JP2004341754A - Character recognition result correcting device and its method, and character recognition result correcting program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a character recognition result correcting device capable of quickly and efficiently correcting a recognition result obtained by recognizing a large amount of character image data segmented from image data. <P>SOLUTION: Character image data, candidate character code of its recognition result and urbanization distance are recorded in a recognized data storage part 203. A display control part 205 reads the character image data whose candidate character codes are the same in the order of the longer urbanization distance, and displays a character image data list. A character correcting part 206 replaces the candidate character code of the selected character image data with the character code of an inputted correct character. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a character recognition result correcting apparatus and method for correcting a character recognition result of character image data included in image data input by an OCR device or the like to correct characters.
[0002]
[Prior art]
In recent years, the demand for paperlessness has increased, and it has become common to digitize existing documents and store and use them.
In order to digitize and save an existing document, a technique of performing character recognition using an OCR device or the like and converting a character included in the document into a text code is used. At present, character recognition is not performed with 100% accuracy. To correct the character recognition result, the character image with the character recognized and the character of the recognition result are displayed side by side, and the wrong part is visually found and corrected using the keyboard. Work is being performed (for example, Patent Document 1).
[0003]
However, there is a problem that it takes time to find and correct an incorrect character.
Therefore, in order to increase the efficiency of correcting a character image that has been erroneously recognized, a technology has been disclosed in which character images of the same character type and the same character code in a predetermined area are collectively displayed and displayed side by side with the character of the recognition result and corrected. (For example, Patent Document 2).
[0004]
[Patent Document 1]
JP-A-1-111285 [0005]
[Patent Document 2]
JP-A-11-007492
[Problems to be solved by the invention]
However, even in the technique of Patent Document 2, it is necessary to sequentially compare the character image and the character of the recognition result, and to correct the character image if it is erroneously recognized. It is necessary to further improve work efficiency.
Accordingly, it is an object of the present invention to provide a character recognition result correcting apparatus and method capable of correcting a large number of character recognition results in a short time.
[0007]
[Means for Solving the Problems]
In order to solve the above-mentioned problems, the present invention provides an extracting unit for extracting character image data for each character from input image data, and a candidate character for recognizing candidate characters from the extracted character image data using a recognition dictionary. Recognizing means, storage means for storing the position of the character image data on the image data, and the character code of the candidate character in association with the character image data, and a character having the same character code stored in the storage means. Extraction display means for extracting image data and displaying it in a list, character correction operation means for correcting misrecognized character image data of the displayed character image data to correct characters,
And a replacement unit configured to replace a character code of a candidate character associated with the erroneously recognized character image data stored in the storage unit with a corrected character code of a correct character.
[0008]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, an embodiment of a character recognition result correcting apparatus according to the present invention will be described with reference to the drawings.
(Embodiment 1)
FIG. 1 is a diagram showing a hardware configuration of an embodiment of a character recognition result correcting apparatus according to the present invention.
[0009]
This character recognition result correction device includes a central processing unit (CPU) 10 for controlling each part of the device, a cathode ray tube (CRT) 11 for displaying the progress and results of processing, a mouse 12, and a keyboard 13 , A scanner 14, a random access memory (RAM) 15, a hard disk 19, and a read-only memory (ROM) 22.
[0010]
The CRT 11 may be a liquid crystal display or the like. The mouse 12, the keyboard 13, and the scanner 14 constitute an input unit.
The RAM 15 includes an entire image storage area 16 for storing image data of a document read by the scanner 14, a recognition data storage area 17 for storing a recognition result for each character image data cut out from the image data, and a work area for the CPU 10. 18
[0011]
The hard disk 19 has a program storage area 20 and an OS storage area 21 for storing an operating system. The program storage area 20 stores the character recognition result correction program of the present embodiment, and the CPU 10 operates according to this program.
The ROM 22 has a startup program storage area 23, a recognition dictionary 24, and a misrecognition dictionary 25.
[0012]
The boot program stored in the boot program storage area 23 is necessary for the operating system to operate.
