JP2008204226A - Form recognition device and its program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To read a character in a form different in a format and the same in kinds though, by detecting a data area without designating position coordinates for a reading area in advance and judging an attribute of each of the data areas. <P>SOLUTION: An item name word dictionary 190, in which a character string and the attribute of the item name are registered, is used. The character string of the item name is extracted by detecting the character strings among the form images and collating with the item name dictionary 190. The character string, succeeded in collation with the item name dictionary 190, is judged as the item name character string and the character string, failed in the collation with the item name dictionary 190, as the data character string, and the attribute of the data character string is judged from the position relation between the item name character string and the data character string. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a form image processing technique, and more particularly to a technique for understanding attributes of written information on a form and character recognition.

  In the form recognition method in the conventional form OCR (Optical Character Reader), there are many cases where a fixed form whose reading position is determined in advance is recognized. As a conventional technique for such a form, there is a “format generator” (for example, see Non-Patent Document 1). With this technology, the character entry position to be read is strictly specified in units of 0.1 mm. Many existing OCR models employ the same format information as the format generator.

  On the other hand, forms such as salary payment reports and receipts cannot be recognized by the above method even for the same type of form, but the number of ruled lines and the position and size of the frames are slightly different for each sheet. For such a form, for example, as shown in Patent Document 1, although the position and size of the frame are different, the arrangement relationship between items is almost constant, and the pre-registered frame structure and form are used. There is a method of specifying a reading region by collating with a frame structure analyzed from an image (Patent Document 1).

  A form with a higher degree of freedom includes a registered notice. For this type of form, if the form is the same type, the item names are almost the same, but the size and number of form frames differ for each sheet. In some cases, the arrangement between items may be different. For such a form, the table structure is first analyzed, the characters in the frame are recognized to detect the frame in which the item name is written, and the right or lower frame is the data area to be read. There is a method of judging and recognizing (Patent Document 2). In the method of Patent Document 2, it is necessary to define in advance the character string of the item name and the correspondence between the item name frame and the data frame.

  Furthermore, Patent Document 3 is a conventional technique for detecting data from a character string of an item name. In the method of Patent Document 3, data having items having a hierarchical relationship such as “arrival-quantity”, “arrival-quantity”, “shipment-quantity”, “shipment-money”, and data of a two-dimensional table structure are targeted. It is said. In this method, recognition is made possible by defining the arrangement relationship between item names in advance.

JP 2004-139484 A Japanese Patent Laid-Open No. 9-319824 JP 2005-275830 A "Hitachi OCR Solution Imaging OCR" Catalog, Hitachi, Ltd., December 2005, P5-6

The characteristics of the form as the recognition target are summarized as follows.
(1) Although the attributes are the same, such as “name” and “name”, the character string of the item name representing the attribute is different for each form.
(2) The positional relationship between the character string of the item name and the character string of the data, including the hierarchical relationship and the two-dimensional relationship, is unknown.
(3) Data of a plurality of attributes are described in one frame.
(4) Item names and data character strings are not enclosed in a frame.
(5) Data of a plurality of items is a set (record), and these data are repeatedly described.
(6) Another character string unnecessary for reading is included in the area where the data is described. For example, this is a case where the character “yen” is written in the amount column.

  When dealing with such a form by the method of Non-Patent Document 1, it is necessary to create format information for all forms and identify which format information is applied for each entered form. In a situation where hundreds to thousands of forms are handled, it is not practical to apply this method from the viewpoint of the cost of creating format information and the accuracy of identification.

  Also, the method of Patent Document 1 is premised on the same arrangement between items (attributes), and thus cannot be applied to the above-mentioned non-standard form.

  Although the method of Patent Document 2 can cope with a part of a non-standard form, it is necessary to define the positional relationship between the item name frame and the data frame representing the attribute and the character string of the item name in advance. It cannot cope with the problems (1) to (6).

