Character extraction apparatus, dictionary production apparatus and character recognition apparatus, using both apparatuses

• US 6,064,769 A
• Filed: 11/05/1998
• Issued: 05/16/2000
• Est. Priority Date: 04/21/1995
• Status: Expired due to Term

• Current Assignee: Ichiro Nakao, Mariko Takenouchi, Saki Takakura, Satoshi Emura
• Original Assignee: Ichiro Nakao, Mariko Takenouchi, Saki Takakura, Satoshi Emura
• Inventors: Takenouchi, Mariko, Emura, Satoshi, Takakura, Saki, Nakao, Ichiro
• Primary Examiner(s): Boudreau, Leo H.
• Assistant Examiner(s): SHERALI, ISHRAT I
• Application Number:
  US09/186,943
• Time in Patent Office: 558 Days
• Field of Search: 382/173, 382/176-177, 382/187, 382/181, 382/190, 382/195, 382/200, 382/203, 382/204, 382/209, 382/224, 382/218, 382/290, 382/229, 382/246
• US Class Current: 382/224
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/148 Segmentation of character regions
• CPC Class Codes: G06V 30/414 Extracting the geometrical structure, e.g. layout tree; Block segmentation, e.g. bounding boxes for graphics or text

Abstract

A character extraction apparatus is provided for extracting character data for each character from a text image which is represented by first pixels corresponding to character images and second pixels corresponding to background images. The character extraction apparatus comprises a character row detecting means for detecting character rows from the text image and obtaining position data of each character row; a pixel array extracting means for extracting arrays of continuous first pixels in an area specified by the character row position data and computing position data of each of the arrays of continuous first pixels; a character array linking means for linking the arrays of continuous first pixels in the area based on the position data of the arrays of continuous first pixels; and a character extracting means for recognizing each set of arrays of continuous first pixels linked by the character array linking means as a character and outputting character data.

Claims

1. A character recognition apparatus, comprising:

a character image extracting means for extracting character images from a text image;

a candidate character selecting means for comparing the character images with standard characters for selecting a plurality of candidate characters with higher matching levels, and assigning first evaluation values to the candidate characters according to the matching levels;

a character rectangle data dictionary for prestoring shape data of circumscribed rectangles of the standard characters;

a character rectangle extracting means for extracting position data of the circumscribed rectangles of the character images extracted by the character image extracting means;

a character rectangle shape data extracting means for extracting normalized shape data from the position data of the circumscribed rectangles of the character images extracted by the character rectangle extracting means;

a character rectangle evaluating means for obtaining a second evaluation value for each candidate character by a certain computation using the normalized shape data extracted by the character rectangle shape data extracting means and the shape data stored in the character rectangle data dictionary; and

a character determining means for determining a character among the plurality of candidate characters selected by the candidate character selecting means based on the first evaluation values and the second evaluation values.

2. The character recognition apparatus of claim 1, wherein the character rectangle data dictionary comprises:

a character rectangle position data registering unit for prestoring data of uppermost positions and lowest positions of the circumscribed rectangles of the standard characters; and

a character rectangle size data registering unit for prestoring data of heights and widths of the circumscribed rectangles of the standard characters.

3. The character recognition apparatus of claim 2, wherein the shape data stored in the character rectangle data dictionary includes a standard uppermost position which is a maximum height of circumscribed rectangles of capitol letters, and includes a base line position which is an average value of lowest positions of continuously printed capitol letters.

4. The character recognition apparatus of claim 3, wherein the character rectangle shape data extracting means comprises:

a part extracting unit for computing a distance between two neighboring circumscribed rectangles of the character images for each combination in a character row based on the position data, and dividing the character row into parts based on the computed distances, each of the parts having character images with the distance less than a threshold value, and the position data consisting of coordinate values of upper-left positions and lower-right positions of the subscribed rectangles of the character images;

a part rectangle shape data extracting unit for extracting uppermost positions and lowest positions of the circumscribed rectangles of the parts extracted by the part extracting unit based on the position data extracted by the character rectangle extracting means;

a relative position data computing unit for obtaining a relative uppermost position and a relative lowest position of each circumscribed rectangle of character image in each part by setting a lowest position of a circumscribed rectangle of the part as a basic value, and computing a height and a width of each of the circumscribed rectangles of the character images from the position data;

