JP6702629B2 - Type OCR system - Google Patents

Description

The present invention relates to a type OCR system for performing OCR processing on a FAX reception document created in a type or a type (for example, handwriting only the date and quantity) except for a handwritten portion.

Up to now, there is a technique disclosed in Patent Document 1 as a technique for automatically sorting an order form, a contact form, and other FAX documents received by FAX for each destination.
More specifically, the destination recognition unit 2 that automatically recognizes the destination of the received FAX, the image database 3 that stores the image data of the FAX that is received in association with the information of the automatically recognized destination, and the FAX of the received FAX. A reception FAX management system is constituted by a character information database 4 for storing character information read from image data by an OCR function and a search means 5 for searching the image database 3 or the character information database 4 by designating a destination. Is.

JP, 2002-165076, A

Such a reception FAX management system is a system developed to sort out a dedicated operator on the assumption that there is a dedicated operator for each destination who is a sender of the received FAX document. .

On the other hand, the applicant of the present application does not aim to automatically sort the received FAX documents for each destination, but the "Global Rule" which is standardly applied to the FAX reception document and the specific reception In order to provide a new type OCR system that has never existed before, which enables the system itself to have a learning function using the "local rule" that is applied only to documents, and which can improve the accuracy of OCR processing, The present invention has been completed.

In order to achieve the above object, the first invention of the present application is a type OCR system for performing OCR processing on a FAX reception document created in a type or a type other than a handwritten part. , OCR processing is performed on the reception document with the transmission source attribute data (TSI code or FAX number, etc.) received by FAX from the transmission source, and the reception document is classified into a target document and a non-purpose document using the classification database. Extraction with a classification function and OCR processing of received documents classified into target documents, and standard rules applied to global rules and local rules applied only to received documents from sources having specific attribute data Extraction function to extract the "item name", "value" and "coordinates" from the received document, and the image of the received document and the extraction function on the monitor screen using the supplemental database and the complementary database that complements the OCR process. The extraction display of the received document is displayed in parallel, the "value" for the "item name" is confirmed, and if it is inappropriate, the correct "value" is corrected, and the OCR area by the extraction function is confirmed. If it is inappropriate, specify the correct OCR area, perform OCR processing again in the specified OCR area, and confirm/correct function to confirm the “value” for the “item name” again, and the extraction confirmed by the confirmation/correction function. Outputs the result in a specified output format and exports it to the target document database, and reflects the OCR area and the corrected value specified by the confirmation/correction function in the extraction database and the complementary database. It is a type OCR system characterized by having a learning function for making it.

A second aspect of the present invention is a type OCR system that performs OCR processing on a received document of a FAX that is created in a type except for the entire type or a handwritten part, and the type OCR system is applied as standard. It has a structure dictionary, an item dictionary, and a complementary dictionary, each of which has a global rule and a local rule which is applied only to a received document from a sender having specific attribute data. It contains document layout information (zone information), information about the positional relationship between items and values, and the number of values, the item dictionary contains information about synonyms and value formats, and the complementary dictionary misrecognizes each item. OCR processing is performed on the receiving document that includes the source attribute data received from the sender by FAX, including the characters that are easy to read and the location information, and uses the structural dictionary, item dictionary, and supplementary dictionary to target the receiving document. For the classification function to classify into the document and the non-purpose document and the OCR processing of the reception document classified into the purpose document, using the structure dictionary/item dictionary/complementary dictionary, the “item name” and “value” from the reception document Display the image of the receiving document and the extracting display of the receiving document extracted by the extracting function side by side on the monitor screen, and display the extraction function to extract the "reliability" and "coordinates", and to display the "value" for the "item name". Confirmation/correction function to confirm the value and correct it to the correct “value”, and the extraction result confirmed by the confirmation/correction function is output in the specified output format, and this is exported to the target document database. It is equipped with a data output function that enables the data to be read, and a learning function that reflects the value corrected by the confirmation/correction function in the structure dictionary, item dictionary, and supplementary dictionary, and is necessary for data conversion without depending on the layout of the target document. It is a type OCR system characterized by enabling extraction of item values.
A second aspect of the present invention is a method for extracting a target document having various layouts by the type OCR system according to claim 2 to extract the values of items necessary for data conversion without depending on the layout of each target document. May be logically defined as a virtual form, and a design function characterized by enabling the extraction function, the confirmation/correction function, and the data output function to be collectively set based on the virtual form.

