JP3335494B2 - Character recognition processor - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition processing apparatus and, more particularly, to a character recognition processing apparatus for automatically and cooperatively performing knowledge processing in character recognition processing of kana characters and kanji written on a form. .

[0002]

2. Description of the Related Art FIG. 6 is a diagram for explaining the background of the present invention.
1 shows an example of a general form used in a character recognition processing device such as an R (optical character reading device).

[0006] Form 1 in FIG. 6 includes a member number,
Application date, postal code, telephone number, address (kana), address (kanji), name (kana), name (kanji), date of birth, occupation,
Age and gender are respectively entered in the designated fields (fields).

[0004] Of these, only the member number is a print field in which print characters are preliminarily written, and the other is a handwritten field in which a form writer writes in handwriting characters. Of the handwritten fields, the application date, zip code, telephone number, date of birth (the latter part), occupation and age are numeric fields that are entered numerically, and the date of birth (the first part),
Gender is a mark field that is filled in by a method of adding a mark (circle) to the pre-filled print characters. Each of the address kana and name kana fields is a kana field filled with kana characters. Each field of the full name kanji is a kanji field filled in with kanji.

[0005] The character recognition processing device reads characters written on a form as shown in FIG. 6 and performs a character recognition process. In this character recognition process, in order to improve the character recognition rate (the ratio of correctly reading the entered characters), for the address and name, a character string once read and recognized from each of the kana and kanji fields (candidate character string ), Predetermined knowledge processing (single knowledge processing) is performed. The independent knowledge process is performed for each field, and is a process of correcting a candidate character string using an address dictionary and a name dictionary prepared in advance. For example, a candidate character string read and recognized from an address kana field is collated with an address kana dictionary, and even if the candidate character string contains an error of two or three characters, this is corrected to a correct character string and output. I do.

Further, as one of the knowledge processes, a cooperative process (cooperative knowledge process) between a kana field and a kanji field for an address and a name is performed. This cooperative knowledge process is a knowledge process that not only performs knowledge processing independently on candidate character strings read from a kana field and a kanji field, but also cooperates two previously specified fields.

In the above-mentioned form 1, for example, "address"
There are two fields for entering a field, a kana field and a kanji field. Thus, for example, the candidate character string obtained from the address kanji field was corrected using the candidate character string obtained from the address kana field, and / or the candidate character string obtained from the address kana field was obtained from the address kana field using the address kanji field. Correct the candidate character string. Thereby, the character recognition rate in the character recognition processing can be increased.

[0008]

According to the above-mentioned prior art, the kana field and the kanji field, which have many types of characters to be entered and are relatively difficult to recognize because they are handwritten, are processed by the cooperative knowledge processing. The character recognition rate can be improved.

However, in the above-mentioned prior art, it is necessary for the operator to specify in advance the kana field and the kanji field to be subjected to the cooperative knowledge processing in the recognition information definition processing.

For example, when performing the character recognition processing on the form shown in FIG. 6, there are two parts where the cooperative knowledge processing should be performed. That is, there are an address kana field and an address kanji field, and a name kana field and a name kanji field. In this case, prior to the character recognition processing, the operator inputs instructing to link the address kana field and the address kanji field in the recognition information definition processing,
An input for instructing that the name kana field and the name kanji field are linked is performed.

The instruction input of the cooperative knowledge processing is performed in the recognition information definition processing as follows. For example, the operator designates each field as an area to be recognized (recognition field), thereafter inputs the attribute of each field, and assigns an item number to each field. Then, for a certain field, if there is a field to be linked with this field, the field to be linked is specified using the item number, and then an instruction to link these fields is input.

As described above, according to the above-described prior art, the operator needs to perform an input for instructing the field to be linked the number of times in the form where the link knowledge processing is to be performed. Therefore, as the number of parts to perform the cooperative knowledge processing increases, the burden on the operator increases accordingly. Further, it is necessary to input this instruction for each type of the form 1, which is a burden on the operator. Further, since the number of times that the operator operates the character recognition processing device is large, the speed of the character recognition processing device itself is increasing, but the efficiency of the character recognition processing cannot be improved. Further, since the input is made by the operator, there is a possibility that an error in inputting the instruction of the cooperative knowledge processing or the like may occur.

An object of the present invention is to provide a character recognition processing apparatus for automatically recognizing fields to be subjected to cooperative knowledge processing and cooperating them. Also, the present invention
It is an object of the present invention to provide a character recognition processing device for automatically recognizing a handwritten kana field and a handwritten kanji field for which cooperative knowledge processing is to be performed and linking them.

[0014]

FIG. 1 is a block diagram showing the principle of the present invention, and shows the configuration of a character recognition processing apparatus according to the present invention. This character recognition processing device reads a character entered in each field of a form 1 having a plurality of predetermined fields, performs a character recognition process on each field, and obtains a character recognition output 6. It is.

