JPH10116321A - Character recognition method and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce an erroneous recognition rate by correcting a collation order so that the collation order of a recognition candidate which is character-recognized becomes the first through the use of a dictionary whose type is the same as the type of the characters on the respective characters of a single character area when the character area is judged to be the single character area where a single type of characters are mentioned. SOLUTION: A character segment part 16 segments respective character pictures, generates character area information, sums up the number of the character pictures contained in the respective character areas and generates an intra-character area character number summed table. A single character area judgment part 22 judges whether the respective character areas are the single character area where the type of the characters contained in the character area is single or not based on character area information of the character area, which is sent from the character segment part 16, or collation dictionary information being the recognition result of the respective character pictures contained in the character area, which is sent from the distance calculation part 20. The type of the dictionary which collation dictionary information of the recognition candidate whose collation order of the respective character pictures is the first shows is set to be the type of the representative dictionary of the character picture.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a character recognition method and a character recognition device.

[0002]

2. Description of the Related Art In a conventional character recognition apparatus, when recognizing a character pattern such as a handwritten character on a medium such as a document or a form to be read, or a printed character (printed character),
Character recognition is performed for each character pattern cut out from the captured image. At the time of character recognition, a feature vector is calculated by performing feature extraction on each cut-out character pattern. Then, the calculated feature vector is compared with the feature vector of the standard pattern stored in the dictionary. As the dictionary of the standard pattern, a plurality of dictionaries such as a dictionary for handwritten characters and a dictionary for printed characters are used. Then, a character category of a feature vector of a standard pattern that is most similar to the calculated feature vector is selected from a plurality of dictionaries, and a recognition result is obtained. The result of this recognition is CR
It is displayed on T. The operator checks the displayed recognition result, and corrects the recognition result when there is an error in the recognition result.

[0003]

However, in the conventional character recognition apparatus, when performing character recognition using a plurality of dictionaries, character recognition is performed using a dictionary of a type different from the character type of the character pattern. was there. For example, a character dictionary may be used for character recognition of a character pattern of handwritten characters. The misrecognition rate of recognition results obtained using a dictionary of a type different from the type of character pattern is generally higher than that of recognition results obtained using a dictionary of the same type as the type of character pattern. Tend to be.

[0004] Therefore, it has been desired to realize a character recognition device and a character recognition method capable of reducing the erroneous recognition rate.

[0005]

By the way, in general, there is often only one kind of character described in one column in a medium on which characters are recorded. In character recognition, one field is usually treated as one character area. Therefore, the type of character described in one character area is often one type. That is, the characters described in the individual character areas are often of one type, such as only handwritten characters or only printed characters. Therefore, the inventor of the present application paid attention to this point.

(First Invention) Therefore, according to the character recognition method of the first invention of the present application, individual character images are cut out from an image captured from a medium in which characters are described, and the individual character images are cut out. The feature of the character image is compared with the feature of the standard character pattern for character recognition stored in each of the plurality of types of dictionaries, and a plurality of standard patterns for the character image are recognized as recognition candidates. In a character recognition method in which a collation order is assigned and selected, and a recognition candidate having the highest collation order among the recognition candidates is used as a recognition result, character region information representing a character region containing a character image is generated. Generating collation dictionary information indicating the type of dictionary in which the recognition result or the standard character pattern of the recognition candidate is stored, based on the character area information and the collation dictionary information,
It is determined whether each character region is a single character region in which the type of character included in the character region is single,
When the character area is a single character area, area representative dictionary information indicating the type of dictionary corresponding to the type of character included in the single character area is generated, and recognition candidates for each character image belonging to the single character area are generated. The matching order of the matching recognition candidates in which the matching dictionary information and the area representative dictionary information match with each other is maintained while maintaining the matching order of the matching recognition candidates. It is characterized in that it is corrected so as to be higher than the matching order of the candidates.

In the character recognition method according to the first aspect of the present invention, preferably, in determining whether or not the character area is a single character area, each character area specified as belonging to the character area based on the character area information is determined. For each character image, the type of the representative dictionary is determined from the type of dictionary indicated by the collation dictionary information for each recognition candidate of the character image, and based on the type of each representative dictionary of each character image included in the character area. It is preferable to determine whether or not the character area is a single character area.

