JP6786658B2 - Document reading system - Google Patents

Description

An embodiment of the present invention relates to a technique for collectively reading a plurality of documents such as an application form and converting them into a character code that can be used by a computer.

Character recognition technology such as OCR (Optical Character Recognition / Reader) is utilized as a tool for digitizing documents and the like.

Japanese Unexamined Patent Publication No. 2002-269253

Provided is a document reading system that reads a plurality of documents having different document types, assigns a character recognition engine suitable for the read document type, and converts them into character codes that can be used by a computer.

The document reading system of the embodiment includes an image generation device that generates image data of a plurality of documents having different document types, and a management device that performs character recognition processing on the image data. The management device registers a plurality of character recognition engines for the character recognition process, and registers one or a plurality of document types with each character recognition engine, and a master control unit associating the document types with the image data. Character recognition processing using the recognition engine allocation control unit that discriminates and assigns the character recognition engine associated with the determined document type to each image data, and the character recognition engine assigned to each image data. It has a recognition processing unit for performing the above. The character recognition engine can include an OCR form recognition engine, a standard form recognition engine, an atypical form recognition engine, and an identity verification document recognition engine, and the recognition engine allocation control unit performs the following processing. The image data of a document to which measures for improving recognition accuracy are applied and / or a document to which a template definition linked with a form ID printed on paper is applied is assigned to the OCR form recognition engine and sorted to the OCR form recognition engine. The image data of a document that has not been created and has a character recognition template pre-registered for each document title is assigned to the standard form recognition engine and sorted into the OCR form recognition engine and the standard form recognition engine. The image data of a document that does not exist and has a character recognition template pre-registered for each item or keyword in the document is assigned to the atypical form recognition engine, and the OCR form recognition engine is used. , Documents that are not sorted by the standard form recognition engine and the non-standard form recognition engine and include the characteristic information of the identity verification document are assigned to the identity verification document recognition engine.

It is the schematic explanatory drawing of the document reading system of 1st Embodiment. It is a figure for demonstrating each process of document type determination and recognition engine selection of the document reading system of 1st Embodiment. It is a block diagram of the document reading management apparatus which comprises the document reading system of 1st Embodiment. In the first embodiment, it is a figure for demonstrating the allocation process of the recognition engine in the case where a plurality of areas having different document types are mixed in one document. It is explanatory drawing of the batch reading process which collectively reads each document of a plurality of different applicants of 1st Embodiment. It is a figure which shows the processing flow of the document reading system of 1st Embodiment. It is a figure which shows the detailed flow of the OCR process of the document reading system of 1st Embodiment. It is a figure which shows an example of the workflow which applied the document reading system of 1st Embodiment. It is a figure which shows an example of the workflow following FIG. It is a figure which shows an example of the workflow following FIG. It is a figure which shows an example of the document (form) of 1st Embodiment. It is a figure which shows an example of the document (form) of 1st Embodiment. It is a figure which shows an example of the document (form) of 1st Embodiment.

Hereinafter, embodiments will be described with reference to the drawings.

(First Embodiment)
FIG. 1 is a schematic explanatory view of a document reading system to which the document reading management device (hereinafter, referred to as a management device) 100 according to the first embodiment is applied. This system digitizes paper-based documents such as tax returns and applications by OCR processing (character recognition processing).

For example, the applicant (filer or submitter) submits application documents (submission documents such as declaration documents and application documents). At that time, the applicant (submitter) can bring the application documents (submission documents) to a predetermined window, or request a delivery company to submit them by a predetermined shipping means. An image generation device 200 is provided at a place where the application documents are received, and the application documents converted into image data by the image generation device 200 are input to the management device 100 and converted into character codes (text).

The image generation device 200 is a scanner that scans a document on a paper medium to generate image data (image data). The image generation device 200 can read a plurality of documents at once, scans a bundle of set documents one by one, and generates image data for each. The data format of the image data is arbitrary, and various formats such as JPEG, GIF, PNG, TIFF, and BMP can be appropriately adopted. Further, the scanned image data may be a PDF format in which a plurality of image data are aggregated into one file at page breaks.

The image generation device 200 may be, for example, a scanning device having a sorter function, a printer or a multifunction device equipped with a scanner function. Further, an image pickup device such as a camera or a terminal (desktop PC, notebook PC, tablet terminal, smartphone) provided with the image pickup device can also be applied as the image generation device 200.

Documents are paper media used in filing, applications, claims, etc., such as tax returns, medical expenses and allowance applications, insurance applications, insurance payment claims, mortgage applications, and financial product transaction applications. These documents can be classified into predetermined document types described later. Documents include printed or handwritten documents and copies.

The document type is associated with a recognition engine described later, and one document type can be associated with one recognition engine or a plurality of recognition engines. On the other hand, one recognition engine can be associated with one or more document types. The document type of the present embodiment is used as a distribution condition of the recognition engine, and is information for controlling so that the document of the predetermined document type is distributed to the predetermined recognition engine.

