JP2012038225A - Character recognition device, character recognition method and character recognition program - Google Patents

Abstract

To realize proper character recognition.
Recognizing that character image A is recognized to identify one or more candidate characters, and for each of the one or more candidate characters, a likelihood that the candidate character is the same character as the character image is calculated. The similarity between the character image B in the history database and the character image to be processed is obtained by referring to the processing unit 14 and the history database including the character image obtained by correcting the character recognition result of the character image. An error list processing unit 16 that corrects the likelihood of the candidate character and an evaluation unit 18 that identifies a character as a recognition result from the candidate character based on the corrected likelihood.
[Selection] Figure 1

Description

  The present case relates to a character recognition device, a character recognition method, and a character recognition program.

  Conventionally, for example, in the exchange transfer process in a financial institution, the following procedure has been performed. (1) The customer (transfer client) fills in the necessary information (transfer destination, amount, etc.) on the transfer slip and makes a transfer request to the sales office. (2) The operator who has received the transfer request reads the image of the transfer slip from the branch office terminal. At this time, the sales office terminal automatically recognizes the read image data and displays the character with the highest recognition rate on the screen as a recognition result. (3) The operator visually checks the displayed characters and the characters entered by the customer on the same screen. At this time, if the operator determines that the characters do not match, the operator corrects the characters to perform transmission processing.

  Here, as a technique for assisting the correction of characters by an operator, there are Patent Documents 1 to 3 and the like. In Patent Document 1, an erroneously recognized character that has been corrected by the operator in the past and a corrected character are stored in association with each other, and when a new recognized character matches the stored erroneously recognized character, A technique for displaying a corrected character associated with a misrecognized character as a candidate character is disclosed.

  Patent Document 2 discloses a technique for storing erroneous recognition frequency information in addition to the contents of Patent Document 1 and displaying candidate characters in order from the most frequently used characters. Patent Document 3 discloses a technique for correcting the rank of candidate characters by correcting the certainty of the recognition result of a character by the accuracy (reliability) of past recognition related to that character. Yes.

JP 59-049675 A Japanese Patent Laid-Open No. 5-26607 JP-A-1-214987

  In Patent Documents 1 to 3, candidate characters are effectively displayed for characters that are generally easily misrecognized by a computer. However, in character recognition, even a character having a relatively high recognition rate may be erroneously recognized due to a handwriting (癖) unique to the client. In such a case, there is a high possibility that the techniques of Patent Documents 1 to 3 cannot be applied.

  Therefore, the present invention has been made in view of the above problems, and an object thereof is to provide a character recognition device, a character recognition method, and a character recognition program that enable highly accurate character recognition.

  The character recognition device described in the present specification is a character recognition device that performs character recognition of a character image, recognizes the character image, specifies one or more candidate characters, and the one or more candidate characters. For each, refer to a candidate character likelihood calculating unit that calculates the likelihood that the candidate character is the same character as the character image, and a history database including the character image in which the character recognition result of the character image is corrected Determining a similarity between the character image of the history database and the character image to be processed, and correcting the candidate character based on the corrected likelihood based on the similarity and the correction unit for correcting the likelihood of the candidate character And a character recognition unit that identifies a character as a recognition result.

  The character recognition method described in the present specification is a character recognition method in which a computer performs character recognition of a character image, and the character image is image-recognized to identify one or more candidate characters, and the one or more A candidate character likelihood calculating step for calculating the likelihood that the candidate character is the same character as the character image for each of the candidate characters, and a history database including the character image in which the result of character recognition of the character image is corrected With reference to, based on the correction step of obtaining the similarity between the character image of the history database and the character image to be processed, correcting the likelihood of the candidate character from the similarity, and the corrected likelihood And a character recognition step of identifying a character as a recognition result from the candidate character.

  The character recognition program described in this specification causes a computer to image-recognize the character image to identify one or more candidate characters, and for each of the one or more candidate characters, the candidate character is the same character as the character image. A candidate character likelihood calculating unit for calculating the likelihood of being a character image, referring to a history database including the character image in which the result of character recognition of the character image is corrected, the character image of the history database, and the processing target A character recognition unit that obtains a similarity with a character image, corrects the likelihood of the candidate character from the similarity, and specifies a character as a recognition result from the candidate character based on the corrected likelihood , A character recognition program that functions as

  The character recognition device, the character recognition method, and the character recognition program described in this specification have an effect of realizing high-accuracy character recognition.