According to the first aspect of the present invention, the following effects can be obtained.
(1) Many (for example, 70 to 80%) fax reception documents can be OCR processed according to a “global rule” that is applied as a standard.
(2) Other (for example, 20 to 30%) FAX reception documents cannot be OCR processed by the "global rule", but can be OCR processed by the "local rule" applied only to a specific reception document.
(3) Therefore, as a result, all FAX reception documents can be efficiently and reliably OCR processed.
(4) In addition, since the present invention does not register the format for each FAX reception document having a different format, the system is not burdened even if the number of FAX reception documents to be processed increases, and rather the learning function is provided. Therefore, there is an effect that the accuracy increases as the OCR processing is performed.

The second aspect of the present invention has the following effects.
(1) Even if the layout of the fax reception document changes, there is no need to redesign each time (without redesigning), and even for fax reception documents of various layouts, the values of the items necessary for data conversion can be ensured. Can be taken out (extracted).
(2) That is, in the conventional form OCR, in the following cases, the target value cannot be extracted as it is and it is necessary to redesign. However, in the present invention, it depends on the layout coordinate of the FAX reception document. Since it does not, it is possible to reliably extract (extract) the value of the item required for data conversion.
・Although the layout is similar for different customers, the layout and fonts are slightly different.
・Fax cannot be received with the registered layout.
・The layout, fonts, and print settings are slightly different for each business partner's base.
-The number of frames and the width of the frame increase or decrease.
・Lines in the NO line increase or decrease.
・Lines will increase or decrease.
(3) Further, according to the type OCR system of the present invention, an object document having various layouts is logically treated as one virtual form in order to extract the values of items necessary for data conversion without depending on the layout of each object document. It is possible to collectively set the extraction function, confirmation/correction function, and data output function (design function).
(4) That is, even if the user does not have special program knowledge, by using this design function, it is possible to easily set the extraction function, the confirmation/correction function, and the data output function. As a result, the opportunities for utilizing the print type OCR system will expand.

The system conceptual diagram about the 1st invention of the invention in this application. "Confirmation/correction screen" relating to the first invention of the present invention Explanatory drawing (1) regarding the 2nd invention of this invention. Explanatory drawing (2) regarding the 2nd invention of this invention. Explanatory drawing (3) regarding the 2nd invention of this invention. Explanatory drawing (4) regarding the 2nd invention of this invention. Explanatory drawing (5) regarding the 2nd invention of this invention. Explanatory drawing (6) regarding the 2nd invention of this invention. The system conceptual diagram regarding the 2nd invention of this invention. The conceptual diagram (1) regarding the "virtual form" of the 2nd invention of this invention. The conceptual diagram (2) regarding the "virtual form" of the 2nd invention of this invention.

First, terms used in the present invention will be described.
[Structure dictionary]
The structure dictionary contains the following information.
□ Document layout information (zone information)
・Coordinates (upper left coordinates, width, height)
-List of item IDs of included items □ Positional relationship between items and values and number of values-Right, bottom, bottom right, no display-Different column/same column/no column-Single/multiple (maximum number)
[Item dictionary]
The item dictionary includes the following information.
□ Item information ・Item ID
・Item attribute ・Synonym group ID
□ Value format ・Character type specification ・Template ・Regular expression □Synonym information ・Synonym group ID
・Item 1, Item 2, Item 3...
Example) Synonym group ID Name item 1 Name item 2 Your name item 3 Your name item 4 Name item 5 Name item 6 Your name item 7 Your name [Complementary dictionary]
The supplementary dictionary contains the following information.
□ Characters that are easily misrecognized and information about the location are stored for each item.
・Item name ・Character position ・Recognized character Example: Item name Date character position 1
Correct character date recognition character mouth