This character recognition processing device includes a single knowledge processing unit 4, a cooperative knowledge processing unit 5, and an automatic cooperative processing unit 3.
The independent knowledge processing unit 4 performs the knowledge processing in the character recognition processing for a predetermined field in the field alone. The cooperative knowledge processing unit 5 performs the knowledge processing in the character recognition processing for a predetermined field by cooperating the field and another field associated therewith. The automatic cooperation processing unit 3 determines, from among a plurality of fields,
A plurality of fields whose positions on the form 1 have a predetermined relationship are searched for and linked together. At this time, in particular, the automatic cooperation processing unit 3 makes the handwritten kana field and the handwritten kanji field cooperate. For a plurality of fields linked by the automatic link processing unit 3, the link knowledge processing unit 5 links these to perform the knowledge processing.

[0016]

The kana field and the kanji field to be linked in the knowledge processing in the character recognition processing, in particular, the handwritten kana field and the handwritten kanji field have a predetermined relationship on the form 1 in most cases. For example, they are provided at a very close position or an adjacent position. Therefore, using this, the automatic cooperation processing unit 3
Automatically retrieves a plurality of fields having a predetermined relationship on the form 1, in particular, a handwritten kana field and a handwritten kanji field, and associates them.

Therefore, according to the character recognition processing device of the present invention, it is possible to automatically associate a plurality of fields having a corresponding relationship on the form 1, particularly a handwritten kana field and a handwritten kanji field. This allows
In most forms 1, a plurality of predetermined fields, in particular, a handwritten kana field and a handwritten kanji field can be linked, and linked knowledge processing can be performed on these fields.

This eliminates the need to specify the kana field and the kanji field to be subjected to the cooperative knowledge processing in the recognition information definition processing. As a result, when performing the character recognition processing, the operator does not need to input an instruction for the cooperative knowledge processing, the burden on the operator can be extremely reduced, and the occurrence of an error in the input of the instruction for the cooperative knowledge processing can be prevented. it can. Further, even when the type of the form 1 is different, it is not necessary to input the instruction of the cooperative knowledge processing every time. In addition, since the number of operations by the operator can be reduced, the efficiency of the character recognition processing can be improved.

[0019]

FIG. 2 is a block diagram of an embodiment, showing a character recognition processing apparatus. The character recognition processing apparatus according to the present embodiment includes a reading processing unit 7 that reads a character written on a form 1, performs a character recognition process on a result read by the reading processing unit 7, and performs a candidate character string 9.
And a knowledge processing unit 11 that performs a knowledge process on the candidate character string 9 to obtain a character recognition output 6. The reading processing unit 7, the character recognition processing unit 8, and the knowledge processing unit 11 are provided in the processing device 2. The processing device 2 includes a CPU (central processing unit) of the character recognition processing device,
It consists of each processing program on the memory of the character recognition processing device. The processing programs include, for example, a reading processing program forming the reading processing section 7, a character recognition processing program forming the character recognition processing section 8, a knowledge processing program forming the knowledge processing section 11, and the like.

When the form 1 is input from a predetermined input device (not shown), the reading processing section 7 reads the printed characters, handwritten characters, and marks entered therein. The form 1 has a plurality of predetermined fields, and characters and the like are entered in each of the fields. The read processing unit 7 reads characters and the like written in each field as the contents of each field. The reading processing unit 7 includes a plurality of forms 1
When the reading of the document is completed, the result of the reading is stored in a predetermined storage area for each form 1, and then the character recognition processing unit 8 is started.

Here, as the form 1, for example, the form shown in FIG. 6 is used. Therefore, this form 1 includes the member number, application date, postal code, telephone number, address (kana), address (kanji), name (kana), name (kanji), date of birth, occupation, year The order and gender are respectively entered in the fields designated in advance.

As described above, only the member number is a print field, and the others are handwritten fields. In the handwritten field, application date, zip code, phone number, date of birth,
The fields of occupation and age are numeric fields, the fields of date of birth and gender are mark fields, the fields of address kana and name kana are kana fields, and the fields of address kanji and name kanji are kanji fields It is. The application date, postal code, address, name, etc. are attributes of the data (that is, knowledge) entered in the respective fields.

When activated by the reading processing unit 7, the character recognizing processing unit 8 performs a character recognizing process on the reading results of the plurality of forms 1 stored in a predetermined storage area, and obtains a candidate character recognizing result. Get character string 9. The candidate character string 9 is stored in a predetermined file in a predetermined format. The candidate character string 9 is obtained for each form 1. The candidate character string 9 for one form 1 (one candidate character string 9) is composed of candidate character strings for each field of one form 1. When the character recognition processing unit 8 obtains the candidate character string 9 by character recognition on the form 1, the character recognition processing unit 8 activates the knowledge processing unit 11 to perform the knowledge processing on the candidate character string 9. Note that the knowledge processing unit 11 may be included in the character recognition processing unit 8.

