JP3455643B2 - Method of updating learning dictionary in character recognition device and character recognition device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a learning dictionary updating method and a character recognition device in a character recognition device (OCR; optical character reading device).

[0002]

2. Description of the Related Art In a conventional OCR, an operator visually checks a character recognition result, and whenever a misread character occurs, a correction input is made to correct the character and correct the result.

[0003]

Therefore, it takes a long time to make a correction, and the operator's load is great. In particular, a character input to the OCR is usually a character printed by a specific printing machine or a specific printer, and thus is often a specific character or character string. In other words, a specific character or character string is always misread, which requires an operator to correct it.

The present invention has been made to solve the problems of the above-mentioned conventional character recognition device, and its purpose is to:
In particular, it is an object of the present invention to provide a learning dictionary update method and a character recognition device in a character recognition device that can reduce the probability of misreading a character printed by a specific printing machine or a specific printer.

[0005]

According to a first aspect of the present invention, there is provided a method for updating a learning dictionary in a character recognition apparatus, wherein a character image is photoelectrically converted to obtain character image data, and character recognition processing is performed on the character image data. When the character recognition processing is performed by referring to the fixed dictionary in which the information to be performed is fixedly stored and the learning dictionary that reflects the correction input, and the correction input is performed on the recognition result , the corrected input character and its Similar
Ranking information in degree and candidate character, misread character and its similarity
As a set of degree information,
It is characterized in that it is registered or not registered in the learning dictionary according to the degree of similarity of the characters that have been correctly input . By this ,
Whether or not to register in the learning dictionary is automatically determined according to the similarity of the characters related to the correction input in the recognition result, and the registration is performed only when necessary, and the corrected and input sentence is registered.
Characters and their similarity, ranking information in candidate characters, and misread sentences
It is possible to create a learning dictionary in which characters and their similarity information are registered as one set .

In the method for updating the learning dictionary in the character recognition apparatus according to claim 2 of the present invention , the recognition processing is included in the one set.
Enter the learning dictionary including the characters before and after the target character.
To record. You can refer to one set including these letters.
Can reduce misreading when the string is predetermined
It

In the one set of learning dictionaries in the method for updating the learning dictionaries in the character recognition apparatus according to the third aspect of the present invention, the character before and after the one character to be recognized and the similarity thereof are included.
It is characterized that it is registered in the learning dictionary including information.
It As a result, these surrounding characters and their similarity information
It is possible to refer to one set including them, and it is possible to reduce misreading when a character string is predetermined.

According to a fourth aspect of the character recognition apparatus of the present invention, a character image data acquisition unit for photoelectrically converting a character image into character image data and information for performing character recognition processing on the character image data are fixed. A stored fixed dictionary,
Corrected input character and its similarity and in the candidate character
One set of ranking information, misread characters and their similarity information
A learning dictionary to be stored , a recognition processing unit for performing character recognition processing on the character image obtained by the character image acquisition unit by referring to the fixed dictionary and the learning dictionary, and a result of the character recognition processing by the recognition processing unit are displayed. Display section for
There is an input unit for inputting information and a correction input from the input unit for the recognition result displayed on the display unit.
And the corrected character and its similarity and candidate character
One set of ranking information, misread characters and their similarity information
Then, in the recognition result, the corrected input character
A control unit for registering or non-registering in the learning dictionary according to the degree of similarity . By this
Te, whether to register against the learning dictionary, is automatically determined depending on the character of the similarity according to the correction input of the recognition result, is only performed registration in the case of the required, be modified input
Character and its similarity, and
It is possible to create a learning dictionary having the content of registered reading characters and their similarity information as one set .

According to the character recognition device of claim 5 of the present invention ,
The learning dictionary contains the characters before and after the one character to be recognized.
It is characterized in that it is registered in the learning dictionary . With this, you can refer to one set including these surrounding characters,
Can reduce misreading when the string is predetermined
It

In the learning dictionary of the character recognition device according to claim 6 of the present invention, the character before and after the one character to be recognized and the like.
It is characterized in that one set including the similarity information is stored . As a result, one set including the characters before and after the one character to be recognized and the similarity information thereof is stored.
Therefore, these are referred to as a whole, and the misreading when the character string is predetermined can be reduced.

In the character recognition apparatus according to claim 7 of the present invention, when the recognition processing unit refers to the learning dictionary and corrects the recognition result, the recognition result character is registered in the learning dictionary. It is characterized in that the information of the preceding and following characters is the same. As a result, when the recognition result is corrected by referring to the learning dictionary, it is required that the information of the characters before and after the character registered as the recognition result character in the learning dictionary match. Misreading in units can be reduced.