Mouth → Date

[Classification database]
It is a database that contains information used to classify FAX received documents. In reality, since the structure dictionary, the item dictionary, and the complementary dictionary are used in the same manner as the extraction database, the classification database does not exist separately from the extraction database. When describing a specific process of the classification function, it means a place where information to be classified is stored.
[Extraction database]
It is a database that contains information used to extract the items in the purpose document. Actually, since the structure dictionary, the item dictionary, and the complementary dictionary are used in the same manner as the classification database, the extraction database does not exist separately from the classification database. When describing a specific process of the extraction function, it means a place where information of items and values to be extracted is stored.
[Complementary database]
It has almost the same meaning as the complementary dictionary. The word “database” has a strong nuance to refer to a vessel, and the dictionary has a strong nuance to refer to the contents contained in a vessel.

[Customer-specific translation master]
When the product number of the customer and the product number of the company are different, it is necessary to replace the product number of the customer with the product number of the company, and the product name is the same. The customer translation master associates the product number and product name of the customer with the product number and product name of the company.
[Usually master by customer]
When the customer specifies when ordering, the common name used by the customer may be used instead of the product name. In some cases, detailed specifications such as color and shape are not described. At that time, it is necessary to identify and replace the product name, product number, branch edition, and specification of the product that is always purchased based on past purchase information. It is the usual master for each customer whether to hold such information.

[Global Rule]
In the structure dictionary, the item dictionary, and the complementary dictionary, the attribute of information that does not depend on a specific customer is "global". The contents of the dictionary having this global attribute are collectively called a global rule.
[Local Rule]
In the structure dictionary, the item dictionary, and the complementary dictionary, the attribute of information for a specific customer is "local". The contents of the dictionary having this local attribute are collectively referred to as a local rule.

Next, an embodiment of the present invention will be described with reference to the drawings.
FIG. 1 illustrates a system conceptual diagram relating to a first invention of the present invention. The type OCR system of the present invention is mainly composed of "classification function", "extraction function", "confirmation/correction function", "data output function", and "learning function". Hereinafter, each function will be described.

First, the "classification function" is a function of classifying a document received by FAX (received document) into a "document (purpose document)" for the purpose of OCR processing or a "non-purpose document (non-purpose document)".
As the classification work, the reception document accompanied by the sender attribute data (TSI code, FAX number, etc.) received from the sender by FAX is wholly OCR processed.
Then, using the classification database, it is classified into "purpose document" and "non-purpose document".
Here, the “classification database” is provided with “definition of terms”, “classification rules” and the like that can classify received documents into target documents or non-target documents. For example, when the purpose document is a “order form”, the order form is defined so that documents having the same meanings as the order form (“order form”, “request form”, etc.) can be classified as the purpose form.

Next, the "extraction function" is a function of extracting the necessary "item name", "value", and "coordinates" from the purpose document.
As the work of extraction, the target document is OCR processed. Then, the "item name", "value", and "coordinates" are extracted using the extraction database and the complementary database.
Here, the extraction database is provided with a “synonym dictionary” capable of extracting “item name” and “value” from the received document, “arrangement of item name and value”, “value pattern for each item”, and the like. That is, when the "item name" is extracted from the purpose document, since the expression of the item name varies, it is absorbed by the "synonym dictionary" and normalized. For example, items such as “name”, “name”, “name”, “name”, “good name” are represented by the item “name”. When extracting the "value (quantity/item name)" corresponding to the "item name", the extraction method uses "item name" as a marker for the text anchor, and its surroundings (up, down, left, right and diagonal: item name and Search and extract from the "value" pattern (value pattern for each item) corresponding to the "item name" in (Definition/Setting in Value Arrangement). In addition, it corresponds also when the "item name" and the "value" are included in the box of the same rectangular area.
Further, the extraction database is divided into a "global rule" which is applied as a standard and a "local rule" which is applied only to a receiving document from a sender having specific attribute data. The global rule is a rule that can be applied to 70% to 80% of the FAX reception document, and the local rule is applied to the remaining 20% to 30% of the FAX reception document to which this global rule cannot be applied. The local rule is stored for each source attribute data (TSI code, FAX number, etc.) received together with the image when receiving the FAX. That is, as long as the documents are received from the same sender, the documents have the same format and the same rule can be applied.

