JP3537570B2 - Space detection method for Japanese-English mixed documents, pitch format determination method, and space detection method for fixed-pitch alphanumeric character strings - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to space detection processing in a character recognition device, and more particularly to space detection processing for a document in a mixture of Japanese and English.

[0002]

2. Description of the Related Art When a document is converted into text data by character recognition so that the document can be reproduced from the text data, it is necessary to detect a space in the document upon character recognition. The following is known as a prior art relating to such space detection in a document.

Prior Art 1: Counting the number of white pixels between characters,
The number of space codes is determined by dividing the counted value by the standard character pitch (Japanese Patent Application No. 63-14282).

Prior art 2: A pitch format (constant pitch / proportional pitch) of a document is determined. Then, a space is detected by a method according to the determined pitch format (Japanese Patent Application No. 1-161176).

Prior art 3: In a Japanese sentence in which Japanese and English characters are mixed, the type of space and the presence or absence of a space are determined by comparing information on the space width between characters and the character width. To determine whether the space is valid or invalid (Japanese Patent Application No. 2-214136).

Prior art 4: In a Japanese sentence containing both Japanese and English characters, the width of the space between characters and the standard size of the character are compared to determine the type and presence or absence of the space. The character space width is corrected, and the corrected value is compared with the standard character size (Japanese Patent Application No. Hei 3-18476).
issue).

[0007]

In a document in which Japanese characters and alphanumeric characters (alphabetic characters and numbers) are mixed, differences in fonts between Japanese characters and alphanumeric characters, differences in character sizes between full-width characters and half-width characters, constant pitch and There is a difference in pitch format called proportional pitch, and such a difference occurs in an unspecified portion in a document. The width of the space also differs depending on the location. This can be easily understood from the print sample shown in FIG. In FIG. 9, in sample 1 in which all pitch characters are printed at a constant pitch, “R
There is a space only between the "h" of the iconh and the next "P". On the other hand, in sample 2 in which alphanumeric characters are printed at a proportional pitch, a space is placed between the "model" of the "new model" and the next "R", and between "h" of the "Ricoh" and the next "P". And the space is narrower than the space of the sample 1.

For such a mixed document of Japanese and English, it has been difficult in the prior art to detect a space with high accuracy.
Also, there has been no known method for accurately determining a pitch format when Japanese characters and alphanumeric characters are mixed in a line. The prior art 2 assumes that the entire document or the entire line is an English sentence, and performs processing in units of English words. Therefore, it is not possible to accurately detect a space in an English word that appears partially in a line. Regarding space detection in fixed pitch character strings,
Because the character spacing of the constant-pitch character string greatly fluctuates, a full-width space can be detected, but even the conventional technology 4 for correcting the character spacing does not improve the detection accuracy of a half-width space. Regarding the space detection in the proportional pitch character string, the space could not be detected even in the prior art 4 because the character pitch of the proportional pitch character string was narrow.

SUMMARY OF THE INVENTION It is an object of the present invention to provide an improved method capable of detecting a space with high accuracy in a Japanese-English mixed document in which Japanese characters and alphanumeric characters are mixed in a line, and suitable for the space detecting method. Pitch format determination method and constant pitch
An object of the present invention is to provide a method for detecting a space in a character string .

[0010]

According to the present invention , there is provided a method for detecting a space in a mixed-language document in a character recognition apparatus for recognizing characters by extracting characters from an image of the document. A target range extracting process of extracting a character string as a target range, a pitch format determining process of determining whether a pitch format is a fixed pitch or a proportional pitch for each target range extracted by the target range extracting process,
A constant pitch character string space detection process for detecting a space for each target range determined as a constant pitch by the pitch format determination process, and a space detection for each target range determined as a proportional pitch by the pitch format determination process it is an basic that it has a proportional pitch string space detecting process.

