method for optical recognition of a multi-language set of letters with diacritics

• US 6,920,247 B1
• Filed: 11/01/2000
• Issued: 07/19/2005
• Est. Priority Date: 06/27/2000
• Status: Expired due to Term

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Cardiff Software, Inc. (Healthstream Incorporated)
• Inventors: Mayzlin, Isaac, Deere, Emily Ann
• Primary Examiner(s): Ahmed, Samir
• Assistant Examiner(s): LE, BRIAN Q
• Application Number:
  US09/704,043
• Time in Patent Office: 1,721 Days
• Field of Search: 382173-180, 382182-198, 382199-200, 382/229, 382/321
• US Class Current: 382/185
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/15 Cutting or merging image elements, e.g. region growing, watershed or clustering-based techniques

Abstract

The present invention is a method for recognizing non-English alpha characters that contain diacritics. An image analysis separates the character into its constituent components. The one or more diacritic components are then distinguished and isolated from the base portion of the character. Optical recognition is performed separately on the base portion. The diacritic is recognized through a special image analysis and pattern recognition algorithms. The image analysis extracts geometric information from the one or more diacritic components. The extracted information is used as input for the pattern recognition algorithms. The output is a code that corresponds to a particular diacritic. The recognized base portion and diacritic are combined and a check is performed for acceptable combinations in a chosen language. By separately recognizing the base portion and diacritic, the character sets used by the recognizer can be narrowed, resulting in greater recognition.

Claims

1. A method of identifying multi-language characters in an optical recognition system, comprising:

digitizing a document having characters printed thereon,

targeting each of the characters for recognition analysis,

separating characters having a single component from characters having more than a single component,

recognizing characters having a single component,

segmenting characters having more than a single component into constituent components, said constituent components including a base component and at least one diacritic component forming a diacritic,

recognizing the base component,

recognizing the diacritic through analysis of the at least one diacritic component, determining through a match analysis whether the diacritic can be used in combination with the base component, and

recognizing the combination of the diacritic and base component in response to a match.

2. The method of claim 1, wherein:

the digitizing step is performed by one of a scanning device, a facsimile machine and a digital camera.

3. The method of claim 1, wherein:

said document is a form document containing a plurality of form fields, each of said form fields being capable of having a character type designator and said targeting step is only performed on said form fields when said character typos designator is an alpha character type designator.

4. The method of claim 1, further comprising:

establishing a rectangle of interest about a least one diacritic component to isolate the diacritic into a separately defined region.

5. The method of claim 4, further comprising:

creating through the digitizing step a binary bitmap having groupings of black pixels and white pixels.

6. The method of claim 5, further comprising:

filtering out black pixel groupings not exceeding a predetermined minimum number of pixels.

7. The method of claim 5, further comprising:

extracting geometric information from the plurality of binary bitmap located within the rectangle of interest by intersecting the diacritic with a plurality of vectors, and

processing the geometric information as input to special pattern recognition algorithms to determine the diacritic most likely corresponding to the input.

8. The method of claim 7, wherein:

the vectors used for extracting geometric information are XMid, YMid, YTop and YBot, where XMid bisects the rectangle of interest along the y-axis, YMid bisects the rectangle of interest along the x-axis, and YTop and YBot intersect the rectangle of interest along the x-axis at predefined distances from the top and bottom of the rectangle of interest.

9. The method of claim 8, further comprising:

recognizing a macron diacritic by the following extracted geometric information;

a diacritic width exceeding twice a diacritic height,

one intersection formed by the vector YMid, and

the diacritic width exceeding the intersective length formed by the vector YMid;

recognizing a diaresis diacritic by the following extracted geometric information;

no intersections formed by the vector XMid;

recognizing a dot above diacritic corresponds by the following extracted geometric information;

three times the number of black pixels exceeding twice the diacritic width multiplied by the diacritic height;

recognizing a grave diacritic corresponds by the following extracted geometric information;

one intersection formed by the vector YTop,

one intersection formed by the vector YBot, and

the x-coordinate of the first intersective midpoint formed by the vector YTop is less than the x-coordinate of the first intersective midpoint formed by the vector YBot;

recognizing an acute diacritic by the following extracted geometric information;

one intersection formed by the vector YTop,

one intersection formed by the vector YBot, and

the x-coordinate of the first intersective midpoint formed by the vector YTop less than the x-coordinate of the first intersective midpoint formed by the vector YBot;

