JP2972443B2 - Character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition apparatus for recognizing characters by extracting characteristic data from a read character image and comparing the extracted characteristic data with a standard character image stored in a dictionary in advance.

[0002]

2. Description of the Related Art As a character recognition device for recognizing character information of a document by computer processing, a CCD (Charge Coupled Device) converts character information to be recognized, for example, Japanese characters.
2. Description of the Related Art An optical character recognition device (OCR) is known, which performs photoelectric conversion by using such a method, cuts out the photoelectrically converted electric signal in units of one character image, and recognizes each character image in accordance with a predetermined recognition logic in a recognition unit. ing.

[0003] As a recognition method, a contour is traced from the starting point of a character, a direction code is given, and this is collated with a similar standard pattern in a dictionary as a direction code sequence, candidates are listed in order of similarity, and characters are listed. There is a recognition method such as outputting a code.

[0004]

In this type of character recognition apparatus, all characters to be recognized have both uppercase and lowercase standard patterns. A recognition method for outputting a result (character code) recognized after the collation is implemented. In this method, since both uppercase and lowercase standard patterns are used, the number of standard patterns increases, and there is a disadvantage that it takes time to collate with the standard patterns.

As described above, when character recognition is performed by extracting contour feature data, it is suitable for recognizing kanji having many contours. However, hiragana, katakana, uppercase and lowercase letters in the alphabet are preferable. Is a similar character with similar shapes, for example, “Y” and “ゅ”, it is difficult to distinguish the outline, and the recognition rate is low.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned drawbacks, speed up the recognition process in character recognition, and provide a character recognition device capable of more accurately distinguishing uppercase and lowercase characters. .

[0007]

According to the present invention, there is provided a reading means for reading an original on which a plurality of characters are displayed, converting the original into image data, and outputting the image data; an image memory for storing image data from the reading means; A character extracting means for sequentially extracting each character image from the image memory, a standard character image to be matched with the extracted character image and a corresponding character code are stored, and a similar character having a geometrically similar uppercase and lowercase character is stored. As for the shape characters, a dictionary memory in which only one of the standard character images of uppercase or lowercase is stored, a detecting means for detecting the size of the extracted character image, and an extracted character based on the storage contents of the dictionary memory A recognition unit that recognizes an image and outputs a character code, a character code from the recognition unit,
A recognition result memory for storing the size from the detection means as a recognition result, based on the contents of the recognition result memory,
The frequency is counted for each character code of the similar character, a histogram of the frequency based on the size of the character having the frequency of 1 or more is created, and whether the division point of the histogram is uppercase or lowercase is used as a threshold value. And a determination unit for determining the character.

In the present invention, the determination means may determine whether the frequency of a character code is less than 1 and a histogram cannot be created for a specific similar character, or if the character size is only uppercase or lowercase. When the threshold value cannot be determined based on the histogram, the determination is performed based on the threshold value of similar characters other than the specific similar character.

[0009]

[0010]

According to the present invention, the character recognition device reads an original on which a plurality of characters are displayed, converts the image into image data, and stores the image data from the reading unit in an image memory. By character extracting means, each character image is sequentially extracted from the image memory, and in the recognizing means,
Recognize the extracted character image based on the contents of the dictionary memory,
Output character code. The dictionary memory stores a standard character image to be matched with the extracted character image and a corresponding character code. For similar characters having a geometrically similar upper case and lower case, one of upper case and lower case is used. Only the standard character image is stored. Further, the character recognizing device detects the size of the extracted character image by the detecting means, and stores it in the recognition result memory together with the character code from the recognizing means. The character recognition device,
In the determining means, based on the contents of the recognition result memory,
The frequency is counted for each character code of the similar character included in one line, a histogram is created based on the size of the character having the frequency of 1 or more, and the division point of the histogram is used as a threshold value to be a capital letter. Or lowercase.

Further, according to the present invention, the character recognizing device may be configured to determine whether a histogram cannot be created for a specific similar character, that is, when the number of similar characters input is less than one, or When it is not possible to determine, that is, when the size of the similar character is large or small, the uppercase and lowercase are determined based on the threshold value of the similar character other than the specific similar character. .

