JP2005301793A - Ocr evaluation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an OCR evaluation system which allows an evaluation result to be easily acquired without requiring a long time. <P>SOLUTION: When receiving an image being an evaluation object generated by a user PC1, and evaluation parameters necessary for image recognition and image evaluation via the Internet 2, an OCR evaluation server 7 recognizes the image being the evaluation object by using the evaluation parameters and determines an evaluation result (correct read, erroneous read, unreadable) of the recognized image and transmits the evaluation result via the Internet 2 in response to a request from the user PC 1. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an OCR evaluation system that provides an OCR evaluation service via a network.

In general, the OCR is composed of a scanner unit and a recognition unit, and is used as an input unit in, for example, a bank transfer service. The scanner unit has means for acquiring a form image using an optical sensor or the like, and the recognition unit is means for correcting or normalizing an image, means for specifying a character position, character recognition, and It comprises means for converting to the above character code, knowledge processing means for selecting a character code from one or more character codes by looking at the relationship with the preceding and succeeding characters, and the like.
The recognition performance of OCR is often evaluated by a recognition rate, and the recognition rate includes a correct reading rate, a misreading rate, and a non-reading rate. The correct reading rate is the proportion of characters that can be read correctly in the whole, the misreading rate is the proportion of reading wrong characters, and the unread rate cannot determine the character code. It is a ratio. In addition, the correct reading rate, the misreading rate, and the unreading rate may be obtained for each character, or may be obtained for each form item or each item item of the form, and the evaluation method is determined according to the business characteristics.
There is a correction work to correct characters that could not be read correctly or characters that could not be determined, to clear data. In some cases, the number of terminals and the number of operators required for the correction work are derived from the recognition rate, and it is sometimes desired to grasp the recognition rate in advance. An OCR development vendor can construct a virtual OCR operation system, provide an OCR recognition rate evaluation service, and grasp the recognition rate in advance. For example, when the fulcrum of a certain bank is increased and a new reading character is added, it is possible to grasp in advance how the recognition rate changes.

By the way, the conventional OCR recognition rate evaluation service as described above has the following problems to be solved.
Since it is performed at the OCR vendor's local site and cannot be directly evaluated by the OCR user, the OCR user is required to obtain evaluation results such as taking the time to write a request to the OCR vendor and accepting the response. It may take time.

  When an image to be evaluated from a user terminal and data necessary for image recognition and image evaluation are received via a public communication network, the OCR evaluation system according to the present invention uses the data to evaluate the image to be evaluated. And an OCR evaluation server for obtaining the evaluation result of the recognition image and transmitting the evaluation result in response to a request from the user terminal.

  In the present invention, when an image to be evaluated from a user terminal and data necessary for image recognition and image evaluation are received via a public communication network, the image to be evaluated is recognized using the data, In addition, since the evaluation result of the recognition image is obtained and the evaluation result is transmitted in response to a request from the user terminal, the user can easily obtain the evaluation result without taking a long time. it can.

Embodiment 1 FIG.
Hereinafter, Embodiment 1 of the present invention will be described with reference to FIGS.
A vendor network 3 shown in FIG. 1 is a network constructed on the OCR vendor side, and is divided into a public segment 5 and a private segment 6 by a gateway 4. The gateway 4 has a firewall function. The OCR evaluation server 7 and the mail server 8, which will be described later, a PC 9 (personal computer) for OCR development installed in the OCR development department, and a system design equipment installed in the system design department. It prevents unauthorized access to devices such as the PC 10 and eliminates virus programs. Also, IP address conversion and port number conversion of packets entering and leaving the private segment 6 are performed. The public segment 5 is a LAN open to the user PC 1 arranged on the Internet 2, and the private segment 6 is a LAN used in the OCR vendor.

  The user PC 1 arranged on the Internet 2 is provided in a bank of a financial institution, for example, and although not shown, means for connecting to the Internet 2, means for setting an HTTP connection with the OCR evaluation server 7, and on the screen Is connected to an OCR (not shown) for reading the contents of a form used for financial transaction business. After starting up the browser, the user PC 1 displays an evaluation parameter screen in which, for example, an evaluation batch list, an image batch, layout information, a recognition package, a correct batch, and an e-mail address necessary for recognition evaluation can be input. An evaluation parameter is created based on various information input according to the screen. When HTTP specifying the OCR evaluation server 7 is input, the OCR evaluation server 7 is connected to the OCR evaluation server 7 via a provider (not shown) on the Internet 2, and the created evaluation parameters are transmitted to the OCR evaluation server 7. To do. The recognition package is an OCR recognition program.

  As shown in FIG. 2, the OCR evaluation server 7 arranged on the vendor network 3 includes an HTTP connection unit 100 for setting an HTTP connection with the user PC 1, an evaluation environment setting unit 101, a recognition evaluation unit 106, Recognition result file 107, totaling unit 109, mail transmission unit 111, evaluation result file 113, evaluation parameter file 150, evaluation batch list file 151, image batch file 152, layout information file 153, and recognition package When the evaluation parameter from the user PC 1 is received through the HTTP connection unit 100, the file 154 and the correct batch file 155 are stored in the evaluation parameter file 150 as shown in FIG. Each file described above is a storage unit provided in the server 7.

  When the evaluation parameter is stored in the evaluation parameter file 150, the evaluation environment setting unit 101 analyzes the evaluation parameter for setting the environment necessary for OCR recognition evaluation, 155, respectively. As shown in FIG. 4, the evaluation batch list file 151 is configured to store a plurality of evaluation batches B001 and B002, and the internal configuration of each evaluation batch B001 and B002 will be described later. A recognition package required for recognition is specified. Note that the evaluation batches B001 and B002 indicate one unit of the evaluation process for the evaluation target image.

