JP2013015909A - Table structure automatic recognition program, table structure automatic recognition method and table structure automatic recognition device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a table structure automatic recognition program, a table structure automatic recognition method and a table structure automatic recognition device capable of highly accurately performing automatic recognition of a table with simple definition of table structures.SOLUTION: A computer executes processing comprising: extracting multiple tables from a document database storing document data including the multiple tables including multiple sections; referring to a table extraction condition database storing a first index of the tables to be extracted and checking the first index and data of the sections of index items included in each of the multiple tables against each other; extracting a table a check result of which satisfies a predetermined condition; and correcting, on the basis of a length of each section included in the table satisfying the predetermined condition, at least one of the length of the section of the index item and the length of the section of a data item corresponding to the index item, used for associating the section of the index item of the table and the section of the data item with each other.

Description

  The present invention relates to a table structure automatic recognition program, a table structure automatic recognition method, and a table structure automatic recognition apparatus that automatically recognize a table structure.

  In software development, a consistency check is performed to confirm consistency between a plurality of design documents including tables defining various requirements. In the consistency check, a table that matches the definition of the table structure is specified in each design document, and items are extracted from the specified table to check whether the items match each other.

  In the conventional consistency check, the table structure is manually defined, and it is difficult to define the structure of the table when the amount of the design document is large. For this reason, conventionally, techniques such as automatic recognition of the table structure are used to reduce the labor involved in defining the table structure.

  As a method for automatically recognizing a table structure, for example, there is a method in which item types such as item names and hierarchical structures and two-dimensional structures are defined in advance, and values corresponding to the items are acquired. In addition, there is a method of detecting a character string from a form image and collating it with the item name dictionary using an item name word dictionary in which item name character strings and attributes are registered. Furthermore, for example, there is a method in which a heading item dictionary representing data attributes is prepared, a rectangular area is calculated based on the data attributes, the distance from the heading, and the positional relationship with respect to the heading, and an item value that minimizes the rectangular area is obtained. Are known.

JP 2005-275830 A JP 2008-204226 A JP 2009-110416 A

  In the conventional method for automatically recognizing a table structure, an attribute must be defined for data that is a general character string, and a function for estimating an attribute for an item is required. In addition, when the headline and the item are associated with each other based on the positional relationship between the headline and the item, there is a possibility that the headline and the item are incorrectly associated with each other when the description amount of the item is large.

  In one embodiment of the present invention, in view of the above circumstances, a table structure automatic recognition program, a table structure automatic recognition method, and a table structure automatic recognition device that automatically recognize a table with high accuracy with a simple table structure definition are provided. It is an object.

  The disclosed technique extracts a plurality of tables from a document database in which document data including a plurality of tables including a plurality of sections is stored, and a table extraction condition in which a first heading of the table to be extracted is stored Referring to the database, the first heading and the data of the section of the heading item included in each of the plurality of tables are collated, a table in which the collation result satisfies a predetermined condition is extracted, and the predetermined condition is Based on the length of each section included in the table to be satisfied, the length of the section of the heading item used for associating the section of the heading item with the section of the data item in the table or the section of the data item corresponding to the heading item Causes the computer to execute a process of correcting at least one of the lengths.

  A method for executing the above processes, an apparatus for executing the program, and a computer-readable storage medium storing the program may be used.

  Automatic table recognition with high accuracy by simple table structure definition.

It is a figure explaining the outline | summary of the automatic recognition of a table structure. It is a figure which shows the hardware structural example of a table structure automatic recognition apparatus. It is a figure explaining an example of functional composition of a table structure automatic recognition device. It is a figure explaining each database stored in the memory | storage part. It is a figure which shows an example of a table extraction condition database. It is a figure which shows an example of an item matching database. It is a flowchart explaining the whole operation | movement of a table structure automatic recognition apparatus. It is a figure explaining extraction of a rectangular area. It is a flowchart explaining the process of a rectangular area extraction part. It is a figure explaining extraction of the similar table | surface which matched table extraction conditions. It is a flowchart explaining the process of a heading item collation part. It is a figure explaining the calculation of a similarity, and the matching of a heading item and a division. It is a 1st figure which shows the example which calculated the similarity for every heading item. It is a 2nd figure which shows the example which calculated the similarity for every heading item. It is a flowchart explaining the detail of the process of step S1112. It is a figure which shows the example which allocated the section with the highest similarity with respect to a heading item list so that a correspondence might become 1: 1. It is a flowchart explaining the process by a data item length correction | amendment part. It is a 1st figure explaining the specific example of correction | amendment of the length of a data item. It is a 2nd figure explaining the specific example of correction | amendment of the length of a data item. It is a flowchart explaining the process by the heading item length correction | amendment part. It is a 1st figure which shows the specific example of correction | amendment of the length of a heading item. It is a 2nd figure which shows the specific example of correction | amendment of the length of a heading item. It is a 3rd figure which shows the specific example of correction | amendment of the length of a heading item. It is a 4th figure which shows the specific example of correction | amendment of the length of a heading item. It is a flowchart explaining the process by the heading corresponding | compatible data specific | specification part. It is a figure explaining specification of the data item corresponding to a heading item. It is a figure explaining the input format of a table | surface. It is a figure which shows the example of an input which takes matching with a heading item and a data item. It is a figure which shows the example of the result of matching.

  In this embodiment, the table structure automatic recognition device automatically recognizes the table structure in which the headline is defined. Automatic recognition of the table structure includes associating table headings with items. The automatic recognition of the table structure is used for, for example, a consistency check for confirming the consistency of the definition of the requirements in each table when each document data includes a table defining the requirements. The document data is, for example, design document data created when performing software design or the like.

  An outline of automatic table structure recognition by the table structure automatic recognition apparatus will be described below with reference to FIG. FIG. 1 is a diagram for explaining an outline of automatic recognition of a table structure.

  In this embodiment, documents included in the design document are classified in advance for each type, and a table heading item included in the document is defined for each type. The table structure automatic recognition apparatus extracts a table including a plurality of cells (sections) from the document for each document type as a rectangular area (step S11). Subsequently, the table structure automatic recognition device collates the data of the heading item in the extracted table, and extracts a table in which the data of the heading item is similar (step S12). Subsequently, the table structure automatic recognition apparatus corrects the length of the data item and the length of the heading item in the extracted table (step S13). The heading item length in this embodiment is the length of the section corresponding to the heading item. Specifically, it is the number of characters in the section corresponding to the heading item. The length of the data item is the length of the section corresponding to the data item. Specifically, it is the number of characters in the section corresponding to the data item. Subsequently, the table structure automatic recognition apparatus specifies data item candidates corresponding to the heading items (step S14).