In the character recognition method of the first invention, it is preferable to use a dictionary for handwritten characters and a dictionary for printed characters.

(Second invention) A second invention according to this application is described.
According to the character recognition apparatus of the present invention, character areas are extracted from a plurality of types of dictionaries each storing a standard pattern for character recognition, and an image captured from a medium in which characters are described, and individual character areas are extracted from the character areas. A character extracting unit for extracting the character image of the character image, a feature extracting unit for extracting the characteristics of each character image, and comparing the character image characteristics with the standard character pattern characteristics to obtain a plurality of standard characters for the character image. In a character recognition device including a pattern and a collation unit that selects a pattern as a recognition candidate and sets a collation order as a recognition candidate, and a collation unit that sets a recognition result having the highest collation order among the recognition candidates as a recognition result, as a character cutout unit, A collation dictionary that includes a character cutout unit that generates character area information indicating a character area in which a character image has been included, and as a collation unit, indicates a type of dictionary in which standard character patterns of recognition candidates are stored. A comparison unit that generates a report, and determines whether each character region is a single character region in which the character type included in the character region is a single character based on the character region information and the collation dictionary information. When the character region is a single character region, the character region includes a single character region determination unit that generates region representative dictionary information indicating a dictionary type corresponding to a character type included in the single character region. Among the recognition candidates of each of the character images to which the matching dictionary information and the area representative dictionary information match, the matching order of the matching recognition candidates that match the matching dictionary information is maintained. It is characterized by comprising a correction unit that corrects the recognition order so that it is higher than the collation order of other recognition candidates that do not match the dictionary information.

In the character recognition device according to the second aspect of the present invention, preferably, the single character area determination unit performs the recognition of each character image for each character image specified to belong to the character area based on the character area information. Determine the type of the representative dictionary from the types of dictionaries indicated by the matching dictionary information about the candidates, and determine whether the character region is a single character region based on the type of each representative dictionary of each character image included in the character region It may be a single character area determination unit that determines whether or not it is not.

In the character recognition device according to the second aspect of the present invention, preferably, the dictionary may include a dictionary for handwritten characters and a dictionary for printed characters.

In each of the inventions according to this application, characters include general symbols.

As described above, according to the first and second aspects of the invention, when the character area is determined to be a single character area in which a single type of character is described, the character of each character in the single character area is The collation order can be corrected so that the collation order of the recognition candidates recognized by the characters using the same type of dictionary is the first. As a result, it is possible to match the type of the character with the type of the dictionary used for the recognition result. For this reason, the false recognition rate can be reduced.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, with reference to the drawings, an embodiment of a character recognition method according to a first invention and a character recognition device according to a second invention according to the present application will be described together. still,
The drawings referred to merely schematically show the sizes, shapes, and arrangements of the components so that these inventions can be understood. Therefore, these inventions are not limited only to the illustrated examples.

(Configuration) First, the configuration of the character recognition device of this embodiment will be described with reference to the block diagram of FIG.

The character recognition device of this embodiment includes a plurality of types of dictionary units 10, scanner units 12, image storage units 14,
Character extraction unit 16, feature extraction unit 18, distance calculation unit 20 as a comparison unit, single character region determination unit 22, and correction unit 24
It has.

The dictionary unit 10 includes a dictionary 10a for handwritten characters and a dictionary 10b for printed characters. The handwritten character dictionary 10a stores standard patterns for character recognition of handwritten characters, and the dictionary 10b for printed characters stores standard patterns for character recognition of printed characters.

The standard pattern of each dictionary is a collection of feature vectors created from handwritten or printed character images of character codes corresponding to the standard pattern. Here, as the feature vector of the standard pattern, an average of feature vectors respectively obtained from a plurality of standard character images having the same character code is used.
Note that there may be a plurality of standard patterns for the same character code, but only one character code corresponds to one standard pattern. Further, the character code does not need to be limited to the JIS code. For example, Uni as a character code
code (Unicode) may be used.