The documents handled in this embodiment are one or more documents for each application. For example, in one application, a set of application documents including an application form, a statement of accounts, a breakdown, attached documents such as identity verification documents, and a certificate may be submitted. In this case, there are a plurality of documents having different document types in the application document set. Therefore, the application documents for each application unit or each applicant can be a document group (bundle of documents) containing a plurality of documents of different document types.

On the other hand, for example, even if it is an insurance application form, the application form may differ depending on the type of insurance such as life insurance, automobile insurance, accident insurance, etc., or the application form may differ depending on the insurance company. The same applies to documents submitted to local governments and government agencies. Therefore, the document group set in the image generation device 200 is also a document group including a plurality of documents of different document types for each application unit or each applicant. In addition, the submitted documents may be completed with one application form. In this case, a plurality of document groups having different applicants and different document types are set in the image generator 200.

As described above, the image generation device 200 of the present embodiment sets a bundle of documents including a plurality of documents of different document types, and sequentially generates each image file by batch continuous scanning. A bundle of documents may be a set of documents submitted by one applicant containing multiple documents of different document types, and a group of multiple documents of the same application form but with different document types. It may be.

The image generator 200 is provided at a store or a base that handles documents. The image data of the document generated by the image generator 200 is transmitted to the management device 100 via a network such as an IP network or a dedicated network (network connection method), or the image generation device is connected to the management device 100 by a cable or the like. The 200 can be directly connected and transmitted (local connection method). Further, the image generation device 200 can be locally connected to the user terminal 300, and in this case, the image data generated by the image generation device 200 is transmitted to the management device 100 through the user terminal 300. can do.

FIG. 2 is a diagram for explaining each process of document type determination and recognition engine selection of the document reading system of the present embodiment. In the present embodiment, the types of documents to be handled and the recognition engine (OCR processing module) suitable for each document type can be registered. Then, the document type is determined for each image data of the document, and a recognition engine for performing OCR processing is assigned to the document. That is, the documents to be OCR processed are set in advance for each of a plurality of recognition engines having different recognition characteristics, and the documents are sorted to the recognition engines that match the recognition characteristics.

In the present embodiment, a recognition engine is assigned to each image data of the document, and OCR processing by the recognition engine is performed in units of image data (page units). In addition, among a plurality of documents, there may be a document to which the same recognition engine is assigned, or a different recognition engine may be assigned to all the documents. Of course, when documents are manually sorted according to the document type in advance and digitized in the sorting unit, the documents are read in the sorting unit and the same recognition engine is assigned to each of a plurality of documents.

Here, the OCR process will be described. The OCR process is a process of converting a paper document on which characters are written into image data (image data), recognizing the characters in the image data and generating character code information, and can be roughly divided into two processes. .. The first is a character extraction analysis process in which layout analysis is performed in the image data, the character area extracted by the layout analysis is extracted, and the characters in the character area are decomposed character by character. The second is a recognition process for recognizing what character the image data of each character obtained by the character extraction analysis process is. It is also possible to recognize characters without decomposing the character strings in the character area into characters one by one.

The recognition process includes, for example, normalization processing for enlarging / reducing a character image to a certain size, and features of a character image (features such as vertical, horizontal, and diagonal lines, length of each line, and positional relationship between lines). Each of extraction processing, matching processing for searching for a character code that matches the feature amount of a character image, and automatic correction processing using a dictionary (word dictionary, etc.) for each character code and / or its character string extracted by the matching processing. Processing is included. Through these processes, the characters in the document are output as character codes. Character-encoded information, that is, character text information can be output for each document in a predetermined file format (txt, csv, xml, docx, xlsx, etc.).

The recognition engine of this embodiment is a program module including each algorithm of character extraction analysis processing and recognition processing, and a plurality of recognition engines whose character extraction analysis processing and / and recognition processing are tuned according to the characteristics of each document. be registered.

In the example of FIG. 2, the OCR form recognition engine performs character recognition processing on a document called an OCR form, which is created on the premise of OCR processing. The OCR form is a form devised so that characters can be easily extracted in the character extraction analysis process of the OCR process. For example, the entry frame and the print guide frame are read by a dropout color (visible to the human eye but read by a scanner). You can print it in a color that does not appear as an image when you use it), add a verify mark such as a 3-point mark to make it easier to detect the top and bottom and tilt at the 3 marked points, or add a reference mark to the form. It is a form with measures to improve the character extraction accuracy (character recognition accuracy) by making it possible to accurately detect the character line position inside and by providing a character frame and margins.