According to a first aspect of the present invention, in the method for detecting a space in a mixed document of Japanese and English, the pitch format determination processing includes:
For each target range, the character rectangle interval ratio, which is a ratio of the character rectangle interval to the standard character size determined based on the clipping information for each target range, is smaller than a predetermined threshold value, the character rectangle interval ratio is smaller. If the number of characters is larger than the threshold and is larger than the character rectangle interval, it is determined that the pitch is proportional, otherwise, it is determined that the pitch is constant.

According to a second aspect of the present invention, in the constant-pitch character string space detection processing, the smaller character rectangular pitch of the character rectangular pitches between the characters before and after the focused character is detected. When the ratio of the character rectangular pitch between the focused characters to the reference pitch is larger than a predetermined threshold value, it is determined that a space exists between the focused characters.

According to a third aspect of the present invention, in the proportional pitch character string space detecting process, a character rectangle interval ratio, which is a ratio of a character rectangle interval to a standard character size determined based on cutout information for each target range, is set. , When it is larger than a predetermined threshold value, it is determined that a space exists between the corresponding characters.

According to a fourth aspect of the present invention, there is provided a character recognition device for recognizing a character by extracting a character from a document image. A character rectangle interval in which the character rectangle interval ratio which is a ratio of the character rectangle interval to the standard character size determined based on the cutout information is smaller than a predetermined threshold is larger than a character rectangle interval in which the character rectangle interval ratio is equal to or larger than the threshold. If there is, a proportional pitch is determined, and if not, a constant pitch is determined.

According to a fifth aspect of the present invention, there is provided a character recognition apparatus for extracting a character from a document image by recognizing a space in a fixed-pitch alphanumeric character string. In contrast, the center of the adjacent character rectangle
The spacing and character rectangle pitch between characters, a reference pitch between characters which focuses the smaller character rectangle pitch ones of character rectangle pitch between the front and rear of characters between target character, attention
Is between the first characters of the alphanumeric character string
Is the character rectangle pitch between characters following the noted character.
The reference pitch is used, and the space between the noticed characters is the end of the alphanumeric character string.
Between the characters of the
Of the character rectangle pitch between the focused characters and the reference pitch (character rectangle pitch / character rectangle pitch).
When the reference pitch is larger than a predetermined threshold value, it is determined that a space exists between the focused characters.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In order to clarify an embodiment of the present invention, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 shows an overall processing flow of an embodiment of the present invention, and FIG.
Shows an example of a device configuration for an embodiment of the present invention. FIG. 3 shows a flow of the space detection processing in FIG. FIG. 4 shows the flow of the pitch format determination processing in FIG. 3, FIG. 5 shows the flow of the proportional pitch character string space detection processing, and FIG. 6 shows the flow of the constant pitch character string space detection processing. 7 and 8 are diagrams for explaining the space detection processing.

Referring first to FIGS. 1 and 2, the overall processing flow will be described in relation to the apparatus configuration. First, document image data is input by the image input unit 200 and stored in the image memory 208 via the bus 213 (step 100). When an image is input, the line / character cutout unit 202 cuts out character lines and characters in the document image in the image memory 208, and stores information on the cutout character lines and character areas in the cutout memory 209 (step). 110). The clipping information stored here includes, for example, the coordinates of the starting point and the ending point of the character line or the starting point coordinates and the width,
The coordinates and size information of the diagonal vertices of the circumscribed rectangle (character rectangle) of the character. Naturally, such information is created and stored in such a manner that the correspondence between each character rectangle and the character line to which it belongs can be identified. The character recognition unit 203 refers to the cut-out information in the cut-out memory 209, takes in the character image from the image memory 208, and compares it with the character dictionary in the character dictionary memory 210 to determine the character code for the character image. Then, it is stored in the result memory 211 (step 120).
Note that the character recognition unit 203 may perform post-processing such as correction of a recognition result by comparison with a character dictionary, using linguistic knowledge such as words and grammar.