  In this embodiment, the table structure included in the document is automatically recognized based on the table header definition and the table structure definition based on the type of the document including the table.

  The table structure automatic recognition device that realizes the above-described table structure automatic recognition is a computer device, and has a hardware configuration as shown in FIG.

  FIG. 2 is a diagram illustrating a hardware configuration example of the table structure automatic recognition apparatus. In the table structure automatic recognition apparatus 100, an input device 11, a display device 12, a main storage device 13, a CPU 14, an interface device 15, an auxiliary storage device 16, and a driver device 17 are connected to each other via a bus B.

  The input device 11, the display device 12, the main storage device 13, the CPU 14, the interface device 15, the auxiliary storage device 16, and the driver device 17 connected to each other via the bus B exchange data with each other under the control of the CPU 14. be able to. The CPU 14 is a central processing unit that controls operation of the entire table structure automatic recognition apparatus 100.

  The interface device 15 receives data from another information processing device and passes the data to the CPU 14. Further, the interface device 15 transmits data to another information processing device in response to an instruction from the CPU 14.

  The auxiliary storage device 16 stores at least a table structure automatic recognition program that causes the table structure automatic recognition device 100 to execute automatic table structure recognition processing as a part of a program that exhibits the function of the table structure automatic recognition device 100. Yes.

  Then, the table structure automatic recognition apparatus 100 is an apparatus having a table structure automatic recognition function when the CPU 14 reads out and executes a table structure automatic recognition program from the auxiliary storage device 16. The table structure automatic recognition program may be stored in the main storage device 13 accessible to the CPU 14. The input device 11 receives data input under the control of the CPU 14. The table structure automatic recognition program can be recorded in the recording medium 18 that can be read by the table structure automatic recognition apparatus 100.

  Examples of the recording medium 18 readable by the table structure automatic recognition apparatus 100 include a magnetic recording medium, an optical disk, a magneto-optical recording medium, and a semiconductor memory. Examples of the magnetic recording medium include an HDD (Hard Disk Drive), a flexible disk (FD), and a magnetic tape (MT). Examples of the optical disc include a DVD (Digital Versatile Disc), a DVD-RAM (Random Access Memory), a CD-ROM (Compact Disc-Read Only Memory), and a CD-R (Recordable) / RW (ReWriteable). Magneto-optical recording media include MO (Magneto-Optical disk). When distributing the table structure automatic recognition program, it is conceivable to sell a portable recording medium 18 such as a DVD or a CD-ROM in which the table structure automatic recognition program is recorded.

  Then, the table structure automatic recognition apparatus 100 that executes the table structure automatic recognition program reads the table structure automatic recognition program from, for example, the recording medium 18 on which the driver device 17 has recorded the table structure automatic recognition program. The CPU 14 stores the read table structure automatic recognition program in the main storage device 13 or the auxiliary storage device 16.

  Then, the table structure automatic recognition device 100 reads the table structure automatic recognition program from the main storage device 13 or the auxiliary storage device 16 which is its own storage device, and executes processing according to the table structure automatic recognition program.

  FIG. 3 is a diagram illustrating an example of a functional configuration of the table structure automatic recognition apparatus. The table structure automatic recognition apparatus 100 according to the present embodiment includes a recognition processing unit 200 and a storage unit 300.

  The recognition processing unit 200 includes a rectangular area extraction unit 210, a heading item matching unit 220, a length correction unit 230, and a heading corresponding item specifying unit 240.

  The rectangular area extraction unit 210 extracts a table from the design document data stored in the design document database 310. The design document data of this embodiment is document data including a plurality of tables, text data, and the like, for example. The document data may be only a table. The design document database 310 of the present embodiment is a so-called document database, and stores design document data that is document data.

  The heading item collation unit 220 collates the data of the heading item of the extracted table with the data of the heading item of another table stored in the design document database 310, and extracts a similar table (hereinafter, similar table). . The length correction unit 230 corrects the length of the heading item and the length of the data item of the extracted table based on the length of the heading item and the length of the data item of the extracted table. The headline corresponding item specifying unit 240 specifies a data item corresponding to the headline item in the extracted table. Details of the processing of each unit will be described later.

  The storage unit 300 includes a design document database 310, a table extraction condition database 320, and an item matching database 330. The storage unit 300 is formed in the main storage device 13 and / or the auxiliary storage device 16, and each of the databases is provided in the main storage device 13 and / or the auxiliary storage device 16.

  Hereinafter, each database stored in the storage unit 300 will be described with reference to FIGS. 4 to 6. FIG. 4 is a diagram illustrating each database stored in the storage unit.

  A portion 41 shown in FIG. 4 describes the design document database 310. The design document data of the embodiment is divided for each document type as shown in a portion 41 and is stored in the design document database 310 for each document type. In this embodiment, the design document data is divided into document types A, B, and C and stored in the design document database 310. The document type is classified, for example, for each table format included in the design document data.

  A portion 42 shown in FIG. 4 explains the table extraction condition database 320. In the table extraction condition database 320 of this embodiment, for each document type, an identifier for identifying a table heading item included in the design document data and data of the table heading item are stored in association with each other. For example, in the table extraction condition database 320, identifiers for identifying the heading items 1 and 2 in the table format of the document type A and the data of the heading items 1 and 2 are stored as the table extraction conditions for each document document A. The table extraction condition database 320 stores the identifier of the heading item 1 and the data of the heading item 1 in the document type B table format.

  A portion 43 shown in FIG. 4 describes the item matching database 330. In the item matching database 330 of the present embodiment, data items expected to be consistent are stored in association with each other in tables included in design document data of different document types. For example, in the item matching database 330, data items 41 of document type A and data items 43 of document type B are stored in association with each other as data items expected to be consistent. A data item 44 included in the document type C is stored in association with the data item 43 that is expected to be consistent with the data item 43 of the document type B.

  FIG. 5 is a diagram illustrating an example of a table extraction condition database. In the table extraction condition database 320 of this embodiment, the document type, the header item identifier, and the header item data are stored in association with each other.

  In the example of FIG. 5, a list of heading items for each document type is stored. For example, the heading item list 321 is a list of heading items of the document type A, and the heading item list 322 is a list of heading items of the document type B. The data of the heading item A-1 in the heading item list 321 is “organization name / role name / charge name”. The data of the heading item B-1 in the heading item list 322 is “organization and role”. In this embodiment, the heading item list for each document type stored in the table extraction condition database 320 is a condition for extracting a table from design document data.