a character classifying unit for extracting a maximum value of the Yelative uppermost positions obtained by the relative position data computing unit, classifying character images whose relative lowest positions are less than a certain ratio of the maximum value into lower extension characters, and classifying other character images into upper characters;

a part base line position extracting unit for obtaining a part base line position of the circumscribed rectangle of the part which is an average value of the relative lowest positions of the circumscribed rectangles of the upper characters in the part;

a standard character position computing unit for computing a standard character uppermost position and a character base line position for each circumscribed rectangle of character image from the relative uppermost position and the relative lowest position by setting the part base line position as a basic value,

and a character rectangle normalizing unit for obtaining normalized shape data which includes the standard character uppermost position, the character base line position and the height and the width of the circumscribed rectangle of the character image computed by setting the maximum value of the heights of the upper characters as a basic value.

5. The character recognition apparatus of claim 4, wherein the character rectangle shape data extracting means further comprises:

a classification correcting unit for classifying all the character images in a part into upper characters when the character images are classified into lower extension characters.

6. The character recognition apparatus of claim 5, wherein the character rectangle evaluating means comprises:

a second evaluation value computing unit for obtaining the second evaluation value for each candidate character which is a sum of absolute values of differences between elements of the normalized shape data and corresponding elements of shape data of a candidate character; and

wherein the character determining means comprises;

a total evaluation value computing unit for summing up the first evaluation values and the second evaluation values and obtaining a total evaluation value for each candidate character, the first evaluation values decreasing as the matching level of the candidate characters rises; and

a determining unit for determining a candidate character which has a minimum total evaluation value.

7. The character recognition apparatus of claim 6, wherein the character rectangle evaluating means comprises:

a character similarity information storage unit for storing information telling whether each of the standard characters has similar characters;

a first judging unit for judging whether each of the plurality of candidate characters selected by the candidate character selecting means has similar characters by referring to the character similarity information storage unit;

a second evaluation value assigning unit for prohibiting the second evaluation value computing unit from obtaining the second evaluation value and assigning certain second evaluation values to all the candidate characters if the first judging unit judges that any of the candidate characters does not have any similar characters.

8. The character recognition apparatus of claim 7, wherein the character rectangle evaluating means further comprises:

a second judging unit for judging whether a candidate character with a highest matching level among those selected by the candidate character selecting means has similar characters by referring to the character similarity information storage unit, wherein the second evaluation value assigning unit prohibits the second evaluation value computing unit from obtaining the second evaluation value and assigns certain second evaluation values to all the candidate characters if the second judging unit judges that the candidate character does not have any similar characters.

12. The character recognition apparatus of claim 3, wherein the character rectangle shape data extracting means comprises:

a part extracting unit for computing a distance between two neighboring circumscribed rectangles of the character images for each combination in a character row based on the position data, and dividing the character row into parts based on the computed distances, each of the parts having character images with the distance less than a threshold value, and the position data including coordinate values of upper-left positions and lower-right positions of the circumscribed rectangles of the character images,

a part rectangle shape data extracting unit for extracting uppermost positions and lowest positions of the circumscribed rectangles of the parts extracted by the part extracting unit based on the position data extracted by the character rectangle extracting means;

a relative position data computing unit for obtaining a relative uppermost position and a relative lowest position of each circumscribed rectangle of character image in each part by setting a lowest position of a circumscribed rectangle of the part as a basic value, and by computing a height and a width of each of the circumscribed rectangles of the character images from the position data;

a character classifying unit for extracting a maximum value of the relative uppermost positions obtained by the relative position data computing unit for each part, classifying character images whose relative lowest positions are less than a certain ratio of the maximum value into lower extension characters, and classifying other character images into upper characters;

a first lowest position correcting unit for judging whether a difference between a relative lowest position of a circumscribed rectangle of a first character image in the part and an initial value is less than a certain value;

if less, extracting the relative lowest position as a corrected lowest position of the circumscribed rectangle of the first character image; and

if not less, extracting the initial value as the same, wherein the initial value is a relative lowest position of the circumscribed rectangle of the first upper character;

a second lowest position correcting unit for judging whether a difference between the corrected lowest position of the circumscribed rectangle of the first character image and a relative lowest position of a circumscribed rectangle of a next character image in the part is less than the certain value;

if less, extracting the relative lowest position as a corrected lowest position of the circumscribed rectangle of the next character image; and

