KR20080006422A - Business form recognition apparatus, and business form recognition program - Google Patents

Abstract

A device and a program for recognizing business forms are provided to recognize the business form having a different size/position of a frame and different order of items even if the business forms have the same kind. A processor detects the frame and character lines from image information of the inputted business form(110). The processor detects the character lines in the detected frame, recognizes an item name of the business form from the detected character line, and detects the frame having the recognized item name as an item name frame(140). The processor detects a data frame corresponding to the recognized item frame based on the recognized item name(150). The processor recognizes the character line included in the detected data frame(160). The processor determines whether the item name having the item name unrecognized in character recognition is found(210). The processor detects the data frame corresponding to the unrecognized item name again based on the unrecognized item name when the unrecognized item name is found(700). The processor recognizes the character line of the detected data frame.

Description

Book recognition device and book recognition program {BUSINESS FORM RECOGNITION APPARATUS, AND BUSINESS FORM RECOGNITION PROGRAM}

1 is a block diagram of a book recognition system according to an embodiment of the present invention.

Fig. 2 is a flowchart showing an outline of a book processing executed by the book recognition apparatus of the embodiment of the present invention.

Fig. 3 is a flowchart showing details of a book processing executed by the book recognition apparatus of the embodiment of the present invention.

4 shows an example of a chart of an embodiment of the present invention.

Fig. 5 is a diagram showing an unread region of a charge of an embodiment of the present invention.

Fig. 6 is a diagram showing an example of adjacency relation between item name frames of the embodiment of the present invention.

7 is a diagram showing an example of a chart of an embodiment of the present invention.

Fig. 8 is a diagram showing an example of adjacency relation between item name frames of the embodiment of the present invention.

9 is a diagram showing an example of a chart of an embodiment of the present invention.

Fig. 10 is a diagram showing an example of adjacency relation between item name frames of the embodiment of the present invention.

Fig. 11 is a diagram showing an item name dictionary in an embodiment of the present invention.

12 is a diagram showing an example of a chart of an embodiment of the present invention.

Fig. 13 is a diagram showing an example of a chart of an embodiment of the present invention.

Fig. 14 is a diagram showing itemized usage knowledge of an embodiment of the present invention.

<Description of the symbols for the main parts of the drawings>

10: input device

20: image input device

30: Book recognition device

40: Recognition dictionary

50: display device

60: Image database

[Patent Document 1] Japanese Patent Application Laid-Open No. 2004-139484

[Patent Document 2] Japanese Patent Application Laid-Open No. 9-319824

[Non-Patent Document 1] The Hitachi OCR Solution Imaging OCR Catalog, Hitachi, Ltd., December 2005 Edition, P5-6

BACKGROUND OF THE INVENTION Field of the Invention The present invention relates to a technology for processing a book image, and more particularly, to a technique for recognizing attributes of information described in a book and letters described in the book.

In the conventional method of recognizing a book by an OCR (Optical Character Reader), there are many methods that can recognize only a formatted book having a predetermined reading position. As a conventional technique recognized by such a formal list, there is a "format generator" (see Non-Patent Document 1, for example). In the technique described in Non-Patent Document 1, the position at which the character to be read is written is strictly specified in 0.1 mm units. Many existing OCRs employ the same format information as the format generator.

On the other hand, in the conventional system, even if the same kind of slips, such as a salary payment report and a prescription, slips, the number of ruled lines, the position of the frame, the size of the frame, etc., are slightly different from each other. In such a book, the arrangement relationship between the items is almost constant, and the frame structure and the size of the frame, etc., which are different from each other, are compared with each other by comparing the frame structure registered in advance with the frame structure analyzed from the book image. There is a method of specifying the read area (see Patent Document 1, for example).

More free forms such as payroll reports and prescriptions are written notices of registration. In the case of a registered notice or the like, the item names are almost the same as long as they are of the same type, but the size and number of the frames of the different forms are different for each sheet. In addition, arrangement | positioning relationship between items may also differ.

Regarding such a list, first, the table structure is analyzed, the characters in the frame are recognized, the frame in which the item name is described is detected, the right or the bottom of the frame in which the item name is written is judged to be the data area to be read, (For example, refer patent document 2). In the method described in Patent Document 2, the character string of the item name and the correspondence between the item name frame and the data frame must be stored in the OCR in advance.

The book recognition device of the present invention uses a book that can be recognized by the technology described in Patent Document 2 as an object of recognition. That is, the present invention provides a book recognition apparatus capable of recognizing a book in which the size and position of the frame are different from each other and the order of the items may be different even with the same kind of book.