It is a figure showing roughly the composition of the transfer slip processing system concerning one embodiment. It is a figure which shows the hardware constitutions of a control apparatus. It is a figure which shows an example of recognition process data. It is a figure which shows an example of error list DB. It is a flowchart which shows a series of processes of a control apparatus. It is a flowchart which shows the specific process of step S12 of FIG. It is a flowchart which shows the specific process of step S32 of FIG. FIG. 8A and FIG. 8B are diagrams showing a state in the middle when the recognition process data is created. It is a figure which shows the state after the format of error list DB. It is a flowchart which shows the specific process of step S48 of FIG. It is FIG. (1) for demonstrating the process of FIG. FIG. 11 is a diagram (No. 2) for describing the process of FIG. 10; FIG. 11 is a diagram (No. 3) for explaining the processing of FIG. 10; It is a flowchart which shows the specific process of step S50 of FIG. FIG. 15 is a diagram (No. 1) for describing the processing of FIG. 14; FIG. 15 is a diagram (No. 2) for explaining the processing of FIG. 14;

  Hereinafter, an embodiment of a transfer slip processing system will be described in detail with reference to FIGS. This transfer voucher processing system 100 is a system installed at a sales office of a bank, for example. As shown in FIG. 1, the transfer slip processing system 100 includes a control device 10 as a character recognition device, a reading device 30, an input device 40, and a display device 50 connected to the control device 10. Yes.

  The control device 10 acquires an image of a transfer slip read (scanned) by the reading device 30, and performs character recognition of a handwritten character described by the transfer requester from the image of the transfer slip. Then, the character recognition result is output to the display device 50. Moreover, the control apparatus 10 is connected to the center which performs a transfer process via a network, and transmits the contents of the transfer slip to the center. The specific configuration of the control device 10 will be described later.

  The reading device 30 is a device that can read an image of a transfer slip such as a scanner. The reading device 30 outputs the reading result to the control device 10. The input device 40 is a user interface including a keyboard and a mouse, and is used when an operator of the transfer slip processing system 100 performs character input, command input, and the like. The display device 50 is a display such as a liquid crystal display or an organic EL display.

  FIG. 2 is a hardware configuration diagram of the control device 10. As shown in FIG. 2, the control device 10 includes a CPU 90, a ROM 92, a RAM 94, a storage unit (HDD (Hard Disk Drive)) 96, an input / output unit 97, and the like. , Connected to the bus 98. In the control device 10, the CPU 90 executes a program (character recognition program) stored in the ROM 92 or the HDD 96, thereby realizing the functions of the respective units in FIG. The input / output unit 97 is connected to the reading device 30, the input device 40, the display device 50, and the like shown in FIG.

  As shown in FIG. 1, the control device 10 functions as an input unit 12, a recognition processing unit 14 as a candidate character likelihood calculation unit, an error list processing unit 16, an evaluation unit 18, an output unit 20, and a transmission unit 22. To do. In the control device 10, the HDD 96 functions as a slip image DB 28 and an error list DB 26 as a history database. A correction unit is realized by a part of the functions of the error list processing unit 16 and the evaluation unit 18, and a character recognition unit is realized by a part of the functions of the evaluation unit 18.

  The input unit 12 receives an image acquired by the reading device 30 and information input from the input device 40. The recognition processing unit 14 performs character recognition on a character image in the image (processing target image) received by the input unit 12 to identify one or more candidate characters, and the one or more candidate characters are identified. A recognition rate (likelihood (also referred to as likelihood or degree of likelihood)) that is the same character as the character image is calculated. The error list processing unit 16 refers to the error list DB 26, obtains the similarity (match rate) between the character image in the DB 26 and the character image included in the image to be processed, and uses the obtained similarity as a candidate. A correction value for correcting the character recognition rate is calculated. The evaluation unit 18 corrects the recognition rate with the correction value, and evaluates the candidate character based on the recognition processing data (see FIG. 3) including the corrected value (hereinafter referred to as the correct answer rate). Perform character recognition. FIG. 3 shows an example of recognition processing data used for evaluation of candidate characters by the evaluation unit 18. The item “character image A” in the recognition processing data of FIG. 3 is a character image input from the input unit 12. The item “candidate characters 1 to 5” means one or more candidate characters obtained by character recognition in the recognition processing unit 14. Each candidate character 1 to 5 is associated with a recognition rate (likelihood). In principle, candidate character 1 is registered in the item “recognition result (provisional)”. In the item “recognition result (book)”, a character to be a recognition result of the character image A as a result of the evaluation by the evaluation unit 18 is registered.