Next, the “complementary function” and “standardization/normalization function” using the complementary database will be described. First, the "complementary function" uses a method of finding a correct character string from an ambiguous character string, such as the N-gram method, when the OCR result includes a character that is erroneously recognized. Is a function that replaces the extracted character string by extracting the closest character string from the registered complementary database. For example, when OCR recognizes a product code with a mixture of alphabets and numbers, uppercase I (eye) and lowercase l (el) and number 1 (ichi), uppercase O (o) and number 0 (zero). , M/m (em) and N/n (en), V/v (buoy) and Y/y (wai) are very difficult to distinguish clearly, but this "complementary function" makes OCR It is expected that the recognition result of can be complemented with the correct product code.

In addition, the "standardization/normalization function" is to convert an independently defined code system or common name or abbreviation that exists for each sender into a standardized code or formal name specified on the receiving side. With the goal. In conversion, in consideration of erroneous character recognition by OCR, a conventional database is searched using a "fuzzy search" typified by the N-gram method similar to the "completion function" to find a matching character string. It is a function that replaces the extraction result by acquiring the corresponding standardized code and official name. With this "standardization/normalization function", it is possible to convert an implicit rule that differs for each sender into a code or name that conforms to the standard rule.

Next, the "confirmation/correction function" is a function of confirming/correcting the "value" and "OCR area" extracted by the extraction function.
The confirmation/correction function is performed on the “confirmation/correction screen” (FIG. 2). The “confirmation/correction screen” is a parallel display of the received image of the target document and the extraction display for displaying the “item name” and “value” data extracted from the received image by the extraction function.
The confirmation function can identify the corresponding part of the received image that corresponds to this by specifying the extracted "item name" and "value" fields of the extraction display (displayed with a color frame, background and letters and numbers). Is displayed in a different color, etc.), making it extremely easy to check.
The correction function performs different processing for the extracted “value” and the extracted “area”. The “area” to be extracted is not wrong, but it is correct if the “value” extracted from it is incorrect (for example, when reading “3” reads “8”). Reenter the "value" in the extract display (manual operator work).
On the other hand, if the "area (region)" to be extracted is mistaken, the correct "area (region)" is designated (selected) on the received image, and the designated region is OCR-processed so that the correct "item name" or ""Value" is extracted, and it is automatically input to the extraction display.

Further, the “data output function” is a function for outputting the checked/corrected data in the form of a predetermined output format and exporting (storing in a database of a target document).

Finally, the "learning function" specifies (selects) the correct "area (region)" on the received image when the "area (region)" to be extracted is mistaken in the confirmation/correction function, At this time, the information in the designated area is stored in the "local (individual) rule" in association with the attribute data (source information) attached to the purpose document. After that, for the purpose document to which the attribute data (source information) is added, the “local (individual) rule” is applied and processed in the extraction function with priority over the “global (standard) rule”. Become.
Further, when the “value” to be extracted is mistaken, the correct “value” is corrected. At this time, the corrected “value” information is stored in the complementary database. After that, in the extraction function, the “value” is automatically complemented and extracted by using this complementing database.