  The image batch file 152 is a file for storing an evaluation target image group recognized by the OCR. As shown in FIG. 5, the image batch file 152 corresponds to each of the images I001, I002, and I003. ) Is stored. The image group includes images evaluated in the past. The evaluation target images are mainly classified into three types, that is, an image of the entire form, an image obtained by cutting out a reading area, and a character pattern image cut out for each character, and are specified in evaluation batches B001 and B002. The layout information file 153 is a file for storing layout information for each form type 1, 2, 3 as shown in FIG. In this layout information, as shown in FIGS. 4B to 4D, the coordinates of the reading areas 01 to 03 on the form set according to the form types 1, 2, and 3, the reading area, and the type ( Kana, kanji, numbers, etc.) are written. Also, the layout information includes a group of reading areas. For example, as shown in FIG. 6E, when the reading areas 01 and 02 are defined as GR1 and the reading area 03 is defined as GR2, Information defined by GR1 and GR2 is written.

  As shown in FIG. 7, the recognition package file 154 is a file that can store recognition packages P001 and P002 of different versions. The recognition packages P001 and P002 are OCR recognition programs as described above, and are composed of a main program, subprograms, parameters, and the like. When recognition packages P001 and P002 of different versions are stored, the difference in recognition performance due to the difference between the recognition packages P001 and P002 can be confirmed. As shown in FIG. 8, in the correct answer batch file 155, correct answer information of the evaluation target image, for example, correct characters (character codes) and character coordinates are written in correspondence with the correct answer information S001 to S003. The OCR recognition performance can be obtained by comparing the correct answer information with the evaluation target image.

  As will be described in detail when the operation is described, the recognition evaluation unit 106 recognizes the evaluation target image for each reading region based on the designation of the evaluation batches B001 and B002 stored in the evaluation batch list file 151, and recognizes the recognized image. The result file 107 is stored. This recognition is performed for each of the recognition packages P001 and P002 specified in the evaluation batches B001 and B002. The totaling unit 109 compares the recognition image stored in the recognition result file 107 with the correct answer information set for the recognition image to determine whether it is correct reading, misreading, or unreading, and the result is obtained. Stored in the evaluation result file 113. When the recognition evaluation of the image to be evaluated by the designation of the evaluation batches B001 and B002 is completed, the details will be described later. Various evaluation tables are created based on this, and performance comparison tables of recognition packages P001 and P002 of different versions are created based on the created evaluation table and stored in the evaluation result file 113. The evaluation table is created for each of the recognition packages P001 and P002 having different versions, and the performance comparison table becomes unnecessary when there is one recognition package. When receiving the evaluation completion message, the mail transmission unit 111 reads the transmission destination mail address from the evaluation parameter file 101 and transmits the evaluation completion message to the transmission destination user PC 1.

  Here, the internal configuration of the evaluation batch described above will be described with reference to FIG. In the figure, (a) is a detailed view showing an evaluation batch B001, and (b) is a detailed view showing an evaluation batch B002. Each of the evaluation batches B001 and B002 includes a batch number indicating the type of the evaluation batch, a batch name indicating the purpose of the evaluation, an evaluation number for specifying the execution order of recognition evaluation, and an evaluation target existing in the image batch of the image batch file 152. Evaluation image designating image, recognition package number designating recognition package stored in recognition package file 154, layout information designating form type stored in layout information file 153, correct information stored in correct batch file 155 The correct information number that specifies is written.

Next, the operation of the first embodiment will be described with reference to FIGS.
After launching the browser, the user PC 1 displays an evaluation parameter screen on which an evaluation batch list, an image batch, layout information, a recognition package, a correct batch, and a self-mail address necessary for OCR recognition evaluation can be input. An evaluation parameter is created based on various information input according to the screen. When HTTP specifying the OCR evaluation server 7 is input, the OCR evaluation server 7 is connected to the OCR evaluation server 7 via a provider (not shown) on the Internet 2, and the created evaluation parameter is sent to the OCR evaluation server 7. Transmit (P1).

  When the evaluation parameter is received via the Internet 2, the OCR evaluation server 7 stores the evaluation parameter in the evaluation parameter file 150 (see FIG. 3), and transmits a recognition message to the user PC 1 to Is prompted to disconnect the connection, and the OCR recognition evaluation is started (P2). This is because the disconnection instruction with the server 7 requires time for OCR recognition evaluation. On the other hand, the evaluation environment setting unit 101 of the OCR evaluation server 7 accesses the evaluation parameter file 150, analyzes the evaluation parameters, and stores them in the predetermined files 151 to 155, respectively. This is to set the environment necessary for recognition evaluation as described above. The evaluation batch list is stored in the evaluation batch list file 151, the image batch is stored in the image batch file 152, and the layout information is recognized in the layout information file 153. The package is stored in the recognition package file 154 and the correct answer batch is stored in the correct answer batch file 155.

  When the environment setting by the evaluation environment setting unit 101 is completed, the recognition evaluation unit 106 accesses the evaluation batch list of the evaluation batch list file 151, reads the first evaluation batch B001, and the evaluation written in the evaluation batch B001. The recognition package P001 described in the column of number 1 is read from the recognition package file 154 and executed, and the recognition of the evaluation target image is started. First, the evaluation target image set in the image I001 described in the column of the evaluation number 1 is read from the image batch file 152, and the reading areas 01 to 03 set in the form type 1 are read from the layout information file 153. Then, for the read evaluation target image, for example, the positions of characters and ruled lines are identified from the read areas 01 to 03 of the read form type 1, and the characters are cut out and recognized, respectively. The character coordinates and recognized characters in 03 are stored in the recognition result file 107. Next, recognition processing for the evaluation target images of the images I002 and I003 is sequentially performed according to the evaluation numbers 2 and 3 described in the evaluation batch B001, and the respective results (character coordinates and recognition characters) are stored in the recognition result file 107. When the recognition process based on the designation of the evaluation batch B001 is completed, the recognition process is performed based on the designation of the evaluation batch B002 shown in FIG. In this example, as shown in FIGS. 9A and 9B, the same form type 1 is recognized by a recognition package P002 of a version different from the above. When the character coordinates and the recognized characters are stored in the recognition result file 107, for example, the batch number, the batch name, and the evaluation number, the character coordinates and the recognized characters of the reading areas 01, 02, 03 for each form type 1, and the evaluation Stores the correct information number set for each number.