  Returning to FIG. 1, the output unit 20 outputs the character recognition result of the evaluation unit 18 to the display device 50. In the error list DB 26, as shown in FIG. 4, the name of the transfer client, the uncorrected characters (character image recognition results displayed on the display device 50 as the character recognition results of the evaluation unit 18), and the operator The corrected character (the character that has been changed to the character image by correction) and the character image at that time (character image B that has been subjected to character recognition) are associated with each other. The error list DB 26 includes a match rate item. However, it is assumed that nothing is initially input in the match rate item.

  Returning to FIG. 1, images of various transfer slips are registered in the slip image DB 28. The image of the transfer slip registered in the slip image DB 28 is obtained when the recognition processing unit 14 acquires only the image of the handwritten character from the image acquired by the reading device 30 (the image of the transfer slip in which the handwritten character is described). Used.

  Next, based on FIGS. 5-16, the process of the transfer slip processing system 100 (mainly control apparatus 10) is demonstrated in detail.

  FIG. 5 is a flowchart showing a series of processing of the control device 10. In the process of FIG. 5, first, in step S10, the recognition processing unit 14 determines whether or not image reading has started. Specifically, the recognition processing unit 14 determines whether or not a transfer slip handwritten by the transfer requester is set in the reading device 30 by an operator and an instruction to start reading is issued by pressing a scan start button or the like. Judging.

  Next, in step S12, the recognition processing unit 14, the error list processing unit 16, and the evaluation unit 18 execute a subroutine for character recognition processing of the transfer slip. In this step S12, each part 14, 16, and 18 performs the process according to the flowchart of FIG.

  In the process of FIG. 6 (step S12), in step S30, the recognition processing unit 14 acquires client information. Here, the recognition processing unit 14 reads and acquires the client information from, for example, an identification card that the operator confirms when a transfer request is issued from the client. However, the present invention is not limited to this, and the recognition processing unit 14 may acquire the client information from, for example, the name of the client described in the transfer slip. Alternatively, the recognition processing unit 14 may acquire the name of the client who is directly input by the operator.

  In step S32, the recognition processing unit 14 executes a recognition processing subroutine. In step S32, a process according to the flowchart of FIG. 7 is executed.

  In the process of FIG. 7, in step S70, the recognition processing unit 14 extracts one character from the transfer slip. In this case, the recognition processing unit 14 takes the difference between the image of the slip stored in the slip image DB 28 and the image of the slip after the client has handwritten, so that among the characters described in the transfer slip, One of these is extracted.

  Next, in step S72, the recognition processing unit 14 extracts a plurality (here, five) of candidate characters 1 to 5 for the extracted character image A. In step S74, the recognition processing unit 14 calculates the recognition rate of each candidate character. The recognition rate in this case can be calculated using distance information obtained from the character feature vector of the candidate character.

  Next, in step S75, the recognition processing unit 14 determines whether the recognition rate of the candidate character 1 is equal to or greater than the threshold value 1. Here, as the threshold value 1, a lower limit value (for example, obtained from a simulation result) of a value shown when character recognition is appropriately performed to some extent can be employed. The case where the judgment here is negative is a case where the recognition rate is lower than the others by correcting characters as shown in the rightmost column of the recognition processing data in FIG. is there. If the determination here is affirmed, the process proceeds to step S76. If the determination is negative, the process proceeds to step S77.

  When the determination in step S75 is affirmed and the process proceeds to step S76, the recognition processing unit 14 sets the candidate character 1 as a recognition result (temporary). On the other hand, if the determination in step S75 is negative and the process proceeds to step S77, the recognition processing unit 14 sets “reject” as the recognition result (temporary). The meaning of “reject” will be described later. Regarding the first character image A, recognition processing data as shown in FIG. 8A can be obtained at the stage after steps S72 to S76 (or S77).

  Next, in step S78, the recognition processing unit 14 determines whether or not the extraction of all the character images A has been completed. If the determination is negative, the process returns to step S70, and the process for the next character image A is performed in the same manner as described above. When the processes for all the character images A are completed, the determination in step S78 is affirmed, and the process proceeds to step S34 in FIG. Note that when the determination in step S78 is affirmative, recognition process data as shown in FIG. 8B can be obtained.

