JP2005128865A - Data recognition device and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a data recognition device that enhances a recognition rate for data recognized from slips or forms and that is capable of accurately recognizing data written on slips or forms, by correcting misrecognition fields. <P>SOLUTION: Slip image data are recognized and computational relationships included in recognized detail lines are predefined by a definition tool 12. Defined computations are performed on the result of recognition on the field of each detail line and the recognition result is compared with the computation result to specify the field where a misrecognition has occurred. The defined computations are performed on the result of recognition on the field supposed to have been correctly recognized, so as to calculate a correct value for the field where the misrecognition is supposed to have occurred. The value for the field where the misrecognition is supposed to have occurred is corrected to provide the correct value. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a data recognition apparatus and a program for reading characters, numbers, and symbols from reflected light of irradiated light and performing a correction operation on numerical values included in an arithmetic expression.

  In recent years, technology for taking characters written on paper into a computer as character codes has been dramatically improved. In general, a method is used in which characters and figures written on paper are read by an image scanner, and character portions in the read image data are converted into character codes by dedicated software. This dedicated software is called OCR software and is widely used for business use and personal use. The recognition rate determines whether the OCR software is good or bad.

  There are two main approaches for improving the recognition rate of OCR software. One is a method of “improving the recognition rate of one character per character”, and the other is a method of “deriving more probable recognition candidates from the preceding and following contexts and field attributes”. A typical example of the latter is address recognition. For example, as shown in FIG. 24, a customer database or a product database is held in advance, and a more probable result is derived while comparing the recognized character string with the database.

  Furthermore, there is a method for increasing the recognition rate from the relationship between a plurality of fields. Specifically, the probability of the recognition result of the telephone number field is compared with the probability of the recognition result of the address field, and the other is corrected based on the more probable field recognition result. Thereby, the recognition rate of OCR software can be improved.

There is also a known method for detecting misrecognition by calculating recognition fields that are in a calculation relationship with each other in order to detect misrecognition of a detail line of a slip and comparing the calculation result and the recognition result (for example, , See Patent Document 1). In this method, as shown in FIG. 25, the calculation of quantity × unit price, which are in a calculation relationship with each other, is performed, and the calculation result is compared with the field of the recognized amount of money to detect misrecognition. In addition, calculation of the amount field is performed, and the erroneous recognition is detected by comparing with the recognized total amount, and the erroneously recognized field is determined based on the detection result. Specifically, “2 (quantity) × 300 (unit price) ≠ 500” is detected in the pineapple row. In the total amount, “500 + 100 + 500 + 2000 + 6000 ≠ 9200” is detected. Then, based on this detection result, it is determined that the field that is likely to be misrecognized is “Pineapple amount”. As a result, it is possible to accurately derive a more reliable field recognition result and a field recognition result that is likely to be erroneously recognized.
JP 2002-56354 A

  However, in the above-described conventional optical character reading device (for example, Patent Document 1), it is possible to detect erroneous recognition but not to correct the erroneous recognition field. Therefore, when an erroneous recognition is detected by the optical character reader, it is necessary to perform a correction operation by an operator. Here, when there are a large number of correction points, there is a problem that the work is complicated, and there are not a few cases where erroneous correction is performed due to human factors such as an operator.

  An object of the present invention is to improve the recognition rate of data recognized from a slip or form, and to recognize the data described in the slip or form with high accuracy by correcting a misrecognition field And providing a program.

In order to solve the above problem, the invention according to claim 1
A data recognition device that reads information entered in a form as image information and performs character recognition of the image information,
A definition file for storing setting information for defining fields included in the image information;
Recognition means for recognizing a field included in the image information based on the setting information;
Input means for inputting an arithmetic relationship between the recognized fields;
A discriminating means for performing an arithmetic operation between recognized fields based on the arithmetic relation, comparing the recognition result of the field with the arithmetic result, and determining whether or not there is a misrecognition;
Calculating means for calculating the value of the field determined to have the misrecognition based on the recognition result of the field determined to have no misrecognition when it is determined that the field recognition result has misrecognition; ,
It is characterized by having.

  According to the first or fourth aspect of the present invention, the information entered in the form is read as image information, the field included in the image information is recognized based on the setting information, and the arithmetic relationship between the recognized fields is input. Based on the input calculation relationship, the calculation between recognized fields is performed, and based on the recognition result and the calculation result, it is determined whether there is a misrecognition in the recognition result of each field. For a field determined to have erroneous recognition, an operation can be performed based on the recognition result of the field determined to have no erroneous recognition, and a correct operation result can be obtained. As a result, it is possible to easily and accurately correct the field detected as erroneous recognition.