When the recognition process by the character recognition unit 203 is completed for the whole or a part of the document image, the space detection process (step 130) starts. This space detection processing (step 130) is performed by the target range extraction unit 20.
4. The process is executed by the pitch format determining unit 205, the constant pitch character string space detecting unit 206, and the proportional pitch character string space detecting unit 207. At this time, the cutout memory 209, the result memory 211, and the work memory 212 are referred to. The space detection result is stored in the result memory 211. When the space detection process is completed, the data in the result memory 211 is output to the outside (step 14).
0), the entire process ends.

The outline of the space detecting process (step 130) is as shown in FIG. First, the target range extraction unit 2
At 04, the character recognition result (character code) in the result memory 211 is referred to, and an alphanumeric character string as a target range of the space detection processing is extracted (step 300) . Information about the extracted target range is stored in the work memory 212. Next, the pitch format determination unit 205 refers to the contents of the cutout memory 209 and the result memory 211 for the characters in the target range, and determines whether the target range is a proportional pitch or a constant pitch (step 301). The determination result is stored in the work memory 212. With respect to the target range determined to be proportional pitch, the proportional pitch character string space detection unit 207 extracts the cutout memory 209 and the result memory 2.
11, the space is detected by the method for proportional pitch, and the detected space is written to the result memory 211 as a space code (step 302). With respect to the target range determined as the constant pitch, the constant pitch character string space detection unit 206 performs space detection by the method for the constant pitch with reference to the contents of the cutout memory 209 and the result memory 211, and detects the detected space. In the result memory 211 as a space code
(Step 303). The specific contents of the processing described above will be described with reference to FIGS.

Consider the case of processing a character line as shown in FIG . This character line has two parts including spaces. One is “Preter 5” printed at proportional pitch (font is Times-Roman).
5 "between" Preter "and" 55 ". The other is “Imagio” and “77” of “Imagio 77” printed at a fixed pitch (font is Courier).
Between. Character recognition for this character line (step 1
The result of 20) does not include a space, and “New copy, Pr
ETER55 and IMAGEIO77 released. "

In the target range extraction process (step 300), a character string in which five or more consecutive alphanumeric characters (alphabetic characters or numbers) are extracted as a target range with reference to a character recognition result that does not include such spaces. . Therefore, in the character line shown in FIG. 7A, as shown in FIG. 7B, two character strings “Preter55” and “Imagio77” are extracted as the target range. Note that the character length for extraction determination is not limited to five characters, and may be changed as appropriate according to the type of document to be processed. Further, the method of determining the target range itself can be appropriately changed.

With respect to the alphanumeric character string extracted as the target range in this way, pitch format determination (step 30)
Perform 1). FIG. 4 is a flowchart thereof.

First, the standard character size for the target range is determined (step 400). Specifically, 1.25 times the maximum character rectangle height (character size A) in the target range and the maximum character rectangle height (character size B) in the character line including the target range are obtained. The larger value of the character sizes A and B is set as the standard character size. However, when the character rectangle of “j” exists in the target range, the character rectangle is not used for determining the standard character size. FIG. 8 is an explanatory diagram of such a standard character size determination.

In the first target range shown in FIG.
Since the height of the character rectangle of "P" is the maximum, a value obtained by multiplying the height by 1.25 and the maximum height of the character rectangle of the "new" in the character line are compared. Character size. Note that the standard character size may be determined by another method.

When the standard character size for the target range is determined as described above, a pair of adjacent character rectangles is selected from the head of the target range using the standard character size, and the space between the character rectangles is selected. The width of the copy, that is, the ratio of the character rectangle interval (see FIG. 7C) to the standard character size (character rectangle interval ratio = character rectangle interval / standard character size) is calculated (steps 405 and 410). The calculated character rectangle interval ratio is stored in the work memory 212. The character rectangle interval is calculated from the coordinates of two character rectangles stored in the cutout memory 209. However, the character rectangle interval may be obtained at the stage of the cutout processing. In that case, the character rectangle interval read from the cutout memory 209 may be used for calculating the character rectangle interval ratio. Next, the character rectangle interval ratio calculated in this manner is set to a predetermined threshold value TH1.
(Step 415). This threshold value TH1 is, for example, 0.17 (although it is not limited to this value, it has been experimentally confirmed that a value around this is appropriate in a general document. However, it is adjusted according to the document to be processed. Is preferred). When the character rectangle interval ratio is smaller than the threshold value TH1, the counter NS (initial value is 0) is incremented by 1 (step 420), and when the character rectangle interval ratio is equal to or greater than the threshold value TH1, the counter NL (initial value is 0) is incremented by one. Is incremented by one (step 421). If the next character rectangle pair remains, step 405 to step 420
Is repeated.