The scanner unit 12 takes in a binarized image of the character from a medium on which the character is described.

In the character extracting section 16, the captured image is decomposed into a character string in a region where characters are arranged. Further, individual character images are cut out from the character string. Then, the character extracting section 16 generates character area information indicating the character area in which each of the extracted character images was included.

In the feature extracting section 18, various features such as the line width, height, horizontal component and vertical component of the character image are extracted as a feature vector of each character image.

In the distance calculation unit 20, the features of the character image are compared with the features of the standard character pattern. As a result, a plurality of standard character patterns for the character image are collated as recognition candidates. Selected. And
The recognition candidate having the highest collation order among the recognition candidates is set as the recognition result. Further, the distance calculation unit 20 generates collation dictionary information indicating the type of dictionary in which the recognition result or the standard character pattern of the recognition candidate is stored.

In the single character area determination section 22, based on the character area information and the collation dictionary information, each character area is determined to be a single character area in which the character type included in the character area is a single type. It is determined whether or not each is not.
Further, when the character area is a single character area, area representative dictionary information indicating the type of dictionary corresponding to the type of character included in the single character area is generated.

In the correction unit 24, among the recognition candidates of each character image included in the single character area, the matching order of the matching recognition candidate whose matching dictionary information matches the area representative dictionary information is determined by the matching recognition candidate. While maintaining the collation order, the modification is performed so that the collation dictionary information and the area representative dictionary information become higher than the collation order of other recognition candidates that do not match.

(Operation) Next, a character recognition method according to this embodiment will be described.

Here, an example of recognizing characters described in the form of the transfer request form shown in FIG. 2 will be described. Also,
FIG. 3 shows the format information of the form shown in FIG. 2, that is, information indicating the position and range of each column. Here, the format information of the first to fifth areas is shown. The first area corresponds to the first row of bank names ("rough banks") on the left side of the transfer request form shown in FIG. The second area corresponds to the branch name (“Hachioji branch”) in the second row on the left side of the transfer request form. The third area corresponds to the column of the company name (“X Electric Co., Ltd.”) in the third row on the left side of the transfer request form. The fourth area corresponds to the name (“Okitaro”) in the fourth row on the left side of the transfer request form. The fifth area corresponds to the first-number (“10000”) column on the right side of the transfer request form.

In performing the character recognition of this form, first, an image of this form is taken in by the scanner unit 12. Then, the scanner unit 12 sends the image data of the read image to the image storage unit 14.

Next, the image storage unit 14 stores the sent image data. Then, the image storage unit 14 notifies the character extracting unit 16 that the new image data has been saved.

Next, the character extracting section 16 receiving this information
Reads out image data stored in the image storage unit 14 and cuts out individual character images.

In extracting a character image, the character extracting section 16 creates position and size information indicating the position and size of the extracted character image in the entire image.

When extracting a character image, the character extracting unit 16 first converts the image data into a character area of the first to fifth areas and a non-character area other than the area based on the format information described above. Divide by unit. When segmenting on a region basis, character region information indicating the correspondence between each character region and image data contained in the character region is created.

FIG. 4 shows an example of image data divided on a region basis. In the first row of FIG. 4, as the area image 1, image data of the type “uneven bank” is shown. Also, 2
The data image of the type “Hachioji branch” is shown as the area image 2 in the lower row. In the third row, as area image 3,
The data image of type "X Denki Co., Ltd." is shown. Also,
On the fourth row, a handwritten data image of “Taro Oki” is shown as the area image 4. In the fifth row, as an area image 5,
9 shows a handwritten “10000” data image.

FIG. 5 shows an example of character area information.
It shows that the first image in the first row in FIG. 5 includes the area image 1. In addition, it indicates that the area image 2 is included in the second area in the second row. In addition, it indicates that the region image 3 is included in the third region in the third row. In addition, it indicates that the area image 4 is included in the fourth area in the fourth row. In addition, it indicates that the fifth area in the fifth row includes the area image 5.

Next, the character extracting section 16 divides each character area into a character string image area (hereinafter also referred to as a unit character string) and cuts out individual character images from this character string.