Even if the position of the frame, the size of the frame, and the like are the same, when the method of Non-Patent Document 1 is applied to different forms, the form information of all the forms is created, and all the form information created in the OCR is stored, and the OCR Must identify which form information to apply to each entered document. In view of the cost of creating the form information and the precision of identification, it is not practical to recognize different forms even if the method described in Non Patent Literature 1 is the same type as the position and size of the frame.

In addition, the method described in Patent Document 1 is also a prerequisite that the arrangement between the items in the list is the same, and therefore, even if the position of the frame, the size of the frame, and the like are the same type, they cannot be applied to different lists.

Although the system described in patent document 2 can be applied to different forms even if the position of a frame, the size of a frame, etc. are the same kind, there exist three subjects at the time of actual operation.

The first problem is that the data frame is not detected when the letter of the item name of the payload is not recognized correctly. The second problem is that, even if it is a character string indicating the same item, for example, "name" and "name", the item name frame is not detected when the notation is different. The third problem is that the item name frame is not detected when it is impossible to detect the form of the table of cards or the line of characters due to noise, broken lines, or the like.

In particular, the first problem is a big problem in actual operation. In general, the letter in which the item name is described is small, and since the contact is frequent between the letter and the letter and between the letter and the ruled line, the probability of the letter being recognized is low. In addition, the probability of recognizing characters of low quality images including crushing, breaking, and noise is further reduced.

According to one typical aspect of the present invention, there is provided a processor for performing arithmetic processing, a storage unit connected to the processor, and an interface connected to an image input device for inputting image information of a document to a document recognition device. In a book recognition device for recognizing image information of a book entered by a device, the processor detects a frame and a character line of the image information of the book entered, detects a character string in the detected frame, and Recognizing the item name in the list from the detected character string, detecting the frame in which the recognized item name exists as an item name frame, detecting a data frame corresponding to the recognized item name, based on the recognized item name, and detecting the An item of recognition processing for recognizing a character string in the data frame, wherein the character string is not recognized in the recognition process. Or not, and if it is determined that there is the unrecognized item name, the data frame corresponding to the unrecognized item name is detected again based on the unrecognized item name, and Characterized in that the character string is recognized.

The present invention will be described with reference to the drawings. In addition, this invention is not limited by the following description.

1 is a diagram illustrating a configuration of a book recognition system according to an embodiment of the present invention.

The book recognition system includes an input device 10, an image input device 20, a book recognition device 30, a recognition dictionary 40, a display device 50, and an image database (DB) 60.

The input device 10 is a device such as a keyboard and a mouse for inputting commands, code data, and the like into the book recognition device 30.

The image input device 20 is a device such as a scanner for inputting a document to the document recognition device 30 using the document as image data.

The book recognition device 30 is a calculator for detecting a reading area of a book input by the image input device 20 and recognizing the letters of the book, and includes a CPU, a memory, and a storage device (not shown).

The recognition dictionary 40 is a dictionary database referred to when the book recognition device 30 recognizes a book. Specifically, the recognition dictionary 40 stores a character recognition dictionary 610 which is referred to when the book recognition device 30 recognizes letters, and the book authentication device 30 detects a book reading area. It stores the book information and the like to be referred to.

The display device 50 is a device such as a display that displays the result of the book recognition by the book recognition device 30.

The image DB 60 stores image data input to the document recognition device 30 by the image input device 20. In addition, the image DB 60 may store in advance image data that is an object recognized by the document recognition device 30 by the image input device 20.

In addition, the present invention may be mounted on an ordinary calculator by software having the same function as the book recognition device 30.

Next, the outline | summary and effect of this invention are demonstrated.

The summary of the present invention is based on the characteristics (item name, arrangement between item name frames, and arrangement relationship between item name frames and data frames) based on a small number of types of books when the book recognition device 30 recognizes a high degree of freedom of formatting. Etc.), and recognizes the item name of the entered document and analyzes the position and attributes of the data frame. In addition, the summary of the present invention is that when the character strings of the item names are incorrectly recognized, when the item names with the same contents are differently displayed, and when the frames are incorrectly detected, the book recognition device 30 may not accurately display the data frames. It is to remedy the situation which is not detected.

Specifically, the item name is recognized, and a data frame is detected from the recognized item name. Then, the character string of the detected data frame is recognized. If there is an item that is not detected in this series of processing, the retry process is repeated in the following manner to save the item that could not be detected.

First, the detected item name frame and data frame are excluded from the object of the retry process, and the retry process is executed with the other area as the object of the retry process. Next, all the item names included in the table are compared with the registered item name dictionary and the recognized item name, thereby specifying an unrecognized item name. Next, a retry process is performed for the specified item name.

The character string of the area where the item name is not recognized is recognized, and the item name is recognized based on the attribute of the data of the character string of the recognized area.