  Although the method of Patent Document 3 has a feature that the hierarchical relationship and two-dimensional relationship of item names can be analyzed, it cannot cope with the problems (1) to (6) as in Patent Document 2. Furthermore, since the hierarchical relationship and the relationship of the two-dimensional table must be defined in advance, for example, the case where the rows and columns of the two-dimensional table are switched cannot be handled.

  Further, since the conventional method recognizes the item name of the data based on the arrangement relationship between the frames, it cannot recognize the form having the description format (3) or (4).

  An object of the present invention is to recognize data without defining the position of a reading item for each form for forms having different formats such as frame size, position and item order despite the same type of form. Therefore, an object of the present invention is to provide a form recognition apparatus capable of automatically analyzing data attributes and a program thereof.

  In order to achieve the above object, in the present invention, a form recognition apparatus comprising a storage unit that stores a recognition dictionary and an item name word dictionary, and a processing unit that performs a form image recognition process, the processing unit includes: Detects an area containing a character string in a form image, detects a character string in the detected area, recognizes the character of the detected character string using a recognition dictionary, and determines the character recognition result of the character string Matches the item name word in the item name word dictionary, and successfully matches the item name word string (hereinafter referred to as the item name character string) and the item name word match. The character string that has not been determined is determined as a character string in which data is described (hereinafter referred to as a data character string), the data is associated with the item name from the arrangement relationship between the item name character string and the data character string, and the data The character string attribute is determined.

  That is, a form recognition program for recognizing characters in a form image, wherein a processing unit of the form recognition apparatus detects a region including a character string in the form image and detects a character string in the region. A step of recognizing the character of the detected character string, a step of collating the character recognition result of the character string with the item name word, and determining that the character string that has been successfully matched with the item name word is an item name character string. A step of determining the character string that has not been successfully matched with the item name word as a data character string, and associating the data with the item name from the arrangement relationship between the item name character string and the data character string, It is configured to execute a step of determining an attribute corresponding to.

  Recognize forms with similar formats but different formats without strict definition. By simply registering the item name character string, it can be recognized while analyzing the attributes of the data in the form.

  Hereinafter, the present invention will be described in more detail with reference to embodiments shown in the drawings. Note that the present invention is not limited thereby. Before describing specific processing contents, an outline of the present invention will be described.

  In the present invention, a character string of data is recognized in a form having a high degree of freedom of format without pre-defining the data description position. For this reason, it is necessary to automatically understand the attribute of the described data. In this embodiment, the word of the item name representing the attribute of the data is registered in the dictionary in advance instead of the data description position. deep. The item name character string is detected by comparing the item name word in this dictionary with the character string in the form. On the other hand, the character string that cannot be matched with the word dictionary of the item name is regarded as a data character string. Furthermore, the correspondence relationship between the data character string and the item name character string is recognized by analyzing the arrangement relationship between the item name character string and the data character string. The attribute of the associated item name is the data string attribute.

  Specifically, the attribute of the data character string is recognized by analyzing the arrangement relation between the item name character strings and between the item name character string and the data character string in the following order.

  (1) The conventional method considers a frame as a reference, and analyzes the arrangement relationship between a frame in which item names are described and a frame in which data is described. In contrast, in this method, processing is performed based on a character string. Specifically, by analyzing the arrangement relationship between the item name character string and the data character string, it is possible to cope with cases where a plurality of items are described in the same frame or when there is no frame.

  (2) The extracted character string is recognized and collated with the word of the item name. The character string that succeeds in collation is the character string of the item name, and the character string that fails in collation is determined to be the data character string. The attributes of the data character string are recognized by analyzing the arrangement relationship between the determined item name character string and the data character string.

  (3) In the conventional method, the layout relationship between item names and data, the hierarchical relationship between item names, and the two-dimensional relationship between item names were defined, whereas in this method, as shown in (2) Recognize data attributes including hierarchical relationships and two-dimensional relationships only from item name words.

(4) A group of data (record) is detected from a list of a plurality of item names and data sets.
By executing part or all of these processes, the present invention can realize form recognition with a high recognition rate.