When activated by the character recognition processing unit 8, the knowledge processing unit 11 performs knowledge processing on the candidate character string 9 and obtains a character recognition output 6. This knowledge processing is performed for each candidate character string 9, and the candidate character string for each field in one candidate character string 9 is used as a processing unit.

The knowledge processing unit 11 includes a kana-kanji automatic cooperation processing unit 3A for performing preprocessing of knowledge processing, a single knowledge processing unit 4 and a cooperative knowledge processing unit 5 for performing knowledge processing according to the result of the preprocessing, and mainly a single knowledge processing unit. 4 and attribute data 10 and a dictionary 14 used by the cooperative knowledge processing unit 5 for knowledge processing.

The kana-kanji automatic cooperation processing unit 3A corresponds to the automatic cooperation processing unit 3 in FIG. 1, and as a pre-processing of the knowledge processing, a candidate character string for each field is subjected to knowledge processing or independent knowledge processing. It is determined whether the target is a target of the cooperative knowledge processing.

First, the kana-kanji automatic cooperation processing unit 3A determines whether or not a candidate character string for each field of the candidate character string 9 is to be subjected to knowledge processing. In this determination process, the kana-kanji automatic cooperation processing unit 3A refers to the attribute data 10 which is originally data used by the single knowledge processing unit 4 and the cooperation knowledge processing unit 5 for knowledge processing.

The attribute data 10 is prepared in advance for each type of the form 1. That is, as shown by the dotted line in the figure, the type of the form 1 and the attribute data 10 correspond one to one. Attribute data 1
0 stores the attribute of each field in one type of form 1. The attributes to be stored include the type of handwriting / printing, which is an attribute determined by the entry method, the type of kana / kanji / numerical / alphabet / mixed, which is an attribute determined by the type of character to be entered, and There are position information, which is an attribute determined by the position, and necessity of knowledge processing, which is an attribute determined by whether knowledge processing is required for the field.

For example, for the form 1 shown in FIG. 6, the attribute of the member number field is set to "print, numeral, position information, and knowledge processing unnecessary". Application date, postal code, phone number,
The attributes of the fields of the date of birth (first half), occupation, and age are "handwritten, numbers, location information, and no knowledge processing required". The attributes of each field of the date of birth (the latter half) and gender are "handwritten, mark, position information, no knowledge processing required". The attributes of the address kana and name kana fields are "handwritten, kana, positional information, knowledge processing required". The attributes of the address kanji and full name kanji fields are "handwritten, kanji,
Location information and knowledge processing are required. "

In addition to the above, an attribute (knowledge attribute) determined by the item number of each field (for example, the item number of the member number field is "1") and the contents of data (knowledge) written in each field ) "Member number"
Is described as an attribute of each field.

Such attribute data 10 is used as an input / output device (not shown) provided in the character recognition processing device in the recognition information defining process executed prior to the start of the character recognition process.
Is input in advance by the operator and stored in a predetermined file. For example, when an operator reads a blank form 1, the character recognition processing device displays the form 1 on a display screen of the input / output device. Then, the operator defines each field from this screen using a mouse or the like, and inputs each attribute (excluding position information) of each defined field from the input / output device. The creation and management of such attribute data 10 is performed by a predetermined processing program.

The position information in the attributes is automatically set by the processing program. That is, when the operator defines each field, the X coordinate (position) and the Y coordinate (position) of the point input on the screen are directly used as the position information of the field.

This position information will be described with reference to FIG. FIG. 3 shows an enlarged part of the form 1 shown in FIG. Position information on the form 1 is determined by an X coordinate (position) and a Y coordinate (position) with reference to the mark 1A. In the figure,
The horizontal direction is defined as the X direction, and the vertical direction (search direction) is defined as the Y direction. When the mark 1A is not printed, the mark 1A is determined based on the end of the form 1. That is, the position of each field on the form 1 is X, which is the start X position of the mark 1A.
The distance from _S, is represented by the distance from Y _S is the starting Y position of the mark 1A.

[0034] For example, in FIG. 3, the starting X location address kana field is a field of 5 th from the head of the form 1 (item number 5) is X _51, starting Y position Y ₅₁
, And the termination Y position is Y _52. Note that the end X position X
₅₂ (not shown) is similarly determined. The same applies to the other fields. The start X position and the start Y position of the position information indicate the position where the field is started when the document 1 is searched in the X direction and the Y direction, respectively.