In addition, since the OCR form is a document created on the premise of OCR processing, the position on the paper on which the characters are written is defined in advance as a template regardless of the above-mentioned measures for improving the character recognition accuracy. It can also be configured to perform character recognition (character extraction analysis) of the position specified by the template. In this case, by printing or writing the form ID, which is an identifier of the template definition information, on the paper, the form ID of the document converted into image data is identified and OCR processing is performed using the corresponding template. can do. Therefore, the documents sorted by the OCR form recognition engine as the character recognition target are the documents for which measures for improving the recognition accuracy are taken and / and the documents to which the template definition associated with the form ID printed on the paper is applied. It should be noted that a document to which measures for improving recognition accuracy can be further configured to perform OCR processing using a template.

Therefore, the documents sorted by the OCR form recognition engine should be selected (sorted) by determining whether there is a measure to improve the character extraction accuracy (character recognition accuracy) such as the 3-point mark, or whether there is a form ID. Can be done. That is, the document type corresponding to the OCR form recognition engine is a document in which a measure for improving the character extraction accuracy such as a 3-point mark exists or an identifier such as a form ID that can identify a form exists.

Next, the standard form recognition engine defines in advance the position on the paper on which characters are entered as a template for each standard form that can be identified by the title of the document or a specific keyword, and the position defined by the template is standardized. Performs recognition processing for recognition (character extraction analysis). The standard form recognition engine searches for a pre-registered character recognition template using the title of the document as a key for documents other than the OCR form, and performs OCR processing using the corresponding character recognition template. Therefore, the documents sorted by the standard form recognition engine (template forms by title) are documents that are not sorted by the OCR form recognition engine and have character recognition templates pre-registered for each document title. It can be selected (sorted) by determining whether or not. The title of the document is, for example, the title name, title or heading of the XX notification form, XX application form, or the like. The document type corresponding to the standard form recognition engine is a document in which a character recognition template pre-registered for each document title exists.

The atypical form recognition engine finds one or more pre-registered headwords (keywords) in documents other than OCR forms and template forms by title, and automatically finds the character strings in the vicinity. Perform keyword search type OCR processing to be extracted. Therefore, the documents sorted by the atypical form recognition engine (template forms by item) are documents that are not sorted into OCR form and template form by title, and are pre-registered for each specific item (keyword) in the document. It can be selected (sorted) by determining whether or not the document has the character recognition template. The document type corresponding to the atypical form recognition engine is a document in which a character recognition template pre-registered for each specific item (keyword) in the document exists.

The identity verification document recognition engine searches for a matching recognition library using the characteristic information of the identity verification document as a key for documents other than the OCR form, the template form by title, and the template form by item, and performs OCR processing. Identity verification documents include multiple types such as driver's license, My Number card, health insurance card, resident ID, and residence card, and a recognition library is registered in advance for each type of identity verification document.

First, by registering titles and items as characteristic information, it is possible to determine which identity verification document is used. Then, after determining which identity verification document, the OCR process is performed using the corresponding recognition library. The recognition library is a tuning of the information and processing used in the above-mentioned character extraction analysis processing and / and recognition processing for each type of identity verification document (characteristic information). For example, in the case of a driver's license, the capital Since fonts (half-width, two-thirds, full-width, etc.) differ depending on the public safety committee of each prefecture, recognition libraries corresponding to these fonts are registered. The same applies to My Number Card, Health Insurance Card, Resident Card, and Residence Card. It is also possible to define and register a template for each type of identity verification document, and register a recognition library including the corresponding template to perform OCR processing.

Documents (document type) sorted by the identity verification document recognition engine are documents that include the characteristic information of the identity verification document, and by determining whether the document has a recognition library pre-registered for each characteristic information. Can be selected (sorted).

As a method for selecting (sorting) documents to be sorted by each recognition engine, the size (vertical / horizontal) of documents (forms) can also be applied. For example, if it is possible to uniquely determine that a document having a specific size is an OCR form, the vertical and horizontal sizes of the document are registered as the target document discrimination condition shown in FIG. It can also be configured to assign a corresponding recognition engine relative to the size of the digitized document.

FIG. 3 is a block diagram of the management device 100. The management device 100 includes a communication device 110, a control device 120, and a storage device 130. The communication device 110 controls data communication with the user terminal 300 via a wireless or wired network. The storage device 130 is a storage area for storing image data of application documents and character code data after recognition processing transmitted from the user terminal 300 or the image generation device 200 for each applicant. As described above, the management device 100 can be configured to be directly connected to the image generation device 200 without going through the user terminal 300.

The control device 120 includes a master control unit 121, a recognition engine allocation control unit 122, a recognition processing unit 123, a recognition result control unit 124, and a data control unit 125.

The master control unit 121 performs registration control of the recognition engine, registration control of the document type corresponding to the recognition engine, and registration control of the template used by each recognition engine. In each registration control, new registration, editing, updating, and deletion are controlled so as to be operable via the administrator terminal 400 through a predetermined screen. For convenience, the mode in which various information registration operations to the management device 100 are performed via the administrator terminal 400 is illustrated, but for example, the user terminal 300 can be configured as the administrator terminal 400. is there.