  FIG. 6 is a diagram illustrating an example of the item matching database. In the item matching database 330 of this embodiment, the heading items of data items expected to be consistent are stored in association with each other. For example, in the example of FIG. 6, in the heading item list 321 of the document type A and the heading item list 322 of the document type B, the heading item A-1 and the heading item B-1 are associated with each other. This indicates that there is a high possibility that the data item corresponding to the heading item A-1 and the data item corresponding to the heading item B-1 match.

  Next, the operation of the table structure automatic recognition apparatus 100 of this embodiment will be described with reference to FIG. FIG. 7 is a flowchart for explaining the overall operation of the table structure automatic recognition apparatus. Details of the processing of each step shown in FIG. 7 will be described later.

  In the table structure automatic recognition apparatus 100 of this embodiment, when the table structure automatic recognition process is started, the rectangular area extraction unit 210 extracts a rectangular area from the design document data to be processed (step S701). The rectangular area extracted here is a table included in the design document data.

  Subsequently, the table structure automatic recognition apparatus 100 uses the headline item list for each document type listed in the table extraction condition database 320 as a table extraction condition, and extracts a similar table that matches the extraction condition by the headline item matching unit 220 ( Step S702). Specifically, for example, a list of heading items included in the design document data of document type A is used as a similar table extraction condition, and a table in which the heading item list matches to some extent in document type A is extracted as a similar table. To do.

  Subsequently, the table structure automatic recognition apparatus 100 corrects the length of the extracted data item of the similar table by the data data item length correction unit 231 included in the length correction unit 230 (step S703). Next, the table structure automatic recognition apparatus 100 corrects the length of the extracted heading item of the similar table by the heading data item length correcting unit 232 of the length correcting unit 230 (step S704).

  Subsequently, the table structure automatic recognition apparatus 100 specifies the data item corresponding to the corrected heading item by the heading correspondence data specifying unit 240 (step S705). Subsequently, the table structure automatic recognition apparatus 100 outputs a list in which the heading item is associated with the data item (step S706).

  Details of the extraction of the rectangular area executed in step S701 will be described below with reference to FIG. FIG. 8 is a diagram for explaining extraction of a rectangular area.

  The design document data 80 shown in FIG. 8 is, for example, design document data of document type A. The design document data 80 includes documents 81, 82,. When the design document data 80 of the document type A is selected as a processing target, the rectangular region extraction unit 210 according to the present embodiment detects and extracts a rectangular region for each document included in the design document data 80. The shape area extraction unit 210 extracts the rectangular area 1 and the rectangular area 2 from the document 81, for example. The shape area extraction unit 210 extracts the rectangular area 3 and the rectangular area 4 from the document 82.

  FIG. 9 is a flowchart for explaining processing of the rectangular area extraction unit.

  The rectangular area extraction unit 210 of the present embodiment selects design document data to be processed from the design document database 310 (step S901). Subsequently, the rectangular area extraction unit 210 sets the pointer to the upper left of the document included in the design document data (step S902). Subsequently, the rectangular area extraction unit 210 scans the pointer to the right end of the document, and determines whether or not the upper left corner of the rectangular area exists (step S903).

  If the rectangular area extraction unit 210 detects a corner (step S904), the process proceeds to step S909 described later. If the corner is not detected in step S904, the rectangular area extraction unit 210 determines whether or not the position of the pointer is at the lower right of the document (step S905). If the position of the pointer is at the lower right of the document in step S905, the rectangular area extraction unit 210 determines whether there is another document included in the design document data (step S906).

  If there is no other document in step S906, the rectangular area extraction unit 210 ends the process. If another document exists in step S906, the rectangular area extraction unit 210 selects the next document (step S907) and returns to step S902. In step S905, when the position of the pointer is not in the lower right of the document, the rectangular area extraction unit 210 moves the pointer to the current next position and moves it to the left end (step S908), and returns to step S903. Specifically, for example, if the current pointer position is k, the rectangular area extraction unit 210 sets the pointer position to k + 1.

  When a corner is detected in step S904, the rectangular area extraction unit 210 scans a line extending right and down from the upper left corner, and detects a lower left corner and an upper right corner (step S909). Subsequently, the rectangular area extraction unit 210 scans a line extending below the upper right corner and detects the lower right corner (step S910). Subsequently, the rectangular area extraction unit 210 stores the position information of the rectangular area in the document (step S911). The position information is the coordinate values of the upper left corner, upper right corner, lower left corner, and lower right corner in the document. Subsequently, the rectangular area extraction unit 210 sets the pointer next to the upper right corner (step S912), and returns to step S903. Specifically, for example, if the current position of the pointer is m, the rectangular area extraction unit 210 sets the pointer to m + 1. In addition, the table | surface which is a rectangular area | region of a present Example has a cell of m column xn row, and was taken as the lump of the cell.

  Next, details of extraction of similar tables that match the table extraction conditions performed in step S702 will be described. FIG. 10 is a diagram for explaining extraction of a similar table that matches a table extraction condition.

  The headline item collation unit 220 of this embodiment extracts a similar table of the table extracted by the rectangular area extraction unit 210 that matches the table extraction condition based on the list of headline items stored in the table extraction condition database 320.

  In the example of FIG. 10, since the design document data 80 to be processed is the document type A, the heading item list 321 of the document type A stored in the table extraction condition database 320 is used as the table extraction condition. The headline item collation unit 230 according to the present embodiment compares the headline item list 321 and the headline item of the table extracted by the rectangular area extraction unit 210 to calculate a similarity, and extracts a table having a similarity greater than or equal to a predetermined threshold. To do. In the example of FIG. 10, rectangular areas 1 to 4 are extracted. In this embodiment, the rectangular area table extracted by the heading item matching unit 220 is called a similarity table.

  Details of the processing of the heading item matching unit 220 of this embodiment will be described below. FIG. 11 is a flowchart for explaining the processing of the heading item matching unit.

  The headline item collation unit 220 according to the present embodiment acquires a table extraction condition corresponding to the document type A of the design document data 80 to be processed from the table extraction condition database 320 (step S1101). The table extraction condition is specifically a heading item list 321 of the document type A. Subsequently, the headline item collation unit 220 sets the headline items included in the table extraction condition as m1, m2,..., Mn (step S1102).