When the processing is completed up to the last character rectangle pair, the processing loop from step 405 to step 420 or 421 is exited, and the routine proceeds to step 425. In this step 425, the value of the counter NS is compared with the value of the counter NL (step 425). The value of the counter NS is the number of character rectangle intervals in which the character rectangle interval ratio in the target range is smaller than the threshold value TH1, that is, smaller than “standard character size * TH1”, and the value of the counter NL is the number of wide character rectangle intervals. . Therefore, in step 425, it is checked whether or not the narrow character rectangle interval is a majority. NS
> NL, that is, when it is determined that the narrow character rectangle interval is the majority, the pitch format of the character string in the target range is set to the proportional pitch, and the flag information is set to the work memory 212.
Is written in association with the target range (step 430),
Otherwise, that is, if the wide character rectangle interval is less than half, the pitch format is set to a constant pitch, and the flag information is written in the work memory 212 in association with the target range (step 435). According to the flag information, a proportional pitch character string space detection process (step 302) or a constant pitch character string space detection process (step 303) is selected as the next process.

In the example shown in FIG. 7A, the first target range "Preter55" is determined to be a proportional pitch as shown in FIG. 7C, and the other target range "Imagio77" is determined to have a constant pitch. Is determined.

The processing content of the space detection processing (step 302) for the target range determined to be proportional pitch will be described. FIG. 5 is a flowchart thereof.

The character rectangle interval ratio of the target range stored in the work memory 212 is extracted, and is taken as a predetermined threshold T.
Compare with H2 (steps 500 and 505). The threshold TH2 is, for example, 0.29 (although not limited to this value,
In a general document, it was confirmed by experiments that the values in this area were appropriate. However, it is preferable to adjust according to the document to be processed.) If the character rectangle spacing ratio is equal to or less than TH2, that is, if the character rectangle spacing is “standard character size * TH
If it is smaller than "2", it is determined that there is no space between the corresponding character rectangle pairs. When the character rectangle interval ratio is larger than TH2, that is, when the character rectangle interval is wider than "standard character size * TH2", it is determined that a space exists between the corresponding character rectangle pairs, and the space between the corresponding recognition result character codes. A space code is inserted (step 510). Similar processing is executed for all character rectangle interval ratios within the target range.

In the example shown in FIG. 7A, as shown in FIG. 7D, "r" at the end of "Preter" in the target range determined to be proportional pitch.
Since only the character rectangle interval ratio between the character and the next “5” exceeds TH2, a space code is inserted at that position.

The contents of the space detection process (step 303) for the target range determined to have a constant pitch will be described. FIG. 6 is a flowchart thereof.

The character rectangle pitch between all characters in the target range is calculated, and the result is stored in the work memory 212 (step 600). That is, the center of a character rectangle adjacent between characters is calculated, and the distance between the centers is determined as the character rectangle pitch between the characters. For this calculation, the coordinate data of the character rectangle stored in the cutout memory 209 is used. However, the center interval between adjacent character rectangles may be calculated in advance in the cutout processing step, stored in the cutout memory 209, and the value thereof may be read.

When the calculation of the character rectangle pitch is completed, one character space to be noticed is selected from the beginning of the target range (step 605). The character rectangle pitch between the following characters is compared with the character rectangle pitch, and the smaller character rectangle pitch is set as the reference pitch for the target character (step 610). When the target character is between the first characters in the target range, since there is no character before the target character, the character rectangular pitch between the characters following the target character is set as the reference pitch. Similarly,
When the space between the characters of interest is between the last characters of the target range,
The character rectangular pitch between the preceding characters is set as the reference pitch.