Part or all of these retry processes can be performed in series or in parallel to provide a book recognition device 30 that is highly likely to be recognized.

FIG. 2 is a flowchart showing the outline of the book processing 100 executed by the book recognition device 30 of the embodiment of the present invention. In addition, the book process 100 demonstrates the detail in FIG.

The book processing 100 executed by the book recognition device 30 is roughly divided into a recognition process 110, a retry preprocess 200, and a retry process 700.

The recognition processing 110 associates item names and data from the image data of the pay book input to the pay book recognition device 30 to recognize the letters of the pay book.

The retry preprocessing 200 detects an item name that could not be recognized by the recognition processing 110.

The retry process 700 recognizes the item name detected by the retry preprocess 200 again. The retry process 700 is a generic term for the retry process A 300, the retry process B 400, and the retry process C 500 in FIG. 3.

Next, the details of the book processing 100 will be described with reference to FIG. 3.

FIG. 3 is a flowchart showing the details of the book processing 100 executed by the book recognition device 30 of the embodiment of the present invention.

First, the recognition dictionary 40 will be described. In the recognition dictionary 40, the character recognition dictionary 610, the item name dictionary 620, the item name frame-data frame arrangement 630, the adjacency relationship 640 between the item name frame, and the item-specific usage knowledge 650 are stored. do.

In the character recognition dictionary 610, character information for detecting a character from the image data of the journal input to the journal recognition device 30 is registered. In the item name dictionary 620, all the item names included in each book are registered for each book.

The item name frame-data frame arrangement 630 stores the arrangement relationship between the item name frame and the data frame. Specifically, the item name frame-data frame arrangement 630 stores arrangement relationships such as whether the data frame is disposed below the item name frame and the data frame is arranged on the left side of the item name frame.

In the adjacent relationship 640 between the item names, the relationship between any item name and the item name adjacent to the item name is stored. In addition, the adjacent relationship 640 between item names demonstrates the detail in FIG. 6, FIG. 8, and FIG.

The item usage knowledge 650 is referred to when the book recognition device 30 detects an item name from a character string of a data frame. As shown in FIG. 14, the item-use knowledge 650 includes the item name 6501 and the description content 6502 as shown in FIG. 14. In the item name 6501, the item name included in the table is registered. In the description content 6502, the content described in the data corresponding to each item name 6501 is registered.

In addition, the description content 6502 corresponding to the item name 6501 "name" included in the item-specific use knowledge 650 may register the first and last name separately, and the description content corresponding to the item address 6501 "address". 6502 may be registered hierarchically by Dodofuken (regional self-government), a municipality (basic self-government), or the like. In addition, "N" contained in description content 6502 corresponding to item name 6501 "telephone number" and "amount of money" represents a number from 0 to 9.

Next, the detail content of the book process 100 is demonstrated.

As described above, the document processing 100 includes a recognition process 110, a retry preprocess 200, a retry process A 300, a retry process B 400, and a retry process C 500. Equipped. In addition, all of the retry processing A 300, the retry processing B 400, and the retry processing C 500 may not be executed. In addition, when the book recognition device 30 executes the retry processing A 300, the retry processing B 400, and the retry processing C 500 in series, the retry processing A 300 and re The order of executing the try processing B 400 and the retry processing C 500 is arbitrary. In addition, the book recognition device 30 may execute the retry processing A 300, the retry processing B 400, and the retry processing C 500 in parallel, and integrate the results of the execution.

If the retry process C 500 is not executed, the item usage knowledge 650 may not be stored in the recognition dictionary 40.

First, when the image data of a book is input to the book recognition device 30, the book recognition device 30 performs preprocessing (120). The preprocess is a process in which the ticket recognition device 30 recognizes a table from the input image data and detects a character row from the recognized table.

Here, "recognizing a table" means to detect the position of a frame from image data input to the book recognition device 30. In addition, a "frame" is a rectangular area enclosed by the vertical ruled line and the horizontal ruled line of the long list. The method described in JP-A 11-53466 can be applied to the method in which the book recognition device 30 recognizes a table.

In addition, there is a method of analyzing the connection component as a method in which the book recognition device 30 detects a character line. Here, the method of analyzing the connection component means that the text device 30 integrates adjacent connection components (lumps in which black pixels are continuous) in the frame in the horizontal direction or the vertical direction, so that the character line in the horizontal direction or the vertical direction is used. Is a method of detecting an image of a character line.