recognizing a caron or breve diacritic by the following extracted geometric information;

two intersections formed by the vector YTop,

the x-coordinate of the first intersective midpoint formed by the vector YTop less than the x-coordinate of the first intersective midpoint formed by the vector YMid, and

the x-coordinate of the first intersective midpoint formed by the vector YMid less than the x-coordinate of the second intersective midpoint formed by the vector YTop;

recognizing a circumflex diacritic by the following extracted geometric information;

two intersections formed by the vector YBot,

the x-coordinate of the first intersective midpoint formed by the vector YBot less than the first intersective midpoint formed by the vector YMid, and

the x-coordinate of the first intersective midpoint formed by the vector YMid less than the x-coordinate of the second intersective midpoint formed by the vector YBot;

recognizing a ring above diacritic by the following extracted geometric information;

two intersections formed by the vector XMid;

recognizing a circumflex diacritic by the following extracted geometric information;

one intersection formed by the vector YTop,

the x-coordinate of the first intersective midpoint formed by the vector YTop greater than the first intersective midpoint formed by the vector YMid, and

the x-coordinate of the first intersective midpoint formed by the vector YTop less than the second intersective midpoint formed by the vector YMid;

recognizing a circumflex diacritic by the following extracted geometric information;

two intersections formed by the vector YTop,

the x-coordinate of the first intersective midpoint formed by the vector YBot less than the first intersective midpoint formed by the vector YTop, and

the x-coordinate of the second intersective midpoint formed by the vector YTop less than the second intersective midpoint formed by the vector YBot;

recognizing a caron and a breve diacritic by the following extracted geometric information;

one intersection formed by the vector YBot,

the x-coordinate of the first intersective midpoint formed by the vector YBot greater than the first intersective midpoint formed by the vector YMid, and

the x-coordinate of the first intersective midpoint formed by the vector YBot less than the second intersective midpoint formed by the vector YMid;

recognizing a caron and a breve diacritic by the following extracted geometric information;

two intersections formed by the vector YTop,

two intersections formed by the vector YBot,

the x-coordinate of the first intersective midpoint formed by the vector YTop less than the first intersective midpoint formed by the vector YBot, and

the x-coordinate of the second intersective midpoint formed by the vector YBot less than the second intersective midpoint formed by the vector YTop.

10. The method of claim 9, further comprising:

sequentially testing the input in a predetermined order with the special pattern recognition algorithm and interrupting the diacritic recognizing step when the diacritic is recognized.

11. The method of claim 1, further comprising:

thickening the character after the separating step to remove white pixel groupings and prevent separation of black pixel groupings if more than a maximum number of anticipated constituent components are found.

12. The method of claim 11, further comprising:

recognizing the character in the same manner as a character having a single component when the number of constituent components exceeds a maximum number of constituent components.

13. The method of claim 1, further comprising:

performing a quantitative image analysis to determine a black pixel count for each of the separated components.

14. The method of claim 13, further comprising:

assigning the constituent component containing the greatest number of black pixels as the base component.

15. The method of claim 14, further comprising:

determining a relative location between two or more diacritic components; and

assigning a label to each of the two or more diacritic components based on their relative location.

16. The method of claim 1, further comprising:

recognizing the base component and the characters having a single component using one of optical character recognition and intelligent character recognition.

17. The method of claim 1, wherein:

the acceptable combinations that exist between the base component and the diacritic are limited by a language of interest.

18. A method of identifying multi-language characters in an optical recognition system, comprising:

digitizing a document having characters printed thereon,

targeting each of the characters for recognition analysis,

separating characters having a single component from characters having more than a single component,

recognizing characters having a single component,

segmenting characters having more than a singe component into constituent components, said constituent components including a base component and at least one diacritic component forming a diacritic,

recognizing the base component,

recognizing the diacritic through analysis of the at least one diacritic component, determining through a match analysis whether the, diacritic can be used in combination with the base component,

recognizing the combination of the diacritic and base component in response to a match,

determining during the match analysis whether the base component is one of a plurality of commonly misrecognized base components,

determining if the commonly misrecognized base component can be matched with the recognized diacritic,

determining the base component that is commonly misrecognized as the commonly misrecognized base component, and

matching the base component to the diacritic when the commonly misrecognized base component does not match with the diacritic.