In a state where the form 1 as shown in FIG. 3 is displayed on the display screen of the input / output device, the operator operates the upper left corner 1B and the lower right corner (diagonal of the upper left corner 1B, not shown) of the field. Position the mouse on and specify the address kana field. At this time, the processing program sets the start X position, start Y position, end X position, and end Y position of the address kana field as attribute information in the attribute data 10.
Write to. This is done for each field. Thereby, the position information of each field is automatically set by the processing program.

With reference to the attribute data 10, the automatic kana-kanji processing unit 3 A knows, among the fields of the form 1, the fields which are handwritten and in which kana or kanji are entered. Target for processing. That is, since the handwritten kana field and the handwritten kanji field have many types of characters to be entered and the character shape depends on the personality of the writer, it is difficult to increase the character recognition rate only by the character recognition process. The candidate character string 9 is corrected. On the other hand, the kana-kanji automatic cooperation processing unit 3A excludes the type field and the number field from the target of the knowledge processing.

For example, in the form 1 shown in FIG. 6, the attribute of the member number field is “print”, so that the member number field is excluded from the subject of the knowledge processing, and the application date, postal code, telephone number,
The fields of date of birth, occupation, age, and gender are also excluded because the attribute is “number”. On the other hand, the fields of address kana and name kana, which are handwritten kana fields, and the fields of address kanji and name kanji, which are handwritten kanji fields, are subjected to knowledge processing.

Next, the kana-kanji automatic cooperation processing unit 3A determines whether the candidate character string for each field determined to be subjected to the knowledge processing is to be subjected to the single knowledge processing or the cooperation knowledge processing. That is, it is determined whether or not a field on the form 1 is linked with another field. Fields that are not targeted for cooperative knowledge processing are targeted for single knowledge processing.

The field to be subjected to the cooperative knowledge processing is a field which is handwritten and in which kana or kanji is entered. For example, in the form 1 shown in FIG. 6, the address kana field and the address kanji field, and the name kana field and the name kanji field are objects to be linked with each other.

In the form 1 shown in FIG. 6, the fields to be subjected to the knowledge processing are all the fields to be subjected to the cooperative knowledge processing. In some cases, fields that should not be included include fields that should not be subjected to the cooperative knowledge processing.

Here, as can be seen from FIGS. 6 and 3,
The field to be subjected to the cooperative knowledge processing has a predetermined position on the form 1, and is located at a very close position or adjacent. Therefore, the kana-kanji automatic cooperation processing unit 3A searches a plurality of fields on the form 1 for a kana field and a kanji field whose positions on the form 1 have a predetermined relationship. In other words, the search is performed using "the position on the form 1 has a predetermined relationship" as a search condition. Then, the kana field and the kanji field extracted as a result of the search are linked. In particular, the handwritten kana field and the handwritten kanji field having such a positional relationship are linked.

In order to search for a kana field and a kanji field whose positions on the form 1 are in a predetermined relationship, the kana-kanji automatic cooperation processing unit 3A includes a horizontal writing recognition field search processing unit 12 and a vertical writing recognition A field search processing unit 13 is provided.

Recognition field search processing unit 12 for horizontal writing
Searches for a field to be linked when the form 1 is written horizontally as shown in FIGS. The horizontal writing recognition field search processing unit 12 searches the horizontal writing form 1 in a downward direction (Y direction) from above, and starts each other on the form 1 from among a plurality of fields on the form 1. A kana field and a kanji field whose positions are substantially the same and whose start Y position of the lower field is very close to the field above it are searched. In other words, the search is performed using the search condition that “the start X positions of each other on the form 1 are substantially the same and the start Y position of the lower field is extremely close to the field above”.

For example, in the form 1 shown in FIG. 3, the address kana field and the address kanji field have a mutual start X
In the position X ₅₁ and X ₆₁ are the same, and, more starting Y position Y ₆₁ below the address kanji field is very close to the end Y-position Y ₅₂ Address Kana fields thereabove. The same applies to the name kana field and the name kanji field. Therefore, these are searched as fields to be linked.

Whether or not the start X positions are substantially the same is determined by determining whether the start X position of the address kana field and the start X
The determination is made based on whether or not the position difference (X ₆₁ −X ₅₁ ) is smaller than a predetermined value A. The predetermined value A is an extremely small value,
Given in advance.

[0046] Furthermore, starting Y position is whether very close position determination, the difference (Y ₆₁ -Y ₅₂₎ is a predetermined value of the starting Y position of the end Y position and address Kanji fields Address Kana Field It depends on whether it is smaller than B or not. Predetermined value B
Is given in advance. Predetermined value B is, for example, 1 one side of the value of the character in the entry region (Y ₆₂ -Y ₆₁₎ or (X ₅₂ -X _51). Note that the predetermined value A is a value sufficiently smaller than the predetermined value B.