Next, the book recognition apparatus 30 recognizes the character string of the item name from the character string contained in the frame detected by the process of step 120 (130). Specifically, the book recognition apparatus 30 cuts out the area | region for every character from the image of the character line detected by the process of step 120. FIG. And the book recognition apparatus 30 references the character recognition dictionary 610, recognizes the pattern of the area | region cut out, and specifies each character, and specifies the character string contained in a frame. Then, the book recognition apparatus 30 compares the specified character string with the character string registered in the item name dictionary 620, and if the specified character string matches the character string registered in the item name dictionary 620, The specified string is recognized as the item name.

In addition, in order to prevent the book recognition apparatus 30 from erroneously recognizing the item name, if the specified character string and the character string registered in the item name dictionary 620 do not completely match, the recognition may not be performed with the item name.

In addition, the method of Unexamined-Japanese-Patent No. 2004-171316 can be applied to the method of specifying the character string contained in a template by the book recognition apparatus 30. As shown in FIG.

Next, the book recognition device 30 detects the frame including the item name and the recognized character string in the process of step 120 as the item name frame (140).

In operation 150, the book recognition apparatus 30 detects a data frame corresponding to each item name frame with reference to the item name frame-data frame arrangement 630. In addition, the book recognition apparatus 30 may detect the data frame corresponding to each item name frame from the item name frame and the arrangement information of the data frame without referring to the item name frame-data frame arrangement 630. When the book recognition apparatus 30 does not refer to the item name frame-data frame arrangement 630, the data is stored in a frame not detected as an item name frame existing around the item name frame (typically, below or to the right of the item name frame). Detects as a template.

Next, the book recognition apparatus 30 recognizes the character string contained in the data frame detected by the process of step 150 (160).

Here, the book recognition device 30 may recognize the character string included in the data frame with reference to the item-specific usage knowledge 650.

Specifically, the book recognition device 30 recognizes a character string included in the data frame while switching the information on the description content 6502 for each item name and comparing it with the information about the description content corresponding to each item name. . That is, the book recognition device 30 estimates the type (eg, number) of character strings to be included in the data frame from the information registered in the description content 6502 of each item name, and calculates the character strings included in the data frame. Recognize.

For example, when the book recognition apparatus 30 recognizes the character string contained in the data frame corresponding to the item name "amount", it refers to the description content 6502 of "amount" of the item name, and the item name " The character string included in the data frame corresponding to "amount of money" recognizes the character string included in the data frame on the premise that it is a number.

On the other hand, when the item name corresponding to the data frame is not registered in the item-use knowledge 650, or when the item-use knowledge 650 is not stored in the recognition dictionary 40, the book recognition device 30 ) Recognizes the character string contained in the data frame corresponding to each item name as it is.

Then, the book recognition device 30 executes the recognition processing 110 for all character lines.

When the recognition process 110 is executed, the book recognition device 30 determines whether there is an item name that is not recognized in the recognition process 110. As a result of the execution, when there is an item name that is not recognized among the item names registered in the item name dictionary 620, the book recognition device 30 executes the retry preprocessing 200. In addition, when all item names are recognized as a result of the recognition processing 110 being executed, the book recognition device 30 ends the book processing 100.

Next, the retry preprocessing 200 executed by the book recognition apparatus 30 when the item name is not recognized as a result of the recognition processing 110 is described.

In the retry preprocessing 200, in the processing executed before the retry processing 700 is executed, as a result of the recognition processing 110, an unrecognized item name and data corresponding to the item name are specified.

By performing the recognition process 110, the book recognition apparatus 30 detects the area | region (henceforth an unread area | region) except the detected item name frame and data frame (210). In addition, an undetected item name frame exists in the unread area.

Next, the book recognition apparatus 30 specifies the item name which is not recognized from the unread area detected by the process of step 210 (220). Specifically, the book recognition apparatus 30 compares the item name registered in the item name dictionary 620 with the recognized item name, and excludes the recognized item name among the item names registered in the item name dictionary 620, thereby not being recognized. Specifies the item name. In addition, the book recognition device 30 may execute the process of step 220 when it is determined whether there is an item name that is not detected in the recognition process 110.

Then, the book recognition device 30 detects the position of the item name frame of the unrecognized item name with reference to the adjacent relationship 640 between the item name frames (230). In addition, when the adjacent relationship 640 between item name frames is not stored in the recognition dictionary 40, the process of step 230 may be abbreviate | omitted.

Here, when a plurality of item name frames which are not detected are detected, and when the adjacent relationship 640 between the item frames is not stored in the recognition dictionary 40, that is, the position of the item name frames which are not detected is unique. If it is not detected, the book recognition device 30 executes the retry process 700. On the other hand, when one item name which is not recognized is detected, the book recognition apparatus 30 returns to step 140 process.

Next, the recognition processing 110 and the retry preprocessing 200 will be described using specific examples.

4 is a diagram showing an example of the list itself of the embodiment of the present invention.