  FIG. 2 shows an example of the hardware configuration of the form recognition apparatus according to the first embodiment of the present invention. In FIG. 1, 10 is an input device that is an input unit for inputting commands and code data, 20 is an image input device that is an image input unit for inputting a form image to be processed, and 30 is character string detection. A form recognition unit for recognizing characters and attributes of data character strings, 40 is a form recognition dictionary in this embodiment, and stores a character recognition dictionary and item name word dictionary, and 50 displays a recognition result. It is a display device which is a display unit. A form image may be input from 60 image databases (DB) instead of 20 image input devices. The recognition dictionary 40 and the image DB 60 are stored in a storage unit (not shown).

  The form recognizing unit 30 is usually composed of a central processing unit (CPU), and the CPU 30 as the form recognizing unit executes the process of the form recognizing program of this embodiment described in detail below. become. The form recognition program executed by the CPU is normally stored in a storage unit (not shown). However, it can also be introduced into the storage unit from the outside via a portable storage medium or a network and stored. Needless to say.

  Details of the processing in the first embodiment will be described below with reference to the drawings. FIG. 1 is a flowchart showing an outline of the form processing by the form processing apparatus 30 of the first embodiment. As described above, this process is usually executed as a program process in the CPU. The same applies hereinafter.

  First, in step 110, an area including a character string is detected from the input form image. Hereinafter, this area is defined as a character string area. This character string area corresponds to an individual frame (cell) in the case of a tabular form. In the case of a non-tabular form or an out-of-line character string, the character string itself or a space for separating the character string corresponds to a frame configured by virtually considering a ruled line.

  Next, in step 120, a character string in each character string area is detected. As a specific example of steps 110 and 120, for example, a method detailed in Japanese Patent Laid-Open No. 11-53466 can be used. For one example of line extraction, a horizontal character line or a vertical character is obtained by integrating adjacent connected components in the region (a block of continuous black pixels) in the horizontal or vertical direction. It is possible to extract rows.

In step 130, the character recognition dictionary 180 is used to recognize individual characters in the character string.
In step 140, the character recognition result in step 130 is collated with the item name word registered in the item name word dictionary 180. As an embodiment of the processes 130 and 140, for example, a method described in Japanese Patent Application Laid-Open No. 2004-171316 can be used.

In step 150, the character string for which word matching succeeded in step 140 is determined as the item name character string.
In step 160, the character string for which word matching failed in step 140 is determined as a data character string. Since word matching has failed, the result of step 130 is used as it is as the character recognition result. However, since there is a possibility that word collation may fail due to an error in character recognition or the like, it may be determined as a data character string candidate and whether or not it is a data character string may be determined based on subsequent processing.

  In step 170, the attribute of each character string is determined from the arrangement relationship between the item name character string and the data character string. Details of attribute recognition will be described in detail with reference to FIGS. 6 to 12 for each item name and data arrangement relationship.

  FIG. 3 and FIG. 4 are examples of forms that have the same type but different frame sizes and item names. The form shown in FIG. 3 includes “bank name”, “branch name”, “account number”, “name”, and “amount” as item names. The form shown in FIG. 4 includes “bank”, “branch”, “account No.”, “name”, and “amount” as item names.

  FIG. 5 shows an example of data stored in the item name word dictionary 190 corresponding to the forms shown in FIGS. Item name words with different written characters are registered, and the same attribute ID is assigned to item name words with the same attribute. For example, the attribute ID of “bank name” and “bank” is “1”. For the item name word to be registered, the form to be processed is observed, and words with the same attribute and different notation are selected. In the present embodiment, all the character strings described as item names are registered, but only a part of the item names may be registered. This has the effect of reducing the number of registered words. For example, by registering only “bank”, word matching succeeds without registering “bank name”. Moreover, although attribute ID was described by the number, a character string may be sufficient.

  Hereinafter, the character string attribute analysis in step 170 of FIG. 1 will be described with reference to FIGS. First, a schematic processing flow of step 170 will be described with reference to FIG. 6, and a detailed processing flow in FIG. 6 will be described with reference to FIGS.