3 to 8 are explanatory views regarding the second invention of the present invention.
A second invention of the present invention uses a structure dictionary/item dictionary/complementary dictionary for a FAX reception document (FAX reception document) created in print characters except for the whole characters or a handwritten part in part. Analyze and extract. For example, taking the document in the figure as an example, as shown in the figure, the FAX reception document is "title zone", "ordering zone", "ordering zone", "main specification zone", "common zone 1", "common zone 2", "common". It is analyzed as "zone 3""common zone 4""common zone 5". Taking the "title zone" as an example of the analysis method, the title is generally placed in the upper center of the document, and the document name is often described (these information is the structure dictionary/item). Accumulated in dictionaries etc.). In the FAX reception document of FIG. 3, since the "order form" is described in the upper middle part of the document, the area described as "order form" can be changed by using the structure dictionary and the item dictionary. Title zone". Similarly, the "ordered party zone" to "common zone 5" are analyzed.

(Classification function)
Then, according to the title (value) described in the “title zone”, it is classified into a target document and a non-target document. If the purpose document is an “Order Form”, the “Order Form” or a synonym for the order form is written in the “Title Zone” (this is the “Order Form” or “Order Form” by using the item dictionary). It can be recognized as a target document.

(Extraction function)
Next, "item name", "value", "reliability" and "coordinates" are extracted from the purpose document. In the case of “order form”, for example, the “item name” is the product number and the “value” is the quantity. Since the product number and quantity are listed in the main detail zone (the information is stored in the structure dictionary), the character type (which can be recognized as "coordinates" in the structure dictionary) from the description content written in the main detail zone ( If it is a product number, it will be written in numbers or English letters, if it is a quantity, it will be written only in numbers or in numbers and units), template (if it is a product number of Company A, XXXNNNN-XN, etc.), "item name" and "value" Is extracted using information such as the positional relationship (the quantity is on the right side or the lower side of the product number: accumulated in the structure dictionary).
In addition, since the quantity etc. are generally written in arithmetic numbers (common to all companies), the global rule that is applied as standard is "quantity is expressed in arithmetic digits" as a structure dictionary/item dictionary.・Prepare for a supplementary dictionary.
On the other hand, the order number etc. cannot be expected to match as a common rule by each trading company, but since it may be common for each company/department, it can be used only in the receiving document from the sender having specific attribute data. As a local rule to be applied, "Structure dictionary of item A is ANNNN-N" and the like are provided in the structure dictionary, item dictionary, and complementary dictionary.
As shown in FIG. 8, when the background of the “item name” is colored or shaded, if it is a general OCR technique, the item name should be recognized and the value extracted from it. Is extremely difficult, but in the present invention, it is possible to extract a value by applying a complementary dictionary, a local rule, or the like.
OCR recognition is also performed by automatically completing and replacing OCR recognition using a complementary dictionary, a translation master for each customer, and a usual master for each customer, and using the result as an extraction result. Support. If the result of OCR is that the "reliability" is low, it will be displayed so that it can be understood by the next confirmation/correction function.

(Confirmation/correction function)
Regarding the confirmation/correction function, the image of the reception document (left) and the extraction display of the reception document extracted by the extraction function (right) are displayed in parallel on the monitor screen, and the "value" for the "item name" is displayed. Check and correct it if it is incorrect (see Figure 2).

FIG. 9 is a system conceptual diagram relating to the second invention of the present invention.
Here, the type of purpose document will be described as "order form". The “classification function” is the same as that of the first invention shown in FIG. 1, and therefore its explanation is omitted. However, since the purpose document is “order book”, it is called order book, order book, or order form. Only the documents that are exposed need to be extracted.