The recognition dictionary 24 registers a set of a standard feature amount of a standard character and its character code. The standard characters are characters to be collated with character image data to be recognized, and include symbols in addition to characters.
[0013]
The misrecognition dictionary 25 will be described later.
FIG. 2 is a configuration diagram of the first embodiment.
The character recognition result correcting device includes a scanner 14, a whole image storage unit 201, a character image data cutout unit 202, a recognition data storage unit 203, a character recognition unit 204, a display control unit 205, and a character correction unit 206. And a control unit 207.
[0014]
The scanner 14 reads the paper on which the document is described and stores the read paper as image data in the overall image storage unit 201. At this time, the image may be stored as a color image or as a black and white binary image.
When the image data is stored in the whole image storage unit 201, the character image data cutout unit 202 cuts out character image data for each character from the image data according to an instruction from the control unit 207, and notifies the character recognition unit 204 of the cutout. Then, the character image data is stored in the recognition data storage unit 203.
[0015]
The character image data cutout unit 202 assigns a document number and a character image number to the image data, and records the position of the character image data on the image data.
The recognition data storage unit 203 records the recognition result of the character image data.
FIG. 3 is a diagram illustrating an example of the character image data recognition result stored in the recognition data storage unit 203.
[0016]
The character image data recognition result 301 includes a document number 302, a character image number 303, character image data 304, a character image position 305, a candidate character code 306, and an urbanization distance 307.
The document number 302 is a number assigned to each sheet of the document read by the scanner 14, that is, each image data.
[0017]
The character image number 303 is a number sequentially given to character image data cut out from the image data.
The character image data 304 is image data of a character cut out as one character.
The character image position 305 is indicated by x and y coordinates of vertices at diagonal positions of a rectangle circumscribing the character image data 304. Here, an upper left vertex and a lower right vertex are used.
[0018]
The document number 302, character image number 303, character image data 304, and character image position 305 are recorded by the character image data cutout unit 202.
The candidate character code 306 and the urbanization distance 307 are recorded by the character recognition unit 204.
The character recognizing unit 204 has a recognition dictionary 24, and upon receiving a notification of the character image data from the character image data extracting unit 202, extracts a feature amount from the character image data, and outputs a standard feature of each standard character in the recognition dictionary 24. Calculate the urbanization distance with the quantity. The standard character with the shortest urbanization distance is recognized as a candidate character, and its character code is recorded in the candidate character code 306 of the character image data recognition result 301 recorded in the recognition data storage unit 203. At the same time, the urbanization distance 307 is also recorded.
[0019]
Here, the urbanization distance 307 indicates the similarity between the character image data to be recognized and the standard character. The shorter the urbanization distance is, the higher the similarity is, and the probability that the character image data is the standard character is high. Get higher. The longer the urbanization distance, the lower the similarity.
For example, the character image data of the character image number “3” is recognized as the candidate character code “2330”, that is, “0”, but the urbanization distance 307 is “2300” and the character image numbers “1” and “2”. Is longer than the urbanization distances “850” and “760” of the character image data “”, and the probability of misrecognition is high.
[0020]
It is needless to say that the similarity does not need to be indicated by the urbanization distance, and other values may be used according to the feature amount of the recognition dictionary and the like.
When correcting the character image data recognition result recorded in the recognition data storage unit 203, the control unit 207 instructs the display control unit 205 to display a list of character image data for each candidate character code.
[0021]
The display control unit 205 selects one of the candidate character codes 306 recorded in the recognition data storage unit 203 and extracts character image data having the same candidate character code. At this time, character image data having a long urbanization distance are sequentially extracted and displayed in a list. The association between the displayed character image data 304 and the character image data recognition result 301 is, for example, associated with a document number 302 and a character image number 303.
[0022]
The character correction unit 206 corrects erroneous recognition of the character image data listed by the display control unit 205.
FIG. 4 is a diagram illustrating an example of a correction screen for incorrectly recognized character image data.
The correction screen 401 displays a character image data list 402 and a text box 403 for inputting a correct character.
