JPH08279022A - Character recognition device - Google Patents

Abstract

PURPOSE: To perform an optimum normalizing process at a high speed as for a character recognition device which recognizes a character by normalizing character data of a document. CONSTITUTION: A rectangle data processing means 13 generates circumscribed rectangle data from image data fetched into an input means 12 and a matching process means 14 processes the circumscribed rectangle data by individual, merging, and separation coordination and temporarily stores the result and rectangle data in a rectangle data processing means 13. After a data extracting process means 15 extracts normal character data from the image data on the basis of the rectangle data, a recognizing process means 16 performs character recognition from the character data.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for normalizing character data of a document to perform character recognition.

[0002]

2. Description of the Related Art In recent years, along with the widespread use of personal computers and facsimiles, the demand for character recognition devices and program software for character recognition has increased, and character recognition is required to be accurate and fast.

Conventionally, a character recognition device performs a normalization process of reading a character as a dot image with a scanner or the like and cutting out character data from the read dot image data. Then, the cut-out dot character data is recognized, and a grammatical check is made as to whether or not the recognized character matches.

The above-described normalization processing is necessary to more reliably perform the next recognition processing, and consideration is given to biased characters. Japanese Patent Laid-Open No. 3-657 discloses a normalization process for such a biased character.
The matters described in Japanese Patent No. 81 are known. This is one of the maximum height or width of the character symbol as the reference side, the other side as the dot existence width of the character symbol pattern, and the target character symbol pattern is cut out in the rectangular area. The size of the target character / symbol pattern in the height direction and the width direction is scaled by the ratio. That is, a character having a dot distribution that is far apart from an average Chinese character such as a punctuation mark "." Is normalized without losing its position information.

[0005]

However, Japanese Unexamined Patent Publication (Kokai) No. 3-6.
The method described in Japanese Patent No. 5781 scans the dots existing in one line from the left side (or the right side) and performs rectangular cutout using the spaces where the dots do not exist as character intervals. There is a problem of causing recognition.

For example, if there is a separated character such as the character "wa" in one character, the normalization process is performed in the separated state. On the other hand, in the case of Chinese characters with small character spacing, they are erroneously cut out as one character. It will be. Further, when an error occurs due to the above-mentioned erroneous recognition, the normalization process is performed again, which takes time.

Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to provide a character recognition device that performs normalization processing at high speed and optimally.

[0008]

FIG. 1 is a block diagram showing the principle of the present invention. In FIG. 1, in order to solve the above problems as described in claim 1, the character recognition device 11 inputs the character data of a document as image data.
Then, the circumscribing rectangle data of the independent constituent part of the character data is created and stored from the image data from the input means 12, and the rectangle data matched based on the circumscribing rectangle data is written. Independently using the rectangular data processing means 13 and predetermined data detected from the character data portion of the image data based on the circumscribed rectangular data read from the rectangular data processing means 13,
By performing the merging and separation processing, the rectangular data corresponding to one character is matched and the rectangular data processing means 13
And the data extraction processing means 15 for extracting the regular character data from the image data taken in by the input means 12 based on the rectangular data of the matching processing means 14, And a recognition processing means 16 for performing character recognition from the regular character data from the character data extraction processing means 15.

[0009]

As described above, in the present invention, the circumscribed rectangular data is created by the rectangular data processing means 13 from the image data taken in the input means 12, and the circumscribed rectangular data is independently, merged or separated by the matching processing means 14. Of the character data, and the character data is stored in the rectangular data processing means 13 as rectangular data, and the normal character data is extracted from the image data by the data extraction processing means 15 based on the rectangular data. Character recognition is performed by the recognition processing means 16 from the data. As a result, optimal normalization processing is performed on separated characters, merged characters, etc., and it is possible to reduce erroneous recognition and increase speed.

[0010]

FIG. 2 shows a functional block diagram of an embodiment of the present invention. FIG. 2 is a functional block diagram of the character recognition device 11, which is configured as a system in which a scanner or the like is connected to a personal computer, for example.

In FIG. 2, document image input means 21
Reads character data of a document as image data, and a device such as a scanner or a facsimile modem for inputting a document as a dot image is used. The read image data is stored and fetched in the image data storage means 22. As the image data storage means 22, for example, a memory (RAM: Random Access Memory) of a personal computer or an HDD (hard disk drive) is used. The document image input means 21 and the image data storage means 22 constitute the input means 12.