  FIG. 6 is a diagram for explaining the processing flow of character string attribute analysis in step 170 of FIG. First, the hierarchical relationship analysis between item name character strings in step 600 is performed. This applies only to the item name character string detected as a result of word matching in step 150 of FIG. The arrangement relationship between these character strings is analyzed to analyze the hierarchical relationship between the item name character strings. Details of this processing will be described with reference to the flowchart of FIG. 7 and the frame arrangement example of FIG.

  The vertical item name-data relationship analysis in step 610 analyzes the relationship between the item name character string and the data character string when the data character string exists below the item name character string, and sets the attribute of the data character string. It is a process to detect. Details of this processing will be described with reference to the flowchart of FIG. 9 and the frame arrangement example of FIG.

  The horizontal item name-data relationship analysis in step 620 analyzes the relationship between the item name character string and the data character string and detects the attribute of the data character string when the data character string exists on the right side of the item name character string. It is processing to do. This process can be realized by changing the process of step 610 in the horizontal direction.

  The repetitive relationship analysis in step 630 is a process of analyzing the data repetitive structure when data character strings having the same attribute are arranged in the vertical and horizontal directions. Details of this processing will be described with reference to FIG.

  Although FIG. 6 shows four types of analysis processes of steps 600, 610, 620, and 630, some processes may be omitted depending on the application. Further, processing described later with reference to FIGS. 13 and 14 may be added.

  FIG. 7 is a diagram for explaining a processing flow of the hierarchical relationship analysis between item name character strings in step 600 of FIG. First, in step 700, the item name character string areas are analyzed one by one. Next, in step 710, it is determined whether another item name character string area exists below the item name character string area. If it exists, step 720 determines whether the left and right ends of the item name character string area to be determined in step 700 match the left and right edges of the item name character string area below the item name character string area. Judgment. If they match, it is determined in step 730 whether the lower item name character string areas are equal in height. The processing of steps 720 and 730 will be specifically described later with reference to FIG. If the heights are equal, in step 740, the hierarchical relationship of the attributes is set with the upper item name character string area as the upper hierarchy and the lower item name character string area as the lower hierarchy. If these processes are completed or if the determinations in steps 710 to 730 are not satisfied, the process target is moved to the next item name character string area in step 750 and the process returns to step 700.

  FIG. 8 is an example in which it is determined in steps 720 and 730 in FIG. 7 that the upper and lower item character string areas have a hierarchical relationship. In steps 700 and 710, a frame including the item name character string 1 is set as an item character string region (800) to be processed. In steps 720 and 730, if two frames, the frame containing the item name character string 21 and the frame containing the item name character string 22, are the lower item name character line areas (810, 820) to be determined, the upper item name The left end of the character string area 810 is coincident with the left end of the lower item name character string area 810. Similarly, the left end of the item name character string area 810 and the right end of the item name character string area 820 are also matched. Meet the conditions. Further, the frame heights of the lower two item name character string areas 810 and 820 are the same, and the condition of step 730 is satisfied. Therefore, the items in the lower hierarchy of the item name character string 1 can be determined as the item name character string 21 and the item name character string 22.

  FIG. 9 is a diagram for explaining the processing flow of the vertical item name-data relationship analysis in step 610 of FIG. First, in step 900, the item name character string areas are analyzed one by one. Next, in step 910, it is determined whether a data character string area exists below the item name character string area. If it exists, it is determined in step 920 whether the left and right ends of the item name character string area to be determined in step 900 match the left and right edges of the lower data character string area. The processing of step 920 will be specifically described later with reference to FIG. If the condition of step 920 is satisfied, it is determined in step 930 that the character string attribute in the lower data character string region is the character string attribute in the upper item name character string region. If these processes are completed or if the determinations in steps 910 and 920 are not satisfied, the process target is moved to the next item name character string area in step 940 and the process returns to step 900. The horizontal item name-data relationship analysis in step 620 of FIG. 6 can be realized by changing the analysis direction of this processing to the horizontal direction.