[0047]

Recognition field search processing unit 12 for horizontal writing
Performs this search as follows: In order to search the form 1 in the Y direction, the attribute data 10 is searched using the Y position of the search target.
See When the start Y position of each field is equal to the search target Y position, this field has been searched. Searched Y position is incremented from the value Y _S. For example, when there are a plurality of fields having the same start Y position such as a member number field and an application date field, the start X position of the attribute data 10 is further referred to and a field having a smaller value (for example, a member number field) ) Is searched first.

Recognition field search processing unit 13 for vertical writing
Searches for a field to be linked when the form 1 is written vertically. The recognition field search processing unit 13 for vertical writing searches for a form for vertical writing from right to left, and finds a form 1 from a plurality of fields on the form.
The starting Y positions of each other are substantially the same, and
A kana field and a kanji field in which the starting X position of the left field is very close to the right field are searched. That is, the search is performed using the search condition that “the start Y positions of the respective fields are substantially the same on the form 1 and the start X position of the left field is extremely close to the right field”. Do.

For example, consider a form in which the form 1 of FIG. 3 or FIG. 6 is rotated by 90 degrees so that the member number field is a vertically written area at the right end of the form 1. X direction and Y
The direction is opposite to the example of FIG. 3 or FIG. In such a form, the address kana field and the address kanji field have the same start Y position, and have the same start X position.
The location is very close. The same applies to the name kana field and the name kanji field. Therefore, these are searched as fields to be linked. The search conditions and the search method are the same as those described above.

With the above processing, the kana-kanji automatic cooperation processing unit 3A links the address kana field and the address kanji field among the fields subjected to the knowledge processing, and converts the name kana field and the name kanji field. Cooperate.

The kana-kanji automatic cooperation processing section 3A performs this cooperation processing as follows. For convenience, the address kana field searched first is the linked side, and the address kanji field searched later is the linked side. Attribute data 10
In, a linking flag is set in the attribute of the address kana field on the linked side, and the item number “6” of the address kanji field on the linked side is set. The same applies to the name field.

The kana-kanji automatic cooperation processing unit 3A transfers control to the independent knowledge processing unit 4 or the cooperation knowledge processing unit 5 according to the result of the above preprocessing. The independent knowledge processing unit 4 or the cooperative knowledge processing unit 5 performs knowledge processing using the dictionary 14.

The dictionary 14 includes a name kana dictionary, a name kanji dictionary, an address kana dictionary, and an address kanji dictionary. For example, the name kana dictionary is a dictionary for knowledge processing of the name kana field, and is composed of data in which existing names and names are described in kana. The same applies to the name kanji dictionary, the address kana dictionary, and the address kanji dictionary.

The independent knowledge processing section 4 performs the knowledge processing in the character recognition processing for a predetermined field in the field alone. That is, in the kana-kanji automatic cooperation processing unit 3A, the following independent knowledge processing is performed on a field that is not linked with another field.

When the independent knowledge processing section 4 receives control from the automatic kana / kanji cooperation section 3A together with the item number of the field to be processed, it performs independent knowledge processing on the field indicated by the item number.

That is, the candidate character string obtained from the field is extracted from the candidate character string 9 using the item number. Also,
The attribute of the item number in the attribute data 10 is referred to and the attribute is known. Then, using the extracted candidate character string, the corresponding dictionary 14 is checked to determine whether or not the candidate character string exists. If the candidate character string exists, it is left as it is. Correction is performed using a column as a candidate character string.

The cooperative knowledge processing section 5 performs the knowledge processing in the character recognition processing for a predetermined field by cooperating the field with another field associated therewith. That is, the following cooperative knowledge processing is performed on the handwritten kana field and the handwritten kanji field linked by the kana / kanji automatic cooperation processing unit 3A.

Specifically, the kana-kanji automatic cooperation processing unit 3A
When the control is passed along with the item number of the field to be processed from, the cooperative knowledge processing for the field indicated by the item number is performed.

For example, when the item number is “7” indicating the name kana field, the candidate character string obtained from the name kana field is extracted from the candidate character string 9. Also, the attribute of item number “7” (the attribute of the name kana field) of the attribute data 10 is referred to, and the name and kana are known as the attributes.
Further, it is known that the cooperation flag is set as the attribute of the name kana field and that the object of the cooperation is the name kanji field of the item number “8”.

Then, using the extracted candidate character string, the name kana dictionary is checked to determine whether or not the candidate character string exists. If the candidate character string exists, it is left as it is. Correction is performed using a close character string as a candidate character string (independent knowledge processing is performed). For example, suppose that the result is “Tanaka”.