When the book shown in FIG. 4 is input to the book recognition device 30, the book recognition device 30 executes the recognition process 110. However, in the table shown in Fig. 4, there is noise generated at the time of printing on the character of "number" of "telephone number". In the process of step 130, the book recognition device 30 cannot recognize a character string with noise in the input book. On the other hand, the book recognition apparatus 30 recognizes the character string contained in the item name of "name", "address", and "money amount". And the book recognition apparatus 30 references these item name dictionary 620, and specifies these item names in the process of step 140. FIG. In addition, in the process of step 150, the book recognition apparatus 30 refers to the item name frame-data frame arrangement 630 and forms a frame arranged under the item name frame of "name", "address", and "amount". It detects as a data frame and recognizes the character string of the detected data frame in the process of step 160.

The book recognition device 30 executes the retry preprocessing 200 because there is an item name not recognized among the item names registered in the item name dictionary 620.

In the process of step 210, the book recognition apparatus 30 detects the area | region 500 enclosed by the thick line shown in FIG. 5 as an unread area | region. Next, in the process of step 220, the book recognition apparatus 30 refers to the item name dictionary 620, and specifies the item name which is not recognized. Here, "name", "address", "telephone number", and "amount of money" are registered in the item name dictionary 620. Since the "name", "address", and "amount" are recognized as the item name in the processing of Step 130, the book recognition device 30 specifies that the item name that is not recognized is "phone number".

Next, in the process of step 230, the book recognition apparatus 30 refers to the adjacency relationship 640 between the item name frames shown in FIG. It is detected that it is arranged between the item name frames of the "money amount".

In addition, the adjacency relationship 640 between the item name frames shown in FIG. 6 includes an item name 6401 and an adjacent item name frame 6402. In the item name 6401, the item name registered in the item name dictionary 620 of each book is registered. The adjacent item name frame 6402 includes "left", "right", "upper", and "lower" indicating an adjacent relationship between the item name frame registered in the item name 6401 and the item name adjacent to the frame. .

When there is no item name adjacent to the item name registered in the item name 6401, "-" is registered in the adjacent item name frame.

Next, a method of detecting the position of the item name "telephone number" specified as the item name not recognized in the process of step 210 will be described.

First, the book recognition apparatus 30 refers to the adjacent item name frame 6402 and detects an item name adjacent to the "telephone number". Here, the "address" and the "money amount" are detected. That is, the "phone number" exists between "address" and "amount".

In addition, since the item name frame of "address" and "amount" is detected in the process of step 140, the position of the item name frame of "telephone number" is uniquely detected.

Therefore, since the position of the item name frame of "telephone number" was uniquely detected, the book recognition device 30 returns to the process of step 140, determines the item name frame again, and then, after the process of step 150, the data frame. Recognize.

Adjacent relations between the item name frames in the case where a list of documents having different adjacencies in item names (for example, a document shown in FIG. 4 and a document shown in FIG. 7) are input to the document recognition device 30 in a mixed manner. As shown in Fig. 8, a plurality of adjacent relations may be registered for one item name.

In addition, as shown in the table shown in FIG. 9, when each item name is adjacent to each other up and down, the adjacency relationship 640 between the item name frames is, as shown in FIG. Adjacent relationships of item names are registered in the "bottom" column.

As described above, when the position of the item name frame which is not recognized is uniquely detected, the book recognition device 30 executes the retry process 700.

First, the retry process A 300 will be described.

The book recognition apparatus 30 recognizes the character string of the frame of the unread area, and recognizes the item name with reference to the item name dictionary 620 (310). When the book recognition device 30 recognizes the item name, the book recognition device 30 returns to the process of step 140.

The process of step 310 is demonstrated using a specific example.

Here, when the form shown in FIG. 4 is input into the form recognition device 30, the thick line frame 500 of the form shown in FIG. 5 detected as a frame of the unread area | region by the process of step 210 is step 310. Is the subject of processing.

In addition, in the process of step 230, although the uppermost frame in the thick line frame 500 was detected as an item name frame, the case where the item name frame was not detected in the thick line frame 500 is demonstrated. In this case, it is assumed that the item name not detected in the processing in step 220 is specified as "phone number".

Since the item recognition device 30 specifies that the item name that has not been detected is "telephone number", the "telephone" registered in the text string recognized in the frame of the unread area and the item name dictionary 620 in the process of step 310. By comparing the number, the character string closest to the "telephone number" among the character strings recognized within the frame of the unread area is recognized as the item name "telephone number".

In addition, when there exists a plurality of item names that are not detected, all item names that are not recognized may be used as the target of verification.