In dividing the character area, the character extracting section 16
Calculates the distribution of black pixels in each area image. Here, since the image is written horizontally, the distribution of black pixels in the vertical direction (y-axis direction) is obtained. Then, a portion where the number of black pixels exceeds the threshold is divided as a character string image. When dividing the character string pixel region, character region information indicating the correspondence between the region, each character string pixel region, and the image data included in the character string pixel region is created.

Here, FIG. 6 shows the distribution of black pixels of the “uneven bank” in the area image 1. The horizontal axis in FIG. 6 represents the number of black points, and the vertical axis represents the y coordinate of the area image. The distribution of black pixels is represented by a curve indicated by I in FIG. Also, in FIG.
The arrow indicated by II indicates the range of the character string image area at the y coordinate.

The left side of FIG. 7 shows the character string image 1 of the area image 1, and the right side of FIG. 7 shows an example of the character area information of the character string image 1. FIG. 7 shows that the character string image 1 is included in the character string 1 (character string image area 1) in the first area. Note that the character string image region corresponds to a portion sandwiched between lines circumscribed above and below the character string image in the region image.

Next, the character extracting section 16 extracts individual character images from the character string image.

In dividing the character string image, the character extracting unit 1
6 finds the distribution of black pixels in each area image. here,
Since the image is a horizontally written character string image, the distribution of black pixels in the horizontal direction (x-axis direction) is obtained. Then, a portion where the number of black pixels exceeds the threshold is divided as a character string image. Further, the character extracting section 16 assigns a unique character image identifier to each of the extracted character images. In addition, character extraction section 1
Reference numeral 6 creates character region information indicating the correspondence between the region, the character string image region, and the image data of the character image.

Here, FIG. 8 shows the distribution of black pixels of the “uneven bank” of the character string image 1. The horizontal axis in FIG. 8 represents the x coordinate of the character string image, and the vertical axis represents the number of black points. And the curve
III represents the distribution of black pixels. An arrow IV indicates the range of the character image area at the x coordinate.

The character images 1 to 4 of the character string image 1 are shown on the left side of FIG. In FIG. 9, the character image area 1 includes the “convex” character image 1. A character image identifier of “character 1” is assigned to the character image 1. Further, the character image area 2 includes the “concave” character image 2. This character image 2 is given a character image identifier of “character 2”. In the character image area 3, the character image 3 of "silver"
It is included. The character image 3 is given a character image identifier of “character 3”. In the character image area 4, "line"
Character image 4 is included. This character image area 4 is provided with a character image identifier of “character 4”.

The right side of FIG. 9 shows an example of character area information of each character image. In FIG. 9, the character image 1 is included in the character image area 1 of the character string 1 (character string image area 1) in the first area. The character image 2 is included in the character image area 2 of the character string 1 in the first area. The character 3 (character image area 3) of the character string 1 in the first area includes the character image 3. The character image 4 is included in the character 4 (character image area 4) of the character string 1 in the first area.

The character extracting section 16 extracts individual character images, creates character area information, and then totals the number of character images included in each character area to collect the number of character images in each character area. (Also referred to as an aggregation table).
The tabulation table is a table indicating how many character image areas each character area is divided into.

In extracting a character image, the character extracting section 16 creates position and size information indicating the position and size of each of the extracted character images in the entire image.

Then, the character extracting section 16 sends the character image identifier and the position size information of each character image to the image storage section 12 in association with each other. The image storage unit 12 stores the sent character image identifier and positional size information. The character cutout unit 16 also sends the character image identifier to the feature extraction unit 18.

Next, the feature extracting unit 18 sends the sent character image identifier to the image storage unit 14. Image storage unit 1
4 sends the character image indicated by the position / size information corresponding to the character image identifier to the feature extracting unit 18. Then, the feature extracting unit 18 extracts the features of the character image. In extracting a feature, the feature extracting unit 18 creates a feature vector.

Then, the feature extracting unit 18 sends the extracted feature vector of the character image together with the character image identifier of the character image to the distance calculating unit 20 as a collating unit.