  Returning to FIG. 6, in step S <b> 34, the error list processing unit 16 reads the top data that has not been read from the error list DB 26. Specifically, the head data (here, the uppermost data) that has not yet been read in the error list DB 26 shown in FIG. 4 is read.

  Next, in step S36, the error list processing unit 16 determines whether data has been read. If the determination here is negative, the process proceeds to step S54. If the determination is positive, the process proceeds to step S38. When the process proceeds to step S38, the error list processing unit 16 determines whether the data is the client data. If the determination in step S38 is negative, the error list processing unit 16 discards the read data in step S42. On the other hand, if the determination in step S38 is affirmative, the error list processing unit 16 acquires the read data in step S40. After step S42 or step S40, the process proceeds to step S44.

  In step S44, the error list processing unit 16 determines whether all data has been read. When judgment here is denied, it transfers to step S34. Then, until the determination in step S44 is affirmed, the processing / determination in steps S34 to S44 is repeated to create (format) an error list DB that collects only the client data. FIG. 9 shows an error list DB after formatting when the client is “Fujitsutarou”.

  When the above process is performed and the determination in step S44 is affirmed, the process proceeds to step S46. In step S46, the evaluation unit 18 determines whether or not one or more pieces of data have been acquired through step S40 once or more. If the determination in step S46 is negative, the process proceeds to step S54. On the other hand, if the determination in step S46 is affirmative, the process proceeds to step S47.

  In step S47, the evaluation unit 18 extracts one character image A. Here, for example, it is assumed that the leftmost character image A in FIG. 8B has been extracted. In the next step S48, the evaluation unit 18 executes a subroutine of the corrected character extraction process 1. In this subroutine, the evaluation unit 18 executes processing according to the flowchart of FIG.

  In step S80 of FIG. 10, the evaluation unit 18 corrects the same corrected character as the candidate character 1 (bold frame in the upper diagram of FIG. 11) (bold broken line frame in the lower diagram of FIG. 11) from the formatted error list DB. A character image B (thick line frame in the lower diagram of FIG. 11) corresponding to is extracted. Next, in step S82, the evaluation unit 18 sets the character image B for which the match rate has not been calculated as a calculation target. Here, the evaluation unit 18 sets a character image B (character image Ba) indicated by a symbol Ba in the lower diagram of FIG. 11 as a calculation target.

  Next, in step S84, the evaluation unit 18 compares the character image A with the character image B (Ba) to be calculated, and calculates a match rate. As the match rate in this case, the match rate (similarity) between the character images A and Ba obtained using a known method can be employed. In FIG. 12, the manner of calculating the match rate is schematically shown by a broken line La. In FIG. 12, 85% is calculated as the match rate between the character image Ba and the character image A.

  In the next step S86, the evaluation unit 18 determines whether or not the match rates of all the extracted character images B have been calculated. Here, since only the match rate with the character image Ba has been calculated, the determination is denied, and the process returns to step S82. Thereafter, the match rate between the character images Bb, Bc... And the character image A is calculated. In FIG. 12, these calculation processes are indicated by broken lines Lb and Lc. Then, at the stage where the match rate with all the character images B extracted in step S80 is calculated, the determination in step S86 is affirmed, and the process proceeds to step S88.

  In step S88, the evaluation unit 18 extracts the maximum one of the calculated match rates. Here, 85% which is the largest numerical value among the numerical values indicated by the thick broken line frame in the lower diagram of FIG. 12 is extracted. Next, in step S90, the evaluation unit 18 acquires a corrected character related to the extracted data with the highest matching rate. Here, as indicated by a thick broken line frame in the lower diagram of FIG. 13, the evaluation unit 18 acquires “1”.

Next, in step S92, the evaluation unit 18 uses the maximum match rate (here, 85%) as a correction value, corrects the recognition rate related to the corrected character based on the following equation (1), and the corrected value: (Here, the accuracy rate is 1).
Correct answer rate 1 = recognition rate for corrected characters × correction value (1)

  The recognition rate related to the corrected character means the recognition rate of the candidate character “1” in the upper diagram of FIG. 13 that matches the corrected character “1” in the lower diagram of FIG. Here, the evaluation unit 18 can acquire the correct answer rate 1 = 66.3% (= 78% × 85%) based on the above equation (1).

  As described above, when the process of step S48 of FIG. 10 is completed, the process proceeds to step S50 (corrected character extraction process 2) of FIG. The process of step S50 is executed along the flowchart of FIG.