That is, the book recognition apparatus 30 recognizes the item name by specifying the unrecognized item name, reducing the item name registered in the item name dictionary 620 matching with the recognized character string in the frame of the unread area. As a result, even if the condition for recognizing the item string is matched with the character string recognized in the frame of the unread area and the item name registered in the item name dictionary 620, the book recognition device 30 is less likely to recognize the item name incorrectly.

Next, the method of alleviating a control condition is demonstrated concretely.

A method of mitigating the first control condition will be described. Even if the character string recognized in the frame of the unread area is one character different from the item name registered in the item name dictionary 620, the book recognition device 30 sets the character string recognized in the frame of the unread area in the item name dictionary. Recognized as an item name registered in (620). Thereby, even if the "number" of the "telephone number" of the unread area | region in the thick line frame 500 shown in FIG. 5 is not recognized because of noise, the ticket recognition apparatus 30 is "the telephone of the unread area | region. Number ”can be recognized as an item name.

A method of mitigating the second control condition will be described. The book recognition apparatus 30 recognizes within the frame of the unread area even when the character string recognized within the frame of the unread area includes characters other than the string of item names registered in the item name dictionary 620. The recognized character string is recognized as an item name registered in the item name dictionary 620. Thus, for example, even if the character string recognized in the frame of the unread area is "」 所 ", and the item name registered in the item name dictionary 620 is" 住所 ", the book recognition device 30 is not allowed. The character string "ご 住所" recognized in the frame of the reading area can be recognized as the item name "住所". As a result, the book recognition device 30 can also recognize the item name not registered in the item name dictionary 620.

A method of mitigating the third control condition will be described. The book recognition apparatus 30 recognizes the character string recognized in the frame of the unread area as the item name even if the character string recognized in the frame of the unread area and the item name registered in the item name dictionary 620 are different from each other. .

Specifically, the recognition dictionary 40 of the book recognition device 30 stores the item name dictionary 620 shown in FIG. The item name dictionary 620 shown in FIG. 11 includes an item name 6201 and notation 6202. In the item name 6201, the item name included in the book is registered. In the notation 6202, a character string recognized by the book recognition device 30 as an item name is registered. In the character string registered in the notation 6202, a character string having the same meaning as that indicated by the item name may be registered. For example, "name" and "name" are registered in notation 6202 of "name" in item name 6201, and "住所" and "ご 住所" in notation 6202 of item "6201". Is registered.

As a result, even if the character string recognized in the frame of the unread area and the item name registered in the item name dictionary 620 are different from each other, the book recognition device 30 displays the character string recognized in the frame of the unread area. When included in the notation 6202 of the dictionary 620, it may be recognized as an item name corresponding to the notation 6202.

In addition, the method of alleviating the 3rd control condition may be applied to the process of step 130.

The processing of step 310 for relieving the matching condition and recognizing the character string of the item name can be applied to the processing of step 130. However, when the method of alleviating the first and second collation conditions is applied to the process of recognizing the character string of the first item name, the book recognition, such as a case where a similar item name is included in the book input to the book recognition device 30, is applied. The device 30 is likely to misrecognize the item name. On the other hand, even if the method of alleviating the third matching condition is applied to the process of step 130, the possibility that the book recognition device 30 recognizes the item name is the same as that of the matching condition of perfect match.

The process of recognizing the character string of the item name by alleviating the matching condition is executed after limiting the recognized item name except for the item name recognized in the recognition process 110. Therefore, the book recognition device 30 is less likely to misrecognize the item name.

Next, the retry processing B 400 will be described.

The book recognition device 30 redetects the frame and the text line from the input image data (410), and recognizes the text string of the item name (420). In addition, the process of step 420 is the same as the process of step 310. When the character string of the item name is detected in the processing of step 420, the processing returns to the processing of step 140.

In addition, when the item name is not recognized by the book recognition device 30, the retry processing B 400 may incorrectly detect a frame or character line in the process of Step 120. And rediscover character lines. As a result, the frame or character line is incorrectly detected, and the item name or data is incorrectly recognized.

The retry process B 400 will be described using a specific example.

When the book shown in FIG. 12 is input to the book recognition device 30, the book shown in FIG. 12 has noise generated when printing, for example, a vertical line in the frame of the “address”. ) Detects the frame by dividing the frame into two vertical lines in the processing of Step 120, and cannot detect the character line of "address".

Therefore, in the process of step 210, the book recognition apparatus 30 detects the thick line frame 1200 containing the item name of "address" as an unread area | region as shown in FIG.

In the unread area detected by the processing in step 210, only the frames divided by the vertical lines have different widths of the frames, so the book recognition device 30 adjusts the frames having different widths to the widths of the different frames. , The frame is redetected and the text lines are redetected (410).