Next, the distance calculation unit 20 includes the feature extraction unit 18
Every time a feature vector of a character image is sent from the device, the feature vector of the character image is compared with the feature vector of the standard character pattern.

In matching, the distance calculation unit 20 calculates the distance between the feature vector of the character image and the feature vector of the standard character pattern. As a result of the distance calculation, the distance calculation unit 20
A plurality of standard character patterns for a character image are selected as recognition candidates with matching order. And the distance calculation unit 2
0 makes the recognition candidate having the highest collation rank among the recognition candidates the recognition result.

Further, the distance calculation section 20 generates collation dictionary information indicating the type of recognition dictionary in which the standard character pattern adopted as the recognition candidate is stored.

Here, FIG. 10A shows an example of a recognition candidate. FIG. 10 shows an example of a recognition candidate for a character image to which a character image identifier of “character 1” is assigned. In the case of this character image, the character code of recognition candidate 1 is “2330”. The character code of recognition candidate 2 is “2339”. The character code of recognition candidate 3 is “2330”
It is. The character code of recognition candidate 4 is “2338”
It is.

FIG. 10B shows an example of collation dictionary information. The matching dictionary information of the recognition candidate 1 of the character image of “character 1” is “dictionary 1”. The matching dictionary information of the recognition candidate 2 is “dictionary 2”. The matching dictionary information of the recognition candidate 3 is “dictionary 2”. The matching dictionary information of the recognition candidate 4 is “dictionary 1”.

Then, the distance calculation unit 20 uses the character code “2330” of the recognition candidate 1 and the collation dictionary information “dictionary 1” of the recognition candidate 1 as the recognition result and the collation dictionary information, respectively, as the single character area determination unit 22. Send to

In the single character area determining section 22, the character area information of the character area sent from the character extracting section 16 and the distance calculating section 2 for each character image included in the character area are included.
Based on the collation dictionary information of the recognition result sent from 0, it is determined whether or not each character area is a single character area in which the type of character included in the character area is single.

Prior to the determination, the single character area determination unit 22
First, the character image belonging to the character area is specified based on the character area information. Therefore, the single character area determination unit 2
2 counts the number of times the matching dictionary information of the character image belonging to each character area has been sent. Then, each time the single character area determination unit 22 counts the number of times, the single character area determination unit 22 reads out the number of characters of the counted character area from the aggregation table of the character extraction unit 16. When the number of characters in the character area is equal to the number of times the matching dictionary information of the character area is sent,
The single character area determination unit 22 starts the determination.

Here, FIG. 11A shows an example of character area information of a character area to be determined. FIG. 11A shows character area information on characters 1 to 5 of the character string 1 in the area 1. In FIG. 11A, since the number of characters in the area 1 is five, all the characters in the character area 1 have been sent when the number of times in the tally table is counted five.

Next, in the determination, the single character area determination unit 22 first determines the type of the representative dictionary from the type of the dictionary indicated by the collation dictionary information for each recognition candidate of the character image for each character image. I do.

Here, FIG. 11B shows an example of the collation dictionary information of the character area to be determined. FIG. 11B shows collation dictionary information on the character images identified by the character information identifiers of the characters 1 to 4, in the order of collation. For character 1, “dictionary 1”, “dictionary 2”, “dictionary 2”, and “dictionary 1” are shown as collation dictionary information in the order of collation order. For character 2, “dictionary 2”, “dictionary 1”, “dictionary 2”, and “dictionary 1” are shown in the order of collation order as collation dictionary information. As for character 3, “dictionary 1”, “dictionary 2”, “dictionary 2”, and “dictionary 1” are shown as collation dictionary information in the order of collation order. For character 4, “dictionary 1”, “dictionary 2”, “dictionary 2”, and “dictionary 1” are shown as collation dictionary information in the order of collation order. For character 5, “dictionary 1”, “dictionary 2”, “dictionary 2”, and “dictionary 1” are shown in the order of collation order as collation dictionary information.