In the recognition engine registration control, for example, a screen for registering the recognition engine master information as shown in FIG. 2 is provided, and the recognition engine OCR processing module and the recognition library are uploaded or already uploaded from the administrator terminal 400. To be able to perform operations such as updating and deleting the OCR processing module and recognition library, and registering, updating, and deleting information related to the association between each recognition engine and the document type (for example, target document discrimination condition (sorting condition)). To control.

The document type registration control for the recognition engine is controlled so that the document types (OCR form, fixed form, atypical, identity verification, free format) associated with each recognition engine can be registered, updated, and deleted. The registered document types can be displayed in a list and can be selected and controlled in the registration control of the recognition engine.

The template registration control used by each recognition engine is to perform layer creation control to set the character recognition area at the character description position of the displayed image data while displaying the captured image data on the screen. .. The template is information in which a plurality of character recognition areas are set according to the character description position of the document and has position information (a set of start position coordinates and end position coordinates of each area) of each character recognition area.

The template registration control also includes an operation of associating with each recognition engine. For example, the OCR form recognition engine associates a template, the standard form recognition engine associates a document title name with a template, and the atypical form recognition engine associates one or more item names (keywords) in a document with a template. Control the association with and so that it can be done.

The recognition engine allocation control unit 122 performs recognition engine distribution processing for each image data (one page) for each of a plurality of documents, and also performs page division processing described later (see FIG. 4). The recognition processing unit 123 performs OCR processing for each page using each recognition engine assigned by the recognition engine allocation control unit 122, and outputs a recognition result (character code). The recognition result control unit 124 stores the image data to be OCR processed and the character code which is the recognition result in the storage device 130 in association with each other. At this time, using the applicant identification information that can uniquely identify the applicant extracted from the document, such as the applicant's name and the applicant's ID code, the correspondence between the image data and its character code is determined for each applicant. It can be divided and stored in the storage device 130.

The data control unit 125 performs an applicant search using the character code after OCR processing, character code editing work for the image data such as data correction and deletion of the character code, correction of the order of the image data, and the like. The editing function for the character code is provided through a predetermined screen. Various operation controls are enabled based on the request from the user terminal 300.

Documents include, for example, an area having a document type to be recognized by the OCR form recognition engine and an area having a document type to be recognized by the standard form recognition engine in one image data (within one page). May be mixed (when there are multiple divided areas). In this way, FIG. 4 is composed of a mixture of an area having a document type to be recognized by the OCR form recognition engine and an area having a document type to be recognized by the standard form recognition engine in one document. It is a figure for demonstrating the allocation process of the recognition engine in the case of. As shown in FIG. 4, for example, the OCR form recognition engine performs both processing for discriminating the document type to be character-recognized and the standard form recognition engine performs both processing for discriminating the document type to be character-recognized, and which document type is used. If the above also applies, the OCR form recognition engine identifies an area having a document type to be recognized as a character and an area having a document type to be recognized by a standard form recognition engine, and each identified area is identified. Acquires the position information (coordinate information) of. Then, the OCR form recognition engine is assigned to the image data to which the position information (coordinate information) of the area to be recognized by the OCR form recognition engine is attached. On the other hand, the standard form recognition engine is assigned to the image data with the position information (coordinate information) of the area to be recognized by the standard form recognition engine. The OCR form engine performs character recognition processing on the area specified by the incidental position information (coordinate information) from the image data allocated in this way. On the other hand, the standard form recognition engine performs character recognition processing on the area specified by the incidental position information (coordinate information) from the image data allocated in this way.

As another method of determining whether or not there are a plurality of document types in one document, for example, if a standard or atypical description area other than the OCR form area exists in the OCR form in advance. Since it may be known, in this case, the existence of a standard or non-standard description area and the position of the description area are registered in advance for each form ID, and as a pre-processing in the recognition engine allocation process. It can also be configured to perform the division process.

On the other hand, by using the processing as shown in FIG. 4, for example, even if a plurality of sheets (multiple types) of documents are included in one image data, the coordinate position of each document is obtained for each document. It is also possible to perform the process of defining, assign an appropriate recognition engine, and configure it to recognize characters within the range of coordinate positions.

In this way, one image data is subjected to OCR processing by a plurality of recognition engines according to the recognition engine that matches the recognition characteristics. As a result, accuracy and speed can be improved.

FIG. 5 is an explanatory diagram of a batch reading process for collectively reading each document of a plurality of different applicants of the present embodiment. At the time of image scanning with the image generator 200, if a set of documents (a bundle of documents) of one applicant is set and scanned each time, the time and effort increases. Therefore, a bundle of documents of a plurality of applicants can be stacked in a predetermined number and set in the image generation device 200, and can be collectively captured. In this case, in order to separate the image data of each of the plurality of applicants, for example, a predetermined partition paper is inserted between the bundle of documents of applicant A and the bundle of documents of applicant B. After batch import, image data can be sorted and saved for each applicant using the partition paper as a clue. In addition, if the order of the documents in a bundle of documents is decided without using a partition paper, the image data can be sorted and saved for each applicant after batch importing using the documents that come to the first page as a mark. it can.