By performing the process of step 410, the book recognition apparatus 30 detects the character line of "address", and similarly to the process of step 310, the matching condition is relaxed to recognize the character string of the item name.

In the table shown in FIG. 12, although the "note" of "address" has a vertical line noise, the book recognition device 30 can recognize the item name as "address" because the check condition is relaxed.

In addition, when a part of the character is recognized as a ruled line and the frame is incorrectly detected, the book recognition device 30 may redetect the frame even if a part of the character detected as the ruled line is not a ruled line.

Next, the case where the book form shown in FIG. 13 is input to the book form recognition apparatus 30 is demonstrated. In the book shown in Fig. 13, since the character string of "phone number" has two lines, and the line of "telephone" and "number" is narrow, the book recognition device 30 performs the "telephone" in the process of step 120. Number "is detected as one character line. For this reason, in the process of step 130, the book recognition apparatus 30 cannot recognize the item name of "telephone number." In addition, in the process of step 210, the book recognition device 30 detects the inside of the thick line frame 1300 including the item name of "telephone number" as an unread area.

In the process of step 410, the book recognition apparatus 30 determines whether the height of the character line of the frame of an unread area | region is higher than the height of another character line, and determines the height of the frame of an unread area | region than the height of another character line. The character line of the frame of the unread area | region of which the height of a character line is high is divided, and a character line is redetected.

Specifically, when the "telephone number" which is two lines of the book shown in FIG. 13 is recognized as one line, since the height of the text line of "phone number" is higher than other text lines, the book recognition device 30 is Two lines of "telephone" and "number" are detected. And the book recognition apparatus 30 connects "telephone" and "number", and recognizes the item name "telephone number" in contrast with the item name dictionary 620.

In addition, although the above-described retry processing B 400 has described an example of recognizing the item name, the retry processing B 400 can be applied even when the book recognition device 30 incorrectly detects a frame and a character line of data.

Next, the retry process C 500 will be described.

When the item recognition device 30 cannot execute the retry process A 300 and the retry process B 400 even if the item name cannot be recognized, the book recognition device 30 recognizes the character string of the data (510). On the basis of this, the item name is recognized (520), and the item name frame is detected (530). And when the item recognition apparatus 30 detects an item name frame, it returns to step 160. FIG.

The retry process C 500 will be described using a specific example.

When the retry process C 500 is executed, the item-specific usage knowledge 650 shown in FIG. 14 is stored in the recognition dictionary 40 of the book recognition device 30.

When the book shown in FIG. 13 is input to the book recognition device 30, in the book shown in FIG. 13, since the letter of "phone number" contacts a ruled line, the book recognition device 30 is a "phone number." Character string cannot be recognized. And the book recognition apparatus 30 detects the thick line frame 1300 which contains the item name of "telephone number" as an unread area | region in the process of step 210 as shown in FIG.

In the process of step 510, the book recognition apparatus 30 recognizes the character string of an unread area | region. The book recognition device 30 then detects the item name 6501 in which the recognized character string is included in the description content 6502 by referring to the item-specific knowledge 650, and recognizes the detected item name as the item name. .

In the book shown in FIG. 13, the character strings of the second and third lines of the unread area are only numbers and hyphens, and the number of digits and the description form are limited. Reference is made to the telephone number as an item name. The technique described in Japanese Patent Laid-Open No. 8-243505 can be applied to collation of a string of numbers having a specific format.

In addition, in the process of step 530, the book recognition apparatus 30 detects the frame of the upper end of an unread area | region, or the frame containing a character string other than a number as a frame of an item name "telephone number". In addition, the upper end of the unread area is detected by the item name frame because the item name usually exists at the upper end or the left end of the table.

In addition, before performing the process of step 530, the book recognition apparatus 30 may perform the process of step 310 again.

In the above-described embodiment, the recognition process 110 is executed to give an example of the case where the number of items that are not recognized is singular. However, even when a plurality of items are not recognized, the retry process 700 is executed. It can apply by increasing the item which can recognize the book recognition apparatus 30. If the item is not recognized even if the retry process 700 is repeatedly executed, the book recognition device 30 cannot recognize the unrecognized item, and thus cannot display the item. Output to 50.

According to one embodiment of the present invention, a similar but different form can be recognized without a strict definition. In addition, even when the character of the item name is unlikely to be recognized, it can be recognized while analyzing the attribute of the table with high precision.