if not less, extracting the corrected lowest position of the circumscribed rectangle of the first character image as the same, wherein this judgement is performed for pairs of neighboring rectangles in order until a circumscribed rectangle of a last character image in the part is corrected;

a standard character position computing unit for computing a standard character uppermost position and a character base line position for each circumscribed rectangle of character image from the relative uppermost position and the relative lowest position by setting the corrected lowest position of the circumscribed rectangle of each character image as a basic value; and

a character rectangle normalizing unit for obtaining normalized shape data which includes the standard character uppermost position, the character base line position and the height and the width of the circumscribed rectangle of the character image computed by setting the maximum value of the heights of the upper characters as a basic value.

13. The character recognition apparatus of claim 12, wherein the character rectangle shape data extracting means further comprises:

a classification correcting unit for classifying all the character images in a part into upper characters when the character images are classified into lower extension characters.

14. The character recognition apparatus of claim 13, wherein the character rectangle evaluating means comprises:

a second evaluation value computing unit for obtaining the second evaluation value for each candidate character which is a sum of absolute values of differences between elements of the normalized shape data and corresponding elements of shape data of a candidate character; and

wherein the character determining means comprises;

a total evaluation value computing unit for summing up the first evaluation values and the second evaluation values and obtaining a total evaluation value for each candidate character, the first evaluation values decreasing as the matching level of the candidate characters rises; and

a determining unit for determining a candidate character which has a minimum total evaluation value.

15. The character recognition apparatus of claim 14, wherein the character rectangle evaluating means comprises:

a character similarity information storage unit for storing information telling whether each of the standard characters has similar characters;

a first judging unit for judging whether each of the plurality of candidate characters selected by the candidate character selecting means has similar characters by referring to the character similarity information storage unit;

a second evaluation value assigning unit for prohibiting the second evaluation value computing unit from obtaining the second evaluation value and assigning certain second evaluation values to all the candidate characters if the first judging unit judges that any of the candidate characters does not have any similar characters.

16. The character recognition apparatus of claim 15, wherein the character rectangle evaluating means further comprises:

a second judging unit for judging whether a candidate character with a highest matching level among those selected by the candidate character selecting means has similar characters by referring to the character similarity information storage unit, wherein the second evaluation value assigning unit prohibits the second evaluation value computing unit from obtaining the second evaluation value and assigns certain second evaluation values to all the candidate characters if the second judging unit judges that the candidate character does not have any similar characters.

17. The character recognition apparatus of claim 16, wherein the threshold value used in the part extracting unit is a half of a maximum value of heights of the circumscribed rectangles of the character images.

18. The character recognition apparatus of claim 17, wherein the certain ratio used by the character classifying unit is 20%.

19. The character recognition apparatus of claim 18, wherein the second evaluation value assigned by the second evaluation value assigning unit is "0".

9. The character recognition apparatus of claim 9, wherein the threshold value used in the part extracting unit is a half of a maximum value of heights of the circumscribed rectangles of the character images.

10. The character recognition apparatus of claim 9, wherein the certain ratio used by the character classifying unit is 20%.

11. The character recognition apparatus of claim 10, wherein the second evaluation value assigned by the second evaluation value assigning unit is "0".

20. A computer-readable recording medium storing a program which allows a computer to function as a character recognition apparatus, the computer-readable recording medium storing:

a character rectangle data dictionary for prestoring shape data circumscribed rectangles of standard characters,

wherein the character recognition apparatus comprises;

a character image extracting means for extracting character images from a text image;

a candidate character selecting means for comparing the character images with the standard characters for selecting a plurality of candidate characters with higher matching levels, and assigning first evaluation values to the candidate characters according to the matching levels;

a character rectangle extracting means for extracting position data of the circumscribed rectangles of the character images extracted by the character image extracting means;

a character rectangle shape data extracting means for extracting normalized shape data from the position data of the circumscribed rectangles of the character images extracted by the character rectangle extracting means;

a character rectangle evaluating means for obtaining a second evaluation value for each candidate character by a certain computation using the normalized shape data extracted by the character rectangle shape data extracting means and the shape data stored in the character rectangle data dictionary; and

a character determining means for determining a character among the plurality of candidate characters selected by the candidate character selecting means based on the first evaluation values and the second evaluation values.