FIG. 6 is a diagram showing a processing flow of the document reading system of the present embodiment. In the following description, a mode in which a set of application documents for each applicant, in which a plurality of documents of different document types exist, is collectively read by the image generator 200 is illustrated.

As shown in FIG. 6, the application documents of the applicant are set in the image generator 200 and scanned to generate image data (S201). The generated image data is output to the user terminal 300 (S202). The user terminal 300 displays a plurality of image data (image data of a plurality of pages) constituting the application documents output from the image generation device 200 (S301).

The user can check a plurality of image data output from the image generator 200 through the user terminal 300, and further delete the image data determined to be unnecessary or delete the image data among the plurality of image data. An operation of rearranging the order (order change operation) can be performed (S302). The image generation device 200 can output image data to the user terminal 300 in order in the order in which the application documents are scanned. In the user terminal 300, the plurality of image data are retained by the scanned order information. Has been done.

After confirming and editing the plurality of image data output from the image generator 200, the user divides the image data for each applicant and saves it locally (S303). At this time, if the user terminal 300 has a time stamp function, a time stamp (for example, a combination of time information issued by a time certification authority and a hash value generated from image data) is used to determine the reception date. ) Is performed, and the time stamp information can be added to each of the plurality of image data classified by the applicant.

The user terminal 300 transmits the image data of the application documents stored locally to the management device 100 (S304). The timing of transmission is arbitrary.

The management device 100 performs the recognition engine registration process (S101), the template registration process (S102), and the like in advance through the administrator terminal 400. Since these processes have been described above, they will be omitted. Next, the management device 100 receives the image data from the user terminal 300 and performs a reception process of storing the image data in the storage device 130 (S103).

The management device 100 performs OCR processing in real time or at an arbitrary timing. The OCR process is configured to include the processes of steps S104 to S107. First, a plurality of image data constituting the application documents are extracted for each applicant, and the image data are sorted page by page, that is, in the order of scanning, the document type identification process (S104), and the recognition engine according to the identified document type. The distribution process (S105), the OCR process (S106) using the recognition engine assigned by the distribution process, and the process (S107) of associating the recognition result (character code) with the image data and saving the image data are performed. The processes of steps S104 to S107 are repeated for the number of pages (the number of image data) to generate and save the character code of each image data constituting the application process.

The management device 100 provides an editing function for the character code after the OCR process. The administrator terminal 400 is displayed with a screen including the image data and the character recognition result, and is controlled so that the character recognition result can be input, edited, and deleted (S401, S108). The management device 100 performs a process of reflecting the editing result (S109).

The user can search for application documents from the user terminal 300, and the management device 100 provides a search function. For example, when there is an inquiry from the applicant, the name, address, etc. of the applicant are input from the user terminal 300 as a search query and transmitted to the management device 100 (S305). The management device 100 (data control unit 125) searches for the stored character code information using the received search query (S110). As a result of the search, the image data group of the application document including the image data corresponding to the hit character code information is extracted and transmitted to the user terminal 300 as the search result (S111). As the search result, the character code information may be output together with the image data.

FIG. 7 is a diagram showing a detailed flow of OCR processing of the document reading system of the present embodiment, and is a detailed flow of OCR processing composed of each processing of steps S104 to S107 in FIG.

The recognition engine allocation control unit 122 extracts the image data of the application documents for each applicant received from the user terminal 300 from the storage device 130, reads the first image data, and performs the document type determination process. First, the recognition engine allocation control unit 122 performs a process of extracting information that can be distinguished from an OCR form such as a three-point mark from the image data (S104A). When the information that can be distinguished from the OCR form can be extracted (YES in S104A), the recognition engine allocation control unit 122 distributes the image data to the OCR form recognition engine (assigns the OCR form recognition engine to the image data). , S105A).

Further, in step S104A, when the information distinguishable from the OCR form cannot be extracted (NO in S104A), the process of extracting the information distinguishable from the template form by title is performed (S104B). When the information that can be distinguished from the template form for each title can be extracted (YES in S104B), the recognition engine allocation control unit 122 distributes the image data to the standard form recognition engine (the standard form recognition engine for the image data). Is assigned, S105B).

Subsequently, in step S104B, when the information that can be distinguished from the template form by title cannot be extracted (NO in S104B), the process of extracting the information that can be distinguished from the template form by item (keyword) is performed (S104C). When the information that can be distinguished from the item-specific template form can be extracted (YES in S104C), the recognition engine allocation control unit 122 distributes the image data to the atypical form recognition engine (atypical form to the image data). Assign a recognition engine, S105C).