  FIG. 10 is an example in which it is determined in step 920 in FIG. 9 that there is an item name-data relationship between the upper and lower character string regions. The example of FIG. 10A is a case where a data frame exists immediately below the item name frame. If the frame including the item name character string is the item character string area (1000) to be processed and the frame 1010 including the data character string is the data character line area in step 920, the left end of the upper item name character string area 1000 is , Both ends of the lower data character string area 1010 coincide. Therefore, the attribute of the character string in 1010 can be determined as the attribute of the character string in 1000. In the example of FIG. 3, it can be determined that the attribute of the data “AAA” is the bank name (the attribute ID of FIG. 5 is “1”).

  The example of FIG. 10B is a case where there are a plurality of data frames immediately below the item name frame. The frame including the item name character string is set as the processing target item character string area (1020), and the frame including the data character strings 1, 2, and 3 is set as the processing target data character string area (1030, 1040, 1050) in step 920. Then, the left end of the upper item name character string area 1020 and the left end of the lower data character string area 1030 are matched, and similarly the right end of the item name character string area 1020 and the data character string area 1050 are also matched. Therefore, the condition of step 920 is satisfied. Furthermore, 1030, 1040, and 1050 are adjacent to each other, and a determination condition that the frame heights coincide may be added. As a result, the attribute of the character string in 1030, 1040, and 1050 can be determined as the attribute of the character string in 1020. In the example of FIG. 4, it can be determined that the attribute of the data in the three frames “(blank)”, “5”, and “000” is the amount of money (the attribute ID of FIG. 5 is “5”).

  FIG. 11 is a diagram for explaining the repeated structural analysis in step 630 of FIG. 11, the arrangement relationship between the item name character string area (1100), which is a frame including the item name character string 1, and the data character string area (1110), which is a frame including the data character string 11, by the processing of FIG. Can be determined that the attribute of the data character string 11 is the attribute of the item name character string 1. Further, when a data character string area 1120 having the same width exists below the data character string area 1110, it is determined that the attribute of the data character string 12 in 1120 is also the attribute of the item name character string 1. The above processing is also executed for the item name character strings 2 and 3, and the repetitive relationship of the data character string area in each item is obtained. Furthermore, each data character string area including the data character string 11, the data character string 21, and the data character string 31 has the same height and determines whether or not they are adjacent to each other. In FIG. 11, in order to satisfy the determination, these data are determined as one set (one record). Similarly, the data character string 12, the data character string 22, and the data character string 32 are determined as one set.

  FIG. 12 is an example of a result of executing the processing of FIG. 1 on the form of FIG. 3 using the item name word dictionary of FIG. The attribute of the data “AAA” is “bank name” existing on the upper side, and its attribute ID is “1”. There is no horizontal attribute. In addition, the five pieces of data “AAA”, “aaa”, “00000000”, “XXX”, and “2000” are determined as one set and are determined as the record number “1”. Similarly, the data in the row with the bank name “BBB” is recognized as the record number “2”. Further, the data “12000” in the bottom row of FIG. 3 is obtained by executing “item name-data relationship analysis” (610, 620) in the vertical direction and the horizontal direction so that the vertical attribute is “money” and the horizontal direction. It can be seen that there is a two-dimensional relationship of “total”. If item names have a hierarchical relationship, hierarchical information may also be added and output. Thus, by recognizing the attributes of each data, even if the forms have different formats, the recognition results can be stored in the same database based on the attributes.

  The data column in FIG. 12 may include not only character recognition results of data but also character string coordinates and character string partial images. This is a usage format in which all data cannot be correctly recognized by character recognition, and only the attributes of the area in which the data is described are recognized for data input without using OCR. In this case, data input can be used for operations such as human punch input.