  When the recognition process by the recognition evaluation unit 106 is completed, the counting unit 109 reads the character coordinates and the recognized characters of the reading areas 01, 02, 03 of the form type 1 set to the evaluation number 1 from the recognition result file 107 and The correct answer information (correct characters and character coordinates) of the correct answer information S001 set to the evaluation number 1 is read from the correct answer batch file 155. Then, the correct characters for the recognized characters in the reading areas 01 to 03 are discriminated from both character coordinates and compared, and any of the recognized characters in the reading areas 01 to 03 in the form type 1 is correctly read, misread or unread. Are stored in the evaluation result file 113 in association with the batch number, batch name, and evaluation number. Further, the character coordinates and the recognized characters of the reading areas 01, 02, 03 of the form type 1 set to the evaluation numbers 2 and 3 are sequentially read from the recognition result file 107, and the correct information (S002, S003) is read in the same manner as described above. ) To determine whether each of the recognized characters is correctly read, misread, or unread, and store each result in the evaluation result file 113 in the same manner as described above. When the evaluation for the recognized character on the evaluation batch B001 is completed, the character coordinates and the recognized character on the evaluation batch B002 are read from the recognition result file 107 and compared with the correct information in the correct batch file 155. This comparison is performed in the order of the evaluation numbers in the same manner as described above, and it is determined whether each of the recognized characters in the reading areas 01 to 03 in each form type 1 is correctly read, misread, or unread, and the result is assigned to the batch number, The result is stored in the evaluation result file 113 in association with the batch name and the evaluation number.

  Thereafter, using the evaluation result on the evaluation batch B001 side as data, the correct reading rate, the misreading rate, and the unreading rate are calculated for each reading region in the order of the evaluation number, and the evaluation unit for the region unit shown in FIG. 11 is created. Further, the correct reading rate, the misreading rate, and the unreading rate are calculated for each group set in advance in the order of the evaluation number and for the form type 1, and an evaluation table for each group shown in FIG. The correct reading rate, the misreading rate, and the unreading rate are calculated for each form type 1 to create an evaluation table for each form shown in FIG. 13, and each evaluation table is stored in the evaluation result file 113. Further, a performance comparison table of recognition packages P001 and P002 of different versions is created based on the created area unit evaluation table (see FIG. 14), and the performance comparison of the recognition packages P001 and P002 is performed based on the group unit evaluation table. A table is created (see FIG. 15), and a performance comparison table for the recognition packages P001 and P002 is created based on the evaluation table for each form (see FIG. 16). Each performance comparison table is stored in the evaluation result file 113. save. Then, when the creation of the evaluation table and the performance comparison table is completed, the mail transmission unit 111 is notified of an evaluation completion message. At this time, the mail transmission unit 111 reads the mail address of the user PC 1 from the evaluation parameter file 101 and transmits an evaluation completion message to the transmission destination.

  On the other hand, when the evaluation completion message is received, the user PC 1 requests and obtains an evaluation result (evaluation table and performance comparison table) from the OCR evaluation server 7 (P3), and the OCR evaluation server 7 responds to the request. Then, the evaluation table and the performance comparison table are read from the evaluation result file 113 and transmitted to the user PC 1 (P4).

  As described above, according to the first embodiment, when an evaluation target image created by the user PC 1 and an evaluation parameter necessary for image recognition and image evaluation are received via the Internet 2, the evaluation parameter is used. Since the evaluation target image is recognized and the evaluation result (correct reading / misreading / non-reading) of the recognition image is obtained and the evaluation result is transmitted in response to a request from the user PC 1, the user side Can easily obtain evaluation results without requiring a long time.

Embodiment 2. FIG.
A second embodiment of the present invention will be described with reference to FIGS. In addition, the same code | symbol is attached | subjected to the same or equivalent part as Embodiment 1 demonstrated in FIG. 1, and description is abbreviate | omitted.
In the second embodiment, as shown in FIG. 17, a Web service server 11, an authentication server 12, and a base information server 13 are added on the vendor network 3. The OCR evaluation server 7 is arranged on the non-public segment 6 similarly to the authentication server 12 and the site information server 13, and the Web service server 11 is installed on the public segment 5.

  As will be described in detail when the operation is described, the Web service server 11 requests and acquires a user ID and password from the user PC 1 when receiving a request for OCR evaluation from the user PC 1, and acquires the acquired user ID and password. When it is transferred to the authentication server 12 for verification and the verification is successful (matched), the user ID is transmitted to the authentication server 12 again to acquire the address of the base information server 13. Thereafter, the user PC 1 is requested for evaluation parameters necessary for recognition evaluation, and the base information server 13 is accessed using the previously acquired address, and the user ID recognition package (OCR of the OCR) associated with the address is obtained. Recognition program) is acquired from the base information server 13. And when the evaluation parameter requested | required of user PC1 is received, a recognition package is attached to an evaluation parameter, it transmits to the OCR evaluation server 7, and OCR evaluation is performed.

  The authentication server 12 has a base information table 12a in which the address of the base information server 13 is written for each user ID as shown in FIG. 18, and when the user ID and password from the Web service server 11 are received. In addition, a user identification information file (not shown) set in advance is accessed to search whether or not the same user ID and password exist. When the same user ID and password do not exist, the Web service When the user authentication failure is notified to the server 11 and the same user ID and password are searched, the Web service server 11 is notified of the user authentication success. When the user ID is received again, the site information table 12 a is accessed, the address of the site information server 13 set for the user ID is searched, and transmitted to the Web service server 11.