Then, the type of dictionary indicated by the collation dictionary information of the recognition candidate having the first collation order of each character image is set as the type of the representative dictionary of the character image. Therefore, in the example shown in FIG. 11B, the type of the representative dictionary of the character 1 is “dictionary 1”, the type of the representative dictionary of the character 2 is “dictionary 2”, and the type of the representative dictionary of the character 3 is “dictionary 2”. , The representative dictionary type of character 4 is “dictionary 1”, and the representative dictionary type of character 5 is “dictionary 1”. Note that the criterion of the representative dictionary is not limited to the criterion of this embodiment. For example, for each character image, among the types of dictionaries shown as the matching dictionary information of the recognition candidates,
The largest number of dictionaries may be used as the representative dictionaries.

Next, the single character area determination section 22 determines whether or not the character area is a single character area based on the type of each representative dictionary of each character image included in the character area (area 1). I do.

For this purpose, the single character area determination unit 22
The number of each type of the representative dictionary of each character image belonging to each character area is compared. If there is no more than two types of representative dictionaries, it is determined that the character area is a single character area. The criterion for determining a single character area is not limited to the criterion of this embodiment.

In the example shown in FIG. 11B, "dictionary 1"
Since there are four representative dictionaries and one representative dictionary of "dictionary 2," region 1 is determined to be a single character region. The characters in the single character area correspond to, for example, only handwritten characters or only printed characters.

When there are a plurality of types of the most representative dictionaries in one character area, it is determined that the character area is not a single character area. This case corresponds to, for example, a case where handwritten characters and printed characters are mixed in one character area.

Further, when the character area is a single character area, the single character area determination section 22 generates area representative dictionary information indicating the type of dictionary corresponding to the type of characters included in the single character area. In the example shown in FIG.
Area representative dictionary information indicating “dictionary 1” is generated.

Then, the single character area determination section 22 sends the determination result and the area representative dictionary information indicating “dictionary 1” to the correction section 24.

The correction unit 24 determines the collation order of the matching recognition candidates whose matching dictionary information matches the area representative dictionary information among the recognition candidates of each character image included in the single character area. While maintaining the order, the matching dictionary information and the area representative dictionary information are corrected so as to be higher than the matching order of other recognition candidates that do not match. It should be noted that the collation order may be corrected at least for characters whose collation dictionary information of the recognition candidate having the first collation rank does not match the area representative dictionary information.

FIGS. 12A to 12C show examples of correcting the collation order of the recognition candidate for character 2 in the character image included in the single character area.

First, FIG. 12A shows the collation dictionary information of each recognition candidate of the character image to which the character image identifier of “character 2” is added. The matching dictionary information of the recognition candidate 1 of the character image of “character 2” is “dictionary 2”. The matching dictionary information of the recognition candidate 2 is “dictionary 1”. The matching dictionary information of the recognition candidate 3 is “dictionary 2”. The matching dictionary information of the recognition candidate 4 is “dictionary 1”.

Next, FIG. 12B shows recognition candidates for a character image to which a character image identifier of "character 2" has been added.
In the case of the character image of “character 2”, the character code of recognition candidate 1 is “2331”. The character code of recognition candidate 2 is “2332”. The character code of recognition candidate 3 is “2333”. The character code of recognition candidate 4 is “2334”. In FIG. 12,
A dictionary and a character code of a recognition candidate based on “Dictionary 1” which is a type of the area representative dictionary are indicated by thick frames.

As shown in FIGS. 12A and 12B, at the stage before correction, the dictionary 2 indicated by the collation dictionary information of the recognition candidate 1 having the first collation rank matches the type of the area representative dictionary. Not. Therefore, the character code “2332” of the recognition candidate 2 and the character code “2334” of the recognition candidate 4, each of which has the dictionary 1 as the collation dictionary information, are ranked higher in the collation order without changing the collation order. Modify as follows.

Here, FIG. 12C shows the recognition candidates after the correction of “character 2”. After the correction, the first place of the collation order is the character code “2332” of “dictionary 1”, and the second place of the collation order is the character code “2334” of “dictionary 1”.

Accordingly, the character code "2
332 ”is the recognition result of the corrected“ character 2 ”.