[0023]
By looking at the character image data list 402, the operator can recognize at a glance what the recognized characters of the character image data are. In the case of the character image data list 402, the numeral “0” is recognized from the character image data.
The display control unit 205 displays the character image data list 402 such that the upper left character image data has a higher probability of being erroneously recognized and the lower right character image data has a lower probability of being erroneously recognized.
[0024]
Therefore, the operator only has to pay particular attention to the character image data in the vicinity of the uppermost stage of the character image data 402 so as to find erroneously recognized character image data.
The operator moves the highlighted cursor 404 to the erroneously recognized character image data with the mouse 12 and inputs the corrected character into the text box 403 using the keyboard 13. For example, the character image data (the first character at the top row) erroneously recognized as “0” is corrected to “1”.
[0025]
The character correction unit 206 receives a notification of the document number and the character image number of the character image data displayed at the position of the cursor 404 from the display control unit 205. The candidate character code specified by the document number and the character image number of the character image data recognition result 301 stored in the recognition data storage unit 203 is replaced with the character code of the character input to the text box 403.
[0026]
This process is repeated for the number of candidate character codes recorded as the candidate character codes 306 of the character image data recognition result 301.
In this way, erroneous recognition of character image data recognized from a large amount of image data can be efficiently corrected in a short time.
When the document is read by the scanner 14 and the image data is stored in the entire image storage unit 201, the control unit 207 instructs the character image data extraction unit 202 to extract character image data for each character. When the recognition of the character image data is completed by the character recognition unit 204 and an end instruction is given by the operator, the display control unit 205 is instructed to display a correction screen.
[0027]
Although the display control unit 205 arranges the character image data with the lowest similarity in order, it is needless to say that the character image data may be arranged in the reverse order.
FIG. 5 shows the contents finally stored as the character recognition correction result in the recognition data storage unit 203.
In the character recognition correction result 501, a document number 502 of the document, a character image position 503 indicating a character position on the image data, and a character code 504 are recorded. When the character code 504 is not corrected as erroneously recognized character image data, the candidate character code 306 of the character image data recognition result 301 is recorded as it is.
[0028]
As a result, the correct character code of the character described at the character image position included in the document is recorded for each document. For example, if the character indicated by the character code is printed in a predetermined font at the size specified by the character image position, the original document input by the scanner 14 is restored.
Next, the operation of the present embodiment will be described with reference to the flowcharts of FIGS.
[0029]
FIG. 6 is a flowchart illustrating the operation of recognizing character image data in a document using the recognition dictionary 24.
First, a document is read by the scanner 14, and image data is stored in the entire image storage unit 201 (S602).
The character image data cutout unit 202 cuts out character image data for each character from the image data, records the character image data and the like in the recognition data storage unit 203, and notifies the character recognition unit 204 of the character image data (S604). .
[0030]
The character recognition unit 204 compares the character with the recognition dictionary 24, calculates the urbanization distance, and recognizes the standard character having the shortest urbanization distance as a candidate character (S606). The character code of the candidate character recognized corresponding to the character image data and the urbanization distance are recorded in the character image data recognition result 301 (S608).
Next, the character recognition unit 204 determines whether or not notification of the character image data has been received from the character image data cutout unit 202 (S610). If the notification has been made, the process returns to S606; if not, the notification has not been made. It moves to S612.
[0031]
Next, the control unit 207 determines whether or not there is an end instruction (S612), and if there is, terminates the process; otherwise, returns to S602.
FIG. 7 is a flowchart for explaining the operation of the correction processing of the character image data recognition result 301 recognized by the character recognition unit 204.
First, the display control unit 205 determines whether the character image data recognition result 301 stored in the recognition data storage unit 203 includes an unprocessed candidate character code for erroneous recognition and correction (S702). If not, the process ends. If there is, one unprocessed candidate character code is selected, and the character image data is read out from the longest urbanization distance in order and displayed as a character image data list (S704).
[0032]
Next, the character correction unit 206 determines whether or not character image data that has been erroneously recognized is selected from the character image data list (S706). If not, the process returns to S702. When selected, the correct character of the selected character image data is input (S708).