  As shown in FIG. 19, the site information server 13 has a recognition package file 13a in which recognition packages P001 and P002 set for each user ID are stored, and the web service server 11 uses the address to identify the recognition package. When receiving the request, the user ID recognition packages P001 and P002 associated with the address are read from the recognition package file 13a and transmitted to the Web service server 11. The recognition packages P001 and P002 are OCR recognition programs of different versions as described in the first embodiment, and are composed of a main program, subprograms, parameters, and the like.

Next, the operation of the second embodiment will be described with reference to the sequence diagram of FIG.
When the user PC 1 and the Web service server 11 are connected via the Internet 2 by setting the HTTP connection from the user PC 1, the Web service server 11 requests the user PC 1 to input an account in P10. At this time, the user PC 1 uses the browser screen launched when setting the HTTP connection as the account input screen (P11), and when the user ID and password of, for example, “U001” are input by the operator's key operation, the user ID (U001) and the password are transmitted to the Web service server 11. When receiving the user ID (U001) and password from the user PC 1, the Web service server 11 transfers them to the authentication server 12.

  When the authentication server 12 receives the user ID (U001) and password, the authentication server 12 accesses a user identification information file (not shown) set in advance and searches whether the same user ID and password exist. When the same user ID and password do not exist, the Web service server 11 is notified of the user authentication failure. When the same user ID and password are searched, the Web service server 11 is notified of the user authentication success. (P17). When the Web service server 11 receives the notification of the authentication failure, the Web service server 11 issues a connection disconnection request to the user PC 1 at P10 to terminate the HTTP connection. When the Web service server 11 receives the notification of the authentication success, the process proceeds from P10 to P12. The user ID (U001) is transmitted to the authentication server 12 again. When the authentication server 12 receives the user ID from the Web service server 11 again, the authentication server 12 accesses the base information table 12a, searches for the address (SV013) of the base information server 13 using the user ID (U001) as a key, and the Web It transmits to the service server 11 (P18).

  When the Web service server 11 receives the address (SV013) set corresponding to the user ID (U001), the Web service server 11 shifts from P12 to P14 and requests an evaluation parameter from the user PC1. On the other hand, when receiving this request, the user PC 1 transmits evaluation parameters (evaluation batch list, image batch, layout information, correct batch, and own mail address) necessary for recognition evaluation to the Web service server 11 (P13). ). This evaluation parameter is created by the operator as described in the first embodiment. In this embodiment, the evaluation parameter does not include a recognition package. This is because the site information server 13 holds the information as described above.

  When the evaluation parameter is received, the Web service server 11 requests the site information server 13 to acquire the recognition package using the previously acquired address (SV013) (P14). Upon receiving this request, the site information server 13 reads the recognition packages P001 and P002 of the user ID (U001) associated with the received address from the recognition package file 13a and provides them to the Web service server 11 (P19). ). If the evaluation parameter acquired from the user PC 1 is insufficient, the Web service server 11 requests the shortage, and if the address of the base information server 13 for the user ID is invalid, , Request the recognition packages P001 and P002 from the user PC 1 having the user ID. Then, when the requested recognition packages P001 and P002 are received, the Web service server 11 shifts from P14 to P15, attaches the recognition packages P001 and P002 to the previously received evaluation parameters, and transmits them to the OCR evaluation server 7. When the in-recognition message from the OCR evaluation server 7 is received, the process proceeds to P16 to request the user PC 1 to disconnect the connection, and the connection with the server 11 is temporarily disconnected.

  On the other hand, the OCR evaluation server 7 stores the evaluation parameters attached with the recognition packages P001 and P002 in the recognition parameter file 150 when received. Thereafter, as in the first embodiment, the evaluation environment setting unit 101 analyzes the evaluation parameters and stores them in the predetermined files 151 to 155 in order to set the environment necessary for recognition evaluation. . Next, the recognition evaluation unit 106 recognizes the evaluation target image for each reading area based on the designation of the evaluation batches B001 and B002 stored in the evaluation batch list file 151, and stores the recognized image in the recognition result file 107. . Thereafter, the counting unit 109 compares the recognition image stored in the recognition result file 107 with the correct answer information set for the recognition image to determine whether it is correct reading, misreading, or unreading, The result is stored in the evaluation result file 113, and when the recognition evaluation of the evaluation target image by the designation of the evaluation batches B001 and B002 is completed, the result of the recognition evaluation stored in the evaluation result file 113 is used as data, and the correct reading rate, The misreading rate and the unreading rate are obtained respectively, and various evaluation tables are created based on the misreading rate and the non-reading rate (see FIGS. 11 to 13), and the performance comparison tables of recognition packages P001 and P002 having different versions based on the created evaluation tables. Is created (see FIGS. 14 to 16) and stored in the evaluation result file 113. When the creation of the evaluation table and the performance comparison table is completed, the mail transmission unit 111 transmits an evaluation completion message to the user PC 1 using the mail address stored in the evaluation parameter file 101 (P20).

  When the evaluation completion message is received, the user PC 1 transmits the user ID (U001) and password again to the Web service server 11 in order to obtain the evaluation result, and the evaluation result (evaluation table and performance comparison table) is transmitted by this transmission. ) Is received and the process ends (P21). On the other hand, the Web service server 11 transfers the received user ID (U001) and password to the authentication server 12 at P10, and when the notification of successful authentication is received by this transfer, moves to P22 and moves to the OCR evaluation server. 7 is requested to obtain the evaluation result. The OCR evaluation server 7 reads the evaluation result corresponding to this request from the evaluation result file 113 and transmits it to the Web service server 11 (P23). At this time, the Web service server 11 transfers the received evaluation result evaluation table and performance comparison table to the user PC 1 and terminates (P22).