In the character recognition device according to claim 8 of the present invention, when the recognition processing unit refers to the learning dictionary and corrects the recognition result, all the characters registered in the learning dictionary for the normal characters are registered. It is characterized in that the information of (1) matches . This ensures that there are no corrections under strict conditions.
To be done.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A character recognition apparatus according to an embodiment of the present invention and a learning dictionary updating method in the character recognition apparatus will be described below with reference to the accompanying drawings. In each drawing, the same constituents are designated by the same reference numerals, and duplicated description will be omitted.
FIG. 1 is a block diagram of a character recognition device according to an embodiment of the present invention. The character recognition device is provided with a character image data acquisition means 1 and is configured to photoelectrically convert a character image into character image data. The character image data acquisition unit 1 includes a reading unit 11 that performs photoelectric conversion,
The normalization unit 12 that performs normalization such as noise removal from the signal obtained by the reading unit 11, and the detection cutout unit 13 that cuts out each character from the image data of the character string output from the normalization unit 12 are provided. Has been. The image data of each character cut out by the detection cutout unit 13 is sent to the recognition processing unit 2.

The recognition processing unit 2 is connected to a fixed dictionary 3 in which information for performing character recognition processing on character image data is fixedly stored, and a learning dictionary 4 for reflecting correction input. The unit 2 refers to the fixed dictionary 3 and the learning dictionary 4 for the character image obtained by the character image acquisition means 1 to perform character recognition processing. The character recognition result is sent to the control unit 5. The control unit 5 includes a display unit 6 such as a CRT display device for displaying information, an input unit 7 such as a keyboard input device for inputting information and commands,
File part 8 for saving final result data of recognition
Is connected to the learning dictionary unit 4, and the contents of the learning dictionary unit 4 are updated.

A flow chart of the character recognition processing by the character recognition device configured as described above is shown in FIG. 2, and the operation will be described with reference to this. First, the character image data acquisition means 1 performs an image (character image) data acquisition process (S1). As a result, image data relating to the characters written on the form or original is obtained, and an image file is created. The image file is created in the recognition processing unit 2 in FIG.

Next, the recognition processing section 2 refers to the image file to read out the character related to the recognition target, and refers to the fixed dictionary 3 to perform the recognition processing of the character related to the recognition target (S).
2). In this recognition processing, a pattern matching method, a feature extraction method, or the like is used, and the similarity of each candidate character is calculated. Then, using the recognition result, the learning dictionary 4 is referred to, and the correction process for the recognition result is performed (S3).

Next, the control unit 5 displays the corrected recognition result on the display unit 6. The operator who sees this confirms the recognition result and corrects the misread character (S
4). For example, move the cursor to the displayed misread character and overwrite the correct character. Learning processing is performed on the result corrected in this way (S
5). That is, the contents of the learning dictionary 4 are updated based on the correction result. This update execution / non-execution is performed depending on how similar the characters related to the correction result are in the recognition result. Be seen. Then, in the next result output process, the correction result is stored in the file unit 8 as a character recognition result file (S6).

Details of the processing shown in FIG. 2 will be described with reference to FIG. In the example of FIG. 3, the OCR form or O
In CR manuscript 21, it is assumed that "... I am a Japanese ...". In the image data acquisition process (S1), the OCR 10 photoelectrically converts the image on the OCR document 21 into image data. Furthermore, OC
The R10 performs character recognition on the image data with reference to the information in the storage areas corresponding to the fixed dictionary 3 and the learning dictionary 4 in FIG. 1 (S2), and displays the recognition result on the display unit 6. In this recognition result, it is assumed that it was misread as "... I am a person from Hiki".

In the next correction process (S3), the operator 22 visually confirms the recognition result displayed on the screen of the display unit 6, and the "Japanese" is mistakenly read as "Hikito". To judge. Therefore, the operator 22 makes a correction input from the input unit 7 to correct “tree” into “book”. Then O
The CR 10 proceeds to the learning process (S5) by inputting a predetermined instruction or automatically, and determines whether the character related to the correction result is similar in the recognition result (based on whether the character is similar or not). ) The registration / non-registration control is performed for the area corresponding to the learning dictionary 4 in the storage area 20 with the corrected character as a normal character. The character information registered in the learning dictionary 4 of the storage area 20 includes misread characters (“tree” in the above), regular characters (“book” in the above), and preceding and following characters (“day” and “person” in the above). , The degree of similarity (the degree of representing the similarity of a character to image data), and the like. When the learning is completed, the storage area 2
The character recognition process using the result of 0 is performed by the OCR 10.