For example, when a scanner is used as the document image input means 21, image data of a document is created by the scanner and input to a predetermined bus of a personal computer, and is binary "1" (for example, a black portion) in the RAM. And “0” (for example, white part) are stored. In this case, RS (Recommended Standar)
d) Serial port such as 232C, Centronics specification, or SCSI (Small Computer System Interfac)
e) or a parallel port such as GPIB (General Purpose Interface Bus) is used. Then, data corresponding to a recording separator such as CRLF is added to the end of one line of the image data.

A dedicated interface board may be provided, and the same applies when a facsimile or facsimile modem is used as the document image input means. Further, it is not necessary for all the image data to be stored in the RAM in the RAM, and the data necessary for the normalization processing and the character recognition processing can be appropriately selected from the image data once stored as a file in the HDD. It may be loaded and stored. In this case, bits and dots are stored in the RAM in association with each other, and one-line length data corresponding to the "paper width" in which the document is written is also stored.

Subsequently, the line width recognition means 23 reads the image data from the image data storage means 22 to create line width data, and the line width data storage means 24 as a part of the RAM (or HDD), for example. It is stored in R, considering small capacity of data, complicated rewriting, etc.
It is desirable to store in AM. The circumscribing rectangle data creating means 25 reads out the image data from the image data storing means 22, and reads out the line width data from the line width data storing means 24 to define a circumscribing rectangle as data of the circumscribing area of the independent constituent part of the character data. Create data,
It is stored in a part of the rectangular data storage means 26. The line width recognition unit 23, the line width data storage unit 24, the circumscribing rectangular data creation unit 25, and the rectangular data storage unit 26 constitute a rectangular data processing unit 13.

On the other hand, the standard rectangular data creating / creating means 27
Reads the line width data from the line width data storage means 24, determines the character width corresponding to the character height of the character data obtained by the line width data, and creates standard rectangular data of full-width characters of the word processor document. RAM (or HD
The data is stored in the standard rectangular data storage means 28 assigned to a part of D).

The rectangular data matching means 29 detects the standard rectangular data read from the standard rectangular data storage means 28, the circumscribed rectangular data read from the rectangular data storage means 26, and the character data portion of the image data. The specified data and the circumscribed rectangular data are processed individually, merged, and separated to be matched with rectangular data corresponding to one character, and are stored in the rectangular data storage means (a part of the memory (or HDD)) 26. Store. That is,
The rectangular data storage means 26 stores both externally connected rectangular data and rectangular data. The standardized rectangular data creating means 27, the standard rectangular data 28, and the rectangular data matching means 29 constitute the matching processing means 14.

The character data extraction means 30 is the data extraction processing means 15, and extracts the normal character data from the image data read from the image data storage means 22 based on the rectangular data read from the rectangular data storage means 26. To do. This regular character data is supplied to the character recognizing means 31 and compared with the reference character data stored here in advance to perform character recognition. The disk 32 stores the recognized character data, and the CRT 32 appropriately displays the recognized character data as a character. This character recognition means 3
1, a disc 32 and a CRT 33 for recognition processing means 1
6 is configured.

The line width recognizing means 23, the circumscribing rectangular data creating means 25, the standard rectangular data creating means 27, the rectangular data matching means 29, the character data extracting means 30, and the character recognizing means 31 are program processes. Therefore, FIG.
FIG. 2 is an explanatory diagram of a process of character extraction by the character recognition device of FIG. In FIG. 3, first, a character line to be normalized is detected (step (S1)), and an independent component part of the character data on the corresponding line is detected as circumscribed rectangle data (S2). All the characters (full-width characters of word processor characters) obtained by the statistical processing of the character height and character width of each independent portion are discriminated and extracted by the standard rectangular data (S3), and some of the partial characters other than this extraction can be extracted. The symbol is extracted (S4). Further, among the partial characters other than those extracted in S4, the split characters are merged (S5), and the contact character portion is cut out (S6). Then, character data other than the above is cut out as half-width characters and all the cut-outs are performed (S7), the process moves to the next line, and the same cut-out processing is repeated.

Here, FIG. 4 is an explanatory diagram of an example of a document for performing the character segmentation of FIG. 3, and FIG. 5 is an explanatory diagram of the character segmentation process of FIG. As shown in FIG. 4 (A), characters are recorded in each line in the document 41. For example, FIG. 4 (B)
The first line shown in "This product has a humidity of 10

[0020]

[Outside 1]

It is assumed that the character segmentation of "35%" is performed, and the segmentation as shown in FIG. 5 is performed. FIG. 5 corresponds to S1 to S7 in FIG.
All the character lines of one line are detected as in (S1),
Independent parts of this character line are detected as in S12 (S2). In this case, the “ha” is detected in pieces.