  As described above, according to the second embodiment, the management of the recognition package on the user side is performed by the Web service server 11 and the site information server 13, so that the user is free from the management of the recognition package. The In addition, since the management of the recognition package is eliminated, the evaluation parameter creation time is reduced, and the evaluation parameter transmission processing time is reduced.

Embodiment 3 FIG.
A third embodiment of the present invention will be described with reference to FIGS. Note that the same or corresponding parts as those in the second embodiment described with reference to FIG.
In the third embodiment, as shown in FIG. 21, an image database server 14 is added on the private segment 6 of the vendor network 3.

  As shown in FIG. 22, the image database server 14 has a user-specific image batch file 14 a in which an image batch is written for each user ID, and sends an image batch acquisition request to which the user ID is attached from the Web service server 11. When received, the image batch set for the user ID is extracted from the user-specific image batch file 14 a and transmitted to the Web service server 11. The images I001 and I002 described in the user-specific image batch file 14a are numbers for designating images to be evaluated (characters and numbers), and each number is associated with the image to be evaluated on the form and written to the file 14a. It is.

  Also, the image database server 14 has an image service menu as shown in FIG. 23, and when the menu read request is received from the user PC 1, the image service menu is displayed on the screen, and any menu is designated. When it is done, the corresponding means is executed. When a user ID, an image batch number, and an image to be evaluated are received after “image registration” is specified in the image service menu, the image batch for each user is associated with the user ID and the image batch number. Register in file 14a. When “image reference” is designated, the corresponding image is read from the user-specific image batch file 14a using the user ID and the image batch number as keys and displayed on the screen of the user PC1.

  If an image to be deleted is selected after “Delete image” is specified in this state, the image is deleted from the user-specific image batch file 14a. If “image correction” is designated when an image is displayed on the screen, the inclination of the image is corrected. This is a function for improving visibility when referring to an image. When “image quality check” is designated, the image quality of each image registered by the user is checked. When an image to be shrunk is selected after “image shrink” is designated, the image is shrunk.

  For example, in the case of correcting the image inclination of a form with a white background as shown in FIG. 24A, the image correction is performed by scanning the inside of the image every one line or every several lines, and from white to black or black. The ruled line is recognized by detecting the change point to white. Then, the coordinates of the four corners of the virtual form are calculated from the outermost positions of the recognized ruled lines, and the center coordinates of the form are calculated from the coordinates of these four points. Next, the inclination angle of the form is calculated from the coordinates of the corners of the form, and the inclination of the image is corrected by rotating the form so that the inclination angle becomes zero around the center coordinates of the form. Further, as shown in FIG. 24B, the rotation correction is performed based on a specified angle from the user PC1.

  As shown in FIG. 25, the image quality check is used to check for blurred lines, blurred characters, crushing characters, and the like. For example, the number of black blocks and the number of black dots (black dot density) in a character rectangle (area) are calculated. If a large number of black blocks are scattered, it is determined as “blurred”, and if there is a portion where the black dot density is high, it is determined as “collapsed”. This image quality check makes it possible to check an image with poor quality. In the image shrink, as shown in FIG. 26 (a), only the read target portion of the registered image is extracted from the form image group to create a region shrink image (b). Create a shrink image. Thereby, the file size of the image to be evaluated (user-specific image batch file 14a) can be reduced, the evaluation time can be shortened, and the capacity of the recording medium (hard disk, backup tape, etc.) can be saved.

Next, the operation when receiving a request for OCR evaluation from the user will be described based on the sequence diagram shown in FIG. In the third embodiment, only the operation between the Web service server 11 and the image database server 14 is performed, and the other parts are the same as those in the second embodiment, and thus the description of the operation is omitted.
When the Web service server 11 receives a desired recognition package from the site information server 13 in P14, the Web service server 11 notifies the image database server 14 of an image batch request with a user ID attached. On the other hand, when the user ID is U001, the image database server 14 extracts the image batch (images I001 and I002) set to the ID from the user-specific image batch file 14a and transmits it to the Web service server 11 ( P19-2). When the Web service server 11 cannot receive a desired image batch from the image database server 14, the Web service server 11 requests the user PC 1 to acquire an image batch as an evaluation parameter is insufficient. In addition, when the image batch (images I001, I002) is received, the Web service server 11 shifts from P14 to P15, and adds the recognition package and the image batch (images I001, I002) to the previously acquired evaluation parameters. Attached and transmitted to the OCR evaluation server 7.

When a character pattern image is used as the evaluation image, the designation of the image batch file is omitted. In such a case, the evaluation environment setting unit 101 of the OCR evaluation server 7 performs the following operation.
(1) Get information on characters to be recognized and evaluated by referring to the correct batch.
(2) A pattern image of the character to be evaluated is acquired from the image database server 14 based on the character information.
(3) Combine the acquired images into one image for each character.
(4) Create layout information according to the combined image.
(5) Create an image batch file and a layout information file based on the combined image and layout information, respectively.
In the above (3), the pattern is synthesized to form one image in order to approximate the actual use conditions. However, the character pattern image may be used as it is without being synthesized.

  As described above, according to the third embodiment, since the image database server 14 manages the images to be evaluated for each user, the user is freed from managing the recognition package and the image batch. Further, since the management of the recognition package and the image batch is eliminated, the evaluation parameter creation time is further reduced, and the evaluation parameter transmission processing time is further reduced.

Embodiment 4 FIG.
Embodiment 4 of the present invention will be described with reference to FIGS. In addition, the same code | symbol is attached | subjected to the same or equivalent part as Embodiment 3 demonstrated in FIG. 21, and description is abbreviate | omitted.
In the fourth embodiment, as shown in FIG. 28, a layout database server 15 is added on the private segment 6 of the vendor network 3.