[0012]

[0013]

FIG. 1 is a block diagram showing a simplified configuration of a character recognition device 16 according to an embodiment of the present invention. The input unit 1 is an input device capable of photoelectric conversion, such as a scanner. The control unit 2 is a part that controls the entire character recognition device 16 and does not need to be independent as a control unit. The display unit 3 displays the result of recognition performed on the input character image in printed characters. The recognition unit 4 collates a standard character image in a recognition dictionary 15 described later with a character image whose features are extracted from the input character image, and outputs the recognition result to a recognition result memory 9 described later.

The language processing section 5 analyzes the result of the recognition section 4 and utilizes the linguistic information to improve the character recognition accuracy. The language processing unit 5 is unnecessary in some cases, and may be included in the recognition unit 4. The output unit 6 is an interface, a floppy disk, RS-23
Output information to 2C, LAN (local area network), etc.

FIG. 2 is a simplified block diagram showing the configuration of control unit 2 shown in FIG. The control unit 2
It comprises a CPU (Central Processing Unit) or a dedicated LSI (Large Scale Integrated Circuit) 7, an image memory 8, a recognition result memory 9, and a case identification memory 10. By the CPU 7 and the case identification memory 10, the case determination unit 1 is used.
1 is configured. The image memory 8 stores the binary image (may be a multi-valued image) obtained by the input unit 1.

The CPU 7 cuts out a character image from the image in the image memory 8 and transfers the cut-out character image and the coordinates (X1, Y1) and (X2, Y2) where the image is located to the recognizing unit 4. Further, the CPU 7 receives a recognition result from the recognition unit 4. Further, the CPU 7 determines uppercase and lowercase letters by a method described later. The image memory 8 includes a work buffer required by the CPU 7. If not included, a work buffer directly connected to the CPU 7 is required.

The recognition result memory 9 stores the recognition results obtained from the recognition unit 4 such as a character code, a position of a character image,
This is a memory for sequentially storing the vertical and horizontal sizes of the character image.
The case identification memory 10 may store a histogram and a frequency table of each character image, and at the same time, may partially use it as a work buffer.

The character extracting section 12 is necessary only when the CPU 7 does not extract a character image corresponding to one character from the image memory 8. When the character cutout unit 12 exists, the output to the work buffer 13 is a character image and coordinates (X1, Y1) and (X2, Y2) where the image is located. The work buffer 13 includes a CPU using the image memory 8.
Reference numeral 7 denotes a memory that performs processing and stores the obtained character image and coordinates, and is used as a work buffer for the collation processing.

The collating unit 14 refers to the recognition dictionary 15 and
Matching is performed according to a predetermined recognition logic, and character codes considered to correspond to the character images are arranged in the recognition result memory 9 of the control unit 2 in order of similarity, and a plurality of character images (from the first place to the n-th place) from the top are displayed. And the size of the character image in the vertical and horizontal directions.

In the present embodiment, the processing in the collating unit 14 is reduced by not putting lowercase letters in the recognition dictionary 15 for similar characters whose uppercase letters and lowercase letters are similar.

FIG. 3 is a diagram showing the relationship between the character image cut out by the CPU 7 and the coordinate axes. The X axis and the Y axis are set in the image memory 8 in advance.
A circumscribed rectangle is selected such that two sides corresponding to the axis and the Y axis are parallel. The point P of this circumscribed rectangle that is closest to the coordinate origin O
(X1, Y1) and the point Q farthest from the coordinate origin O
(X2, Y2) are stored in the recognition result memory 9 as position data of the character image.

FIG. 4 is a flowchart showing the operation of character recognition in the character recognition device 16. First, in step k1, the input unit 1 receives image data from an input device such as a scanner, and stores the image data in the image memory 8 of the control unit 2. In the next step k2, if there is character image data in the image memory 8, the process proceeds to step k3, where the CPU 7 or the character cutout unit 12 cuts out one character image with the circumscribed rectangle as shown in FIG. At the same time, the image memory 8 stores the row and column of the original where the character image is located.