Subsequently, as shown in S13 (S3), full-width characters are extracted by the above-described statistical processing (details will be described later),
As shown in S14 (S4), the extractable symbols “,” other than the double-byte characters are extracted. After that, as shown in S15 (S5), the split character "ha" is merged and S1
6 (S6), the contact character is cut out (not present in this example). Then, the characters other than the cutout are used as half-width characters and the cutout is performed as in S17 (S7).

Next, the details of character segmentation will be described. FIG. 6 shows an explanatory diagram of creating circumscribed rectangle data. As a preliminary step, in addition to the area where the image data of the image data storage means 22 is stored, an empty data area 22a (the size is the same as the paper width) and an empty flag area for storing character data as binary data on the RAM. 22b are prepared. In FIG. 6, for example, the case of the above-mentioned “book” will be described. As shown in FIG. 6, when there is no dot for each line of the dot row, it is set to "0", and once it is confirmed that there is a dot, "1" is detected.

Once the presence of dots can be confirmed, the address on the RAM corresponding to the left end in the row direction is written, regarded as the beginning of a row, and a flag is set. The flag indicates the existence of dots, and when the dots cannot be confirmed when the lines are successively checked one by one in the row direction, the flag is set there. Then, the address corresponding to the left end of the line when the flag is set is stored.

On the other hand, when a line is scanned, the presence of dots is detected. Therefore, when a dot appears, a bit corresponding to the "distance from the left end of the row" is set. As a result, as shown in FIG. 6, data is created such that the dot distribution of characters is viewed in the row direction (vertical direction of the paper). This data is line width data indicating the range of characters in the line direction, and the difference between the dot start address and the dot end address is the character width. Figure 6
Data such as will increase every time a line is recognized. The line width data is RA as the line width data storage means 24.
It is stored in another area (not shown) on M.

By this processing, it is possible to automatically remove the noise even if the data has a lot of noise by checking the dot existence rate for each row data.
That is, by examining the distribution of line width data, data with extremely small line width can be regarded as noise. After removing the noise component in this way, the average value of the line width is calculated again to obtain the average value of the character height. Then, the line width data and the character width data are combined to create circumscribed rectangle data.

Therefore, FIG. 7 shows an explanatory view of the independent portion detection by the circumscribed rectangle data. Further, FIG. 8 shows an explanatory view of storing the circumscribed rectangle data in the memory (or RAM).
FIG. 7 shows the character data for one line shown in FIG.
3 shows the circumscribed rectangular area of the circumscribed rectangular data created by the above method. Then, the circumscribed rectangle data of FIG. 7 is created as a table on the memory (or RAM) as shown in FIG. 8, and the head address and offset (maximum 7) width (character width) are set for all independent parts. , A full-width character flag and a half-width character flag are stored. The full-width character flag and the half-width character flag are sequentially set in the matching process described later.

The matching process for recognizing full-width characters, separated characters, merged characters and half-width characters will be described.
In this case, the average value of the character height obtained as described above is regarded as it is as the average value of the character width. This is especially true for type characters such as word processor documents, where the ratio of height to width is approximately 1: 1 for double-byte characters (double-byte characters).
This is based on the fact that the character pitches of printed characters are substantially uniform and the heights of printed characters are approximately average.

To recognize double-byte characters, first, character width data m
And average value of character width data (that is, average value of character height)
Data whose ratio x with h matches 0.7 <x <1.4 is extracted, and a flag (full-width character flag in FIG. 8) is set on them. This is based on the prediction that the characters contained in this range will be regular double-byte characters.

For recognition of characters other than full-width characters, an average value is calculated only for the character width of the flagged rectangular data, a more accurate average value W of the character width is calculated, and the maximum value Wm is calculated.
Extract ax. For the rectangular data for which the flag is not set, the average value W and the maximum value Wmax obtained above are compared to perform character separation processing. That is, W ≦ m ≦
In case of Wmax, it is extracted as one character, and 2W ≦ m ≦ 2W
In case of max, it is separated into two characters and extracted. And m
When> 2 Wmax, the circumscribed rectangle data recognized by the extraction process is separated by dividing the character into the integer part of (m / Wmax) and a flag is set. Then, the data for the ones for which the flag is not yet set matches the condition of m <W. That is, the circumscribed rectangle data is less than the width of one full-width character.