  According to the second aspect of the present invention, since the inputted calculation relation is stored in the storage means as calculation information, for a form having the same format, the form is recognized based on the calculation relation defined in the past. Can do. Thereby, business efficiency can be improved significantly.

  According to the third aspect of the present invention, the field in which the recognition result and the calculation result are different and the probability of the recognition are stored, and whether or not the recognized field has a misrecognition based on the certainty of the recognition. In order to discriminate, it is possible to accurately discriminate between a field having no erroneous recognition and a field that has been erroneously recognized. As a result, the recognition rate of the data recognition apparatus can be improved, and an accurate value can be calculated for the misrecognized field.

  Hereinafter, the best mode for carrying out the present invention will be described in detail with reference to the drawings. However, the scope of the invention is not limited to the illustrated examples.

First, the configuration of the present embodiment will be described.
FIG. 1 is a diagram showing a schematic configuration of a data recognition apparatus 1 in the present embodiment. As shown in FIG. 1, the data recognition device 1 includes a reading device 11, a definition tool 12, a definition file 13, an OCR recognition engine 14, a business application 15, a storage device 16, an input device 17, a display device 18, and the like. Is done.

  The reading device 11 acquires a slip or a form as image data, and outputs the acquired image data to the definition tool 12 and the OCR recognition engine 14. FIG. 2 shows a configuration block diagram of the reading device 11. As shown in FIG. 2, the reading device 11 includes a reading unit 11a, an image processing unit 11b, a memory unit 11c, a scanner control unit 11d, and an interface unit 11e.

  The reading unit 11a includes a light source unit 111 that irradiates light on a slip, a CCD (Charge Coupled Device) 112 that reads light reflected from the slip, and an A / D conversion unit 113 that performs analog / digital conversion on an image signal read by the CCD 112. And. The reading unit 11a irradiates the slip with light from the light source unit 111 and detects the reflected light by the CCD 112, thereby acquiring the image data of the slip.

  The image processing unit 11b performs image processing such as binarization, smoothing, and noise removal on the image data read by the reading unit 11a. The memory unit 11c stores the image data of the slip that has undergone image processing in the image processing unit 11b.

  The scanner control unit 11d reads various control programs stored in the storage device 16, develops them in the RAM, and controls each unit of the reading device 10 according to these control programs. The interface unit 11f outputs the slip image data stored in the memory unit 11c to the definition tool 12, the OCR recognition engine 14, and the display device 18 under the control of the scanner control unit 11d.

  Returning to FIG. 1, the definition tool 12 has a role for defining setting information such as the positions of recognition fields scattered on the image data of the read slip, the number and type of characters of the recognition fields, and the arithmetic relationship between the recognition fields. It is a tool that bears. Specifically, the definition tool 12 reads various processing programs from the storage device 16 and executes main processing, image reading processing, field setting processing, and calculation setting processing described later. Details of each process will be described later.

  Hereinafter, information defined by the definition tool 12 will be described with reference to FIGS. FIG. 3A shows a setting screen 181 for setting the position of the recognition field, the number of characters in the recognition field, and the character type from the read slip. As shown in FIG. 3A, for example, a field 181a for inputting a customer name, a field 181b for inputting an address, and a field 181c for inputting a telephone number are provided in the upper part of the setting screen 181. Here, when the range of each field (for example, a rectangular area) is designated by the mouse or the like provided in the input device 17, the definition tool 12 causes the position information (for example, the upper left coordinate and the lower right coordinate) of the designated rectangular area. ) Is acquired, and the designated rectangular area is set as the recognition field.

  When the rectangular area of the field is designated, a setting screen 182 shown in FIG. 3B is displayed. In this setting screen 182, the number of characters and the character type of the corresponding recognition field are set. That is, the number of characters and the character type input on the setting screen 182 are acquired in association with the designated field range, and stored in the definition file 13 as setting information.

  In the lower part of the setting screen 181, a field 181d for inputting a product name, a field 181e for inputting a quantity, a field 181f for inputting a unit price, a field 181g for inputting an amount, and a field 181h for inputting a total amount are provided. Yes. In the fields 181d to 181h, the position of the recognition field, the number of characters in the recognition field, and the character type are similarly set on the setting screens 181 and 182.

  Furthermore, in the fields 181d to 181h shown in the lower part of the setting screen 181, calculation relationships are established, and these calculation relationships are defined by the definition tool 12. With reference to FIG. 4, the calculation relationship defined by the definition tool 12 will be described.

FIG. 4A is a diagram illustrating a data configuration example of a slip having two types of calculation relationships. First, as the first calculation relationship, as shown in FIG.
Amount field f (4) = quantity field f (2) × unit price field f (3)
The calculation relationship is established. This calculation relationship is converted into a data format that can be stored in the definition file 13 as calculation information that defines “f (4) · 1 = f (2) · 1xf (3) · 1”. Stored in For example, the calculation information described above is converted into a stack format using reverse Polish notation, as indicated by data 181i in FIG. The data format is not limited to this.