The character correction unit 206 replaces the candidate character code of the character image data recognition result 301 recorded in the recognition data storage unit 203 with a character code of a correct character (S710), and returns to S706.
[0033]
(Embodiment 2)
Next, a second embodiment of the character recognition result correcting apparatus according to the present invention will be described. The configuration of the character recognition result correction device is substantially the same as the configuration of the first embodiment (see FIG. 2), except that the character correction unit 206 has the misrecognition dictionary 25. Therefore, a configuration unique to the present embodiment will be described.
[0034]
FIG. 8 is a diagram illustrating an example of the misrecognition dictionary included in the character correction unit 206. The misrecognition dictionary 25 registers, for each character code 801, an erroneously recognized character 802 and a character code 803 which are liable to be erroneously recognized as the character empirically. For example, the number “0” may be erroneously recognized from the numbers “6” and “1”, the alphabets “O”, “o”, “a”, “D”, and the Greek letters “σ” and “Θ”. It is shown that.
[0035]
When the character correcting unit 206 receives the notification of the candidate character code from the control unit 207, the character correcting unit 206 reads a plurality of erroneously recognized characters registered in the matching character code in the erroneously recognized dictionary 25, and notifies the display control unit 205 of the same.
The control unit 207 receives the completion of the recognition of the character image data by the character recognition unit 204, and instructs the display control unit 205 to display a correction screen. When receiving the notification of the selected candidate character code from the display control unit 205, it notifies the character correcting unit 206 of the candidate character code.
[0036]
The display control unit 205 displays the correction screen on the CRT 11 when receiving the display instruction of the correction screen from the control unit 207 and receiving the notification of the erroneously recognized character from the character correction unit 206.
FIG. 9 is a diagram illustrating an example of the correction screen displayed on the CRT 11.
The correction screen 901 displays a character image data list 402 and a plurality of misrecognized characters 902 corresponding to the candidate character codes. The misrecognized character 902 is a character group having a high probability of being erroneously recognized as a character of the candidate character code displayed in the character image data list 402.
[0037]
The misrecognized character 902 is displayed near the top of the character image data 402, that is, near the character image data with low similarity.
More specifically, the correction operation of the candidate character will be described. When the character image data 905 of the fourth character at the top of the character image data list 402 is erroneously recognized, the operator uses the mouse to select among the erroneously recognized characters 902. A correct character “6” 903 is selected and dragged and dropped onto character image data 905 as indicated by an arrow 904.
[0038]
The character correction unit 206 reads the character code of the selected character “6” from the misrecognition dictionary 25 and replaces the candidate character code recorded in the recognition data storage unit 203 with the read character code.
In this way, if characters having a high probability of being erroneously recognized are displayed as erroneously recognized characters 902, the complexity of inputting characters to the text box 403 using a keyboard as in Embodiment 1 can be reduced. By avoiding this, the erroneously recognized character can be corrected efficiently.
[0039]
The display control unit 205 may change the display of the character image data 905 to the display of the erroneously recognized character “6” 903 so that it can be confirmed that the erroneously recognized character image data has been corrected. .
Next, the operation of the present embodiment will be briefly described. The only difference is the operation of the first embodiment and the operations of S704 to S708 of the flowchart shown in FIG.
[0040]
The display control unit 205 selects one unprocessed candidate character code, reads out the character image data in order from the longest urbanization distance and arranges the character image data list, and displays a correction screen on which a misrecognized character is displayed. indicate.
The character correction unit 206 determines whether a correct character has been selected from the erroneously recognized characters and whether drag and drop has been performed on the erroneously recognized character image data. If drag and drop is performed, the process returns to step S710; otherwise, the process returns to step S702.
[0041]
In each of the above embodiments, erroneous recognition and correction of handwritten characters described in a document have been described. However, the present invention is not limited to handwritten characters, but can be applied to printed printed characters.
Although the JIS character code is used as the character code, other character codes may be used as a matter of course.