Next, a value obtained by dividing the character rectangular pitch between the characters of interest by the reference pitch is calculated as a character rectangular pitch ratio (= character rectangular pitch / reference pitch) (step 615).
Then, it is determined whether the character rectangle pitch ratio is larger than a predetermined threshold value TH3, that is, whether the character rectangle pitch between the characters of interest is wider than “reference pitch * TH3” (step 62).
0). The threshold value TH3 is, for example, 1.8 (although the value is not limited to this value, in a general document, a value around this can detect both a full-width space in a full-width character string and a half-width space in a half-width character string. Has been confirmed by experiment, but it is preferable to adjust according to the document to be processed).

If the character rectangle pitch ratio is TH3 or less,
It is determined that there is no space between the target characters. If the character rectangle pitch ratio is larger than TH3, it is determined that there is a space between the characters of interest.
A space code is inserted between the corresponding character codes in 11 (step 625). Similar processing is repeated up to the last character in the target range.

In the example shown in FIG. 7 (a), as shown in FIG. 7 (d), the last "o" of "Imagio" and the next "7" in the target range determined to have a constant pitch. Since the character rectangular pitch ratio exceeds TH3 only between the characters, it is determined that there is a space between the characters, and a space code is inserted.

[0037]

The main effects of the present invention are listed below. (1) An alphanumeric character string is extracted as a target range from the character recognition result, a pitch format is determined for each target range, and a space detection process for each pitch format is performed for each target range. High-precision space detection is possible for mixed-language Japanese and English documents in which alphanumeric strings and Japanese strings are mixed, and alphanumeric strings with fixed pitch and proportional pitch are mixed. become.

(2) The standard character size is determined for each of the alphanumeric character strings extracted as the target range, and the pitch format is determined by focusing on the relative width ratio of the character rectangle interval to the standard character size in the alphanumeric character string. Even if alphanumeric character strings are mixed with Japanese character strings and full-width and half-width alphanumeric character strings are mixed, the pitch format of alphanumeric character strings is fixed pitch or proportional. It is possible to accurately determine whether the pitch is a pitch, and therefore, it is possible to accurately apply space detection processing that matches the pitch format to alphanumeric character strings in mixed Japanese and English documents, so that alphanumeric character strings with fixed pitch and proportional pitch Can be detected with high precision in a document in which is mixed.

(3) Since the character rectangular pitch of the constant-pitch alphanumeric character string varies greatly depending on the character, even if the relative width of the character rectangular pitch and the fixed reference pitch, which does not reflect this variation, is examined, the space is not changed. Cannot be detected accurately. According to the present invention , the smaller one of the character rectangular pitches between the preceding and succeeding characters between the noted characters is used as the reference pitch for the noted characters. Spaces in a character string can be detected with high precision, and therefore, space detection accuracy can be improved for a document in which alphanumeric character strings having a constant pitch are mixed.

(4) Since a proportional pitch alphanumeric character string has a narrow character rectangle interval, it is difficult to detect a space by a method of comparing a standard character size with a character rectangle interval without considering a pitch format or a character type. According to the present invention ,
Since the standard character size is determined for each proportional pitch alphanumeric character string and the ratio of the character rectangle spacing to the standard character size is thresholded, spaces in alphanumeric character strings with proportional pitch can be detected accurately, and Space detection accuracy of a document in which alphanumeric character strings of pitches are mixed can be improved.

[Brief description of the drawings]

FIG. 1 is a flowchart showing an overall processing flow of an embodiment of the present invention.

FIG. 2 is a block diagram showing an example of a device configuration for one embodiment of the present invention.

FIG. 3 is a flowchart showing an outline of a space detection process in FIG. 1;

4 is a flowchart showing the contents of a pitch format determination process in FIG.