FIG. 4 shows details of the correction process during the above process, which will be described. The image data 23 obtained by the character image data acquisition means 1 is subjected to recognition processing using the fixed dictionary 3, and character data “ADC” is obtained.
Shall be obtained. Here, the printed image is “ABC”, which means that “B” is misread as “D”. Next, the correction process is performed, but the correction process is premised on that the corresponding character (here, “D”) is present in the learning dictionary. Therefore, the recognition processing unit 2 uses S11.
As shown in, the learning dictionary 4 is searched. Here, the target character “D” is registered, it is determined to be “present”, and the process proceeds to S12.

As a result of the recognition performed by the recognition processing unit 2 on the image data relating to the print character "B", the candidate character "D" is obtained.
As a result of this recognition, a plurality of candidate characters are obtained as shown in the candidate character table 25. Then, the character "D" having the highest degree of similarity is output as a candidate character.

On the other hand, the learning information corresponding to the character "D" registered in the learning dictionary 4 is as shown in the character information table 24. That is, for the misread character “D”, the preceding character is “A”, the subsequent character is “C”, the similarity of the misread character is “95”, and the correct character (misread character is The correct letter instead of is "B",
The similarity of the regular character is "85", and the rank of the regular character is "3". In other words, in the past history, the character “B” has been misread, and at that time,
It indicates that it is misread as "D". Then, the candidate characters of the recognition result (corresponding to the candidate character table 25) at that time include “D” and “B”, and the respective similarity degrees are “95” and “85”. “B” indicates that the rank was “3”.

Therefore, the recognition processing unit 3 refers to the character data related to the recognition result this time, and recognizes the candidate character "D" to be recognized.
Detect the preceding and following characters of. Here, the preceding character “A” and the succeeding character “C” are detected, and it is detected that they match the information in the character information table 24 of the learning dictionary 4.
Further, the recognition processing unit 3 refers to the candidate character table 25 to detect the similarity “95” of the misread character “D”, the similarity “85” and the rank “3” of the regular character “B”, It is detected that the information matches the information in the character information table 24 of the learning dictionary 4. That is, it is detected that the situation regarding the candidate character “D” related to the recognition target this time is exactly the same as the information in the character information table 24 regarding the misread character “D” in the learning dictionary 4. Therefore, assuming that all the conditions are matched, the candidate character “D” is corrected to “B” and output.

As described above, when it is detected that the situation regarding the candidate character relating to the recognition target this time is exactly the same as the information in the character information table regarding the misread character of the learning dictionary 4, the same is performed by performing the correction. It is possible to properly correct misreading of characters when printed by a printing machine or printer. In the present embodiment, the condition of complete coincidence is set as the condition, but the similarity is set to a predetermined range (for example, “5” before and after), and correction is performed within that range. May be. Therefore, in the recognition result of this time, the similarity of "D" which is a misread character is "100 to 90",
If the similarity of the regular character "B" is "90-80" in the recognition result this time, the correction is performed. When the width is given in this way, even if the characters printed by the same printing machine or printer are recognized, the calculation result of the degree of similarity may be slightly different each time due to variations in printing conditions. Can be remedied and appropriate corrections can be made.

In the above example, the condition is that the preceding and following characters match. Therefore, in the case of the first character, the previous character is "none", and in the case of the last character the latter character is "none".
Becomes However, in the embodiment in which the conditions are relaxed, the preceding and following characters are not referred to. Even in such a case, the similarity regarding the candidate character relating to the recognition target is conditioned on the condition that the situation regarding the candidate character relating to the recognition target this time is exactly the same as the information in the character information table regarding the misread character in the learning dictionary 4. Appropriate correction will be made by tightening the conditions.

Here, the file structure of the learning dictionary 4 is changed to the sixth
Shown in the figure and described. The files of learning dictionary 4 are
It is composed of a head portion, a pointer portion, and a data portion. In the header part, type of character code (alphanumeric, kanji,
(Kana, symbol, etc.), the same number of records + 1 record is assigned, and the record number corresponds to the character code. For example, the character code of "D" in the JIS code is 2234, but the 2234th record is the information position of "D" in the header part. Further, the final record in the header section records the final registration position information (the last record number registered in the learning file), and this record is referred to at the time of registration and updated after registration. In the record, reference positions of the pointer portion and the data portion are recorded as the learning information of the character code, and zero (0) is recorded when the learning information targeted by the character code does not exist. However, by recording zero (0) in all records in the initial state, updating when learning information does not exist becomes unnecessary.

In the pointer portion, when the next candidate exists for the learning information of the character code, the reference position of the pointer portion of the next candidate is recorded. Zero (0) is recorded if there is no next episode. Similar to the above-mentioned header part, by recording zero (0) in all the records in the initial state, the update when the next candidate does not exist becomes unnecessary. Although the number of records in the pointer portion is arbitrary, the learning information can be recorded by the set number of records.