  In step S102 of FIG. 14, the evaluation unit 18 extracts the character image B and the uncorrected character of data in which the same character as the candidate character 1 (here “7”) is the corrected character. . Specifically, the evaluation unit 18 uses the character image of the corrected character (thick line frame in the lower diagram of FIG. 15) and the character before correction (both of the lower characters of FIG. 15) as the candidate character 1 (“7”). (Refer to a thick broken line frame).

  Next, in step S104, the evaluation unit 18 sets the character image B for which the match rate has not been calculated as a calculation target. Here, it is assumed that the character image BA in the lower diagram of FIG. Next, in step S106, the evaluation unit 18 compares the character image A with the character image B to be calculated (here, the character image BA), and calculates a match rate. As a result, the matching rate between the character image BA and the character image A is calculated to be 70% as shown in the lower diagram of FIG.

  Next, in step S108, the evaluation unit 18 determines whether or not the match rates of all extracted character images B have been calculated. Here, since the match rate of the character image BA has only been calculated, the determination is denied and the process returns to step S104.

  Thereafter, the processes and determinations in steps S104 to S108 are repeated until the determination in step S108 is affirmed. If the determination in step S108 is affirmed, the process proceeds to step S110, and the evaluation unit 18 extracts the maximum one of the calculated match rates. Here, 70% shown in the lower diagram of FIG. 16 is extracted.

Next, in step S112, the evaluation unit 18 corrects the recognition rate of the candidate character 1 based on the following equation (2) using the maximum matching rate as a correction value, and the corrected value (hereinafter referred to as correct answer rate 2). ) To get.
Correct answer rate 2 = recognition rate for candidate character 1 × correction value (2)

  Here, the evaluation unit 18 acquires the correct answer rate 2 = 56% (= 80% × 70%) from the above equation (2).

  As described above, when all the processes in step S50 in FIG. 14 are completed, the process proceeds to step S52 in FIG.

  In step S <b> 52 of FIG. 6, the evaluation unit 18 determines whether the correct answer ratios 1 and 2 are equal to or greater than the threshold value 2. The case where this determination is denied means a case where the character is corrected in the character image A as shown in the rightmost column of FIG. That is, as the threshold 2, it is possible to statistically acquire the correct answer rates 1 and 2 when the characters are corrected, and use a predetermined value (for example, an average value or a maximum value) obtained from the values. When judgment here is denied, it transfers to step S54. As described above, when the determinations in steps S36 and S46 are negative, the process proceeds to step S54.

  In step S54, the evaluation unit 18 sets the recognition result (temporary) as the recognition result (main). For example, if the recognition result (temporary) is set to reject in step S77 in FIG. 7, the determination in step S52 is likely to be denied. In such a case, in step S54, a rejection is set for the recognition result (book). Further, when the determination in step S36 or step S46 is negative and the process proceeds to step S54, the process using the error list DB 26 cannot be performed, so the recognition result (provisional) is directly used as the recognition result. Set to (Book). In addition, after the process of step S54 is performed, it transfers to step S62.

  On the other hand, when the determination is affirmed in step S52 and the process proceeds to step S56, the evaluation unit 18 determines whether or not the correct answer rate 1 is equal to or higher than the correct answer rate 2. If the determination here is affirmative, that is, if the correct answer rate 1 is greater than or equal to 2, the correct character in the corrected character extraction process 1 is set as the recognition result (book). On the other hand, if the determination in step S56 is negative, the process proceeds to step S60, and the evaluation unit 18 sets the candidate character 1 as a recognition result (book). In addition, after the process of step S58 or step S60 is performed, it transfers to step S62.

  In step S62, the evaluation unit 18 determines whether or not the recognition results (books) for all the character images A have been set. If the determination here is negative, the process returns to step S47, and the processing / determination of steps S47 to S62 for the next character image A is performed. Then, when the determination in step S62 is affirmed, all the processes in FIG. 6 are terminated, and the process proceeds to step S14 in FIG. If the determination in step S62 is affirmed, the evaluation unit 18 has obtained recognition processing data as shown in FIG.

  In step S14 of FIG. 5, the output unit 20 outputs the recognition result to the display device 50 of FIG. Here, when a rejection is set in the recognition result (book), a display indicating that the recognition could not be performed, for example, a symbol such as “?” Is displayed. Next, in step S16, the error list processing unit 16 determines whether or not the recognition result has been corrected by the operator. When judgment here is affirmed, it transfers to step S18. In step S18, the error list processing unit 16 updates the error list DB 26. That is, the error list processing unit 16 newly registers the client name (Fujitsutaro), the characters before and after correction, and the character image B in the error list DB 26 shown in FIG.