  Subsequently, the heading item collation unit 220 acquires a similarity threshold of the table extraction condition (step S1103). This threshold value is stored in advance in the storage unit 300 of the table structure automatic recognition apparatus 100.

  Next, the headline item collation unit 220 sets h1, h2,..., Hn as rectangular regions corresponding to the table type of the document type A in the rectangular region extracted by the rectangular region extracting unit 210 (step S1104). Subsequently, the headline item matching unit 220 extracts one rectangular area (step S1105), divides it into sections, and sets each section as k1, k2,..., Kn (step S1106). A section is an area surrounded by a line in a rectangular area.

  Next, the headline item collation unit 220 extracts one headline item included in the headline item list 321 (step S1107), and the similarity between the extracted headline item and each character string in the sections k1, k2,. The degree is calculated (step S1108). The similarity in the present embodiment is a value based on the number of characters that match between the extracted headline item and the character string in the section. Details of the calculation of the similarity will be described later.

  Subsequently, the heading item matching unit 220 stores the calculation result (step S1109). The calculation result is stored in the storage unit 300 or the like as a set in which three values of a heading item, a section (any one of k1, k2,..., Kn) and similarity are associated. Subsequently, the heading item collation unit 220 determines whether or not there are remaining heading items (step S1110). If there are remaining heading items in step S1110, the heading item matching unit 220 extracts the next heading item (step S1111), and repeats the processing from step S1108 onward.

  If there is no remaining headline item in step 1110, the headline item collation unit 220 allocates the section having the highest similarity to the headline item list in the rectangular area so that the correspondence is 1: 1 (step 1112). . In step S1112, the pairs may be rearranged in descending order of similarity so that the heading item and the section are in a one-to-one relationship. Details of the processing in step S1112 will be described later.

  Subsequently, the headline item collation unit 220 obtains an average of the degrees of similarity in which the headline item and the section correspond to each other one by one, and when the average is equal to or greater than the threshold, the character string in the rectangular area, the headline item, and the section is obtained. It outputs as a similar table (step 1113). The heading item matching unit 220 determines whether or not there is a remaining rectangular area (step 1114). If there is a remaining rectangular area in step S1114, the next rectangular area is extracted (step 1115), and the process returns to step S1106. If there is no remaining rectangular area in step S1114, the heading item collation unit 220 ends the process.

  Hereinafter, with reference to FIG. 12, the calculation of the similarity and the association between the heading item and the section by the heading item matching unit 220 in steps S1108 and S1109 will be described. FIG. 12 is a diagram for explaining the calculation of the similarity and the association between the heading item and the section.

  FIG. 12 shows a case where, for example, the heading item list 321 which is the table extraction condition of the document type A is collated with the rectangular areas 1 to N. The rectangular area output from the document type A is divided into sections as shown in the area 121.

  The heading item matching unit 220 checks the number of characters that match the heading item A-1 in the heading item list 321 with respect to the character strings in the sections 1 to 8 of the rectangular area 1. Then, the similarity between the heading item A-1 and the section 1 is calculated by the similarity calculation formula shown in Expression (1), and the similarity is stored in the storage unit 300 or the like.

  Similarly, the heading item collation unit 220 checks the number of characters that match the heading item A-2 with respect to the character strings of the sections 1 to 8. And the similarity of heading item A-1 and the division 1 is calculated by the similarity calculation formula shown in Formula (1). The heading item collation unit 220 performs the above-described processing for all heading items included in the heading item list 321 for each section of the rectangular regions 1 to N.

  FIG. 13 is a first diagram illustrating an example of calculating the similarity for each heading item. In FIG. 13, the calculation result 131 of the similarity with heading item A-1 to A-4 with respect to the divisions 1-8 of the rectangular area 1 is shown. In the present embodiment, the heading item, section, and similarity shown in FIG. 13 are stored as one set.

  FIG. 14 is a second diagram illustrating an example of calculating the similarity for each heading item. In FIG. 14, the calculation result 141 shows the similarity between the heading items A-1 to A-4 for the sections 1 to 8 of the rectangular area N.

  Next, with reference to FIG. 15, the details of the allocation by the headline item collation unit 220 in step S1112 will be described. FIG. 15 is a flowchart illustrating details of the process in step S1112.

  The headline item collation unit 220 sorts the sets stored in step S1109 in descending order of similarity (step S1501). Subsequently, the heading item matching unit 220 sets the pointer to the first set (step S1502). Subsequently, the heading item collation unit 220 outputs the heading item of the set pointed to by the current pointer, the section, and the similarity (step S1503). That is, the set having the highest similarity between the heading item and the section is output. In this embodiment, this output is a one-to-one allocation result.

  Subsequently, the heading item collation unit 220 deletes the output set from the similarity calculation result (step S1504). In this example, the information is deleted, but an output prohibition flag may be attached to the corresponding group so that it is not output. Subsequently, the heading item collation unit 220 advances the pointer by one (step S1505). If there is a pair pointed to by the pointer in step S1505 (step S1506), the headline item collation unit 220 returns to step S1503. In step S1505, if there is no pair pointed to by the pointer, the headline item matching unit 220 ends the process of step S1112.

  FIG. 16 is a diagram illustrating an example in which a section having the highest similarity is assigned to the heading item list so that the correspondence is 1: 1. In FIG. 16, the result of performing the processing described with reference to FIG. 15 on the calculation result 131 is shown as the allocation result 161.

  In the allocation result 161, the section most similar to the heading item A-1 in the rectangular area 1 is the section 2, the section most similar to the heading item A-2 is the section 3, and is most similar to the heading item A-3. The section is section 4, and the section most similar to the heading item A-4 is section 1. In the present embodiment, groups other than the allocation result 161 may be deleted from the storage unit 300.

  The headline item collation unit 220 according to the present embodiment obtains the above-described allocation result for all the rectangular areas extracted from the design document data 80 of the document type A. Then, the headline item collation unit 220 extracts a rectangular area in which the average value of the similarity included in the allocation result 161 is equal to or greater than the threshold, and stores the extracted rectangular area in the storage unit 300 as a similarity table. For example, in the case of the allocation result 161 illustrated in FIG. 16, the rectangular area 1 is held as the similarity table 1 when the average value of the four similarities 1, 1, 0.8, and 0.5 is equal to or greater than the threshold value.