In the next "extraction function", the value of the required item is extracted from the input document using the extraction database (structure dictionary/item dictionary/completion dictionary). In the "order form", customer name, order date and time, product number, product name, unit price, quantity, total amount, desired delivery date, etc. are items to be extracted.
The extraction function handles the OCR result by setting "item name, value, reliability, coordinate". An example is “customer name, XX company, 85%, (972, 246)-(1331, 308)”. Although it is necessary to consider that the OCR engine has different interpretations of reliability and the range of values returned by the coordinates differs depending on the resolution of the image, these four pieces of information are treated as being supposed to be obtained by some method.
As shown in the figure, if the first OCR processing performed by the classification function or the first OCR processing is insufficient, the second OCR processing is performed based on the zone information, and the relationship between the zone and the item is determined. Extract the relationship between items and values. For example, zone A is located at the upper right side of the middle part of the order form, and there are "company name", "TEL", and "FAX" as items, so this is extracted as the order source and the relationship between the item and value is extracted. Extract. Zone B is located in the lower half of the order form and is extracted as the order quantity.

Regarding this extraction function, a complementing function (replacement dictionary) may be used to replace a character erroneously recognized by OCR with a correct character. When replacing a character with a correct character, it is premised on the operation of holding a misrecognition pattern for each word instead of each character and correcting the word to be correct.
For example, “Starting date main temple” → “Order date and time”.
Further, a customer-specific translation master may be used to provide a function (standardization/normalization function) of replacing the customer's product number or product name with the company's product number or product name.

The subsequent “confirmation/correction function” is a function of allowing a person to visually confirm a character with low reliability in the OCR result and correct it to a correct character. In general, an image to be recognized is displayed on one side of the screen and a character of the recognition result is displayed on the other side of the screen, and the character of the recognition result can be corrected.
In addition, the “data output function” specifies the format to be finally output after the OCR recognition and confirmation/correction work, and outputs the data to a relational database or CSV file.
Further, the "learning function" stores the corrected contents in the complementary database (complementary dictionary) in the confirmation/correction function, and makes effective use when complementing the subsequent OCR recognition result.

Further, the second invention has a feature of "virtual form".
The “virtual form” will be described with reference to FIGS. 10 and 11.
As shown in FIG. 10, conventionally, work has been performed with the processing content set for each document (document having the same layout).
On the other hand, the present invention (second invention) has a technical idea of consolidating each document into a logically defined “virtual form”, and the above-mentioned structure dictionary Item dictionary/complementary dictionary or global rule/local rule is used to extract the value of the item required for data conversion without depending on the layout of each document (Fig. 11: Relationship between virtual form and various dictionaries) ).
Therefore, conventionally, setting was required for each document (document having the same layout) and for each work (for example, extraction, confirmation/correction, output), but in the present invention, a theoretically defined virtual form is required. By setting the definition content of, it is naturally unnecessary to set each document (each layout) and each work. In other words, in the present invention, since each work is being carried out to create a virtual form with a focus on the virtual form, once the definition (setting) of the virtual form is decided, other settings are basically unnecessary. is there.

The virtual form is a form that does not depend on a specific layout, and is a form created for each purpose of the document such as “order form”, “quotation request form”, “work request form”, and includes the following information.
□ Document type (order form, request for quotation, contact form, etc.)
-Document type ID
-Document type name-Document type description □ List of items to be used When defining a virtual form, item information is central. If it is an "Order Form", items such as document name, order form number, order form date, order originator, order originator phone number, order originator FAX number, product number, product name, unit price, quantity, total amount, consumption tax, etc. are required. Therefore, the information of these items is held as a reference method (item ID) to the item dictionary.
□ Corresponding structural dictionary The virtual form does not have information about the layout of the document or information about the layout of the item values. Since these pieces of information are in the structure dictionary, reference information (document type IDEA) to the corresponding structure dictionary is held.
□ Confirmation/correction screen layout Layout that improves the efficiency of visually checking the contents by arranging general form items that do not depend on the physical arrangement of specific individual form items □ Output format Settings that are output to a relational database or CSV file in a unified format of date and amount that does not depend on the format of the items described in each individual form □ Logical check Total of the amount written in the statement And the amount written in the total amount column match, whether the desired delivery date is a sufficient number of days after the order date, etc. Logical check logic settings □ DB lookup There are several options to enter. Set the selectable items when limited. Settings such as payment method can be selected from "cash, credit card, cash on delivery, transfer after delivery"