  Next, in step S20, the transmission unit 22 determines whether or not the transmission button has been pressed by the operator. The transmission button is a button that is pressed when the operator confirms that the content of the recognition result displayed on the display device 50 or the corrected recognition result is correct. If the determination in step S20 is negative, the process returns to step S16. On the other hand, if the determination in step S20 is affirmed, the process proceeds to step S22. In step S22, the transmission unit 22 transmits the recognition result displayed on the display device 50 or the corrected recognition result (referred to as transfer data) to the center that performs the transfer process.

  As described above in detail, according to the present embodiment, the recognition processing unit 14 recognizes the character image A to be processed, identifies candidate characters 1 to 5, and specifies the candidate characters 1 to 5. Each recognition rate (likelihood) is calculated, and the error list processing unit 16 refers to the error list DB 26 including the character image B in which the result of character recognition has been corrected to another character in the past. From the match rate (similarity) between the character image B in the error list DB 26 and the character image A, a correction value for correcting the recognition rate is calculated. And the evaluation part 18 correct | amends a recognition rate using the calculated correction value, calculates | requires a correct answer rate, and determines a recognition result (book) based on the said correct answer rate. Thereby, in this embodiment, the recognition rate can be corrected in consideration of the correction history of similar character images among character images recognized in the past. In this way, by considering the correction history of similar character images, the recognition rate is relatively high. However, even for characters that may be erroneously recognized due to the handwriting (跡) unique to the client, Character recognition can be performed with high accuracy.

  In the present embodiment, the error list DB 26 has an image of characters that have been recognized in the past, an uncorrected character related to the recognition result, and a corrected character, in association with the error list processing unit. 16 obtains from the error list DB 26 one or more character images B in which the candidate character having the highest recognition rate matches the character before correction, and matches each of the one or more character images B with the character image A. The rate (similarity) is calculated. Then, the maximum value of the matching rate (similarity) is calculated, and the maximum value of the matching rate is calculated as a correction value of the recognition rate of candidate characters that match the corrected character associated with the character image B having the maximum matching rate. And As a result, in this embodiment, the evaluation unit 18 sets the recognition rate of the candidate character that matches the corrected character associated with the character image B having the highest match rate with the character image A as the maximum match rate. Since the correct rate 1 can be calculated after correction, it is possible to calculate the correct rate of the candidate character that matches the corrected character with high accuracy.

  In the present embodiment, the error list processing unit 16 acquires one or more character images B in which the candidate character having the highest recognition rate matches the candidate character after correction from the error list DB 26, and the one or more character images B are obtained. The maximum value of the match rate between each character image B and the character image A is calculated, and the maximum value of the match rate is used as the correction value for the recognition rate of the candidate character having the highest recognition rate. Thereby, in this embodiment, the evaluation unit 18 determines the recognition rate of the candidate character having the highest recognition rate as the character image B and the character image A in which the candidate character having the highest recognition rate matches the corrected candidate character. Since the correct rate 2 can be calculated by correcting with the maximum value of the match rate, it is possible to calculate the correct rate of the candidate character with the highest recognition rate with high accuracy. Not limited to this, the error list processing unit 16 calculates a match rate between each character image B and the character image A, and uses a value obtained by some calculation using the match level as a recognition rate correction value. Also good.

  Further, in the present embodiment, the correct answer rate 1 and the correct answer rate 2 are compared, and when the correct answer rate 1 ≧ the correct answer rate 2, the candidate character related to the correct answer rate 1 is set as the recognition result (book), and the correct answer rate 1 <correct answer When the rate is 2, the candidate character related to the accuracy rate 2 is set as the recognition result (book). Thereby, a recognition result (this) can be determined by a simple method of simple comparison of correct answer rates 1 and 2. Here, in the present embodiment, based on the case of the past character image B similar to the character image A, in the case of the character image A, the possibility of correcting the recognition result (temporary) character to another character; Since it is determined which is more likely to be corrected from another character to a recognition result (provisional) character, the recognition result (book) can be appropriately determined.

  Further, in the present embodiment, the client's past data is extracted from the error list DB 26, and the recognition result (book) is determined using the extracted data. Thus, the recognition result (book) of characters having various characteristics can be appropriately determined for each client.