  Next, details of the correction of the length of the data item of the similar table performed in step S703 will be described. In the table structure automatic recognition apparatus 100 of the present embodiment, the data item length correction unit 231 corrects the length of the extracted data item of the similar table. FIG. 17 is a flowchart for explaining processing by the data item length correction unit.

  The data item length correction unit 231 prepares a similarity table corresponding to the document type of the processing result in step S702 (step S1701). Subsequently, the data item length correction unit 231 takes out one similarity table (step S1702). Subsequently, the data item length correction unit 231 divides the similarity table for each section (step S1703). Subsequently, the data data item length correcting unit 231 acquires data item data for each section and stores the length of the data item in the storage unit 300 or the like (step S1704). The length of the data item is the length of the section corresponding to the data item, and specifically indicates the number of characters in the section.

  Subsequently, the data data item length correction unit 231 determines whether or not there are remaining similar tables (step S1705). If a similar table remains in step S1705, the data item length correction unit 231 extracts the next similar table (step S1706) and returns to step S1703. If there is no remaining similar table in step S1706, the length of the data item in the similar table is normalized using the length of the acquired data item (step S1707). Specifically, the data item length correction unit 231 may obtain an average value of the data length for each data item of each similar table and correct the length of each data item to be this average value. .

  FIG. 18 is a first diagram illustrating a specific example of correcting the length of a data item. FIG. 18 shows a case where rectangular areas 1 to 4 of the document type A are extracted as the similarity table. FIG. 18A is a diagram showing the correspondence between the heading item and the data item, and FIG. 18B is a diagram for explaining the correction of the length of the data item.

  The data item length correction unit 231 according to the present embodiment sets a partition other than the partition to which the heading items A-1 to A-4 are allocated by the heading item matching unit 220 as a data item partition. For example, in the similarity table of document type A, as shown in FIG. 18A, heading items A-1 to A-4 are allocated to sections 1 to 4. Therefore, the sections 5 to 8 are sections of the data items 1 to 4 corresponding to the heading items A-1 to A-4.

  As shown in FIG. 18B, the data item length correction unit 231 acquires the number of characters in the sections 5 to 8 of the rectangular areas 1 to 4 as the length of the data item in each section. In the rectangular area 1, since the character string in the section 5 is 9 characters, the length of the data item in the section 5 is 9. In this embodiment, a space, a code, etc. are counted as one character. Since the number of characters in the section 6 of the rectangular area 1 is 4, the length of the data item in the section 6 is 4.

  The data item length correction unit 231 holds the data item lengths of the sections 5 to 8 in the storage unit 300 that has acquired the extracted rectangular regions (similar tables).

  FIG. 19 is a second diagram illustrating a specific example of correction of the length of a data item. FIG. 19 shows an example in which the length of the data item for each rectangular area is corrected from the length of the data item corrected by the data item length correction unit 231 in FIG. 19A shows an example of the average value of the lengths of the data items in the sections 5 to 8, and FIG. 19B shows an example in which the length of the data items in the rectangular area 1 is corrected.

  In this embodiment, the average of the lengths of the data items in each section is the length of the corrected data item. As shown in FIG. 19A, the total length of the data items of the section 5 in the rectangular areas 1 to 4 is 36. Therefore, the data item length correction unit 231 uses the average value 9 as the length of the corrected data item in the section 5. The data item length correction unit 231 similarly calculates the length of the corrected data item for the sections 6 to 8. In the calculation of the average value, the value of the second decimal place is rounded up. The calculated length of the corrected data item is stored in the storage unit 300 or the like in association with each section.

  The data item length correction unit 231 corrects the length of the character strings in the sections 5 to 8 based on the corrected data item length. By this correction, for example, as shown in FIG. 19B, the length of the character string of the data item in the section 5 of the rectangular area 1 becomes 9, the length of the character string of the data item in the section 6 becomes 4, and the section The length of the character string of the data item 7 is 4.5, and the length of the character string of the data item of the section 8 is 3.5.

  Next, details of the correction of the length of the heading item in the similarity table performed in step S704 will be described. In the table structure automatic recognition apparatus 100 of the present embodiment, the heading item length correction unit 232 corrects the length of the extracted heading item of the similar table. FIG. 20 is a flowchart for explaining processing by the heading item length correction unit.

  The heading item length correction unit 232 according to the present exemplary embodiment acquires a table extraction condition for a document type that is a target of heading item length correction (step S2001). Here, the document type to be corrected is the document type A, and the table extraction condition of the document type A is the target table extraction condition. That is, the target table extraction condition is the heading item list 321 of the document type A.

  Subsequently, the heading item length correction unit 232 refers to the item matching database 330 and acquires a heading item list (table extraction condition) of the document type expected to be consistent with the document type A (step S2002). Here, the document type expected to be consistent with the document type A is the document type B, and the table extraction condition of the document type B is the correspondence table extraction condition. That is, the correspondence table extraction condition is a heading item list 322 of the document type B.

  Subsequently, the heading item length correction unit 232 prepares a similar table extracted from the design document data of the document type B based on the correspondence table extraction condition (step S2003). Next, the heading item length correction unit 232 extracts one heading item from the target table extraction condition (step S2004). Specifically, for example, one heading item is extracted from the heading item list 321.

  Subsequently, the heading item length correction unit 232 determines whether or not a heading item corresponding to the document type for which consistency is expected exists in the correspondence table extraction condition (step S2005). Specifically, for example, it is determined whether or not there is a heading item that can be expected to be consistent with the heading item extracted in step S2004 in the heading item list 322 of the document type B.

  If there is a corresponding heading item in step S2005, the heading item length correction unit 232 calculates an average value of the lengths of heading items that match the corresponding heading item in the similar table for the correspondence table extraction condition ( Step S2006). Specifically, for example, a headline item that can be expected to be consistent with the headline item extracted in step S2004 is extracted from the similar table extracted from the design document data of document type B using the headline item list 322 as a table extraction condition. Calculate the average number of characters in the heading item. If no corresponding heading item exists in step S2005, the process proceeds to step S2008 described later.

  Subsequently, the heading item length correction unit 232 corrects the length of the heading item extracted in step S2004 using the average value of the heading item calculated in step S2006 (step S2007). Subsequently, the heading item length correction unit 232 determines whether or not there are remaining heading items (step S2008). If there are remaining heading items in step S2008, the heading item length correction unit 232 extracts the next heading item (step S2009) and returns to step S2005. If there is no remaining heading item in step S2008, the heading item length correction unit 232 ends the process.