  As shown in FIG. 29, the layout database server 15 has a user-specific layout information file 15a in which layout information is written for each user ID, and receives a layout acquisition request attached with the user ID from the Web service server 11. The form types 1 and 2 set for the user ID are extracted from the user-specific layout information file 15a and transmitted to the Web service server 11. Form types 1 and 2 described in the user-specific layout information file 15a are numbers for designating the form images to be evaluated, and each number and form image are associated with each other and written in the file 15a.

  Further, the layout database server 15 has a layout service menu as shown in FIG. 30, and when a menu read request is received from the user PC 1, the service menu is displayed on the screen, and any menu is designated. When it is, the corresponding means is executed. When the user ID, the form type, and the layout of the form to be evaluated are received after “layout registration” is specified in the layout service menu, the layout information for each user is associated with the user ID and the form type. Register in file 15a. When “reference layout” is designated, the corresponding layout is read from the user-specific layout information file 15a using the user ID and form type as keys and displayed on the screen of the user PC1. This function enables the user side to obtain a standard layout.

  In this state, when “delete layout” is designated and a layout to be deleted is selected, the form is deleted from the user-specific layout information file 15a. When “Create Layout” is specified and, for example, an image of the entire form as shown in FIG. 31A is received, a reading area is set and extracted from the image by ruled line detection, and the layout of the form is created. (Refer to (b)). This function frees the user from creating layouts. The created layout can be registered in the user-specific layout information file 15a by designating “layout registration” in the menu.

  When the layout displayed on the screen by “reference layout” is shifted as shown in FIG. 32A and “layout correction” is designated, reading is performed from the image by ruled line detection as described above. The area is corrected (see (b)). This function makes it possible to correct a roughly created layout to an accurate position, and facilitates layout creation by the user. In this case as well, the layout of the corrected reading area is registered when “layout registration” is designated. When the layout of the shrink image is selected after “Layout Shrink” is designated, the image is shrunk (see FIGS. 33A and 33B). This function makes it easy to create a shrink image layout. In particular, with regard to the character pattern image, it is cut out and accumulated each time from the form and the image for each area in the past, and the pattern variation is enhanced by filtering processing such as blurring, crushing, and distortion.

Next, the operation when a request for OCR evaluation is received from the user side will be described based on the sequence diagram shown in FIG. In the fourth embodiment, only the operation between the Web service server 11 and the layout database server 15 is performed, and the other parts are the same as those in the third embodiment, and thus the description of the operation is omitted.
When the Web service server 11 receives a desired image batch from the image database server 14 in P14, it notifies the layout database server 15 of a layout request with a user ID attached. On the other hand, when the user ID is U001, the layout database server 15 extracts the layout information (form type 1 and 2) set to the ID from the user-specific layout file 15a and transmits it to the Web service server 11 ( P19-3). If the Web service server 11 does not receive provision of desired layout information from the layout database server 15, the Web service server 11 requests the user PC 1 to acquire layout information as insufficient evaluation parameters. Further, when the layout information (form type 1, 2) is received, the Web service server 11 shifts from P14 to P15, and recognizes the recognition package, image batch, and layout information (form type) in the previously acquired evaluation parameter. 1, 2) is attached and transmitted to the OCR evaluation server 7.

  As described above, according to the fourth embodiment, since the layout database server 15 manages layout information for each user, the user is freed from managing recognition packages, image batches, and layout information. In addition, since management of recognition packages, image batches, and layout information is eliminated, the labor required for creating evaluation parameters is reduced, and the transmission processing time of evaluation parameters is further shortened compared to the third embodiment.

Embodiment 5 FIG.
Embodiment 5 of the present invention will be described with reference to FIGS. The same or corresponding parts as those in the fourth embodiment described with reference to FIG.
In the fifth embodiment, as shown in FIG. 35, a correct database server 16 is added on the private segment 6 of the vendor network 3.

  As shown in FIG. 36, the correct database server 16 has a correct information file 16a for each user in which correct information is written for each user ID, and sends a correct information acquisition request to which the user ID is attached from the Web service server 11. When received, correct information S001 and S002 set for the user ID are extracted from the correct information file 16a for each user and transmitted to the Web service server 11. Correct answer information S001 and S002 described in the correct answer information file for each user 16a are numbers for specifying correct images, and the numbers and form images are associated with each other and written in the file 16a.

  Further, the correct database server 16 has a correct service menu as shown in FIG. 37, and when the menu read request is received from the user PC 1, the service menu is displayed on the screen, and any menu is designated. When it is, the corresponding means is executed. When a user ID, a correct information number, and a correct image are received after “correct registration” is specified in the layout service menu, the image is associated with the user ID and the correct information number and registered in the user-specific correct information file 16a. To do. When “reference to correct answer” is designated, the corresponding correct answer information is read from the user-specific correct answer information file 16a using the user ID and the correct answer information number as keys and displayed on the screen of the user PC1.

  In this state, when “correct answer deletion” is designated and the correct image to be deleted is selected, the image is deleted from the user-specific correct information file 16a. When “Create correct answer” is designated and an image as shown in FIG. 38A read by a scanner, for example, is received, the reading area is searched by ruled line detection, and the image in that area is corrected. Created as information (see (b)). By using this function, the user is freed from creating correct information. The created correct answer information can be registered in the correct answer information file 16a for each user by designating “correct answer registration” in the menu.

  After the “correction correction” is designated, as shown in FIGS. 39A and 39B, when the correct image and the image that requires correction are received, the correct image is determined based on the correct image. to correct. With this function, it is possible to correct a rough image of a correct answer correctly, and a user can easily create a correct image. In this case as well, when “correct answer registration” is designated, the corrected correct image is registered as correct answer information.