Next, a candidate character string obtained from the name / kanji field to be linked is extracted from the candidate character string 9 and its attribute is known from the attribute data 10. Then, using the extracted candidate character string, the name / kanji dictionary is checked to determine whether or not the candidate character string exists. If the candidate character string exists, it is left as it is. Correction is performed using a column as a candidate character string (independent knowledge processing is performed). For example, assume that the result is “Yunaka”.

Here, it is assumed that the candidate character string “Tanaka” obtained from the name kana field is correct, and the candidate character string “Yunaka” obtained from the name kanji field is incorrect. It is also assumed that these character strings are both present in the full name kana dictionary and the full name kanji dictionary. In this case, the candidate character string cannot be corrected by the independent knowledge processing.

The cooperation knowledge processing unit 5 determines that the candidate character string “Tanaka” obtained from the name kana field is three characters,
The candidate character string “Tanaka” is determined to be correct based on the fact that the reading of the candidate character string “Yunaka” obtained from the full name kanji field is the four characters “Yoshinaka”. This is because the number of characters in the candidate character string obtained from the name kana field cannot be wrong.

Next, the cooperative knowledge processing unit 5 refers to the full name kanji dictionary using the candidate character string “Tanaka” determined to be correct, and extracts a character form close to the candidate character string “Yunaka” obtained from the full name kanji field. The character string “Tanaka” whose reading is “Tanaka” is corrected to be a candidate character string.

The explanation of the above knowledge processing is as follows.
This is an example, and in practice, the single knowledge processing unit 4 and the cooperative knowledge processing unit 5 perform various other knowledge processes. FIG. 4 is a flow of the knowledge processing, and shows the processing performed by the kana-kanji automatic cooperation processing unit 3A and the recognition field search processing unit 12 for horizontal writing. In this processing flow, when the form 1 is searched, the handwritten kana field is searched before the handwritten kanji field among the fields to be linked,
Further, a case will be described in which the form 1 in which these are successively searched is processed.

When the character recognition processing unit 8 activates the knowledge processing unit 11, the form 1 to be processed by the knowledge processing unit 11 is
(S1). If horizontal writing is not performed, the processing shown in FIG. 5 is performed. In the case of horizontal writing, control is passed to the recognition field search processing unit 12 for horizontal writing.

The horizontal writing recognition field search processing unit 12 to which the control has been passed searches the form 1 to be processed in a downward direction from the top to search for a field (S2).
When a field is detected by this search, the recognition field search processing unit 12 for horizontal writing refers to the attribute data 10 and checks whether or not the field is a handwritten Kana field (S3).

If the field is not a handwritten kana field, the horizontal writing recognition field search processing section 12 notifies the knowledge processing section 11 of this fact, and the knowledge processing section 11 checks whether or not the field searched in S2 is the last field (S4). ). For this purpose, the maximum value of the item number of the form 1 is set in a predetermined register in advance, and 1 is subtracted from the contents of the register every time the field is searched in S2 and S6. Register contents are 0
If it becomes, it is the last field.

If it is the last field, the knowledge processing unit 1
1 ends the knowledge processing. If not the last field,
The knowledge processing unit 11 is used for the horizontal writing recognition field search processing unit 1
Then, control is passed to S2 and S2 and subsequent steps are repeated.

In the case of a handwritten kana field,
The horizontal writing recognition field search processing unit 12 refers to the attribute data 10 and checks whether or not the field searched in S2 is a knowledge field (S5).

If it is not a knowledge field, S4 and subsequent steps are executed for the field searched in S2. If the field is a knowledge field, the horizontal writing recognition field search processing unit 12 searches the form 1 further downward from the position of the field searched in S2 to search for the next field (S6). Normally, the next field is, as shown in FIG. 3, the start X position of which is the X of the field searched in S2.
The position is substantially the same as the position, and the start Y position is very close to the Y position of the field searched in S2.

When the next field is detected by this search, the horizontal writing recognition field search processing unit 12 refers to the attribute data 10 and checks whether or not the next field is a handwritten kanji field (S7).

If the field is not a handwritten kanji field, the horizontal writing recognition field search processing unit 12 passes control to the single knowledge processing unit 4, and the single knowledge processing unit 4 determines whether the field searched in S2 and the field searched in S6 Each independently executes independent knowledge processing (S8). After the end of S8, S4 and subsequent steps are executed for the field searched in S6, which is the field searched later.

In the case of a handwritten kanji field,
The horizontal writing recognition field search processing unit 12 refers to the attribute data 10 and checks whether the next field is a knowledge field and the knowledge attribute is the same as the field searched in S2 (S9).

If it is a knowledge field and the knowledge attributes are not the same, S8 is executed. That is, the horizontal writing recognition field search processing unit 12 passes control to the single knowledge processing unit 4, and the single knowledge processing unit 4 independently executes the single knowledge processing for the field searched in S2 and the field searched in S6. Is executed (S8). After the end of S8,
S4 and subsequent steps are executed for the field retrieved in S6, which is the field retrieved later.