In step k4, the CPU 7 uses the work buffer in the image memory 8 or the work buffer 13 to extract character image data of one character from image data of one character, and position data of the character image ( X
1, Y1) and (X2, Y2) are output to the recognition unit 4.
In step k5, the collating unit 14 of the recognizing unit 4 collates the input character data sent from the work buffer 13 with the standard character pattern, and determines the character code corresponding to the standard character pattern with the similarity to the input character data. In the next step k6, the character codes from the first place to the n-th place are output to the recognition result memory 9. However, the standard character pattern used for collation is only uppercase (or only lowercase) for similar characters. When the recognition process for one character is completed, the process returns to step k2, and for the next one character image data in the image memory 8, steps k2 to k2 are performed.
The processing of k6 is performed.

In step k2, if there is no character image data in the image memory 8, the operation proceeds to the next step k7.
If there is any recognition result data in the recognition result memory 9 in step k7, the process proceeds to step k8, and if there is no data, the process ends.

In step k8, the CPU 7 creates a frequency table to be used by the upper / lower case determination unit 11, and creates a histogram of the size of the extracted character image of similar characters based on the frequency table. Next, in step k9, the CPU
Reference numeral 7 determines the size of the character image at the division point serving as a reference for determining whether the case is uppercase or lowercase, that is, the threshold value c, from the histogram created as described above.

At step k10, the CPU 7 determines at step k
The value c of the division point obtained in 9 is compared with the size of the character calculated from the position coordinates of each character image, and it is determined whether the character is uppercase or lowercase. In step k11, the CPU 7 outputs the recognition result from the display unit 3 or the output unit 6 after correcting the corresponding character code with respect to the recognition result in step k6 determined to be lowercase.

FIG. 5 is a flowchart showing a procedure for creating a frequency table executed by the upper / lower case judging section 11, and FIG.
FIG. 6 is a diagram showing a configuration example of data of a frequency table created by the operation shown in the flowchart of FIG. 5. First, step m1
Clear the frequency column and the size column of the table shown in FIG. The categories for which the frequency table is created are similar characters whose uppercase and lowercase are similar, that is, "A,
I, u, e, o, tsu, ya, yu, yo, wow, a, i, u,
D, o, t, ya, yu, yo, wa, ka, ke, C, O, S,
V, W, X, Z ".

At step m2, it is checked whether or not there is recognition result data in the recognition result memory 9. If there is no data in the recognition result memory 9, the process ends. At step m3, the recognition result of each character is recognized from the recognition result memory 9;
For example, a character code and coordinates are read. Step m
In 4, the character code read from the recognition result memory 9 is "A, I, U, E, O, T, Y, Y, Y, W, A, I,
U, d, o, t, ya, yu, yo, wa, ka, ke, C, O,
If it is not a similar character such as "S, V, W, X, Z", the process returns to step m2. If the character code is a similar character, the process proceeds to the next step m5. At step m5, 1 is added to the frequency field of the character code of the corresponding similar character, and counting is performed. In step m6, the vertical and horizontal sizes obtained by subtracting from the coordinates are added to the end of the corresponding size column, and step m6 is performed.
Return to 2.

FIG. 7 is a flow chart showing an operation for creating a histogram based on the data of the frequency table of similar characters shown in FIG. 6 and obtaining a division point for distinguishing between uppercase and lowercase characters, and FIG. 8 is shown in FIG. 4 is a histogram created based on data of a frequency table of similar characters. First, it is determined in step n1 whether or not there is data in the frequency table. If there is no data, the process is terminated. If there is data, the process proceeds to the next step n2, where data of one line of the character code in the frequency table of FIG. 6 is read. If the frequency of the read line is less than 1 at step n3, the process returns to step n1 to read the data of the line of the next character code. If the frequency of the line of the character code read in step n3 is 1 or more, step n4
Using the size described in the table, a histogram is created in which the horizontal axis is the size and the vertical axis is the number, that is, the frequency as shown in FIG. 8, and the values a and b as shown in FIG. Ask.