Next, the operation when receiving a request for OCR evaluation from the user will be described based on the sequence diagram shown in FIG. In the fifth embodiment, only the operation between the Web service server 11 and the correct database server 16 is performed, and the other parts are the same as those in the fourth embodiment, and thus the description of the operation is omitted.
When the Web service server 11 receives desired layout information from the layout database server 15 in P14, it notifies the correct database server 16 of a correct information request with a user ID attached. On the other hand, when the user ID is U001, the correct answer database server 16 extracts the correct answer information S001 and S002 set for the ID from the correct answer information file 16a for each user and transmits it to the Web service server 11 (P19-4). ). If the Web service server 11 cannot receive the desired correct answer information from the correct database server 16, the Web service server 11 requests the user PC 1 to acquire correct answer information as an evaluation parameter is insufficient. Further, when the correct answer information S001, S002 is received, the Web service server 11 shifts from P14 to P15, and adds the recognition package, image batch, layout information, and correct answer information S001, S002 to the previously acquired evaluation parameters. Attached and transmitted to the OCR evaluation server 7.

  As described above, according to the fifth embodiment, since the correct answer database server 16 manages correct answer information for each user, the user is free from management of recognition packages, image batches, layout information, and correct answer information. The In addition, since management of recognition packages, image batches, layout information, and correct answer information is eliminated, the labor required to create evaluation parameters is further reduced, and the transmission process time of evaluation parameters is further reduced as compared with the fourth embodiment.

Embodiment 6 FIG.
A sixth embodiment of the present invention will be described with reference to FIGS. In addition, the same code | symbol is attached | subjected to the same or equivalent part as Embodiment 5 demonstrated in FIG. 35, and description is abbreviate | omitted.
In the sixth embodiment, as shown in FIG. 41, an analysis service server 17 is added on the private segment 6 of the vendor network 3.

  The analysis service server 17 analyzes a misread or unread image among the images recognized by the OCR evaluation server 7, and creates an analysis table as shown in FIG. 42, for example. In this analysis table, “batch number” when evaluated, “evaluation number”, “reading area number” of the form to be evaluated, “correct image” based on correct information, “recognition recognized by OCR evaluation server 7” The “result image”, the “determination result” by the OCR evaluation server 7, the “region image” to be evaluated, the “cause”, and the like are shown.

  The analysis service server 17 has an analysis service menu as shown in FIG. 43, and when the menu read request is received from the user PC 1, the service menu is displayed on the screen, and any menu is designated. When that happens, the function is executed. When “analysis result reference” is designated, the created analysis table is displayed on the screen of the user PC 1. When “analysis result creation” is designated, it is recognized by the OCR evaluation server 7 as described above. The misread or unread image is analyzed among the detected images, and the analysis table is displayed on the screen.

Next, the operation when a request for OCR evaluation is received from the user will be described based on the sequence diagram shown in FIG. In the sixth embodiment, only the operation between the OCR evaluation server 11 and the analysis service server 17 is performed, and the other parts are the same as those in the fifth embodiment, and thus the description of the operation is omitted.
In P20, the OCR evaluation server 11 requests the analysis service server 17 to perform analysis when the recognition evaluation (correct reading, misreading and non-reading) of the evaluation target image by designating the evaluation batch is completed. At this time, the batch number, evaluation number, reading area number of the form to be evaluated, correct image based on correct information, recognized recognition result image, determination result (misread or unread), evaluation target area An image or the like is notified to the analysis service server 17 as data. On the other hand, when data required for analysis is received, the analysis service server 17 creates an analysis table as shown in FIG. 42 based on this data. When creating this analysis table, if the determination result is unread, write the cause of unread in the “Cause” column. When the creation of the analysis table is completed, the OCR evaluation server 11 is notified (P24).

  As described above, according to the sixth embodiment, misread or unread images among the images recognized by the OCR evaluation server 7 are analyzed and the result is notified to the user side. Can understand the cause of misreading and unreading, and can deal with it in advance depending on the cause.

It is a block diagram of the OCR evaluation system in Embodiment 1 of this invention. It is a block diagram of an OCR evaluation server. It is a block diagram of an evaluation parameter file. It is a block diagram of an evaluation batch list. It is a block diagram of an image batch. It is explanatory drawing of layout information. It is a block diagram of a recognition package file. It is a block diagram of a correct batch file. It is a detailed view of an evaluation batch. 6 is a sequence diagram showing an operation of the OCR evaluation system in the first embodiment. FIG. It is a figure which shows the evaluation table | surface of a region unit. It is a figure which shows the evaluation table | surface of a group unit. It is a figure which shows the evaluation table | surface of a form unit. It is a figure of the performance comparison table with respect to the evaluation result of a region unit. It is a figure of the performance comparison table | surface with respect to the evaluation result of a group unit. It is a figure of the performance comparison table | surface with respect to the evaluation result of a form unit. It is a block diagram of the OCR evaluation system in Embodiment 2 of this invention. It is a figure which shows a base information table. It is a block diagram of the recognition package file classified by user. FIG. 10 is a sequence diagram showing an operation of the OCR evaluation system in the second embodiment. It is a block diagram of the OCR evaluation system in Embodiment 3 of this invention. It is a block diagram of an image batch file classified by user. It is a figure which shows an image service menu. It is a figure which shows the example of image correction. It is a figure which shows the image quality check example. It is a figure which shows the example of a shrink image. FIG. 10 is a sequence diagram showing an operation of the OCR evaluation system in the third embodiment. It is a block diagram of the OCR evaluation system in Embodiment 4 of this invention. It is a block diagram of the layout information file classified by user. It is a figure which shows a layout service menu. It is a figure which shows the example of layout creation. It is a figure which shows the example of layout correction | amendment. It is a figure which shows the example of layout shrink. FIG. 10 is a sequence diagram showing an operation of the OCR evaluation system in the fourth embodiment. It is a block diagram of the OCR evaluation system in Embodiment 5 of this invention. It is a block diagram of the correct answer information file classified by user. It is a figure which shows a correct answer service menu. It is a figure which shows the example of correct answer preparation. It is a figure which shows the example of correct correction. FIG. 16 is a sequence diagram showing an operation of the OCR evaluation system in the fifth embodiment. It is a block diagram of the OCR evaluation system in Embodiment 6 of this invention. It is a figure which shows an analysis table. It is a figure which shows an analysis service menu. FIG. 20 is a sequence diagram showing an operation of the OCR evaluation system in the sixth embodiment.