Here, assuming that “dictionary 1” is a dictionary for handwritten characters and “dictionary 2” is a dictionary for printed characters, before correction, the recognition candidate 1 for “character 2” of handwritten characters is The character code was "2331" stored in the dictionary. The misrecognition rate of recognition results obtained using a dictionary of a type different from the type of character pattern is generally higher than that of recognition results obtained using a dictionary of the same type as the type of character pattern. Therefore, when the recognition candidate 1 of “character 2” is used as the recognition result, the probability of erroneous recognition increases.

On the other hand, after the correction, the recognition candidate with the highest matching order of the handwritten character “character 2” is the character with the highest matching order among the recognition candidates stored in the handwriting dictionary. The code was "2332". Therefore, by correcting the collation order, it is possible to reduce the erroneous recognition rate compared to before the correction.

(Modification) In the character recognition method and the character recognition device according to the present invention, the method for determining a single character area and the single character area determination unit are not limited to the method and the apparatus according to the above-described embodiment. Absent.

For example, in determining whether or not a character area is a single character area, the matching dictionary information is used for each recognition candidate of each character image specified to belong to the character area based on the character area information. May be determined based on the type of dictionary indicated by.

Further, when determining whether or not the character area is a single character area, the single character area determination unit determines whether the character image belongs to the character area based on the character area information. A single character area determination unit that determines whether the character area is a single character area based on the type of dictionary indicated by the collation dictionary information for each recognition candidate may be used.

More specifically, for example, for all of the character images belonging to the same character area or for the recognition candidates from the top to a certain order, the number of dictionaries indicated by all the collation dictionary information is compared for each type. Then, when there is no more than two types of dictionaries included in the largest number, the character area may be determined to be a single character area.

Next, it is determined whether or not “area 1” shown in FIGS. 11A and 11B is a single character area by this determination method. For the top four recognition candidates for the character images of "character 1" to "character 5" shown in FIG. 11B, the numbers for each type of dictionary indicated by all the collation dictionary information are totaled and compared. . As a result, "dictionary 1" is 1
0 times and “dictionary 2” also 10 times. Accordingly, since there are two types of dictionaries included in the largest number, it is determined that “area 1” is not a single character area in this determination method.

In the above-described embodiment, only examples in which these inventions are configured under specific conditions have been described. However, these inventions can be subjected to many changes and modifications.
For example, in the above-described embodiment, two types of dictionaries, a handwritten dictionary and a type dictionary, are used as the dictionaries. However, the types of dictionaries need not be limited to these two types. For example, a dictionary for Mincho and a dictionary for Gothic may be used as the type dictionary. Also, the type of dictionary is not limited.

It is not necessary to limit the method of creating a dictionary standard pattern to the method used in this embodiment. For example, the feature extraction unit may create a feature vector using the feature extraction method used, or a feature vector obtained from one character image may be used as a standard pattern as it is.

Further, in the above-described embodiment, an example of a one-line character area has been described. However, the number of lines in the character area is not limited to one line, and the present invention is also applicable to a character area including a plurality of lines. can do.

[0083]

According to the character recognition method of the first invention and the character recognition device of the second invention, when a character area is determined to be a single character area in which a single type of character is described, a single character For each character in the area, the collation order of the recognition candidates recognized by using the dictionary of the same type as the type of the character is 1
The collation order can be modified so as to be ranked. As a result, it is possible to match the type of the character with the type of the dictionary used for the recognition result. For this reason, the false recognition rate can be reduced.

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining a configuration of a character recognition device according to an embodiment;

FIG. 2 is an example of a form to be recognized in the embodiment.

FIG. 3 is an example of format information according to the embodiment;

FIG. 4 is an example of image data divided on a region basis according to the embodiment;

FIG. 5 is an example of character area information in the embodiment.

FIG. 6 is a distribution of a region image and its black pixels in the embodiment.

FIG. 7 is an example of a character string image and character area information according to the embodiment.

FIG. 8 is an example of a character string image and a distribution of black pixels thereof in the embodiment.

FIG. 9 is an example of a character image and character area information according to the embodiment.