Next, as the second calculation relationship, as shown in FIG.
Total field sum (1) = total field f (4) .1 + total field f (4) .2 + total field f (4) .3 + total field f (4) .4 + total field f (4) .5
The calculation relationship is established. This operation relation is an operation that defines "sum (1) = f (4) ・ 1 + f (4) ・ 2 + f (4) ・ 3 + f (4) ・ 4 + f (4) ・ 5" Information is converted into a data format that can be stored in the definition file 13 and stored in the definition file 13 described later.

  The definition file 13 stores setting information and calculation information defined by the definition tool 12 described above. A data configuration example of each piece of information defined by the definition tool 12 will be described with reference to FIGS. As shown in FIG. 5 to FIG. 6, the definition file 13 constructs a hierarchically connected structure and stores it in each piece of information in the structure.

  As shown in FIG. 5, a slip information structure 12a is constructed as the topmost structure. The slip information structure 12a stores basic information, header information, and detailed information. Also, a basic information structure 12b is constructed below the basic information, and the slip size (horizontal) and slip size (vertical) of the read slip are stored.

  Below the header information, a header information structure 12c is constructed. The header information structure 12c includes an issuer company code, issuer company code display information, issuer company name, display information, issuer TEL number, and issue. The original TEL number display information, the issuer FAX number, and the issuer FAX number display information are stored.

  Further, as shown in FIG. 6, corresponding display display information structures 12i to 12l are constructed below the items of the header information structure 12c. In the display information structures 12i to 12l, as information indicating the position where the corresponding field is displayed on the display screen, lower left coordinates (horizontal), lower left coordinates (vertical), lower right coordinates (horizontal), lower right coordinates ( Vertical) is stored.

  Returning to FIG. 5, a detail information structure 12d is constructed below the detail information. The detail information structure 12d includes the number of detail rows, the number of fields, field information, field operation information, the total number of fields, and the total field information. , Total field calculation information is stored.

  Below the detailed information structure 12d, there is a field information structure 12m (FIG. 6) corresponding to the field information, a field operation information structure 12e (FIG. 5) corresponding to the field operation information, and total field information. Correspondingly, a total field operation information structure 12q (FIG. 6) and a total field operation information structure 12g (FIG. 6) are constructed corresponding to the total field operation information.

  As shown in FIG. 6, the field information structure 12m stores the number of characters, the character type, the field display information, the probability of recognition, the recognition candidate, the recognition result, and the next field information. Further, a display information structure 12n is constructed below the field display information, and a field information structure 12o is constructed below the next field information.

  Similarly, the total field information structure 12q stores the number of characters, character type, field display information, probability of recognition, recognition candidate, recognition result, and the following total field information. Further, a display information structure 12r is constructed below the field display information, and a total field information structure 12s is constructed below the next field information.

  Further, as shown in FIG. 5, the field calculation information structure 12e stores calculation information and next calculation information, and the total field calculation information structure 12g stores total calculation information and next total calculation information. Is stored.

  Returning to FIG. 1, the OCR recognition engine 14 includes a data processing unit 14 a, a data management unit 14 b, and the like, acquires each information defined by the definition tool 12 from the definition file 13, and is input from the reading device 11. In addition to performing various recognition processing of the image data of the slip, the processed information is managed.

  The data processing unit 14a acquires the image data of the slip from the reading device 11, recognizes the field based on the information defined by the definition tool 12 from the acquired image data, and stores the recognized field in the data management unit 14b. . The data processing unit 14a acquires various programs from the storage device 16, and performs main processing, verification processing, erroneous recognition field detection processing, and erroneous recognition field correction processing, which will be described later, for information stored in the data management unit 14b. Execute. Details of each process will be described later.

  The data management unit 14b classifies and stores the fields recognized by the data processing unit 14a into field information structures, and stores information generated in each process executed by the data processing unit 14a described above in each structure. Store and manage. A data configuration example of information stored in the data management unit 14b will be described with reference to FIGS. As illustrated in FIGS. 7 to 8, the data management unit 14 b stores each piece of information in a hierarchically connected structure.

  FIG. 7 is a diagram illustrating a data configuration example of information regarding fields recognized from the image data of the slip. As shown in FIG. 7, a slip information structure 14c is constructed as the uppermost structure, and basic information, header information, and detail information are stored in the slip information structure 14c. Further, a detail information structure 14d is constructed corresponding to the detail information below the slip information structure 14c. The detail information structure 14d stores the number of detail rows, the number of fields, field information, field operation information, the total field number, total field information, and total field operation information.