Further, in step S104C, when the information that can be distinguished from the itemized template form cannot be extracted (NO in S104C), the process of extracting the information that can be distinguished from the identity verification document is performed (S104D). When the information that can be distinguished from the identity verification document can be extracted (YES in S104D), the recognition engine allocation control unit 122 distributes the image data to the identity verification document recognition engine (identification document recognition to the image data). Assign an engine, S105D).

The recognition processing unit 123 performs OCR processing using the assigned recognition engine on the image data to which the recognition engine can be assigned through the steps S104A to 104D and steps S105A to S105D (S106). Then, the recognition result control unit 124 associates the recognition result (character code) obtained by the OCR process with the image data, and stores the recognition result in the storage device 130 (S107). These processes are repeated for all the image data constituting the application document (S1108), and the process is completed.

The processing order of steps S104A to 104D can be changed as appropriate. Further, the recognition engine allocation control unit 122 extracts information that can be identified as an OCR form, information that can be identified as a template form by title, information that can be identified as a template form by item, and information that can be identified as an identity verification document. In, for example, a simple OCR process can be executed to extract information such as a three-point mark, a title, an item, and characteristic information of an identity verification document from the image data (preliminary OCR process for determining the document type). ).

On the other hand, through steps S104A to 104D, for the document (image data) to which the recognition engine could not be assigned (NO in S104D), the process proceeds to the manual setting process due to a reading error (S1101).

Specifically, the recognition engine allocation control unit 122 can display the image data in which the reading error has occurred on the screen of the administrator terminal 400 and control the recognition engine so that it can be manually selected (S1102). In addition, it is possible to control the creation / correction of the correspondence between the recognition engine and the document type (S1103). For example, it is possible to create and register the correspondence between the title or item of the image data in which the reading error has occurred and the recognition engine. As a result, the recognition engine can be automatically assigned to the image data in which a reading error has occurred in the future. Alternatively, there is also a method of displaying the image data in which a reading error has occurred on the screen of the administrator terminal 400 and specifying the character recognition target area (reading location) by tracing the image data displayed on the screen with a finger. .. This specification method corresponds to free-form OCR processing (free-form character recognition processing).

Further, the recognition engine allocation control unit 122 can display the image data in which the reading error has occurred on the screen of the administrator terminal 400, and can perform various controls for modifying and newly creating the template (S1104). For example, a new template is registered for the image data in which a reading error occurs, and the registered template (including titles and items) is controlled so as to be associated with the recognition engine. With this configuration, the template to which the recognition engine can be assigned can be expanded, and the recognition engine can be automatically assigned to the image data in which a reading error has occurred in the future.

When the recognition engine allocation is completed by each of the processes of steps S1102, S1103, and S1104, the recognition engine master information is updated and the correspondence between the recognition engine, the document type, and the template is learned (S1107). Then, the process proceeds to step S106, and the recognition engine allocation control unit 122 executes the OCR process using the manually assigned recognition engine. The processes of steps S1102 to S1104 and S1107 are executed by the master control unit 121.

Further, the manual setting process due to a reading error can be configured to include the processes of steps S1105 and S1106. Step S1105 is an OCR process skip process that saves the image data as it is without performing the OCR process on the image data of the reading error. Step S1106 is an OCR processing skip for deleting the image data of the reading error. In the case of step S1106, the process is to delete the image data itself of the reading error from the plurality of image data constituting the application document.

FIG. 8 is a diagram showing an example of a workflow to which the document reading system of the present embodiment is applied. 9 is a diagram showing an example of the workflow following FIG. 8, and FIG. 10 is a diagram showing an example of the workflow following FIG. An example of this workflow includes, for example, one or more application reception offices (windows such as stores and government offices) and a processing center, and the management device 100 is installed in the processing center.

An image generation device 200 and a user terminal 300 are installed at the application reception office, and the user terminal 300 is connected to the management device 100 of the processing center by a network via a dedicated network or an IP network. The processing center is also provided with an image generation device 200 and a user terminal 300, and the user terminal 300 of the processing center is connected to the management device 100. Further, the processing center is provided with an administrator terminal 400, which is connected to the management device 100.

First, at the application reception office, the set of application documents brought by the applicant is scanned by the image generator 200. Then, on the user terminal 300, the user can check each image data of the generated application document set, delete unnecessary image data, and change the order of the image data arranged in the scanning order. .. Then, the image data group of the application document set for each applicant is stored in the storage device of the user terminal 300 and transmitted to the management device 100 at an arbitrary timing.

On the other hand, the application documents sent by the applicant are physically transported to the processing center. The user of the processing center sets each of the transported application documents in the image generation device 200 and performs image data generation processing. Also in the processing center, the user terminal 300 can check each image data of the generated application document set, delete unnecessary image data, and change the order of the image data arranged in the scanning order. Similar to the application reception office, the image data group of the application document set for each applicant is directly stored in the management device 100 without being transmitted to the management device 100 at an arbitrary timing or stored in the storage device of the user terminal 300. It can be stored in the device 130. In the example of FIG. 8, as shown by the alternate long and short dash line, as a workflow, the application document set of the received paper medium is processed without generating the image data and transmitting it to the management device 100 at the application reception office. It is also possible to transport the center.