  In the above embodiment, the case where the evaluation unit 18 determines the recognition result (the book) based on the result of comparing the correct answer rate 1 and the correct answer rate 2 has been described, but the present invention is not limited to this. For example, the evaluation unit 18 may rank candidate characters related to the correct answer rate 1 and candidate characters related to the correct answer rate 2 based on a result of comparing the correct answer rate 1 and the correct answer rate 2. In this case, the output unit 20 may display the higher rank on the display device 50 and additionally display the lower rank as a correction candidate. That is, in the said embodiment, the evaluation part 18 will not ask | require the display aspect, if predetermined | prescribed character recognition is performed based on the corrected recognition rate (correction rate), and the result is displayed.

  In the above embodiment, the case where the error list processing unit 16 extracts only the client data from the error list DB 26 and performs the above-described processing has been described (see FIG. 9). However, the present invention is not limited to this, and the above-described processing may be performed using all data registered in the error list DB 26 (data of the client and data other than the client).

  Further, in the above embodiment, as shown in the above formulas (1) and (2), the correction value (maximum value of the match rate) is integrated with the recognition rate to calculate the correct answer rate. It is not limited to. For example, the correct answer rate may be calculated by a predetermined calculation using the recognition rate and the correction value.

  In the embodiment described above, the recognition rate is calculated using the distance information of the character feature vector. However, the present invention is not limited to this. The recognition rate may be calculated based on grammatical implications. For example, in the column where the number is likely to be written, if the katakana is recognized, the recognition rate can be lowered. Moreover, when the place name related to the place name before and after is recognized in the place where the address should be described, the recognition degree can be increased.

  The above processing functions can be realized by a computer. In that case, a program describing the processing contents of the functions that the processing apparatus should have is provided. By executing the program on a computer, the above processing functions are realized on the computer. The program describing the processing contents can be recorded on a computer-readable recording medium.

  When the program is distributed, for example, it is sold in the form of a portable recording medium such as a DVD (Digital Versatile Disc) or a CD-ROM (Compact Disc Read Only Memory) on which the program is recorded. It is also possible to store the program in a storage device of a server computer and transfer the program from the server computer to another computer via a network.

  The computer that executes the program stores, for example, the program recorded on the portable recording medium or the program transferred from the server computer in its own storage device. Then, the computer reads the program from its own storage device and executes processing according to the program. The computer can also read the program directly from the portable recording medium and execute processing according to the program. Further, each time the program is transferred from the server computer, the computer can sequentially execute processing according to the received program.

  The above-described embodiment is an example of a preferred embodiment of the present invention. However, the present invention is not limited to this, and various modifications can be made without departing from the scope of the present invention.