  Further, a field information structure 14e corresponding to the field information is constructed below the detailed information structure 14d. The field information structure 14e includes the number of characters, character type, field display information, probability of recognition, and recognition candidates. The recognition result and the following field information are stored. Corresponding to the field operation information, a field operation information structure 14h is constructed, and operation information and the next operation information are stored in the field operation information structure 14h.

  FIG. 8 is a diagram illustrating a data configuration example of data management information for specifying a misrecognized field among recognized fields. The verification management information 14j is constructed in the uppermost structure, and the verification management structure 14k is constructed in the lower layer of the verification management information 14j. The verification management structure 14k stores abnormal field information and normal field information.

  In addition, an abnormal field management structure 141 corresponding to the abnormal field information is constructed below the check management structure 14k. The abnormal field management structure 14l stores illegal field information and the next management structure. The If invalid field information is stored in the abnormal field management structure 141 as a result of the verification process described later, the field information structure (FIG. 7) of the field whose verification result is abnormal is pointed.

  Further, a normal field management structure 14h corresponding to normal field information is constructed below the check management structure 14k, and normal field information and the next management structure are stored in the normal field management structure 14h. The If normal field information is stored in the normal field management structure 14h as a result of the verification process described later, the field information structure (FIG. 7) of the field whose verification result is normal is pointed out.

  Returning to FIG. 1, the business application 15 is an application that performs various business processes using the processed data in the OCR recognition engine 14.

  The storage device 16 has a recording medium (not shown) in which a program, data, and the like are stored in advance, and this recording medium is constituted by a magnetic or optical recording medium or a semiconductor memory. This recording medium is fixedly attached to the recording device 16 or is detachably mounted. This recording medium is processed by the system program, various application programs corresponding to the system, and various processing programs. Stored data and the like are stored.

  Some or all of the programs, data, and the like stored in the recording medium are transmitted from other devices such as a server and a client via a network line transmission medium such as a WAN (Wide Area Network) and a LAN (Local Area Network). The recording medium may be received and stored from the transmission control unit 14, and the recording medium may be a recording medium of a server constructed on a network. Furthermore, the program may be transmitted to a server or a client via a transmission medium such as a network line and installed in these devices.

  The input device 17 includes a keyboard having cursor keys, numeric input keys, and various function keys, and outputs a pressing signal corresponding to the key pressed on the keyboard to the reading device 11, the definition tool 12, or the data processing unit 14a. To do. Note that the input device 17 may include a pointing device such as a mouse and a touch panel, and other input devices as necessary.

  The display device 18 includes a display screen made up of a CRT (Cathode Ray Tube), an LCD (Liquid Crystal Display) or the like, and displays display information such as image data read from the reading device 11 and input instructions input from the input device 17. Display on the screen.

Next, the operation of the present embodiment will be described.
A program for realizing each function described in each flowchart described later is stored in the storage device 16 of the data recognition device 1 in the form of a readable program code, and the definition tool 12 or the data processing unit 14a The operations according to the program code are sequentially executed.

  First, processing executed by the definition tool 12 will be described with reference to FIGS. FIG. 9 is a flowchart showing main processing executed by the definition tool 12. As shown in FIG. 9, when a process to be executed via the input device 17 is selected (step S1), the definition tool 12 determines whether or not it is an end selection, and when it is not an end selection (step S2; NO), it is determined whether or not the image reading process is selected (step S3). Here, when the image reading process is selected (step S3; YES), the definition tool 12 reads an image reading process program from the storage device 16, and an image reading process (FIG. 10) described later is executed (FIG. 10). Step S6). The image reading process will be described with reference to FIG.

  FIG. 10 is a flowchart showing an image reading process executed as a part of the main process. As shown in FIG. 10, when the slip image data (image) is read from the reading device 11 (step S11), the definition tool 12 reads the read image data into a predetermined address of the storage device 16 (main storage device). (Step S12). Further, the read image data is displayed on the display device 18 (step S13), and after the image reading process is completed, the process proceeds to the main process.

  Returning to FIG. 9, when the field setting process is selected in step S1 (step S4; YES), the definition tool 12 reads the field setting process program from the storage device 16, and the field setting process (FIG. 11) described later. ) Is executed. The field setting process will be described with reference to FIG.

  FIG. 11 is a flowchart showing a field setting process executed as part of the main process. As shown in FIG. 11, when the field setting process is executed, an initialization process (FIG. 12) is executed. The initialization process will be described with reference to FIG.

  As shown in FIG. 12, the register is initialized and set as “CNT = 0, SX = 0, SY = 0, EX = 0, ET = 0, CN = 0, CK = 0” (step S41). Next, the field information structure is secured, and the address is set in “Mlinep → Fieldp” and the register FP (step S42). Further, the display information structure is secured, and the address is set in “Mlinep → Fieldp → FieldDispp” and the register FDP (step S43). Then, after the initialization process is completed, the process proceeds to the field setting process.