Explanation of symbols

1 User PC, 2 Internet, 3 Vendor Network, 4 Gateway, 5 Public Segment, 6 Private Segment, 7 OCR Evaluation Server, 8 Mail Server, 9 OCR Development PC, 10 System Design PC, 11 Web Service Server, 12 Authentication server, 13 site information server, 14 image database server, 15 layout database server, 16 correct database server, 17 analysis service server.

Claims (11)

  When an image to be evaluated from a user terminal and data necessary for image recognition and image evaluation are received via a public communication network, the image to be evaluated is recognized using the data, and the recognized image An OCR evaluation system comprising an OCR evaluation server that obtains the evaluation result of and transmits the evaluation result in response to a request from the user terminal. A web service server, an authentication server having at least user identification information, a base information server having a recognition program for recognizing an image to be evaluated, and an OCR evaluation server are arranged on a network constructed in a predetermined area,
When the user identification information from the user terminal is received via the public communication network, the Web service server forwards the authentication information to the authentication server, and when the authentication is successful, When an image and data necessary for image evaluation are requested and the image to be evaluated and data necessary for image evaluation are received by this request, the recognition program is read from the base information server, and the image to be evaluated and Notify the OCR assessment server with the data,
The OCR evaluation server recognizes an image to be evaluated based on a recognition program and obtains a recognition image evaluation result using the data. An image database server arranged on the network and having an image to be evaluated as data is arranged on the network,
When the authentication of the user identification information by the authentication server is successful, the Web service server requests data required for image evaluation from the user terminal, and when the data is received by this request, the base information server 3. The OCR evaluation system according to claim 2, wherein a recognition program is read from the image database, an image to be evaluated is read from the image database server, and the OCR evaluation server is notified together with the data. A layout database server having layout information of an image to be evaluated among data necessary for image evaluation is arranged on the network,
When the authentication of the user identification information by the authentication server is successful, the Web service server requests the user terminal for data necessary for image evaluation excluding the recognition program and layout information, and the data is received by this request. When the recognition program is read from the site information server, the image to be evaluated is read from the image database server, the layout information is read from the layout database server, and notified to the OCR evaluation server together with the data,
4. The OCR evaluation server according to claim 3, wherein the OCR evaluation server recognizes an image to be evaluated by applying layout information based on a recognition program, and obtains an evaluation result of the recognized image using the data. Evaluation system. A correct database server having correct information to be compared with an image to be evaluated among data necessary for image evaluation is arranged on the network,
When the authentication of user identification information by the authentication server is successful, the Web service server reads a recognition program from the site information server, an image to be evaluated from the image database server, and layout information from the layout database server. In addition, the correct answer information is read from the correct answer database server and notified to each OCR evaluation server,
5. The OCR evaluation server according to claim 4, wherein the OCR evaluation server recognizes an image to be evaluated by applying layout information based on a recognition program, and obtains an evaluation result of the recognized image based on correct answer information. Evaluation system.   The result of analyzing the misread and unread of the recognized image among the evaluation results of the recognized image determined by the OCR evaluation server placed on the network is output, and the analysis result is transmitted in response to a request from the user terminal The OCR evaluation system according to claim 1, further comprising an analysis service server that performs the analysis.   The web service server is arranged in a public segment of the network, the OCR evaluation server, the authentication server, the site information server, the image database server, the layout database server, the correct database server, The OCR evaluation system according to claim 6, wherein an analysis service server is arranged in a private segment of the network.   The image database server registers an image to be evaluated in response to a registration request from the user terminal, transmits the image to be evaluated registered in response to a reference request from the user terminal to the user terminal, In response to a delete request, the image to be evaluated is deleted, the image quality of the image to be evaluated is checked and corrected in response to an image quality check request from the user terminal, and is applied in response to a shrink request from the user terminal. The OCR evaluation system according to claim 3, wherein the image to be evaluated is shrunk.   The layout database server registers an evaluation target image layout in response to a layout registration request from a user terminal, and transmits the evaluation target image layout registered in response to a layout reference request from the user terminal to the user terminal. In response to the layout deletion request from the user terminal, the corresponding layout of the evaluation target image is deleted, and in response to the layout creation request from the user terminal, the corresponding evaluation target image layout is created. 9. The layout of an image to be evaluated is corrected in response to a layout correction request, and the layout of the image to be evaluated is shrunk in response to a layout shrink request from a user terminal. OCR evaluation system described in 1.   The correct database server registers correct information of an image to be evaluated in response to a correct registration request from a user terminal, transmits correct information registered in response to a correct reference request from the user terminal to the user terminal, and In response to a correct deletion request from the terminal, the corresponding correct answer information is deleted, in response to a correct answer creation request from the user terminal, correct answer information of the image to be evaluated is generated, and in response to a correct correction request from the user terminal 10. The OCR evaluation system according to claim 3, wherein the correct answer information is corrected. The analysis service server transmits the analysis result of the recognition image to the user terminal in response to the analysis result reference request from the user terminal, analyzes the recognition image in response to the analysis result creation request from the user terminal, and outputs the result. The OCR evaluation system according to claim 3, wherein the OCR evaluation system transmits to the user terminal.

Images