Here, FIG. 9 shows a circumscribed rectangle of full-width characters or less.
An explanatory view of recognition of data is shown. FIG. 9A shows an example for explanation.
For example, it shows a "custom" character, and the character width m of the independent part
The circumscribed rectangle data adjacent to the circumscribed rectangle of (<W)
Distance to W₁, W₂, ... W _n(W in FIG. 9₁~ W_Four)
As W₁If ≧ Wmax, it seems to be a symbol
Therefore, it is extracted as one character. Also, W ≦ W_n≤ Wm
In the case of ax (n = 2, 3, 4, ...), W_nUp to 1 character
To extract. W in FIG. 9 (A)_FourUp to 1 character
It is shown to extract.

As shown in FIG. 9B, this is "ha".
Then, up to W ₂ will be extracted as one character. Therefore, the data shown in (4) and (5) of FIG. 8 are recognized as full-width characters as shown in FIG. 9 (C) by matching and are flagged, and the rectangular data matching means 29 stores the rectangular data. Writing to the means 26 is performed as rectangular data.

The circumscribed rectangle data which has not been flagged by the above processing is judged as a character having a rule different from Kanji such as half-width alphanumeric characters and is extracted as an independent character and a flag (half-width character flag) is set. Is written. In FIG. 8, "1", "0", "3", and "5" are set as one character and a half-width character flag is set.

In this way, the circumscribed rectangular data is modified and written and stored as rectangular data in the rectangular data storage means 26. Then, the character data extraction unit 30 extracts character data from the image data read from the image data storage unit 22 based on the rectangular data, and appropriately enlarges or reduces the character data so that the character recognition unit 31 in the subsequent stage can easily recognize the character data. The scaling factor for the conversion processing is determined and the processing is transferred to the character recognition means 31.

In this way, the separated characters and the merged characters can be optimally normalized, and since the program processing is not large, high-speed processing can be realized. In addition, in the case of a document in which character fonts of different sizes are used in one line, normalization processing is optimally performed by first separating the characters in one line with characters of the same size and then processing. be able to. For example, a character height position for one line is sequentially stored for each dot (flag) of the flag area 22b, and statistical processing such as normal distribution is performed on the character height position to generate character fonts of a plurality of sizes. The character height of each font when used can be obtained.

[0036]

As described above, according to the present invention, the circumscribed rectangular data is created by the rectangular data processing means from the image data taken in the input means, and the circumscribed rectangular data is independently or merged by the matching processing means. By performing the process of separation matching,
The rectangular data is stored in the rectangular data processing means as a rectangular data,
After the normal character data is extracted from the image data based on the rectangular data by the data extraction processing means, the recognition processing means performs character recognition from the character data, so that the optimum normal character for separated characters and merged characters can be obtained. As a result of the conversion processing being performed, erroneous recognition is reduced and speeding up can be achieved.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

FIG. 2 is a functional block diagram of an embodiment of the present invention.

FIG. 3 is an explanatory diagram of a process of extracting a character by the character recognition device of FIG.

FIG. 4 is an explanatory diagram of an example of a document that cuts out characters in FIG.

FIG. 5 is an explanatory diagram of a character cutting process of FIG.

FIG. 6 is an explanatory diagram of creating circumscribed rectangle data.

[Fig. 7] Bounding rectangle

FIG. 8 is an explanatory diagram of storing circumscribed rectangle data in a memory.

FIG. 9 is an explanatory diagram of recognition of circumscribed rectangle data other than full-width characters.

[Explanation of symbols]

 11 Character Recognition Device 12 Input Means 13 Rectangular Data Processing Means 14 Matching Processing Means 15 Data Extraction Processing Means 16 Recognition Processing Means 21 Character Image Input Means 22 Image Data Storage Means 23 Line Width Recognition Means 24 Line Width Data Storage Means 25 Circumscribing Rectangle Data creating means 26 Rectangular data storing means 27 Standard rectangular data storing means 28 Standard rectangular data matching means 29 Rectangular data matching means 30 Character data extracting means 31 Character recognition means

Claims (1)

[Claims] 1. Input means for capturing character data of a document as image data, and circumscribing rectangle data of independent constituent parts of the character data is created and stored from the image data from the input means. Rectangular data processing means in which rectangular data matched based on the circumscribed rectangular data is written, and a predetermined data detected from the character data portion of the image data based on the circumscribed rectangular data read from the rectangular data processing means. A matching processing unit that matches the rectangular data corresponding to one character and stores it in the rectangular data processing unit by performing independent processing, merging, and separation processing using the data, and the rectangular data of the matching processing unit. And data extraction processing means for extracting the regular character data from the image data taken in by the input means, A recognition processing means for performing character recognition from the regular character data from the character data extraction processing means.