  Returning to FIG. 11, when the process to be executed is selected via the input device 17 (step S22), the definition tool 12 determines whether or not it is an end selection, and when it is not an end selection (step S23; NO). Then, it is determined whether or not a position / size setting instruction has been selected (step S24).

  Here, when an instruction to set the position size is input (step S24; YES), that is, when a rectangular area is specified by the mouse or the like on the setting screen 181 shown in FIG. The upper left coordinates (SX, SY) and lower right coordinates (EX, EY) of the rectangular area thus obtained are acquired (step S25). Next, a setting screen 182 (FIG. 3B) for setting the maximum number of characters and the character type of the designated rectangular area is displayed on the display device 18 (step S26), and the maximum number of characters input on the setting screen 182 ( CN) and character type (CK) are acquired (steps S27 and S28).

  Then, each acquired information is stored at a predetermined address (step S28). That is, the acquired character type (CN) is stored in the number of characters (Chanum) of the field information structure (FP), and the acquired character type (CK) is stored in the character type (Charkind) of the field information structure (FP). . The upper left coordinate (SX, SY), the upper left coordinate horizontal (SX), the upper left coordinate vertical (SY), the lower right coordinate horizontal (EX), and the lower right coordinate vertical (EY) of the display structure information body (FDP) are obtained. ) And lower right coordinates (EX, EY) are stored.

  Next, in order to define the next field, an area reservation / register update process is executed (step S30). In FIG. 13, the execution field information structure is secured as part of the field setting process (step S45), and the address is set to “FP → Nextp” (step S45). Next, “FP” is changed to “FP → Nextp” (step S46). Further, the display information structure is secured and the address is set to “FP → FieldDispp” (step S47). Next, “FDP” is changed to “FP → FieldDispp” (step S46). Then, after the area reservation / register update process is completed, the process proceeds to the field setting process.

  Returning to FIG. 11, the register counter is incremented (+1) (step S31), the process proceeds to step S22, and when the next process is selected (step S22), the processes of steps S24 to S31 are repeatedly executed. The On the other hand, when the end process is selected (step S23; YES), a post process for ending the field setting process is executed (step S32). The post-processing will be described with reference to FIG.

  As shown in FIG. 14, first, it is determined whether or not the counter is 0. If the counter is 0 (step S51; YES), the memory pointed to by “Mlinp → Fieldp → FieldDispp” is released. (Step S52). Next, the memory pointed to by “Mlinp → Feildp” is released (step S53), and the field setting process ends.

  Subsequently, returning to FIG. 9, when calculation setting processing is selected in step S <b> 1 (step S <b> 5; YES), the definition tool 12 reads the calculation setting processing program from the storage device 16, and calculation setting processing described later. Is executed (step S8). The calculation setting process will be described with reference to FIG.

  FIG. 15 is a flowchart showing calculation setting processing executed as part of the main processing. As shown in FIG. 15, in executing the calculation setting process, an initialization process (FIG. 16) is executed (step S61). The initialization process will be described with reference to FIG.

  FIG. 16 is a flowchart showing an initialization process executed as a part of the calculation setting process. As shown in FIG. 16, the register is initialized and the counter is set to 0 (step S71). Next, the field information structure is secured, and the address is set in “Mlinep → FieldCulp” and the register FCP (step S72). Further, the display information structure is secured, and the address is set in “FCP → FieldCulp” and the register FCP (step S73). Then, after the initialization process is completed, the process proceeds to the calculation setting process.

  Returning to FIG. 15, when the process to be executed is selected via the input device 17 (step S62), the definition tool 12 determines whether or not it is an end selection, and when it is not an end selection (step S63; NO). It is determined whether or not a recognition field has been selected (step S64).

  When the recognition field is selected (step S64; YES), the field information structure of the selected field is specified (step S65). Next, “FP” is added to the FCP stack (step S66), and “=” is added to the FCP stack (step S67). Subsequently, the definition tool 12 reads the calculation information setting process program from the storage device 16 and executes the calculation information setting process (step S68). The calculation information setting process will be described with reference to FIG.

  FIG. 17 is a flowchart showing calculation information setting processing executed as part of the calculation setting processing. As shown in FIG. 17, a recognition field is selected via the input device 17 (step S82; YES), and the field information structure (FP) of the selected recognition field is specified (step S83), and is added to the FCP stack. “FP” is added (step S84).

  Next, it is determined whether or not an operator has been input (step S85). If an operator has been input (step S85; YES), the operator input to the FCP stack is added (step S86). Subsequently, it is determined whether or not an end instruction has been selected. If no end instruction is input (step S81; NO), the processes of steps S82 to S86 are repeated. When the end instruction is selected (step S81; YES), the calculation information setting process is ended, and the process proceeds to the calculation setting process.

