Character recognition method and system

• US 5,034,991 A
• Filed: 04/06/1990
• Issued: 07/23/1991
• Est. Priority Date: 04/10/1989
• Status: Expired due to Fees

• Current Assignee: Hitachi Keiyo Engineering Company Limited, Hitachi, Ltd.
• Original Assignee: Hitachi Keiyo Engineering Company Limited, Hitachi, Ltd.
• Inventors: Yano, Souichi, Hata, Seiji, Hagimae, Kinuyo
• Primary Examiner(s): Boudreau, Leo H.
• Application Number:
  US07/505,586
• Time in Patent Office: 473 Days
• Field of Search: 382/21, 382/37, 382/38, 382/3, 382/11, 382/13, 382/30, 382/34, 382/36, 382/39
• US Class Current: 382/209
• CPC Class Codes: G06V 30/10 Character recognition
• CPC Class Codes: G06V 30/1478 of characters or characters lines
• CPC Class Codes: G06V 30/158 using character size, text spacings or pitch estimation
• CPC Class Codes: G06V 30/162 Quantising the image signal
• CPC Class Codes: G06V 30/18086 by performing operations within image blocks or by using histograms
• CPC Class Codes: G06V 30/184 by analysing segments intersecting the pattern
• CPC Class Codes: G06V 30/196 using sequential comparisons of the image signals with a plurality of references

Abstract

Character recognition method and system in which a character indicated in a printed, stamped, carved or other form is two-dimensionally imaged and stored as image data and the stored image data is subjected to an image processing to recognize the character. The recognition of the character is performed in such a manner that each time the comparison of plural kinds of feature vectors extracted from the character to be recognized and a dictionary vector of each candidate character in a group of candidate characters preliminarily prepared is made for one of the plural kinds of feature vectors, a candidate character having its dictionary vector away from the extracted feature vector by a distance not smaller than a predetermined value is excluded from the candidate character group. The dictionary vector for each candidate character is defined as an average vector for a variety of fonts. A difference between the dictionary vector and the feature vector extracted from the character to be recognized is estimated by virtue of a deviation vector for the variety of fonts to produce an estimated value. The exclusion from the candidate character group is judged on the basis of the estimated values each of which is cumulatively produced each time the estimation for the difference is made.

Claims

1. A character recognition method of recognizing a typical character, including an alphanumeric character or a symbol, by two-dimensionally imaging the character, storing the two-dimensional image as image data and subjecting the stored image data to an image processing;

comprising the steps of;

extracting different kinds of feature vectors from the two-dimensional image of the character to be recognized;

comparing one of the extracted different kinds of feature vectors with a dictionary vector for each candidate character in a group of candidate characters preliminarily prepared to determine a distance between the dictionary vector and the one kind of feature vector;

estimating the distance determined by said comparing step to exclude from the candidate character group a candidate character having its dictionary vector spaced from the one kind of feature vector by a distance not smaller than a predetermined value;

repeating said comparing and estimating steps for each of the remaining kinds of feature vectors until the number of candidate characters included in said candidate character group becomes equal to or smaller than a predetermined value; and

identifying the character to be recognized on the basis of restricted candidate characters obtained as a result of said repeating step.

2. A character recognition method according to claim 1, further comprising a step of storing as the dictionary vector of each candidate character an average feature vector for a variety of fonts and a deviation feature vector for the variety of fonts.

3. A character recognition method according to claim 2, wherein said comparing step includes a substep of calculating a difference between said average feature vector of the dictionary vector and said extracted feature vector, and said estimating step includes a substep of estimating the distance calculated by said calculating substep by reference to said deviation feature vector of the dictionary vector to produce a cumulative estimated value each time said repeating step is carried out.

4. A character recognition method according to claim 3, wherein said estimating step determines a candidate character to be excluded from said candidate character group on the basis of said cumulative estimated value.

5. A character recognition method in which a typical character indicated is recognized by an image processing after two-dimensional imaging of the character and the storage thereof as image data, comprising the steps of:

comparing one of plural kinds of feature vectors extracted from the character to be recognized and a dictionary vector of each candidate character in a group of candidate characters preliminarily prepared to determine a distance between the one kind of feature vector and the dictionary vector;

estimating said distance to exclude from the candidate character group a candidate character having its dictionary vector spaced from the one feature vector by a distance not smaller than a predetermined value;

repeating said comparing and estimating steps for each of the remaining kinds of feature vectors until the number of candidate characters in said candidate character groups becomes equal to or smaller than a predetermined value; and

recognizing the character to be recognized on the basis of restricted candidate characters obtained as a result of said repeating step.

6. A character recognition method according to claim 5, wherein the dictionary vector of each of said candidate characters is defined as an average vector for a variety of fonts, the distance between the feature vector extracted from the character to be recognized and the dictionary vector determined in said comparing step is estimated in said estimating step by virtue of a deviation vector for the variety of fonts to produce an estimated value, and the exclusion from said candidate character group in said estimating step is judged on the basis of the estimated values each of which is cumulatively produced each time said estimating step is made.