As shown in FIG. 9, the management device 100 of the processing center mainly performs the OCR process (document reading process) shown in FIG. 7, and stores the character code information and the image data in association with each other. Other detailed workflows are as described above and will be omitted.

Then, in FIG. 10, the application reception office can access the management device 100 through the user terminal 300 and search for application documents. The same applies to the processing center. In addition, the processing center can edit the character code information through the administrator terminal 400.

As described above, the document reading system of the present embodiment can significantly reduce the work labor in digitizing the paper medium in the workflow including the application reception office and the processing center. That is, since the document reading system of the present embodiment automatically assigns a recognition engine to a plurality of types of documents included in the document group and performs OCR processing, document sorting work when scanning with the image generator 200, etc. No need for trouble. In the past, it was necessary to perform OCR processing by performing sorting work for each document type according to the recognition engine, which was very laborious and costly. However, in the present embodiment, since the types of a plurality of documents constituting the set of documents are discriminated and the optimum recognition engine is automatically sorted, the work effort is reduced while realizing the character recognition processing with high accuracy. Can be done.

Further, in the conventional manual sorting for each document type, conversely, the document types that do not match the recognition engine have not been digitized. For this reason, non-digitized documents are information that is not linked to a set of documents, and since they were purposely stored on paper media, it is necessary to physically search for non-digitized documents when needed, which is extremely difficult. It was a hassle. On the other hand, in the present embodiment, the document that is not character-coded is also stored as it is as image data, and is associated and saved together with another image file that is character-coded. Therefore, it is possible to easily extract a desired document from a set of documents using the applicant as a key, and it is possible to reduce work labor.

In particular, in this embodiment, since one document discriminates the document type in units of one image data (one page) and assigns a recognition engine, high-precision character recognition processing is realized while suppressing the complexity of the OCR processing itself. To. That is, if a plurality of recognition engines are assigned to one document, it is necessary to discriminate the area for each recognition engine included in one document, which complicates the process. However, in the present embodiment, the processing performance of the recognition engine can be maximized, and high-precision reading can be realized while reducing human labor by controlling the distribution of the recognition engine on a page-by-page basis.

11 to 13 are diagrams showing an example of a document (form) of the present embodiment. FIG. 11 shows an example of a three-point form including a three-point mark and a form ID, and an example of a form by title. Further, FIG. 12 shows an example of an item-specific (specific keyword) form and an example of an identity verification form. FIG. 13 shows an example of a free format form.

For example, a bundle of application documents is a combination of the application form of the three-point form shown in FIG. 11 or the specific keyword form shown in FIG. 12 and the identity verification form shown in FIG. 12 in application units (or applicant units). Can include. Further, the title identification form of FIG. 11 and the free-form form of FIG. 13 can be included in a bundle of documents of a set of application documents as attachments to a predetermined application form.

Although the present embodiment has been described above, each function of the document reading management device 100 can be distributed and configured in a plurality of devices as needed. Further, each function constituting the document reading management device 100 can be realized by a program, a computer program prepared in advance for realizing each function is stored in the auxiliary storage device, and a control unit such as a CPU assists. The functions of each unit can be operated by reading the program stored in the storage device into the main storage device and executing the program read into the main storage device by the control unit.

Further, the above program can be provided to a computer in a state of being recorded on a computer-readable recording medium. Computer-readable recording media include optical discs such as CD-ROMs, phase-changing optical discs such as DVD-ROMs, magneto-optical disks such as MO (Magnet Optical) and MD (Mini Disk), floppy (registered trademark) disks, and the like. Examples include magnetic disks such as removable hard disks, compact flash (registered trademarks), smart media, SD memory cards, and memory cards such as memory sticks. In addition, a hardware device such as an integrated circuit (IC chip or the like) specially designed and configured for the purpose of the present invention is also included as a recording medium.

Although the embodiments of the present invention have been described, the embodiments are presented as examples and are not intended to limit the scope of the invention. This novel embodiment can be implemented in various other embodiments, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are also included in the scope of the invention described in the claims and the equivalent scope thereof.