  Returning to FIG. 15, when the calculation information setting process is completed (step S68), the counter of the register is incremented (step S69), the process proceeds to step S62, and the next process is selected (step S62). Here, when the next recognition field is selected, the processes of steps S65 to S69 are repeated. On the other hand, when the end process is selected (step S63; YES), a post process for ending the calculation setting process is executed (step S70). The post-processing will be described with reference to FIG.

  As shown in FIG. 18, first, it is determined whether or not the counter is 0. If the counter is 0 (step S87; YES), the memory pointed to by “Mlinp → FieldCulp → Culinfop” is released. (Step S88). Next, the memory pointed to by “Mlinp → FeildCulp” is released (step S89), and the calculation setting process ends.

  Returning to FIG. 9, when the image reading process (FIG. 10), the field setting process (FIG. 11), the calculation setting process (FIG. 18), and the like are completed, the process proceeds to step S <b> 1 and the next process is selected. Here, when the termination process is selected (step S2), the main process executed by the definition tool 12 is terminated.

  Next, each process executed by the data processing unit 14a will be described with reference to FIGS. FIG. 19 is a flowchart showing main processing executed by the data processing unit 14a. As shown in FIG. 19, “FCP” is changed to “Mlinep → FieldCulp” (step S91). Next, the normal field management structure area is secured by the data processing unit 14a, and the address is set in “KMNP → rightp” and the register rightp (step S92).

  Also, an abnormal field management structure area is secured, and the address is set in “KMNP → badp” and the register Badp (step S92). Next, it is determined whether or not the FCP is 0 (step S94). If the FCP is not 0 (step S94; NO), the verification processing program is read from the storage device 16, and the verification processing described later (FIG. 20). Is executed. The verification process will be described with reference to FIG.

  FIG. 20 is a flowchart showing a verification process executed as part of the main process. As shown in FIG. 20, “SP” is set to “FCP-> Culinfop” (step S111). Next, for example, an operation stacked in the reverse Polish notation on the stack pointed by SP is executed (step S112). Then, it is determined whether or not the calculation result is correct (step S113).

  If the calculation result is correct (step S113; YES), the field management structure is constructed with the argument “Rightp” (step S114). On the other hand, if the operation result is incorrect (step S113; NO), the field management structure is constructed with the argument “Badp” (step S114). With reference to FIG. 21, the construction process of the field management structure will be described.

  FIG. 21 is a flowchart showing a field management structure construction process executed as a part of the verification process. As shown in FIG. 21, the data processing unit 14a extracts the stack pointer SP and registers “ Is stored in "RR" (step S121). Next, it is determined whether or not “RR” is “=” (step S122). If “RR” is not “=” (step S122; NO), whether or not “RR” is an operator. Is discriminated (step S123). When “RR” is an operator (step S123; YES), the process proceeds to step S121.

  On the other hand, when “RR” is not an operator (step S123; NO), “PP → fp” is set to “RR” (step S124). Then, the field management structure is secured, and the addresses are set to “PP → Nextp” and “PP” (step S124). Then, the stack point SP is extracted, and it is determined whether or not “RR” is “=” (step S122). Here, when “RR” is “=” (step S122), the construction process of the field management structure is ended.

  In the construction process of the field management structure described above, when the calculation result is correct in the verification process, the address of the field information structure of the field constituting the calculation result is stored in the normal field management structure. If there is an error in the operation result in the verification process, the address of the field information structure of the field constituting the operation result is stored in the abnormal structure field management structure.

  Therefore, the field management structure construction process does not include the misrecognized field in the field listed in the normal field management structure, but the field listed in the abnormal field management structure is correctly recognized. This field is also included.

  Next, returning to FIG. 19, when the verification process ends, “FCP” is changed to “FCP → FieldNextp”, and the process proceeds to step S94. If it is determined in step S94 that FCP is 0 (step S94; YES), “FCP” is changed to “Mlinep → FieldCulp” (step S97). Next, it is determined whether or not “KMNP → badp” is 0 (step S98). If “KMNP → badp” is 0 (step S98; YES), the main process is terminated.

  On the other hand, if “KMNP → badp” is not 0 (step S98; NO), the data processing unit 14a reads an erroneous recognition field detection processing program from the storage device 16, and an erroneous recognition field detection processing (step S99) described later. Is executed. This misrecognized field detection process is a process of removing a correctly recognized field from the fields listed in the abnormal field structure in the above-described field management structure construction process. The misrecognition field detection process will be described with reference to FIG.