7. A character recognition method according to claim 5, wherein the character to be recognized is indicated in plural as a character row which is to be recognized in units of one character through said comparing, estimating, repeating and recognizing steps, and said recognizing step includes collating the recognized character row with each of character rows preliminarily held as reference data, preferential recognition and collation being made to a specified portion of said character row while the recognition and collation to the remaining portion of said character row is made only when the result of collation of the specified portion with the held reference data is affirmative.

8. A character recognition method according to claim 5, wherein said repeating step is terminated at the point of time when the candidate characters have been considerably restricted by the exclusion from said candidate character group at said estimating step.

9. A character recognition method according to claim 5, wherein said group of candidate characters is preliminarily limited to a group of characters of a specified kind.

10. A character recognition method according to claim 5, the recognition of the character is made in accordance with a recognition code designator.

11. A character recognition method according to claim 5, wherein the character to be recognized may be indicated in plural as a character row, and said method further comprises a step of establishing a window to embrace the character or character row to be recognized, a step of processing a multi-valued two-dimensional image of the character row in said window to produce a histogram for brightness in said window, and a step of determining a peak point and two skirts of the brightness histogram to discriminate as the brightness of a background of the character row a portion of said brightness histogram around said peak point and as the brightness of the character row itself a portion of said brightness histogram which extends on the side of one of said two skirts far away from said peak point.

12. A character recognition method according to claim 5, wherein the character to be recognized may be indicated in plural as a character row, and said method further comprises a step of establishing a window to embrace the character row to be recognized, and a step of binarizing a many-valued two-dimensional image of the character row in said window by determining a peak point and two skirts of a histogram for brightness in said window, connecting said peak point and one of said two skirts far away from said peak point by a straight line and defining a threshold value for binarization on the basis of the brightness at a point on the brightness histogram farthest from said straight line.

13. A character recognition method according to claim 12, further comprising a step of extracting the characters in a character height direction from a binary two-dimensional image in said window obtained by said binarizing step and said character extracting step includes dividing said binary two-dimensional image into a plurality of sections in a character width direction at intervals equal to or greater than one time of a character width, producing a projection distribution for the frequency of appearance of character pixels of a binary image to the character width direction for each divisional section, and determining a character existing position in the character height direction from the projection distribution for each divisional section.

14. A character recognition method according to claim 13, wherein the character existing position in the character height direction determined for each divisional section is approximated on the assumption that it lies between two parallel straight lines which are established in said window on the basis of said character existing position.

15. A character recognition method according to claim 13, wherein the extraction in the character height direction in each divisional section is made by approximate straight lines established on the basis of two parallel straight lines, and a projection distribution for the frequency of appearance of character pixels of a binary image to the character height direction is produced in each divisional section to extract the character row in the character width direction on the basis of the projection distribution.

16. A character recognition method according to claim 15, wherein the extraction of the character row in the character width direction is made by dividing the projection distributions to the character width direction produced for the divisional sections into a plurality of groups in the character width direction in accordance with a predetermined small threshold value while dividing virtually and equally an area from a group start point of the leading group of the groups and a group end point of the tailing group of the groups into a plurality of regions in accordance with the presumed number of characters, whereby in each divisional region a narrower group having a width smaller than a predetermined width, if any, is incorporated into a wider group which has a width not smaller than said predetermined width and exists in the vicinity of said narrower group and on the side of one two virtual division points on opposite sides of said narrower group which is far away from said narrower group.

17. A character recognition method according to claim 16, wherein if a group having a width greater than a character width is extracted from the groups after the incorporation processing, the number of characters included in that group is estimated and that group is virtually and equally segmented in accordance with the estimated number of characters to thereby define the minimum point existing in the vicinity of each segmentation point as a new character extraction point in the character width direction.

18. A character recognition system for recognizing a typical character including an alphanumeric character or a symbol by two-dimensionally imaging the character, storing the two-dimensional image as image data and subjecting the stored image data to an image processing, comprising:

extracting means for extracting a plurality of kinds of feature vectors from a two-dimensional image of the character to be recognized;

comparing means coupled with said extracting means for comparing one of the extracted plural kinds of feature vectors with a dictionary vector for each candidate character in a group of candidate characters preliminarily prepared to determine a distance between the dictionary vector and said one kind of feature vector;

estimating means responsive to said comparing means for estimating said distance to exclude from the candidate character group a candidate character having its dictionary vector spaced from the one feature vector by a distance not smaller than a predetermined value;

controlling means for repeating the operations of said comparing means and said estimating means for each of the remaining kinds of feature vectors until the number of candidate characters n said candidate character group becomes equal to or smaller than a predetermined value; and

identifying means coupled with said controlling means and said estimating means for identifying the character to be recognized on the basis of restricted candidate characters obtained by the operation of said controlling means.

19. A character recognition system according to claim 18, further comprising a memory for storing as the dictionary vector of each candidate character an average feature vector for a variety of fonts and a deviation feature vector for the variety of fonts.

20. A character recognition system according to claim 19, wherein said comparing means includes means for calculating a difference between said average feature vector of the dictionary vector and said extracted feature vector, and said estimating means includes means for estimating the distance calculated by said calculating means by reference to said deviation feature vector of the dictionary vector to produce a cumulative estimated value each time the operations of said comparing means and said estimating means are repeated.