Method for repairing images for optical character recognition performing different repair operations based on measured image characteristics

• US 5,142,589 A
• Filed: 12/21/1990
• Issued: 08/25/1992
• Est. Priority Date: 12/21/1990
• Status: Expired due to Fees

• Current Assignee: Environmental Research Institute Incorporated
• Original Assignee: Environmental Research Institute of Michigan
• Inventors: Beyer, James, Lougheed, Robert
• Primary Examiner(s): Moore, David K.
• Assistant Examiner(s): Stellrecht, Barry
• Application Number:
  US07/632,010
• Time in Patent Office: 613 Days
• Field of Search: 382/9, 382/13, 382/55, 382/54, 382/18
• US Class Current: 382/102
• CPC Class Codes: G06T 5/30 Erosion or dilatation, e.g. thinning
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/168 Smoothing or thinning of the pattern; Skeletonisation

Abstract

A computer system and method for repairing address blocks and ZIP codes numerical groups before segmentation of the ZIP code in an optical character recognizer. The system repairs the image based upon classification of the image as normal or otherwise. If the class is otherwise, background inclusions and gaps are filled in. If the class is normal, small noise pixel elements are eliminated. Once the ZIP code is located, non-horizontal and horizontal gaps are filled in between strokes and small strokes. The amount of dilation of the strokes and small strokes depends on the size of the strokes. The strokes are dilated and then the dilated area is eroded and shaped to conform to the original strokes. The repair connects strokes before segmentation of the ZIP code block into individual digits.

Claims

1. A method of optical character recognition for indicating a plurality of characters formed by strokes composed of a digital image of pixels, said method comprising the steps of:

identifying and quantifying small background inclusions in said character strokes;

identifying and quantifying small gaps between character strokes;

identifying small image inclusions surrounded by background; and

repairing said digital image based upon a quantity of small background inclusions or small gaps identified by (a) filling in small background inclusions and gaps of said character strokes if said quantity of small background inclusions is greater than a predetermined hole threshold or the quantity of said small gaps is greater than a predetermined first gap threshold, (b) otherwise eliminating said small image inclusions from said digital pixel image if said quantities are equal to or smaller than said respective thresholds.

2. The method of optical character recognition as defined in claim 1 wherein:

said hole quantity of small background inclusions being a ratio of the number of pixels identified as background inclusions divided by the number of pixels identified as foreground of the digital image; and

said first gap quantity of small gaps being a ratio of a number of pixels identified as part of a small gap divided by the number of pixels identified as foreground of the digital image.

3. The method of optical character recognition as defined in claim 1 wherein:

said step of identifying and quantifying small gaps identifies gaps in horizontally disposed lines.

4. The method of optical character recognition as defined in claim 1 further comprising the steps of:

skeletonizing said character strokes into skeletal images;

identifying and quantifying skeletal ends of said skeletal images; and

said step of repairing said digital image includes filling in said small background inclusions and gaps of said character strokes if said quantity of small gaps is greater than a predetermined second gap threshold that is less than said first gap threshold and a quantity of said skeletal ends is greater than a predetermined end threshold.

5. The method of optical character recognition as defined in claim 4 wherein:

said hole quantity of small background inclusions being a ratio of the number of pixels identified as small background inclusions divided by the number of pixels identified as part of the character strokes;

said quantity of small gaps being a ratio of the number of pixels identified as part of a small gap divided by the number of pixels identified as part of the character strokes; and

said quantity of skeletal ends being a ratio of a number of pixels identified as end pixels divided by the number of pixels identified as part of the skeletal images.

6. The method of optical character recognition as defined in claim 5 further comprising the steps of:

identifying and quantifying T-junctions of said skeletal images;

said step of repairing said digital image includes filling in said small background inclusions and small gaps of said character strokes if a quantity of T-junctions is greater than a predetermined T-junction threshold; and

said quantity of T-junctions being a ratio of the number of pixels of said digital image identified as T-junctions divided by the number of pixels identified as part of the skeletal images.

7. The method of optical character as defined in claim 1 further comprising the steps of:

skeletonizing said character strokes into skeletal images;

identifying and quantifying T-junction of said skeletal images; and

said step of repairing said digital image includes filling in said small background inclusions and small gaps of said character strokes if a quantity of said T-junctions is greater than a predetermined T-junction threshold.

8. The method of optical character recognition for indicating a plurality of characters formed by strokes composed by a digital image of pixels;

said method comprising the steps of;

identifying small background inclusions in said character strokes;

identifying small gaps between said character strokes;

identifying small foreground inclusions;

classifying said digital image as a second class image based upon an identification of at least threshold quantities of small background inclusions or small gaps and as a first class image if said threshold quantities of small background inclusion or small gaps are not identified;

repairing said plurality of characters if said classifying step identified said digital image as a second class image by filling in said small background inclusions and said small gaps; and

eliminating said small foreground inclusions if said classifying step identified said digital image as a first class image.

9. The method of optical character recognition as defined in claim 8 wherein:

said step of identifying and quantifying small gaps identifies gaps in horizontally disposed lines.

10. The method of optical character recognition for indicating a plurality of characters composed of strokes, the method including segmenting the plurality of characters into individual character images, and recognizing the individual character images, the improvement of repairing the plurality of characters comprising the steps of:

filling in background inclusions in the strokes;

skeletonizing the plurality of characters into skeletal images;

eliminating small spurs in said skeletal images to form filtered skeletal images;

identifying said filtered skeletal images in at least two different sizes;

identifying ends and junctions of said filtered skeletal images;

dilating said identified ends and junctions of said filtered skeletal images over the strokes in the digital image and dilating said identified ends and junctions beyond the strokes in the digital image an amount based upon the identified size of said filtered skeletal image; and

eroding and shaping said dilated ends and junctions to preserve connectivity between strokes that have been connected to said dilating step.