100 Document reading management device 110 Communication device 120 Control device 121 Master control unit 122 Recognition engine allocation control unit 123 Recognition processing unit 124 Recognition result control unit 125 Data control unit 130 Storage device

Claims (6)

A document reading system including an image generator that generates image data of a plurality of documents having different document types and a management device that performs character recognition processing on the image data.
The management device is
A master control unit that registers a plurality of character recognition engines for character recognition processing and associates one or more document types with the character recognition engine.
A recognition engine allocation control unit that determines the document type of the image data and assigns the character recognition engine associated with the determined document type to the image data.
Wherein each image data, possess a recognition processing unit that performs a character recognition process using the character recognition engine assigned, and
The character recognition engine includes an OCR form recognition engine, a standard form recognition engine, an atypical form recognition engine, and an identity verification document recognition engine.
The recognition engine allocation control unit
Allocate the image data of the document to which the template definition associated with the form ID printed on the paper and / or the document to improve the recognition accuracy is applied to the OCR form recognition engine.
The image data of a document that is not sorted by the OCR form recognition engine and has a character recognition template pre-registered for each document title is assigned to the standard form recognition engine.
The image data of a document that is not sorted by the OCR form recognition engine and the standard form recognition engine and has a character recognition template pre-registered for one or more items or keywords in the document. Assigned to the atypical form recognition engine,
Documents that are not sorted by the OCR form recognition engine, the standard form recognition engine, and the non-standard form recognition engine, and which include the characteristic information of the identity verification document, are assigned to the identity verification document recognition engine.
A document reading system characterized by this. The image generator generates an image data group of a set of submitted documents composed of a plurality of documents having different document types for each submitted document or for each document submitter.
The recognition engine allocation control unit determines the document type of each image data of the image data group for each submitted document or for each document submitter, and uses the character recognition engine associated with the determined document type as the submitted document. The document reading system according to claim 1, wherein the document reading system is assigned to each set of image data. The recognition engine allocation control unit determines whether or not there are areas corresponding to a plurality of document types in one image data, and when it is determined that there are areas corresponding to a plurality of document types. It is characterized in that the position information of each area corresponding to the plurality of document types is acquired, and the character recognition engine associated with the determined document type is assigned to the image data and the acquired position information. The document reading system according to claim 1 or 2. The recognition engine allocation control unit receives image data of a reading error that could not be distributed to the OCR form recognition engine, the standard form recognition engine, the non-standard form recognition engine, and the identity verification document recognition engine.
(A) Selection process for selecting one of the recognition engines for image data that has a reading error.
(B) The first skip process of saving the image data as it is without performing the character recognition process for the image data in which a reading error occurs.
(C) A second skip process for deleting image data that has a reading error without performing the character recognition process.
Perform at least one of
The master control unit
(D) First master processing for newly creating and editing the correspondence between the recognition engine and the document type for the image data in which a reading error has occurred.
(E) Second master processing for newly creating and editing the character recognition template for the image data in which a reading error has occurred.
(F) Third master processing for designating a character recognition target area for image data that has a reading error,
Document reading system according to claim 1 performs at least one of, it characterized that. A document reading management device that performs character recognition processing for image data of multiple documents with different document types.
A master control unit that registers a plurality of character recognition engines for character recognition processing and associates one or more document types with the character recognition engine.
A recognition engine allocation control unit that determines the document type of the image data and assigns the character recognition engine associated with the determined document type to the image data.
Each image data has a recognition processing unit that performs character recognition processing using the assigned character recognition engine .
The character recognition engine includes an OCR form recognition engine, a standard form recognition engine, an atypical form recognition engine, and an identity verification document recognition engine.
The recognition engine allocation control unit
Allocate the image data of the document to which the template definition associated with the form ID printed on the paper and / or the document to improve the recognition accuracy is applied to the OCR form recognition engine.
The image data of a document that is not sorted by the OCR form recognition engine and has a character recognition template pre-registered for each document title is assigned to the standard form recognition engine.
The image data of a document that is not sorted by the OCR form recognition engine and the standard form recognition engine and has a character recognition template pre-registered for one or more items or keywords in the document. Assigned to the atypical form recognition engine,
Documents that are not sorted by the OCR form recognition engine, the standard form recognition engine, and the non-standard form recognition engine, and which include the characteristic information of the identity verification document, are assigned to the identity verification document recognition engine.
A document reading management device characterized by this. A program executed by a computer that performs character recognition processing on image data of a plurality of documents having different document types.
A first function of registering a plurality of character recognition engines for character recognition processing and associating one or more document types with the character recognition engine, and
A second function of determining the document type of the image data and assigning the character recognition engine associated with the determined document type to the image data.
For each of the image data, a third function of performing character recognition processing using the assigned character recognition engine is realized .
The character recognition engine includes an OCR form recognition engine, a standard form recognition engine, an atypical form recognition engine, and an identity verification document recognition engine.
The second function is
Allocate the image data of the document to which the template definition associated with the form ID printed on the paper and / or the document to improve the recognition accuracy is applied to the OCR form recognition engine.
The image data of a document that is not sorted by the OCR form recognition engine and has a character recognition template pre-registered for each document title is assigned to the standard form recognition engine.
The image data of a document that is not sorted by the OCR form recognition engine and the standard form recognition engine and has a character recognition template pre-registered for one or more items or keywords in the document. Assigned to the atypical form recognition engine,
Documents that are not sorted by the OCR form recognition engine, the standard form recognition engine, and the non-standard form recognition engine, and which include the characteristic information of the identity verification document, are assigned to the identity verification document recognition engine.
A program characterized by that .