[0042]
【The invention's effect】
As described above, the present invention provides an extracting unit that extracts character image data for each character from input image data, and a candidate character recognition unit that recognizes candidate characters from the extracted character image data using a recognition dictionary. Means, storage means for storing the position of the character image data on the image data, and the character code of the candidate character in association with the character image data, and a character image having the same character code stored in the storage means. Extraction display means for extracting data and displaying the list, character correction operation means for correcting incorrectly recognized character image data among the character image data displayed in the list to correct characters, and erroneous recognition stored in the storage means Replacement means for replacing the character code of the candidate character associated with the corrected character image data with the corrected character code of the correct character. With such a configuration, character image data included in a large amount of image data is corrected for each character code, so that character image data that has been erroneously recognized can be corrected to a correct character efficiently in a short time.
[0043]
Further, the candidate character recognizing means obtains a similarity indicating the likelihood of recognition of the recognized candidate character, and the storage means stores the character image data further in association with the similarity of the candidate character. The extraction display means extracts character image data in order of high or low similarity and displays it in a list. With such a configuration, the operator can find erroneously recognized character image data by focusing on the first or last character image data displayed in the list.
[0044]
The candidate character recognizing means obtains a distance between a standard feature amount of a character registered in a recognition dictionary and a feature amount of the character image data, recognizes the shortest character as a candidate character, and stores the shortest character in the storage unit. The similarity is the distance, and the extraction display means extracts and displays the list in ascending order of the distance. With such a configuration, the operator can find the recognized character image data by paying attention only to the first character image data displayed in the list.
[0045]
In addition, the character correction operation unit reads a plurality of characters corresponding to a character code and a plurality of characters corresponding to a character code of a candidate character from the misrecognition dictionary. An erroneously recognized character display section displayed together with a list display of candidate characters, and a correction operation for associating one character image data displayed in the list with one character displayed on the erroneously recognized character display section to correct the one character character Part. With such a configuration, the character image data that is erroneously recognized in the character image data displayed in the list can be easily corrected to a correct character.
[0046]
The correction operation unit is an operation in which one character is dragged and dropped onto one character image data by an operator. With such a configuration, the operator can correct the erroneously recognized character image data to correct characters by a simple operation.
Also, the present invention provides a cutting step of cutting out character image data for each character from the input image data, a candidate character recognition step of recognizing candidate characters from the cut out character image data using a recognition dictionary, A recording step of recording the position of the character image data on the image data and the character code of the candidate character in the storage area in association with the character image data, and the character image data having the same character code recorded in the storage area Extracting and displaying a list of characters, a character correction operation step of correcting character image data that is erroneously recognized among the character image data displayed in the list to correct characters, and an erroneous recognition stored in the storage area. Replacing the character code of the candidate character associated with the corrected character image data with the corrected character code of the correct character. It is. According to such a method, the character image data that has been erroneously recognized can be efficiently corrected to a correct character in a short time.
[0047]
Further, the present invention is a character recognition result correction program applied to a character recognition result correction device incorporating a computer, wherein a cutout step of cutting out character image data for each character from input image data; Character recognition step of recognizing a candidate character from a character image data using a recognition dictionary, a storage area in which the position of the character image data on the image data and the character code of the candidate character are associated with the character image data. A recording step of recording the character image data having the same character code recorded in the storage area, and an extraction display step of extracting and displaying a list of character image data; A character correcting operation step of correcting a character to a correct character, and the character correction data associated with the erroneously recognized character image data stored in the storage area. It is set to be executed each step of the replacement step of replacing the correct character code of the character is corrected character code of the complement characters to the computer. By applying such a character recognition result correction program to a character recognition result correction device, it is possible to efficiently correct incorrectly recognized character image data into correct characters in a short time.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating a hardware configuration of a character recognition result correcting apparatus according to a first embodiment of the present invention.
FIG. 2 is a configuration diagram of the above embodiment.
FIG. 3 is a diagram illustrating an example of recognition data stored in the recognition data storage unit according to the embodiment in association with character image data;
FIG. 4 is a diagram showing an example of a correction screen according to the embodiment.
FIG. 5 is a diagram showing an example of a character recognition result corrected in the embodiment.