Further, at step n5, c = (a +
b) Determine a value c that satisfies / 2. At this time, if the input document image is written horizontally, the value c is obtained using a horizontal size histogram. In the case of vertical writing, the value c is determined from the vertical size histogram.

When the frequency of data for one line is less than 1 for a specific character in the frequency table of similar characters and a histogram cannot be created for the specific character, or when the value a, When b cannot be detected, the value of the division point is used as the division point for the value of the division point obtained with respect to the data of other similar characters in the category.

After the value c of the division point is determined, the contents of the recognition result memory 9 are compared with the vertical or horizontal size and the value c of the division point, and are converted into lowercase character codes. If the size is smaller than the value c, it becomes lowercase.

FIG. 9 shows a case where a document 20 including characters is read,
Once stored as a character image and recognized,
It is a figure showing signs that it is outputted in the form similar to. FIG.
1A shows an example of a document 20 read by the input unit 1. An input unit 1 which is an input device such as a scanner includes an original 2
The image 21 in the area surrounded by the virtual line 22 of 0 is read and output as a binary image to the image memory 8 by photoelectric conversion or the like. A binary image is obtained by obtaining a histogram of the amplitude of an input signal, that is, a light intensity or a print density, and obtaining a certain level,
That is, the image is represented as 1 when the amplitude is larger than the threshold value and as 0 when the amplitude is smaller than the threshold value. For example, when the density of a certain point is 1, it is expressed in black, and when it is 0, it is expressed in white.

FIG. 9B shows a part of the binary image 21 inputted into the image memory 8. In the image memory 8, 2
The value image 21 is stored corresponding to the X coordinate and the Y coordinate,
The CPU 4 or the character cutout unit 12 sets the coordinates (X1, X1) of the points P and Q representing the position of the
Along with Y1) and (X2, Y2), the vertical and horizontal sizes of one character calculated from the coordinate values and the number and line of the character to be cut out are output to the recognition result memory 9.

FIG. 9C shows an example in which the read document 20 is subjected to recognition processing by the character recognition device 16 and then printed out on the recording paper 32 via the output unit 6. Character recognition device 1
The characters recognized in step 6 are arranged on the basis of row and column data 30 indicating positions in the recognition result memory 8 shown in FIG.

FIG. 10 is a view showing data relating to each input character stored in the recognition result memory 9.

FIG. 10A shows input character data stored in the recognition result memory 9 after the collating unit 14 has collated with the recognition dictionary 15. In the recognition result memory 9, an address 24 of a storage area in which various data corresponding to each one character is stored, and a result obtained when the cut-out character image is compared with the recognition dictionary 15 by the matching unit 14. The first character code 25 to the n-th character code 26 arranged and output in the order of similarity, and the position coordinates P (X1, Y1) of the cut-out one-character image shown in FIGS. 3 and 9 (2). 27 and Q (X2, Y2) 28, vertical and horizontal size 2 of one character image
9. Data such as a position (row and column) 30 indicating the row and column of the input image 21 where the character is input is stored in correspondence with each character.

FIG. 10B is a diagram in which the first character code 25 in FIG. 10A is extracted. Regarding the first character code output as the recognition result, FIG.
Character code 25 shown in, since the standard pattern of the upper case only for matching shape similar characters are used, the character code 0536 uppercase "Tsu" is given for lower case _"Tsu". Similarly, for the lowercase letter " _i ",
The character code 0504 of the capital letter “A” is given.

FIG. 10 (3) shows a case where an uppercase / lowercase determination unit 11 determines whether an input character image is uppercase or lowercase based on a threshold value obtained from a histogram. , The first character code 31.

[0040] In FIG. 10 (2), respectively, had been given a character code of a capital _{"Tsu", "I"} are respectively healed written in lower case letters of the code 0535,0503.

As described above, the recognition dictionary 15 stores a standard pattern consisting of only uppercase characters for similar characters having uppercase and lowercase characters that are geometrically similar to each other as a standard pattern. Does not determine whether uppercase or lowercase by the collating unit 14 in character recognition. As a result, the number of standard patterns to be compared can be reduced, and character recognition can be performed faster. In addition, for similar characters, whether uppercase or lowercase is determined based on the size of the extracted character image, it is possible to increase the recognition accuracy of similar characters that are difficult to accurately recognize in comparison with a standard pattern.