FIG. 10A is an example of a recognition candidate according to the embodiment, and FIG. 10B is an example of collation dictionary information.

FIG. 11A is an example of character area information in the embodiment, and FIG. 11B is an example of collation dictionary information.

FIGS. 12A to 12C are examples of correcting the collation order;

[Explanation of symbols]

 10: Dictionary unit 10a: Dictionary for handwritten characters 10b: Dictionary for printed characters 12: Scanner unit 14: Image storage unit 16: Character extraction unit 18: Feature extraction unit 20: Distance calculation unit 22: Single character area determination unit 24 : Correction department

Claims (6)

[Claims] 1. A character image is cut out from an image captured from a medium in which characters are described, and features of each character image are extracted. The features of the character image are stored in a plurality of types of dictionaries. Is compared with the features of the standard character pattern for character recognition, a plurality of standard patterns for the character image are selected as recognition candidates with a collating order, and the most significant collating order among the recognition candidates is selected. In a character recognition method that uses a high recognition candidate as a recognition result, the dictionary in which character region information representing a character region in which the character image is included is generated, and the recognition result or a standard character pattern of the recognition candidate is stored. Based on the character area information and the collation dictionary information,
It is determined whether each of the character regions is a single character region in which the type of characters included in the character region is single, and if the character region is the single character region, the single character region is determined. Generating region representative dictionary information indicating the type of dictionary corresponding to the type of character included in the character string, and among the recognition candidates for each character image belonging to the single character region, the collation dictionary information and the region representative dictionary information match. The matching order of the matching recognition candidates is corrected to be higher than the matching order of other recognition candidates whose matching dictionary information and the area representative dictionary information do not match while maintaining the matching order of the matching recognition candidates. A character recognition method. 2. The character recognition method according to claim 1, wherein in determining whether the character area is the single character area, it is specified that the character area belongs to the character area based on the character area information. For each character image, the type of the representative dictionary is determined from the types of dictionaries indicated by the matching dictionary information for each recognition candidate of the character image, and the type of each representative dictionary of each character image included in the character area And determining whether the character area is the single character area based on the following. 3. The character recognition method according to claim 1, wherein a dictionary for handwritten characters and a dictionary for printed characters are used as the dictionary. 4. A character area is extracted from a plurality of types of dictionaries each storing a standard pattern for character recognition, and an image captured from a medium in which characters are described, and individual character images are extracted from the character area. A character extracting unit that extracts a character, a feature extracting unit that extracts a characteristic of each character image, and a plurality of standard character patterns corresponding to the character image by comparing the characteristic of the character image with the characteristic of the standard character pattern. Is selected as a recognition candidate with matching order, and
A character recognition device comprising: a recognition unit that uses a recognition candidate having the highest matching order among the recognition candidates as a recognition result; and a character region indicating a character region containing the character image as the character cutout unit. A character extracting unit that generates information; a collating unit that generates collation dictionary information indicating a type of the dictionary in which the standard character pattern of the recognition candidate is stored; Based on the collation dictionary information,
It is determined whether each of the character regions is a single character region in which the type of characters included in the character region is single, and if the character region is the single character region, the single character region is determined. A single character area determination unit that generates area representative dictionary information indicating the type of dictionary corresponding to the type of character included in the recognition dictionary information and the collation dictionary information among recognition candidates for each character image belonging to the single character area. The matching order of the matching recognition candidates that match the area representative dictionary information is compared with the matching order of other recognition candidates whose matching dictionary information does not match the area representative dictionary information while maintaining the matching order of the matching recognition candidates. A character recognition device comprising a correction unit for correcting the height to be higher than the height. 5. The character recognition device according to claim 4, wherein the single character area determination unit includes: for each character image identified as belonging to the character area based on the character area information, The type of the representative dictionary is determined from the types of dictionaries indicated by the collation dictionary information for the recognition candidates, and the character area is changed to the single character based on the type of each representative dictionary of each character image included in the character area. A character recognition device, comprising: a single character area determination unit that determines whether the area is an area. 6. The character recognition device according to claim 4, wherein the dictionary includes a dictionary for handwritten characters and a dictionary for printed characters.