  FIG. 13 is an example of analysis in the case where there are a plurality of items in the same frame in the character string attribute analysis in step 170 of FIG. This processing may be added to the flow of FIG. In the conventional method for associating data with item names such as Patent Document 3 and the like, a frame is used as a basic unit for attribute assignment, and therefore a plurality of items are included in the same frame, such as “bank” and “branch” in FIG. If this is the case, the attribute of each data string cannot be assigned. In this method, as a result of the analysis in step 610 or step 620 of FIG. 6, there are a plurality of item name character strings in the item name character string area and a plurality of data character strings in the data character string area. Are assigned different attributes corresponding to individual data strings. In the case of FIG. 13, the attribute of “data character string 1” is the attribute of “item name character string 1”, and the attribute of “data character string 2” is the attribute of “item name character string 2”. The relative position within the frame can be used as the reference for association. In the example of FIG. 13, the upper character strings in the frame are associated with the lower character strings. For this arrangement relationship, it is also possible to use criteria such as the left and right arrangement and the number of lines in the frame.

  FIG. 14 shows an example of analysis in the case where both the item name character string and the data character string exist in the same frame in the character string attribute analysis in step 170 of FIG. This processing may be added to the flow of FIG. In the conventional method of associating data with item names such as Patent Document 3 and the like, the frame is the basic unit of attribute assignment, and therefore data cannot be attributed even in this example. In this case, when there is no data character string area adjacent to the frame including the item name character string and the data character string exists in the same frame, the item name character string and the data character string are associated with each other. Further, this association can be made not only within the same frame but also between a plurality of character strings outside the frame.

  Furthermore, when only a part of the same character string has been successfully matched with the item name word, the character string can be divided into a part that has been successfully matched and the other, and the above-described correspondence can be made.

  FIG. 15 shows an example of displaying the recognition result of the first example. In this example, the attribute data that the operator does not want to show is hidden. FIG. 15 shows an example in which the data in the name column is hidden in the form shown in FIG. This function enables an operator who performs punch input and confirmation of recognition results to see only necessary data without looking at personal information.

  FIG. 16 shows an example of displaying the recognition result of the first example. In this example, only the data of a specific attribute is displayed as a list. FIG. 16 is an example in which only the data of the attribute of the bank name is displayed for the form 1 for the form of FIG. 3 and the form 2 for the form of FIG. The data display may be a recognition result or a partial image of the character string. In this example, the data of a plurality of forms is displayed as a list, but only the data of one form may be displayed. In addition, data with a plurality of attributes may be displayed.

  FIG. 17 is a diagram for explaining a preprinted character recognition method according to another embodiment. In this embodiment, preprinted characters such as “yen” are also registered in the item name word dictionary 190 of FIG. If the collation with the pre-printed word is successful in the item name character string collation in step 140, the character string is determined to be pre-printed in step 150. As a result, in the character string attribute analysis in step 170, it can be determined that the character string of “yen” is pre-printing having a monetary attribute. This process has the effect of preventing erroneous recognition of preprinted characters as data.

The figure which shows the flow of the form recognition process in a 1st Example. The block diagram which shows schematic structure of the form recognition apparatus in connection with a 1st Example. The figure which shows an example of the process target form in a 1st Example. The figure which shows an example of the process target form in a 1st Example. The figure which shows an example of the item name word dictionary in a 1st Example. The figure which shows an example of the character string attribute analysis processing flow of a 1st Example. The figure which shows an example of the hierarchical relationship analysis process flow between the item name character strings of FIG. 6 of a 1st Example. The figure which shows the example of the object of the hierarchical relationship analysis process between the item name character strings in a 1st Example. The figure which shows an example of the vertical direction item name-data relationship analysis processing flow of FIG. 6 of a 1st Example. The figure which shows the example of the object of the vertical direction item name-data relationship analysis process in a 1st Example. The figure which shows the example of the object of the repetition relationship analysis process in a 1st Example. The figure which shows an example of the recognition result of the form of FIG. 3 and FIG. 4 in a 1st Example. The figure which shows an example in which several items are described in the same frame. The figure which shows the example in which both the character string of an item name and the character string of data are described in the same frame. The figure which shows the example which hides the specific item in a 1st Example. The figure which shows the example which displays only the data of the specific attribute in a 1st Example as a list display. The figure which shows the example in which both the character string of pre-printing and the character string of data are described in the same frame in another Example.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... Input device, 20 ... Image input device, 30 ... Form recognition part (CPU), 40 ... Recognition dictionary, 50 ... Display device, 60 ... Image DB, 180 ... Character recognition dictionary, 190 ... Item name word dictionary, 800, 810, 820, 1000, 1020, 1100 ... item name character string, 1010, 1110, 1120 ... data character string.