FIG. 6 is a flowchart illustrating an operation of character recognition according to the embodiment.
FIG. 7 is a flowchart illustrating an operation of correcting a recognition result of character image data according to the embodiment.
FIG. 8 is a diagram illustrating an example of the contents of a misrecognition dictionary included in a character correction unit according to a second embodiment of the character recognition result correction apparatus according to the present invention.
FIG. 9 is a diagram illustrating an example of a correction screen according to the embodiment.
[Explanation of symbols]
10 CPU
11 CRT
12 Mouse 13 Keyboard 14 Scanner 15 RAM
19 Hard disk 22 ROM
24 recognition dictionary 25 misrecognition dictionary 201 whole image storage unit 202 character image data cutout unit 203 recognition data storage unit 204 character recognition unit 205 display control unit 206 character correction unit 207 control unit

Claims (7)

Extracting means for extracting character image data for each character from the input image data;
Candidate character recognition means for recognizing candidate characters from the cut-out character image data using a recognition dictionary;
A storage unit that stores the position of the character image data on the image data and the character code of the candidate character in association with the character image data.
Extraction display means for extracting character image data having the same character code stored in the storage means and displaying the extracted character image data in a list,
Character correction operation means for correcting misrecognized character image data among the character image data displayed in the list to correct characters,
A character recognition result correction method comprising: replacing a character code of a candidate character associated with the erroneously recognized character image data stored in the storage means with a corrected character code of a correct character. apparatus. The candidate character recognition means obtains a similarity indicating the likelihood of recognition of the recognized candidate character,
The storage unit further stores the character image data in association with a similarity of a candidate character,
2. The character recognition result correcting apparatus according to claim 1, wherein the extraction display unit extracts the character image data in order of high or low similarity and displays the list in a list. The candidate character recognition means obtains a distance between a standard feature amount of a character registered in a recognition dictionary and a feature amount of the character image data, and recognizes the shortest character as a candidate character,
The similarity stored in the storage means is the distance,
3. The character recognition result correcting device according to claim 2, wherein the extraction display unit extracts the objects in order from the one with the longest distance and displays the list. The character correction operation means includes:
A misrecognition dictionary that registers a plurality of characters with a high probability of misrecognition corresponding to the character code;
An erroneously recognized character display unit that reads a plurality of characters corresponding to the character codes of the candidate characters from the erroneously recognized dictionary and displays the candidate characters together with a list of candidate characters;
5. The character recognition device according to claim 4, further comprising: a correction operation unit configured to associate the one character image data displayed in the list with one character displayed on the misrecognized character display unit and correct the character image data into one character. Result correction device. The character recognition result correcting apparatus according to claim 4, wherein the correction operation unit is an operation of dragging and dropping one character to one character image data by an operator. Cutting out character image data for each character from the input image data;
A candidate character recognition step of recognizing candidate characters from the cut-out character image data using a recognition dictionary;
A recording step of recording a position on the image data of the character image data and a character code of a candidate character in a storage area in association with the character image data;
An extraction display step of extracting character image data having the same character code recorded in the storage area and displaying a list thereof;
A character correction operation step of correcting character image data that is misrecognized among the character image data displayed in the list to correct characters,
Replacing the character code of the candidate character associated with the erroneously recognized character image data stored in the storage area with the corrected character code of the correct character. Method. A character recognition result correction program applied to a character recognition result correction device incorporating a computer,
Cutting out character image data for each character from the input image data;
A candidate character recognition step of recognizing candidate characters from the cut-out character image data using a recognition dictionary;
A recording step of recording a position on the image data of the character image data and a character code of a candidate character in a storage area in association with the character image data;
An extraction display step of extracting character image data having the same character code recorded in the storage area and displaying a list thereof;
A character correction operation step of correcting character image data that is misrecognized among the character image data displayed in the list to correct characters,
A character recognition result for causing a computer to execute each step of replacing a character code of a candidate character associated with the erroneously recognized character image data stored in the storage area with a corrected character code of a correct character Fix.

Images