11. The method of optical character recognition as defined in claim 10 said step of identifying said filtered skeletal images in at least two different sizes includes identifying said filtered skeletal images smaller than a predetermined fraction of the height of said plurality of characters as small skeletal images and identifying said filtered skeletal images larger than said predetermined fraction as large skeletal images.

12. The method of optical character recognition as defined in claim 11 wherein said step of dilating said identified ends and junctions dilates said small skeletal images a greater amount than said large skeletal images.

13. The method of optical character recognition as defined in claim 10 wherein said step of dilating said identified ends and junction dilates smaller sized filtered skeletal images a greater amount than larger sized filtered skeletal images.

14. A method for selectively connecting and repairing strokes of character groups composed of a digital image of pixels, said method comprising the steps of:

filling in small holes within said strokes of said character groups;

skeletonizing said strokes of said character groups into skeletal images;

filtering out small spurs of each skeletal image to form filtered skeletal images;

identifying said filtered skeletal images in at least two different sizes;

identifying end sections and junctions of each filtered skeletal image;

dilating the junction sections and end sections beyond said stroke an amount based on the size of said filtered skeletal image; and

eroding and shaping said dilated junctions and end sections to preserve connectivity between strokes that have been connected by said dilating step.

15. A method for selectively joining, fusing and repairing strokes of characters formed by a digital pixel image, said method comprising the steps of:

skeletonizing said strokes of said characters into a skeletal image;

filtering out small spurs in each skeletal image to form filtered skeletal images; and

dilating and eroding said filtered skeletal images in a controlled fashion based upon the size of said filtered skeletal images relative to the height of the characters for selectively connecting two nearby strokes.

16. A method of optical character recognition for indicating a plurality of characters formed of strokes composed of a digital image of pixels wherein said plurality of characters forms a portion of said digital image;

said method comprising the steps of;

identifying small background inclusions in the character strokes of said digital image;

identifying small gaps between character strokes of said digital image;

identifying small foreground inclusions of said digital image;

classifying said digital image as a second class image based upon the identification of a threshold quantity of small background inclusions in the character strokes or a threshold quantity of small gaps between the character strokes and as a first class image if said threshold quantities are not identified;

repairing said second class digital image by filling in said small background inclusion and said small gaps to form part of said digital image;

repairing said first class digital image by eliminating small image inclusions from said digital image;

repairing said plurality of characters in said portion of said digital image by selectively joining strokes that are slightly disconnected and filling in small background inclusions in said character strokes;

segmenting said plurality of repaired characters into individual character images; and

recognizing said individual character images.

17. The method of optical character as defined in claim 16 wherein:

said threshold quantity of small background inclusions being a predetermined ratio of the number of pixels of said digital image identified as small background inclusions divided by the number of pixels identified as part of the character strokes of said digital image; and

said threshold quantity of small gaps being a predetermined ratio of the number of pixels of said digital image identified as part of a small gap divided by the number of pixels identified as part of the character strokes of said digital image.

18. The method of optical character recognition as identified in claim 16 wherein:

said step of repairing said plurality of characters in said portion of said digital images includes;

filling small background inclusions in said strokes of said plurality of characters;

skeletonizing said strokes of said character groups into a skeletal image;

filtering out small spurs of each skeletal image to form filtered skeletal images;

identifying said filtered skeletal images in at least two different sizes;

identifying end sections and junctions of each filtered skeletal image;

dilating the end sections and junctions beyond said strokes an amount based on the size of the filtered skeletal image; and

eroding and shaping said extended end sections and junctions in order to selectively join two nearby strokes.

19. A method for repairing a digital image forming characters composed of strokes, said method comprising the steps of:

identifying small background inclusion in said strokes;

identifying small gaps between said strokes;

identifying small foreground inclusions of said digital image surrounded by a background;

quantifying said small background inclusions to determine if a predetermined threshold quantity of small background inclusions is exceeded;

quantifying said small gaps to determine if a predetermined threshold quantity of small gaps is exceeded; and

repairing said digital image by filling in said small gaps and said small background inclusions if either of said predetermined thresholds are exceeded and if neither predetermined threshold is exceeded by eliminating said small foreground inclusions.

20. A method for repairing a digital image forming characters composed of strokes, said method comprising the steps of:

identifying defects in said digital image;

classifying said digital image in one of at least two classes based on the amount and kind of identified defects;

repairing said digital image based on the class selected by said classifying means;

selecting a portion of said digital image; and

repairing said selected portion of said digital image by selectively joining strokes that are slightly disconnected.

21. A method of optical character recognition for indicating a plurality of characters formed of strokes composed of a digital image of pixels, said method comprising the steps of:

identifying and quantifying small background inclusions in said character strokes;

identifying and quantifying small gaps between character strokes;

identifying small image inclusions surrounded by background; and

repairing said digital image by filling in said small background inclusions and gaps of said character strokes if either said quantity of small background inclusions is greater than a predetermined hole threshold or the quantity of said small gaps is greater than a predetermined first gap threshold.

22. A method of optical character recognition for indicating a plurality of characters formed by strokes composed of a digital image of pixels, said method comprising the steps of:

identifying and quantifying small background inclusions in said character strokes;

identifying and quantifying small gaps between character strokes;

identifying small image inclusions surrounded by background; and

repairing said digital image by eliminating said small image inclusions from said digital pixel image if neither said quantity of small background inclusions is greater than a predetermined hole threshold nor the quantity of said small gaps is greater than a predetermined first gap threshold.