Generally, in the case of type OCR, since a color image or a high-resolution image is premised, the expected OCR result cannot be extracted unless the target image has ideal quality. However, according to the present invention, since it is possible to extract an OCR result from a document having a poor image quality in a black-and-white image or a low-resolution image by FAX or copying, a FAX order form, invoice, or dot printer It is possible to extract the data from the slips and the necessary data from the black-and-white copy and gray-scale copy of the identification document (passport/driver's license) that cannot be attached with the original. Also, since it has a design function (design function), if you set it according to your needs, you can efficiently extract and utilize the target data.

Claims (3)

A type OCR system for performing OCR processing on a FAX received document created in type except for the entire type or a handwritten part in a part,
A classification function of performing OCR processing on a reception document accompanied by sender attribute data received by FAX from a transmission source and classifying the reception document into a target document and a non-purpose document by using a classification database.
OCR processing of received documents classified into object documents, and extraction database and OCR having global rules applied as standard and local rules applied only to received documents from sources having specific attribute data An extraction function for extracting the "item name", "value", and "coordinates" from the received document by using a complementary database that complements the processing,
Displays in parallel and extracted display of the received document that has been extracted by the image extraction function of the receiving written on the monitor screen, check the "value" for the "item name", correct by the operator if improper "value ", and confirm the OCR area by the extraction function, and if it is inappropriate, specify the correct OCR area of "item name" or "value" by the operator , and perform OCR processing again in the specified OCR area, Confirmation/correction function to confirm the "value" for the "item name" again,
A data output function that outputs the extraction result confirmed by the confirmation/correction function in a predetermined output format and exports it to the database of the target document,
The information in the OCR area specified by the confirmation/correction function is associated with the attribute data attached to the target document and stored in the local rule, and the information on the corrected value is stored in the supplemental database. , Learning function to reflect in extraction database and complementary database,
A type OCR system characterized by being equipped with. A type OCR system for performing OCR processing on a FAX received document created in type except for the entire type or a handwritten part in a part,
The type OCR system has a structure dictionary, an item dictionary, and a complementary dictionary, each of which has a global rule that is applied as a standard and a local rule that is applied only to a receiving document from a sender having specific attribute data. Then
The structure dictionary includes a document type, zone information that is document area information, information about the positional relationship between items and values, and the number of values,
The item dictionary contains information about synonyms, value formats,
The supplementary dictionary contains, for each item, characters that are easily misrecognized and information about the location,
A classification function that performs OCR processing on a reception document accompanied by sender attribute data received by FAX from a transmission source and classifies the reception document into a target document and a non-target document by using a structure dictionary/item dictionary/complementary dictionary. ,
For the OCR processing of the receiving documents classified into the target documents, using the structure dictionary/item dictionary/complementary dictionary , based on the zone information from the relevant receiving document , the relationship between the zone and the item And the extraction function to extract "item name""value""reliability""coordinates",
Displays in parallel and extracted display of the received document that has been extracted by the image extraction function of the receiving written on the monitor screen, check the "value" for the "item name", correct by the operator if improper "value Confirmation/correction function to correct
A data output function that outputs the extraction result confirmed by the confirmation/correction function in a predetermined output format and exports it to the database of the target document,
A learning function that reflects the value corrected by the confirmation/correction function in the structure dictionary/item dictionary/completion dictionary,
And a type OCR system capable of extracting the values of items required for data conversion without depending only on the layout coordinates of the target document. The type OCR system according to claim 2 makes it possible to extract a target document having various layouts into logical values for each purpose of a document in order to extract the values of items necessary for data conversion without depending only on the layout coordinates of the individual target documents. The design function is characterized in that the extraction function, the confirmation/correction function, and the data output function can be collectively set by setting the definition contents of the virtual form so that it can be aggregated into one virtual form defined in Type OCR system with.