Claims (10)

A form recognition device comprising a storage unit for storing a recognition dictionary and an item name word dictionary, and a processing unit for recognizing a form image,
The processor is
In the form image, an area including a character string is detected,
Detect the character string in the detected area,
Recognizing characters of the detected character string using the recognition dictionary;
Collating the character recognition result of the character string with the item name word in the item name word dictionary;
The character string that succeeded in item name word matching is determined as the item name character string, and the character string that did not succeed in item name word matching is determined as the data character string.
A form recognition apparatus that associates data with an item name from an arrangement relationship between the item name character string and the data character string and determines an attribute corresponding to the item name. The form recognition device according to claim 1,
The processor is
When determining the attribute corresponding to the item name by associating data with the item name from the arrangement relationship between the item name character string and the data character string, the item name attribute is determined from the item name character string arrangement. A form recognition device that analyzes the hierarchical relationship between the two. The form recognition device according to claim 1,
When determining an attribute corresponding to the item name by associating data with the item name from the arrangement relationship between the item name character string and the data character string,
A form recognition apparatus that analyzes the attribute of the data character string from the vertical or horizontal arrangement relationship between the item name character string and the data character string. The form recognition device according to claim 3,
When determining an attribute corresponding to the item name by associating data with the item name from the arrangement relationship between the item name character string and the data character string,
A form recognition device that analyzes the repetitive structure of data having the same attribute from the arrangement relation of the data character string. The form recognition device according to claim 1,
When determining an attribute corresponding to the item name by associating data with the item name from the arrangement relationship between the item name character string and the data character string,
When the data character strings having a plurality of attributes exist in the same area, the data character strings are associated with each other by associating the arrangement of the item name character strings with the arrangement of the data character strings in the area. A form recognition device that analyzes column attributes. The form recognition device according to claim 1,
When determining an attribute corresponding to the item name by associating data with the item name from the arrangement relationship between the item name character string and the data character string,
A form recognition apparatus that determines that both attributes are the same when both the item name character string and the data character string exist in the same area. The form recognition device according to claim 1,
When determining an attribute corresponding to the item name by associating data with the item name from the arrangement relationship between the item name character string and the data character string,
A form recognition device for determining a character string in which pre-printed characters are described by storing item name words and pre-printed words in an item name word dictionary. A form recognition program that is executed by a form recognition device including a storage unit that stores a recognition dictionary and an item name word dictionary, and a processing unit that performs a form image recognition process,
In the processing unit,
Detecting an area including a character string in a form image;
Detecting a character string in the region;
Recognizing the detected character of the character string;
Collating a character recognition result of a character string with an item name word stored in the storage unit;
Determining a character string that has been successfully matched with an item name word as an item name character string;
Determining a character string that has not been successfully collated as an item name as a data character string;
A form recognition program for executing a step of determining an attribute corresponding to the item name by associating data with the item name based on an arrangement relationship between the item name character string and the data character string. A program for recognizing a form according to claim 8,
In the step of determining the attribute,
A form recognition program comprising a step of analyzing a hierarchical relationship between attributes of the item name from an arrangement of the item name character string. A program for recognizing a form according to claim 8,
In the step of determining the attribute,
A form recognition program comprising a step of analyzing an attribute of the data character string from a vertical or horizontal arrangement relationship between the item name character string and the data character string.

Images