In addition, the following additional notes are disclosed regarding the above description.
(Supplementary note 1) A character recognition device for character recognition of a character image, wherein the character image is image recognized to identify one or more candidate characters, and for each of the one or more candidate characters, the candidate character is a character The character image of the history database is referred to by referring to a candidate character likelihood calculating unit that calculates a likelihood that the character is the same as the image, and a history database including the character image in which the result of character recognition of the character image is corrected. And a correction unit that corrects the likelihood of the candidate character from the similarity, and a character that is a recognition result from the candidate character based on the corrected likelihood A character recognition device comprising: a character recognition unit to be identified.
(Supplementary Note 2) The history database has a character image that has undergone character recognition, a recognition result of the character image, and a character that has been made a character of the character image by correction, and the correction unit includes the history image A character image whose character image recognition result is the same as the candidate character is obtained from the database, a similarity between the character image and the character image to be processed is calculated, and the similarity is used to calculate the character image. The character recognition apparatus according to appendix 1, wherein the likelihood of the candidate character is corrected.
(Additional remark 3) The said correction | amendment part acquires the character image which makes a candidate character the same character as a character image from the said history database, calculates the similarity degree of the said character image and the character image which is the said process target, The character recognition device according to appendix 2, wherein the similarity is used as a correction value for the likelihood of the candidate character.
(Supplementary Note 4) A character recognition method in which a computer performs character recognition of a character image, and recognizes the character image to identify one or more candidate characters, and each of the one or more candidate characters A candidate character likelihood calculating step for calculating a likelihood that the character is the same character as the character image, and a history database including the character image in which the result of character recognition of the character image is corrected, the history database A correction step of correcting the likelihood of the candidate character from the similarity, and a recognition result from the candidate character based on the corrected likelihood And a character recognition method for specifying a character to be recognized.
(Supplementary Note 5) The history database has a character image that has undergone character recognition, a recognition result of the character image, and a character that has been made a character of the character image by correction, and in the correction step, a computer , Obtaining a character image whose character image recognition result is the same as the candidate character from the history database, calculating a similarity between the character image and the character image to be processed, and using the similarity The character recognition method according to appendix 4, wherein the likelihood of the candidate character is corrected.
(Additional remark 6) In the said correction process, a computer acquires the character image which makes a candidate character the same character as a character image from the said history database, The similarity degree of the said character image and the character image which is the said process target is obtained. The character recognition method according to claim 5, wherein the character recognition is calculated and used as a correction value for the likelihood of the candidate character using the similarity.
(Supplementary note 7) The computer recognizes the character image and identifies one or more candidate characters, and for each of the one or more candidate characters, the likelihood that the candidate character is the same character as the character image A candidate character likelihood calculating unit for calculating, referring to a history database including the character image in which the result of character recognition of the character image is corrected, the similarity between the character image of the history database and the character image to be processed is determined. And a correction unit that corrects the likelihood of the candidate character from the similarity, and a character recognition unit that specifies a character as a recognition result from the candidate character based on the corrected likelihood. A character recognition program.
(Supplementary Note 8) The history database has a character image that has undergone character recognition, a recognition result of the character image, and a character that has been made a character of the character image by correction, and the correction unit includes the history image A character image whose character image recognition result is the same as the candidate character is obtained from the database, a similarity between the character image and the character image to be processed is calculated, and the similarity is used to calculate the character image. The character recognition program according to appendix 7, wherein the likelihood of a candidate character is corrected.
(Additional remark 9) The said correction | amendment part acquires the character image which makes a candidate character the same character as a character image from the said history database, calculates the similarity degree of the said character image and the character image which is the said process target, 9. The character recognition program according to appendix 8, wherein the similarity is used as a correction value for the likelihood of the candidate character.

10 Control device (character recognition device)
14 Recognition Processing Unit (Candidate Character Likelihood Calculation Unit)
16 Error list processing part (part of correction part)
18 Evaluation part (part of correction part, character recognition part)
26 Error list DB (history database)

Claims (5)

A character recognition device for character recognition of a character image,
Recognizing the character image to identify one or more candidate characters, and for each of the one or more candidate characters, calculate the likelihood that the candidate character is the same character as the character image And
Referring to a history database including the character image in which the result of character recognition of the character image is corrected, the similarity between the character image in the history database and the character image to be processed is obtained, and the candidate is obtained from the similarity. A correction unit for correcting the likelihood of characters;
A character recognition device comprising: a character recognition unit that identifies a character as a recognition result from the candidate character based on the corrected likelihood. The history database has a character image that has undergone character recognition, a recognition result of the character image, and a character that has been made a character of the character image by correction,
The correction unit acquires, from the history database, a character image whose character image recognition result is the same as the candidate character, calculates a similarity between the character image and the character image to be processed, The character recognition apparatus according to claim 1, wherein the likelihood of the candidate character is corrected using a similarity.   The correction unit acquires a character image in which the candidate character is the same character as the character image from the history database, calculates a similarity between the character image and the character image to be processed, and calculates the similarity The character recognition apparatus according to claim 2, wherein the character recognition device is used as a correction value for the likelihood of the candidate character. A character recognition method in which a computer performs character recognition of a character image,
Recognizing the character image to identify one or more candidate characters, and for each of the one or more candidate characters, calculate the likelihood that the candidate character is the same character as the character image Process,
Referring to a history database including the character image in which the result of character recognition of the character image is corrected, the similarity between the character image in the history database and the character image to be processed is obtained, and the candidate is obtained from the similarity. A correction step for correcting the likelihood of the character;
And a character recognition step of identifying a character as a recognition result from the candidate character based on the corrected likelihood. Computer
Recognizing the character image to identify one or more candidate characters, and for each of the one or more candidate characters, calculate the likelihood that the candidate character is the same character as the character image Part,
Referring to a history database including the character image in which the result of character recognition of the character image is corrected, the similarity between the character image in the history database and the character image to be processed is obtained, and the candidate is obtained from the similarity. A character recognition program that functions as a correction unit that corrects the likelihood of a character, and a character recognition unit that specifies a character as a recognition result from the candidate character based on the corrected likelihood.

Images