If the knowledge fields are the same and the knowledge attributes are the same, the recognition field search processing unit for horizontal writing 12
Notifies the kana-kanji automatic cooperation processing unit 3A of this fact, and the kana-kanji automatic cooperation processing unit 3A associates the kanji field searched in S6 with the kana field searched in S2 (S10).

Thereafter, the horizontal writing recognition field search processing unit 12 passes control to the cooperative knowledge processing unit 5, and the cooperative knowledge processing unit 5 executes cooperative knowledge processing (S11). After this linkage knowledge processing, S6, which is the field searched later
Step S4 and subsequent steps are executed for the field searched by.

FIG. 5 is a flowchart of the knowledge processing, in which the kana-kanji automatic cooperation processing unit and the recognition field search processing unit 1 for vertical writing are used.
3 shows the processing performed. This processing flow shows processing executed after S1 of FIG. Therefore, a case will be described in which the handwritten Kana field is searched before the handwritten Kanji field, and the form 1 in which these are continuously searched is processed.

Form 1 to be processed in S1 of FIG.
Is recognized as vertical writing, the kana-kanji automatic cooperation processing unit 3A passes control to the recognition field search processing unit 13 for vertical writing.

The recognition field search processing unit 13 for vertical writing to which control has been passed searches the form 1 to be processed from right to left to search for a field (S1).
2). When a field is detected by this search, the recognition field search processing unit 13 for vertical writing checks whether or not the field is a handwritten Kana field, similarly to S3 (S13).

If the field is not a handwritten kana field, the knowledge processing unit 11 checks whether or not the field searched in S12 is the last field, as in S4 (S14). If it is the last field, the knowledge processing section 11 ends the knowledge processing. If it is not the last field, the knowledge processing unit 11 passes control to the recognition field search processing unit 13 for vertical writing, and
Step S12 and subsequent steps are repeated.

In the case of a handwritten Kana field,
The recognition field search processing unit 13 for vertical writing checks whether the field searched in S12 is a knowledge field as in S5 (S15).

If it is not a knowledge field, S14 and subsequent steps are executed. If it is a knowledge field, the recognition field search processing unit 13 for vertical writing searches the form 1 further leftward from the position of the field searched in S12 to search for the presence of the next field (S16). The next field is
The start Y position is substantially the same as the Y position of the field searched in S12, and the start X position is extremely close to the X position of the field searched in S12.

When the next field is detected by this search, the recognition field search processing unit 13 for vertical writing checks whether or not the next field is a handwritten kanji field as in S7 (S17).

If it is not a handwritten kanji field, the recognition field search processing unit 13 for vertical writing passes control to the single knowledge processing unit 4, and the single knowledge processing unit 4 executes single knowledge processing (S18). Thereafter, S14 and subsequent steps are executed.

In the case of a handwritten kanji field,
The recognition field search processing unit 13 for vertical writing checks whether the next field is a knowledge field and the knowledge attribute is the same as the field searched in S12, as in S9 (S19).

If it is a knowledge field and the knowledge attributes are not the same, S18 is executed. That is, the independent knowledge processing unit 4 executes the independent knowledge processing (S18). If it is a knowledge field and the knowledge attributes are the same, as in S10, the kana-kanji automatic cooperation processing unit 3A associates the kanji field searched in S16 with the kana field searched in S12 (S20). .

After that, the recognition field search processing unit 13 for vertical writing activates the cooperative knowledge processing unit 5, and the cooperative knowledge processing unit 5
Executes the cooperative knowledge processing (S21). After this, S12
Do the following:

The processing flows shown in FIGS. 4 and 5 show an example of the automatic cooperation processing in the knowledge processing.
Various changes can be made in accordance with each attribute of the field in the form 1. That is, this processing flow is for the case where the form 1 in which the handwritten kana field is searched for before the handwritten kanji field is searched and these are successively searched is processed. Therefore, in the form 1, when the order of detection is reversed or when the search is not performed continuously, the processing flow is changed.

Next, the cooperative processing based on the processing flow of FIG. 4 for the forms shown in FIGS. 3 and 6 will be briefly described. Since this form 1 is written horizontally, the knowledge processing unit 1
1 recognizes that the form 1 to be processed in S1 is horizontal writing, and passes control to the horizontal writing recognition field search processing unit 12. The recognition field search processing unit 12 for horizontal writing first detects the member number field in S2, but since this is a print field and not a handwritten Kana field (S3), no knowledge processing is performed. next,
After S4, the application date field is detected in S2, but since this is also a numeric field and not a handwritten Kana field (S3), no knowledge processing is performed. The same applies to the postal code field and the telephone number field.