Claims (16)

A processor for performing arithmetic processing, a storage unit connected to the processor, and an interface connected to an image input device for inputting the image information of the document to the document recognition device, wherein the image information of the document entered by the image input device is provided. In the book recognition device for recognizing, The processor, Detecting a frame and a character line of the image information of the input list; Detecting a character string in the detected frame, recognizing an item name in a table from the detected character string, A frame in which the recognized item name exists is detected as an item name frame, Based on the recognized item name, detecting a data frame corresponding to the recognized item name, Perform a recognition process of recognizing a character string in the detected data frame, In the recognition processing, it is determined whether there is an item name for which a character string is not recognized, If it is determined that the unrecognized item name is found, the data frame corresponding to the unrecognized item name is detected again based on the unrecognized item name, And recognizes a character string of the detected data frame. The method of claim 1, And the processor detects the unrecognized area by excluding the detected item name frame and the detected data frame from the input image information. The method of claim 1, The storage unit stores an item name dictionary in which item names to be included in the table are registered. The processor, When the recognition processing is executed, the item name registered in the item name dictionary is compared with the recognized item name. And recognizing the unrecognized item name by excluding the recognized item name among the item names registered in the item name dictionary. The method of claim 1, The processor, If it is determined that there is an unrecognized item name, an unrecognized area is detected from the image information of the input list except for the detected item name frame and the detected data frame, Detect a character string in the unrecognized region, And recognizing the unrecognized item name from the detected character string under conditions different from the step of recognizing the item name in the recognition processing. The method of claim 4, wherein The processor, In the step of recognizing the item name in the recognition process, by comparing the detected character string with the item name registered in the item name dictionary, the item name in the table is recognized, If it is determined that the item name is not recognized, Detect a character string in the unrecognized region, The condition for matching the detected character string with the item name registered in the item name dictionary is relaxed than the condition for matching the detected character string in the recognition process with the item name registered in the item name dictionary to recognize the unrecognized item name. A book recognition device, characterized in that. The method of claim 4, wherein The processor, Detecting the frame included in the unrecognized region again from the detected frame, Detecting a character string in the again detected frame, And recognize the unrecognized item name from the detected character string. The method of claim 4, wherein The processor, In the detected character line, detecting a character line included in the unrecognized area again, Detect a character string based on the again detected character line, And recognize the unrecognized item name from the detected character string. The method of claim 4, wherein The processor, Recognize a character string from the unrecognized area, And recognizes the most reliable item name from the recognized character string. The image recognition device is inputted to the document recognition device having a processor for performing arithmetic processing, a storage unit connected to the processor, and an interface connected to an image input device for inputting image information of the document to the document recognition device. As a program for recognizing image information of a book, The program is stored in the ticket recognition device. Detecting a frame and a character line of the image information of the input list; Detecting a character string in the detected frame to recognize an item name in a table from the detected character string, Detecting a frame in which the recognized item name exists, as an item name frame, Based on the recognized item name, detecting a data frame corresponding to the recognized item name, Perform a recognition process of recognizing a character string in the detected data frame, In the recognition processing, it is determined whether there is an item name for which a character string is not recognized, Based on the unrecognized item name, again detecting a data frame corresponding to the unrecognized item name, And a character string of the detected data frame. The method of claim 9, And the program detects the unrecognized area by excluding the detected item name frame and the detected data frame from the input image information. The method of claim 9, In the storage unit provided in the book recognition device, an item name dictionary in which an item name to be included in the book is registered is registered, The program is in the ticket recognition device, When the recognition processing is executed, the item name registered in the item name dictionary is compared with the recognized item name, And recognizing the unrecognized item name by excluding the recognized item name from the item names registered in the item name dictionary. The method of claim 9, The program is in the ticket recognition device, If it is determined that there is an unrecognized item name, the detected item name frame and the detected data frame are excluded from the image information of the input list to detect an unrecognized area, Detect a character string in the unrecognized region, And recognizing the unrecognized item name from the detected character string under conditions different from the step of recognizing the item name in the recognition processing. The method of claim 12, The program is in the ticket recognition device, In the step of recognizing the item name in the recognition process, by contrasting the detected character string with the item name registered in the item name dictionary, the item name in the table is recognized, If it is determined that the item name is not recognized, Detect a character string in the unrecognized region, The condition for matching the detected character string with the item name registered in the item name dictionary is relaxed than the condition for matching the detected character string in the recognition process with the item name registered in the item name dictionary to recognize the unrecognized item name. The program characterized in that the. The method of claim 12, The program is in the ticket recognition device, In the detected frame, the frame included in the unrecognized area is detected again, Detect a character string in the again detected frame, And recognize the unrecognized item name from the detected character string. The method of claim 12, The program is in the ticket recognition device, In the detected character line, detecting a character line included in the unrecognized area again, Detect a character string based on the again detected character line, And recognize the unrecognized item name from the detected character string. The method of claim 12, The program is in the ticket recognition device, Recognize a string from the unrecognized region, And recognizing the most reliable item name from the recognized character string.

Images