  Hereinafter, the correction of the length of the heading item will be described in detail with reference to FIGS. FIG. 21 is a first diagram illustrating a specific example of correcting the length of a heading item.

  FIG. 21 shows a case where the length of the heading item in the similarity table of document type A is corrected. Based on the item matching database 330, the heading item length correction unit 232 determines that the heading item A-1 of the document type A and the heading item B-1 of the document type B are expected to be consistent. The heading item length correction unit 232 determines that the heading item A-3 and the heading item C-1 of the document type C are expected to be consistent.

  Then, the heading item length correction unit 232 acquires the heading item list 322 of the document type B from the table extraction condition database 320 as the table extraction condition. Then, the processing in steps S701 and S702 in FIG. 7 is performed on the design document data of the document type B, and the similarity table of the document type B is extracted. The heading item length correction unit 232 also acquires the heading item list 323 for the document type C as a table extraction condition, and extracts a similar table of the document type C.

  FIG. 22 is a second diagram illustrating a specific example of correction of the length of a heading item. When the heading item length correction unit 232 extracts the similar table of the document type B and the similar table of the document type C, the heading item length correction unit 232 extracts one heading item from the heading item list 321 of the document type A and extracts the heading item into the document types B and C. It is determined whether there is a heading item that can be expected to be consistent with the heading item.

  In FIG. 22, for example, when the heading item A-1 is extracted from the heading item list 321, the heading item A-1 can be expected to be consistent with the heading item B-1, and the heading item A-3 is consistent with the heading item C-1. It turns out that sex can be expected. It can also be seen that there are no heading items that can be expected to be consistent with the heading items A-2 and A-4 in the document types B and C.

  FIG. 23 is a third diagram illustrating a specific example of correcting the length of a heading item. When the heading item that can be expected to be consistent with the heading item of the document type A is found in the document types B and C, the heading item length correction unit 232 calculates the average length of the corresponding heading item in the similarity table of each document type. calculate.

  23A is a diagram for explaining the calculation of the average value of the lengths of the heading items B-1 in the similar table of document type B. FIG. 23B shows the heading items in the similar table of document type C. It is a figure explaining calculation of the average value of the length of C-1.

  As shown in FIG. 23A, the document type B is a heading item in which the heading item B-1 is expected to be consistent with the heading item A-1. Therefore, the heading item length correction unit 232 extracts the number of characters of the heading item B-1 from each similar table of the document type B, and obtains this average value. In the example of FIG. 23A, three similar tables are extracted from the document type B. Therefore, the heading item length correction unit 232 calculates an average value of the lengths (number of character strings) of the heading item B-1 of the three similar tables.

As shown in FIG. 23B, the document type C is a heading item in which the heading item C-1 is expected to be consistent with the heading item A-3. Therefore, the heading item length correction unit 232 extracts the number of characters of the heading item C-1 from each similarity table of the document type C, and obtains this average value. This average value is stored in the storage unit 300 or the like.
In the example of FIG. 23B, three similar tables are extracted from the document type C. Therefore, the heading item length correction unit 232 calculates the average value of the lengths (number of character strings) of the heading item C-1 of the three similar tables. In the calculation of the average value, the numbers after the decimal point are rounded down.

  FIG. 24 is a fourth diagram illustrating a specific example of correcting the length of a heading item. FIG. 24A is a diagram for explaining the correction method, and FIG. 24B is a diagram showing an example in which the heading item in the rectangular area 1 is corrected.

  The heading item length correction unit 232 of this embodiment uses the length of the heading item of the document type A and the average value of the lengths of heading items expected to be consistent with other document types. Correct the length of the heading item. Specifically, as shown in FIG. 24A, the length of the heading item of the document type A and 1/2 of the sum of the average lengths of heading items expected to be consistent with other document types. The value is the length of the heading item after correction.

  In FIG. 24, there are a heading item A-1 having a heading item expected to be consistent with the heading item B-1 of the document type B, and a heading item expected to be consistent with the heading item C-1 of the document type C. A certain heading item A-3 is corrected. The length of the heading item A-1 is 15, and the average length of the heading item B-1 is 11. Therefore, the corrected length of the heading item A-1 is 13.

  The length of the heading item A-3 is 6, and the average length of the heading item C-1 is 4. Therefore, the corrected length of the heading item A-3 is 5.

  As a result of the correction, the heading item in the rectangular area 1 has a length of 13 for the heading item A-1, a length of 3 for the heading item A-2, and a heading item A-3 as shown in FIG. Is 5 and heading item A-4 is 2 in length.

  In this embodiment, for example, when the heading item expected to be consistent with the heading item A-1 of the document type A does not exist in the heading item list of the other document types, the document type similar to the document type A and the table format. The headline item is corrected using the average value of the number of characters of similar headline items. For example, in the item matching database 330, the heading item A-4 of the document type A does not have a heading item that is expected to be consistent. Therefore, the heading item length correction unit 232 includes a heading item B- similar to the heading item A-4 in the heading item list 322 of the document type B including the heading item expected to be consistent with the heading item A-1. The average value of the number of characters of 5 and the number of characters of the heading item A-4 may be used for the correction of the heading item A-4.

  In this embodiment, the description has been made assuming that both the correction of the length of the heading item and the correction of the length of the data item are performed, but the present invention is not limited to this. For example, the length of either the heading item or the data item may be corrected. When either one of the corrections is performed, it is preferable to correct the length of the data item.

  Next, specific details of the data item corresponding to the corrected heading item performed in step S705 will be described. In the table structure automatic recognition apparatus 100 of the present embodiment, the headline corresponding data specifying unit 240 specifies a data item corresponding to the corrected headline item. FIG. 25 is a flowchart for explaining processing by the headline-corresponding data specifying unit.

  The headline-corresponding data specifying unit 240 acquires a corrected similarity table in which the data item and the headline item are corrected (step S2501). Subsequently, the headline corresponding data specifying unit 240 extracts one headline item from the acquired similarity table (step S2502). Subsequently, the headline-corresponding data specifying unit 240 calculates the area of the created region based on the combination of the extracted headline item and each data item in the similar table (step S2503).

  Specifically, the headline-corresponding data specifying unit 240 regards the area created by the heading item and the data item as a trapezoid having the length of the heading item and the length of the data item as the upper side and the lower side, respectively. (Length + data item length) × height ÷ 2. At this time, the height may be calculated, for example, assuming that the height of one section is 1.

  The headline-corresponding data specifying unit 240 performs the calculation of this area for all data items for one heading item, and saves the data in the storage unit 300 or the like as a set in which the heading item, the data item, and the area are associated with each other. (Step S2504).