FIG. 5 is a flowchart showing the content of a proportional pitch character string space detection process in FIG. 3;

FIG. 6 is a flowchart showing the contents of a constant pitch character string space detection process in FIG. 3;

FIG. 7 is a diagram illustrating a specific example of space detection.

FIG. 8 is an explanatory diagram of a method of determining a standard character size for pitch format determination.

FIG. 9 is a diagram for clarifying a problem in space detection of a mixed document in Japanese and English.

[Explanation of symbols]

200 Image input unit 202 Line / character cutout unit 203 character recognition unit 204 Target range extraction unit 205 Pitch format judgment unit 206 Constant pitch character string space detector 207 Proportional pitch character string space detector 208 Image memory 209 Cutout memory 210 character dictionary memory 211 Result memory 212 work memory 213 bus

Claims (5)

(57) [Claims] 1. A character recognition apparatus for extracting and recognizing characters from an image of a document, a target range extracting process for extracting an alphanumeric character string as a target range based on a character recognition result of a document in a mixture of English and Japanese, and the target range A pitch format determination process for determining whether the pitch format is a fixed pitch or a proportional pitch for each target range extracted by the extraction process, and a space is detected for each target range determined as a constant pitch by the pitch format determination process. Constant pitch character string space detection processing, and proportional pitch character string space detection processing for detecting a space for each target range determined as proportional pitch by the pitch format determination processing,
In the pitch format determination processing, for each target range,
Standard characters determined based on clipping information for each target range
Character rectangle spacing ratio, which is the ratio of character rectangle spacing to size
Is smaller than the predetermined threshold.
The shape spacing ratio is greater than the character rectangle spacing that is greater than or equal to the threshold.
If it is proportional pitch, if not
Characterized as a fixed pitch
Pace detection method. 2. The method according to claim 2, wherein the pitch format is determined by a constant pitch.
When it is determined, the constant pitch character string space detection processing
Of the character rectangle between the characters before and after the character of interest.
Between the characters focusing on the smaller character rectangle pitch
Of the character rectangle pitch between the focused characters
Attention when the ratio to the reference pitch is larger than a predetermined threshold
Characteristic that there is a space between characters
2. A method for detecting spaces in mixed Japanese and English documents according to claim 1.
Law. 3. The method according to claim 1, wherein said pitch format determination processing
When it is determined that the pitch is proportional, the proportional pitch
Switch character string space detection processing,
For standard character size determined based on clipping information
The character rectangle space ratio, which is the ratio of the character rectangle space
Space between corresponding characters when greater than value
2. The Japanese-English language according to claim 1, wherein
Space detection method for mixed documents . 4. A character is cut out from a document image and recognized.
In the character recognition device, each recognized alphanumeric character string
Is determined based on the cut-out information for each alphanumeric character string.
Is the ratio of the character rectangle spacing to the standard character size
The character rectangle spacing whose rectangle spacing ratio is smaller than the predetermined threshold
Is greater than the character rectangle spacing that is greater than or equal to the threshold.
If it is a number, judge it as proportional pitch, so
Otherwise, the pitch is determined to be a constant pitch.
Format determination method. 5. Character recognition by extracting characters from an image of a document
In the character recognition device, each alphanumeric sentence of a fixed pitch recognized
The distance between the centers of adjacent character rectangles in the character string
And the character rectangle pitch between the characters before and after
The smaller of the character rectangle pitches
The reference pitch between the noted characters is used as the reference pitch.
If it is between the first characters of the alphanumeric character string,
The character rectangular pitch between the characters after the character is used as the reference pitch,
If the space between the noticed characters is the last character of the alphanumeric string
The character rectangle pitch between the characters before the noted character
Is the reference pitch, and the character rectangular pitch between
Where the ratio to the reference pitch is (character rectangle pitch / reference pitch)
When the value is larger than the specified threshold, a space
Fixed-pitch alphanumeric sentences characterized by determining that
How to detect space in strings.