  FIG. 22 is a flowchart showing a misrecognition field detection process executed as part of the main process. As shown in FIG. 22, “Rightp” is “KMNP → rightp”, “Badp” is “KMNP → badp”, register “B” is “Badp”, and register “R” is “Rightp” (step) S131-S134). Next, it is determined whether or not “R” is 0 (step S135). If “R” is 0 (step S135; YES), “B” is changed to “B → Nextp”, and the process proceeds to step S134. Transition.

  On the other hand, if “R” is not 0 (step S135; NO), it is determined whether or not “B” is 0 (step S137). If “B” is 0 (step S137; YES), this The erroneous recognition field detection process is terminated. If “B” is not 0 (step S137; NO), it is determined whether “R → fp” and “B → fp” are equal (step S138).

  When “R → fp” and “B → fp” are equal (step S138; YES), “B → fp” is set to 0 (step S139). On the other hand, if “R → fp” and “B → fp” are not equal (step S138; NO), “R” is set to “R → Nextp” (step S140). Then, the process proceeds to step S135, and the above-described processing is repeatedly executed until “B” becomes zero.

  Returning to FIG. 19, when the erroneous recognition field detection process is completed (step S99), the abnormal field management structure is re-chained (step S100). Specifically, the abnormal field management structures chained in “KMNP → badp” are sorted in the order of the low probability of recognition stored in “KMNP → badp → fp → RecLebel” and re-chaining is performed. Is called.

  Next, when re-chaining is performed, an erroneous recognition field correction processing program is read from the storage device 16 by the data processing unit 14a, and an erroneous recognition field correction processing (FIG. 23) described later is executed. The misrecognition field correction process will be described with reference to FIG.

  FIG. 23 is a flowchart showing a misrecognition field correction process executed as part of the main process. As shown in FIG. 23, “Ilp” is “KMNP → FieldCulp”, and “Badp” is “KMNP → badp”. The register “I” is “Ilp” and the register “B” is “Badp” (steps S141 to S144). Next, it is determined whether or not “B” is 0 (step S145). If “B” is 0 (step S145; YES), “I” is changed to “I → FieldNextp” (step S146). The process proceeds to step S144.

  On the other hand, if “B” is not 0 (step S145; NO), it is determined whether or not “I” is 0 (step S147). If “I” is 0 (step S147; YES), this The erroneous recognition field correction process is terminated. If “I” is not 0 (step S147; NO), it is determined whether or not the same address as “R → fp” exists in the stack pointed to by “I → Colinfop” (step S148).

  If the same address as “R → fp” exists in the stack pointed to by “I → Colinfop” (step S148; YES), the correct value of the input field managed by “B → fp” in the “I → Culinfop” stack is set. It is obtained from the calculation formula of the stack and stored in “B → fp → RecReslt” (step S149). That is, the calculation result stored in the stack is used to calculate the operation result of the least probable field using the recognition result of the other field.

  If the same address as “R → fp” does not exist in the stack pointed to by “I → Culinfop” (step S148; NO), “B” is changed to “B → Nextp” (step S150), and the process proceeds to step S145. To do. Then, the above-described processing is repeatedly executed until “I” becomes zero.

  Returning to FIG. 19, when the misrecognition field recognition process is completed (step S <b> 101), the process proceeds to step S <b> 91, and the processes of steps S <b> 91 to S <b> 101 are repeatedly executed. S98; YES), the main process is terminated.

  As described above, the image data of the slip is recognized, and the calculation relationship included in the recognized detail line is defined by the definition tool 12 in advance. Then, a defined calculation is performed on the recognition result of each field in the detail line, and the field in which erroneous recognition occurs is specified by comparing the recognition result and the calculation result. In addition, an operation defined for the recognition result of the field that is presumed to be correctly recognized is performed to calculate a correct value of the field that is presumed to be erroneously recognized. Then, the value of the field estimated to have been erroneously recognized is corrected as the calculated correct value.

  As a result, not only erroneous recognition is detected from the recognition result recognized from the image data, but also the correct value of the field presumed to be misrecognized is calculated based on the recognition result of the field presumed to be correctly recognized. be able to. Therefore, it is possible to easily and accurately correct the erroneously recognized field, and to improve the recognition rate in the data recognition apparatus 1.

  Further, the probability of the recognized field is stored based on a plurality of arithmetic relationships, and the value of the field that is presumed to be erroneously recognized is calculated based on the recognition result of the presumed field that is correctly recognized. Therefore, based on a more accurate value, the value of the field estimated to have been erroneously recognized can be corrected, and the recognition rate of the data recognition apparatus 1 can be improved.

  In addition, since the definition tool 12 can recognize the fields from the acquired image data of the slip and define the arithmetic relations between the fields, the present invention is applied to slips, forms, and the like including various arithmetic relations. Therefore, it is possible to provide a highly versatile data recognition device 1.