  The headline corresponding data specifying unit 240 determines whether or not there are remaining headline items (step S2505). If there is a remaining headline item in step S2505, the headline-corresponding data specifying unit 240 extracts the next headline item (step S2506) and returns to step S2503.

  If there is no remaining headline item in step S2505, the headline-corresponding data specifying unit 240 excludes the calculation result corresponding to the predetermined condition (step S2507). In the present embodiment, the calculation result of one data item corresponding to a plurality of heading items and the calculation result in which the number of intersections between the heading item and the line segment connecting the data items exceeds a certain threshold are excluded. The constant threshold value of the number of intersections may be given as a fixed ratio with respect to the number of heading items in the similarity table, for example. The constant threshold value for the number of intersections may be changed according to the number of heading items. The constant threshold value of the number of intersections is stored in advance in the storage unit 300 or the like.

  Subsequently, the headline corresponding data specifying unit 240 selects a group having the smallest area for each headline item from the calculation result (step S2508). The heading item corresponding data item specifying unit 240 outputs the selected set of heading items and data items as a final result (step S2509). By the processing of the heading item corresponding data item specifying unit 240, the data item corresponding to the heading item in the similar table is specified.

  FIG. 26 is a diagram for explaining identification of a data item corresponding to a heading item. In FIG. 26, when the constant threshold of the number of intersections is set to 1, for example, FIG. 26A shows a case where the number of intersections of the line segment connecting the heading item and the data item exceeds the predetermined threshold. B) shows the case where the number of intersections of the line connecting the heading item and the data item does not exceed a certain threshold.

  In FIG. 26A, the section corresponding to the data item that creates the area having the smallest area with respect to the section 261 corresponding to the heading item is the section 262. However, the line segment H1 connecting the heading item of the section 261 and the data item of the section 262 intersects with the line segment connecting the other heading item and the data item at three points. Therefore, since the number of intersections is 3 and exceeds a certain threshold, the data item in the partition 262 is not selected as the data item corresponding to the heading item in the partition 261.

  In FIG. 26B, when the data item whose number of intersections exceeds a certain threshold value is excluded, the partition that creates the header item of the partition 261 and the area of the smallest area is the partition 263. A line segment H2 connecting the heading item of the section 261 and the data item of the section 263 does not intersect with a line segment connecting another heading item and the data item. Therefore, the number of intersections is 0 and does not exceed a certain threshold value, so the data item of the partition 263 is specified as a data item corresponding to the heading item of the partition 261.

  As described above, in this embodiment, in addition to the area of the area created by the heading item and the data item being minimized, the number of intersections of line segments connecting the heading item and the data item is limited. This reflects the observation that in the interpretation of the table, the greater the number of intersections, the greater the cognitive burden.

  In the present embodiment, the accuracy of automatic recognition of the table structure can be improved by providing a restriction on the number of intersections in this way.

  Next, an input format and an output format of the table structure when the table structure automatic recognition apparatus 100 according to this embodiment performs automatic table structure recognition will be described.

  FIG. 27 is a diagram illustrating a table input format. FIG. 27A shows an example of a table, and FIG. 27B shows an example of an input format of the table structure shown in FIG.

  In this embodiment, the heading item is represented by “H: coordinate: character string length”, the data item is represented by “coordinate: character string length”, and the height of the region is represented by “coordinate Y: height”. In this embodiment, the coordinates are expressed in, for example, MSExcel (registered trademark) format.

  In table 2 shown in FIG. 27A, sections indicated by coordinates F2, G2, G3, and H3 are heading items, and a character string of the heading item is input to each section. In table 2, the remaining sections are data items, and a character string of the data item is input to each section. Therefore, table2 indicates the heading item as “H: F2: 2 H: G2: 3 H: G3: 2” as shown in FIG. Data items without H: at the head are data items, and the data items are indicated by “F5: 4 G4: 16 G5: 13 H5: 5”. In addition, the height of each section is indicated by “2: 1 3: 1 4: 2 5: 2”. For example, a Y coordinate of 2 is interpreted as a height of 1.

  FIG. 28 is a diagram illustrating an input example for associating a heading item with a data item. When inputting the table structures of a plurality of tables to the table structure automatic recognition apparatus 100 of the present embodiment, the table structures of the plurality of tables may be input as one data as shown in FIG. FIG. 28 shows data including a table structure from table1 to table6.

  Next, with reference to FIG. 29, a format for associating heading items and data items output from the table structure automatic recognition apparatus 100 of the present embodiment will be described. FIG. 29 is a diagram illustrating an example of the result of association.

  In this embodiment, the result of associating the heading item with the data item is output in the form of “number of intersections # area size # individual area 1 # individual area 2 #... # Individual area n”. The Each area is represented by “individual area size_area height_heading item_data item”, the heading item is represented by “H: coordinate: character string length”, and the data item is represented by “coordinate: It is represented by “character string length”. The area size is an area of a rectangular area forming the table, and each area size is an area for each partition included in the table.

  For example, in the table 2 in FIG. 29, the number of intersections is 1, the area size is 106.5, the area size of the area 1 is 18, the height of the area is 6, the coordinates of the heading item are F2, the length of the heading item is 2, The coordinate of the data item corresponding to the item is F5, and the length of the data item is 4. The area size of area 2 is 24, the area height is 6, the coordinates of the heading item are G2, the length of the heading item is 3, the coordinate of the data item corresponding to the heading item is H5, and the length of the data item is 5 It becomes.

  As described above, in this embodiment, the data item corresponding to the heading item can be specified only by inputting the coordinates and the character string length of the heading item and the data item. Therefore, automatic table recognition can be performed with high accuracy by a simple table structure definition.