Next, the horizontal writing recognition field search processing section 12 detects an address kana field in S2.
Since this is a handwritten Kana field (S3) and a knowledge field (S5), another field is searched downward in S6 to detect an address kanji field. This is a handwritten kanji field (S
7) Since the knowledge attribute is the same for "address" (S
9), the horizontal writing recognition field search processing unit 12 executes S1
At 0, the address kanji field and the address kana field are linked. Thereby, the cooperative knowledge processing is performed on the address kanji field and the address kana field (S11).

Next, the recognition field search processing unit 12 for horizontal writing detects the name kana field in S2.
Since this is a handwritten kana field (S3) and a knowledge field (S5), another field is searched downward in S6 to detect a name kanji field. This is a handwritten kanji field (S
7) Since the knowledge attribute is the same for "name" (S
9), the horizontal writing recognition field search processing unit 12 executes S1
At 0, the name kanji field and the name kana field are linked. Thereby, the cooperative knowledge processing is performed for the name kanji field and the name kana field (S11).

[0094]

As described above, according to the present invention,
In the character recognition processing device, the handwritten kana field and the handwritten kanji field are automatically set by utilizing that the positions of the handwritten kana field and the handwritten kanji field to be subjected to the cooperative knowledge processing on the form are in a predetermined relationship. Since most of the forms can be automatically linked to perform cooperative knowledge processing, there is no need to input instructions for cooperative knowledge processing, thereby reducing the burden on the operator. In addition, it is possible to prevent an error in instruction input or the like, thereby improving the efficiency of character recognition processing.

[Brief description of the drawings]

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

FIG. 2 is a configuration diagram of an embodiment.

FIG. 3 is an explanatory view of an embodiment.

FIG. 4 is a knowledge processing flow.

FIG. 5 is a knowledge processing flow.

FIG. 6 is a background explanatory diagram of the present invention.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 form 2 processing device 3 automatic cooperation processing unit 3A kana-kanji automatic cooperation processing unit 4 independent knowledge processing unit 5 cooperation knowledge processing unit 6 character recognition output 7 reading processing unit 8 character recognition processing unit 9 candidate character string 10 attribute data 11 knowledge processing Unit 12 Recognition field search processing unit for horizontal writing 13 Recognition field search processing unit for vertical writing 14 Dictionary

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-6-266888 (JP, A) JP-A-5-258099 (JP, A) JP-A-5-189614 (JP, A) JP-A-63-1988 34680 (JP, A) (58) Field surveyed (Int. Cl. ⁷ , DB name) G06K 9/03, 9/72

Claims (2)

(57) [Claims] An apparatus comprising a plurality of predetermined fields.
Read the characters entered in each field of the form
A character recognition processor that performs character recognition processing for this
In the given fieldIn that fieldCharacter recognition
Perform knowledge processing in the knowledge processing in the field alone
A single knowledge processing unit,In that fieldCharacter recognition
Knowledge processing in knowledge processing, its fields and this
Should be done in cooperation with other linked fields
Whether the information instructing the fact corresponds to the form to be processed
Corresponding to the field to be processed,Selectable
The holding part of the described attribute data, and theIn that fieldCharacter recognition
Knowledge processing in knowledge processingCharacters in
And other fields associated with itCharacters inAnd in
A cooperative knowledge processing unit that performs coordination, and a position on a form is determined in advance from among a plurality of fields.
Search for multiple fields in a related relationship and combine them.
This is an automatic cooperation processing unit that
Search from the top to the bottom of the
Field, the start X position of each other on the
Qualitatively identical and starting Y position of lower field
Handwriting in a position very close to the field above it
Search for Kana field and handwritten Kanji field
Recognition field search processing unit for horizontal writing and form for vertical writing
From the right to the left, and search multiple
From the field, start Y position of each other on the form
Start of the field that is substantially the same and to the left
X position is very close to the field to the right
Search handwritten kana field and handwritten kanji field
With the vertical writing recognition field search processing unit
Automatic cooperation processing unit having two field search processing units
And the automatic cooperation processing unitBut beforeRecognition field detection for horizontal writing
Processing by the search processing unit and / or the vertical writing recognition file.
Linked knowledge processing based on the processing by the field search
Dictate the need forAndThe cooperation knowledge processing unitThe said
Check the attribute data described corresponding to the field
ClickCorresponding multiple fieldsInside Characterabout
The feature is that the knowledge processing is performed in cooperation with
Character recognition processor. 2. The character according to claim 1, wherein the automatic cooperation processing unit comprises a kana-kanji automatic cooperation processing unit, and the kana-kanji automatic cooperation processing unit associates a handwritten kana field and a handwritten kanji field. Recognition processing device.