  The description in the present embodiment is an example of a suitable data recognition apparatus 1 according to the present invention, and the present invention is not limited to this. For example, in the present embodiment, the case where the data recognition apparatus 1 reads the image data of a slip and recognizes information entered in the slip has been described as an example. However, other forms, estimates, design documents, etc. May be used.

  The configuration example of the data stored in the definition file 13 or the data management unit 14b is an example, and is not limited to the above-described example. For example, the calculation information has been described using a method of storing the stack using reverse Polish notation, but the present invention is not limited to this. In the present embodiment, the total field is processed as one of the field calculation information, but the total field and the field calculation information may be handled separately.

  In addition, it is needless to say that the detailed configuration and detailed operation of the data recognition apparatus 1 in the present embodiment can be appropriately changed without departing from the spirit of the present invention.

It is a block diagram which shows the functional structure of the data recognition apparatus 1 in embodiment which applied this invention. It is a block diagram which shows the functional structure of the reader 11 shown in FIG. (A) It is a figure which shows an example of the setting screen 181 displayed on the display apparatus 18 of FIG. (B) It is a figure which shows an example of the setting screen 181 displayed on the display apparatus 18 of FIG. It is a figure explaining the method of defining a calculation relationship by the definition tool. FIG. 4 is a diagram illustrating an example of a partial data configuration stored in a definition file 13. FIG. 4 is a diagram illustrating an example of a partial data configuration stored in a definition file 13. It is a figure which shows the example of a part of data structure managed by the data management part 14b. It is a figure which shows the example of a part of data structure managed by the data management part 14b. 4 is a flowchart showing main processing executed by a definition tool 12. 5 is a flowchart showing an image reading process executed by a definition tool 12. 6 is a flowchart showing a field setting process executed by the definition tool 12. 6 is a flowchart showing an initialization process executed by the definition tool 12. 5 is a flowchart showing area reservation / register update processing executed by a definition tool 12; 3 is a flowchart showing post-processing executed by a definition tool 12. 4 is a flowchart showing calculation setting processing executed by a definition tool 12. 6 is a flowchart showing an initialization process executed by the definition tool 12. 10 is a flowchart showing calculation information setting processing executed by the definition tool 12; 3 is a flowchart showing post-processing executed by a definition tool 12. It is a flowchart which shows the main process performed by the data processing part 14a. It is a flowchart which shows the verification process performed by the data processing part 14a. It is a flowchart which shows the construction process of the field management structure performed by the data processing part 14a. It is a flowchart which shows the misrecognition field detection process performed by the data processing part 14a. It is a flowchart which shows the misrecognition field correction process performed by the data processing part 14a. It is a figure explaining the method of improving a recognition rate with reference to a database in the conventional data recognition apparatus. It is a figure explaining the method of performing a calculation to a recognition result and improving a recognition rate in the conventional data recognition device.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Data recognition apparatus 11 Reading apparatus 12 Definition tool 13 Definition file 14 OCR recognition engine 14a Data processing part 14b Data management part 15 Business application 16 Storage apparatus 17 Input apparatus 18 Display apparatus

Claims (4)

A data recognition device that reads information entered in a form as image information and performs character recognition of the image information,
A definition file for storing setting information for defining fields included in the image information;
Recognition means for recognizing a field included in the image information based on the setting information;
Input means for inputting an arithmetic relationship between the recognized fields;
A discriminating means for performing an arithmetic operation between recognized fields based on the arithmetic relationship, comparing the recognition result of the field with the arithmetic result, and determining whether or not there is a misrecognition;
Calculating means for calculating the value of the field determined to have the misrecognition based on the recognition result of the field determined to have no misrecognition when it is determined that the field recognition result has the misrecognition; ,
A data recognition apparatus comprising:   2. The data recognition apparatus according to claim 1, wherein the definition file stores a calculation relationship between fields input by the input unit as calculation information. Based on the comparison result between the recognition result of the field and the calculation result, the storage unit stores a field in which the recognition result and the calculation result are different and the probability of the recognition,
The determination means compares the probability of recognition of a field in which the recognition result stored in the storage means and the calculation result are different, and determines whether or not the recognized field has a false recognition. The data recognition apparatus according to claim 1. On the computer,
A function of storing setting information for defining a field included in the image information;
A function of recognizing a field included in the image information based on the setting information;
A function of inputting an arithmetic relationship between the recognized fields;
A function of performing an operation between the recognized fields based on the operation relationship, comparing the field recognition result and the operation result, and determining whether or not there is a recognition error;
A function for calculating the value of the field determined to have the erroneous recognition based on the recognition result of the field determined to have no erroneous recognition when it is determined that the recognition result of the field has erroneous recognition;
A program to realize

Images