The present invention includes configurations as described in the following supplementary notes.
(Appendix 1)
Extracting the plurality of tables from a document database storing document data including a plurality of tables including a plurality of partitions;
With reference to the table extraction condition database in which the first heading of the table to be extracted is stored, the first heading is collated with the data of the section of the heading item included in each of the plurality of tables,
Extracting a table in which the result of the matching satisfies a predetermined condition;
Based on the length of each section included in the table satisfying the predetermined condition, the length of the section of the heading item used for associating the section of the heading item with the section of the data item in the table or data corresponding to the heading item An automatic table structure recognition program for causing a computer to execute processing for correcting at least one of the lengths of the section of an item.
(Appendix 2)
In the correction process,
For each section of the data item included in the table that satisfies the predetermined condition, obtain the length of the section of the data item to normalize the length of the section of the data item,
The program for automatically recognizing a table structure according to supplementary note 1, which causes a computer to execute a process of correcting a length of a section of a data item used for the association to the normalized value.
(Appendix 3)
In the correction process,
With reference to the item matching database in which the headings that match the data of the data item section are stored in association with each other, the second heading associated with the first heading is obtained;
Collating the first heading and the second heading with the data of the section of the heading item included in each of the plurality of tables;
Extracting a second table in which the result of the matching satisfies a predetermined condition;
The table according to appendix 1 or 2, which causes the computer to execute processing for correcting the length of the section of the heading item used for the association using the average value of the length of the section of the heading item included in the second table. Automatic structure recognition program.
(Appendix 4)
Obtaining a line connecting each of the sections of the heading item of the table satisfying the predetermined condition and each of the sections of the data item;
Identify the combination of the heading item and the data item of the table in which the number of times the line segments cross each other is a predetermined number of times or less;
The table structure automatic recognition program according to any one of appendices 1 to 3, which causes a computer to execute a process of associating and outputting the identified heading item and data item.
(Appendix 5)
The table structure automatic recognition program according to appendix 4, wherein the predetermined number of times is set to be a predetermined ratio with respect to the number of heading items in the table that satisfy the predetermined condition.
(Appendix 6)
In the verification,
Processing for calculating the similarity of each of the plurality of tables based on the ratio of the heading item data stored in the table extraction condition database and the heading item data of the plurality of tables extracted from the document database When,
The table structure automatic recognition program according to any one of appendices 1 to 5, which causes a computer to execute a process of extracting a table having the similarity equal to or greater than a predetermined threshold from the plurality of tables.
(Appendix 7)
A table structure automatic recognition method in which a computer automatically recognizes a table structure,
Extracting the plurality of tables from a document database storing document data including a plurality of tables including a plurality of partitions;
With reference to the table extraction condition database in which the first heading of the table to be extracted is stored, the first heading is collated with the data of the section of the heading item included in each of the plurality of tables,
Extracting a table in which the result of the matching satisfies a predetermined condition;
Based on the length of each section included in the table satisfying the predetermined condition, the length of the section of the heading item used for associating the section of the heading item with the section of the data item in the table or data corresponding to the heading item A table structure automatic recognition method for correcting at least one of the lengths of item sections.
(Appendix 8)
A table structure automatic recognition device that automatically recognizes a table structure,
A rectangular area extraction unit for extracting the plurality of tables from a document database storing document data including a plurality of tables including a plurality of sections;
Referencing the table extraction condition database in which the first heading of the table to be extracted is stored, collating the first heading with the data of the section of the heading item included in each of the plurality of tables, and A header item matching unit that extracts a table in which the result of
Based on the length of each section included in the table satisfying the predetermined condition, the length of the section of the heading item used for associating the section of the heading item with the section of the data item in the table or data corresponding to the heading item A table structure automatic recognition apparatus comprising: a length correction unit that corrects at least one of the lengths of the sections of items.

DESCRIPTION OF SYMBOLS 100 Table structure automatic recognition apparatus 200 Recognition processing part 210 Rectangular area extraction part 220 Heading item collation part 230 Length correction part 231 Data item length correction part 232 Heading item length correction part 240 Heading corresponding | compatible data specification part 300 Storage part 310 Design Database 320 Table extraction condition database 330 Item consistency database

Claims (6)

Extracting the plurality of tables from a document database storing document data including a plurality of tables including a plurality of partitions;
With reference to the table extraction condition database in which the first heading of the table to be extracted is stored, the first heading is collated with the data of the section of the heading item included in each of the plurality of tables,
Extracting a table in which the result of the matching satisfies a predetermined condition;
Based on the length of each section included in the table satisfying the predetermined condition, the length of the section of the heading item used for associating the section of the heading item with the section of the data item in the table or data corresponding to the heading item An automatic table structure recognition program for causing a computer to execute processing for correcting at least one of the lengths of the section of an item. In the correction process,
For each section of the data item included in the table that satisfies the predetermined condition, obtain the length of the section of the data item to normalize the length of the section of the data item,
The program for automatically recognizing a table structure according to claim 1, which causes a computer to execute a process of correcting a length of a section of a data item used for the association to the normalized value. In the correction process,
With reference to the item matching database in which the headings that match the data of the data item section are stored in association with each other, the second heading associated with the first heading is obtained;
Collating the first heading and the second heading with the data of the section of the heading item included in each of the plurality of tables;
Extracting a second table in which the result of the matching satisfies a predetermined condition;
The computer according to claim 1, wherein the computer executes a process of correcting the length of the section of the heading item used for the association using the average value of the length of the section of the heading item included in the second table. Table structure automatic recognition program. Obtaining a line connecting each of the sections of the heading item of the table satisfying the predetermined condition and each of the sections of the data item;
Identify the combination of the heading item and the data item of the table in which the number of times the line segments cross each other is a predetermined number of times or less;
The table structure automatic recognition program according to any one of claims 1 to 3, which causes a computer to execute a process of outputting the identified heading item and a data item in association with each other. A table structure automatic recognition method in which a computer automatically recognizes a table structure,
Extracting the plurality of tables from a document database storing document data including a plurality of tables including a plurality of partitions;
With reference to the table extraction condition database in which the first heading of the table to be extracted is stored, the first heading is collated with the data of the section of the heading item included in each of the plurality of tables,
Extracting a table in which the result of the matching satisfies a predetermined condition;
Based on the length of each section included in the table satisfying the predetermined condition, the length of the section of the heading item used for associating the section of the heading item with the section of the data item in the table or data corresponding to the heading item A table structure automatic recognition method for correcting at least one of the lengths of item sections. A table structure automatic recognition device that automatically recognizes a table structure,
A rectangular area extraction unit for extracting the plurality of tables from a document database storing document data including a plurality of tables including a plurality of sections;
Referencing the table extraction condition database in which the first heading of the table to be extracted is stored, collating the first heading with the data of the section of the heading item included in each of the plurality of tables, and A header item matching unit that extracts a table in which the result of
Based on the length of each section included in the table satisfying the predetermined condition, the length of the section of the heading item used for associating the section of the heading item with the section of the data item in the table or data corresponding to the heading item A table structure automatic recognition apparatus comprising: a length correction unit that corrects at least one of the lengths of the sections of items.

Images