[0042]

As described above, according to the present invention, one of uppercase and lowercase characters is stored in a dictionary memory as a standard character image with respect to similar characters having a geometrically similar uppercase and lowercase character. A standard character image is stored, and the recognition unit does not determine whether uppercase or lowercase characters are similar for similar characters. The standard character image is
Regardless of whether the characters are similar characters or not, matching is performed for all extracted character images, so that the number of standard character images can be reduced and character recognition can be speeded up. For similar characters, the size of the extracted character image is separately detected by the detection means, and based on the contents of the recognition result memory, the similar character characters included in one line of the frequency table of similar characters are respectively detected. Count the frequency for each character code, create a histogram of the frequency based on the size of the character whose frequency is 1 or more, find the division point, and use this size as a threshold to determine whether it is uppercase or lowercase. Since the determination is made, it is possible to more accurately recognize the similar-shaped character, which is conventionally difficult to accurately recognize in the collation with the standard character image.

According to the present invention, when a histogram cannot be created for a specific similar character or when a threshold value cannot be determined based on the histogram, the threshold of similar characters other than the specific similar character is determined. Since it is determined whether the extracted character image is uppercase or lowercase based on the value, the frequency of occurrence of erroneous character recognition can be reduced.

[0044]

[Brief description of the drawings]

FIG. 1 is a simplified block diagram showing a configuration of a character recognition device 16 according to an embodiment of the present invention.

FIG. 2 is a simplified block diagram showing a configuration of a control unit 2 shown in FIG.

FIG. 3 is a diagram illustrating a relationship between a character image cut out by a CPU 7 and coordinate axes.

FIG. 4 is a flowchart showing an operation of character recognition in the character recognition device 16;

FIG. 5 is a flowchart illustrating a procedure for creating a frequency table, which is executed by the case determination unit 11;

FIG. 6 is a diagram showing a configuration example of data of a frequency table created by the operation shown in the flowchart of FIG. 5;

FIG. 7 is a flowchart showing a procedure for obtaining a division point c for upper / lower case discrimination.

8 is a histogram created based on the data of the frequency table shown in FIG.

FIG. 9 is a diagram showing a state in which a document 20 including characters is read, temporarily stored as a character image, recognized, and output again in the same form as the document 20.

FIG. 10 is a diagram showing an example of data relating to each input character stored in a recognition result memory 9;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input part 2 Control part 4 Recognition part 7 CPU 8 Image memory 9 Recognition result memory 10 Uppercase / lowercase identification memory 11 Uppercase / lowercase determination part 12 Character extraction part 14 Collation part 15 Recognition dictionary 16 Character recognition device

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06K 9/46 G06K 9/62

Claims (2)

(57) [Claims] A reading unit that reads a document on which a plurality of characters are displayed, converts the document into image data, and outputs the image data; an image memory that stores image data from the reading unit; A character extracting means for sequentially extracting, and a standard character image to be matched with the extracted character image and a corresponding character code are stored. For similar characters having a geometrically similar uppercase and lowercase character, uppercase or lowercase characters are used. A dictionary memory in which only one of the standard character images is stored, a detecting means for detecting the size of the extracted character image, and a character code which recognizes the extracted character image based on the storage contents of the dictionary memory and converts the character code. A recognition means for outputting, a recognition result memory for storing a character code from the recognition means, and a size from the detection means as a recognition result; , The frequency is counted for each character code of the similar character, a histogram of the frequency based on the size of the character having the frequency of 1 or more is created, and the division point of the histogram is used as a threshold to determine whether the character is uppercase. A character recognition device comprising: a determination unit configured to determine whether the character is a lowercase letter. 2. The method according to claim 1, wherein the frequency of the character code is less than 1 and a histogram cannot be created for the specific similar character, or the character size is only uppercase or lowercase and the histogram is not included. The character recognition device according to claim 1, wherein when a threshold value cannot be determined based on the threshold value, a determination is made based on a threshold value of a similar character other than the specific similar character.