In the data section, learning information (preceding character, rear character, misread character similarity,
Regular characters, regular character ranking, regular character similarity) are recorded. The number of records is the same as the number of records in the pointer section.

Next, the operation when referring to the file of the learning dictionary 4 will be described with reference to FIG. Transition from the beginning of the file to the position of the character code record (Fig. 7) of the character "D". Explaining in the example (ex.) Of Fig. 7, the transition destination is the 2234th record. ) Data (0002 in FIG. 7) is read in. This data is the position information of the pointer part and the data part, and "0" is recorded only when 0 is recorded.

Here, since 0002 is read,
Next, it transits to the position of the O002 record of the data part (FIG. 7). Read the learning information (previous character, rear character, misread character similarity, authentic character, authentic character rank, authentic character similarity) of this data part, and compare it with the content of the recognition result candidate character table, etc. For example, the search ends here, and if they do not match, the process moves to the 0002th record (FIG. 7) of the pointer part.

Then, the data of the 0002th record (FIG. 7) of the pointer portion is read. Since the data here is 0014, next, the learning information of the 0014th record (FIG. 7) of the data part is read out and compared with the contents of the recognition result candidate character table and the like. If the result of the comparison is a match, the search is ended here, and if they are not a match, a transition is made to the 0014th record (FIG. 7) of the pointer part.

The position information of the 0014th record (FIG. 7) of the pointer part is OO1F, and next, the transition is made to the 001Fth record of the data part (FIG. 7) to compare the learning information. here,
Further, if they do not match, the data is read out by transiting to the 001Fth record of the pointer part (FIG. 7). In this example, the data of the 001Fth record of the pointer part (FIG. 7) is zero (0 ), The next candidate does not exist, and the reference process for the character ends. If the next candidate does not exist, the character replacement (correction process) is not performed, and the character recognition result is directly output to the correction process.

FIG. 5 shows details of the learning process during the process of the character recognition apparatus, which will be described. It is assumed that the image data 23 obtained by the character image data acquisition unit 1 is subjected to recognition processing including correction processing using the fixed dictionary 3 and the learning dictionary 4 to obtain character data “ADC”. Here, the printed image is "AB
"C", which means that "B" was misread as "D".

In the correction process S4, the operator corrects "D" to "B" by making a correction input from the input section. Next, the control unit 5 automatically advances to the learning process S5 by inputting a predetermined instruction or when the correction of one page of the OCR document is completed, and the learning dictionary 4 is selected from the character data and the candidate character table 25.
The learning information related to the contents of the character information table 24 to be registered is acquired.

That is, the misread character "D" and the correct character "B" are obtained based on the history information in which the misread character "D" is replaced with the correct character "B". Further, the character “A” and the character “C” after the correct character “B” are obtained from the character data after the correction processing. Then, the similarity “95” of the misread character “D” is detected from the candidate character table 25, the similarity “85” and the rank “3” of the regular character “B” are obtained, and the content of the illustrated character information table 24 is obtained. Acquire learning information related to. The similarity of the regular character “B”, that is, the similarity of the corrected input character (regular character) in the recognition result of the image data is larger than zero. The learning information is registered in the learning dictionary 4.

In the above, in the recognition result performed on the image data, registration is made in the learning dictionary 4 on the condition that the similarity of the corrected character (regular character) is larger than zero. The degree is a predetermined value (for example, "4
0 ”) or more, the learning dictionary 4 may be registered. By doing so, the degree to which the learning dictionary 4 is not updated can be increased if it is not so similar. The contents of the character information table 24 are as follows.
Not limited to the above contents, there is an embodiment in which the preceding and following characters are not registered and an embodiment in which the regular character rank is not registered. There is also an embodiment in which the degree of similarity, the rank, etc. of the preceding and following characters are registered. Even in this case, appropriate character recognition processing is performed using the learning dictionary 4. In addition, in the embodiment in which the similarity and rank of the preceding and following characters are registered, it is a condition for correction that they match or the similarity falls within a predetermined range.

Next, the processing operation when registering in the file of the learning dictionary 4 will be described with reference to FIG. From the beginning of the file, the character code record of the character "D" (Fig. 8
). Explaining with the example (ex.) Of FIG. 8, the transition destination is the 2234th record. Transition destination (Fig. 8)
Data 0002 is read. The data here is the position information of the pointer portion and the data portion, and becomes "no candidate" only when 0 is recorded.

In this example, since the data is 0002, next, the transition is made to the 0002th record (FIG. 8) of the pointer portion, and the data 0014 relating to the stored contents is read. Since the data 0014 is the position information in the pointer portion and the data portion of the next candidate, next, transition is made to the 0014th record (FIG. 8) of the pointer portion, and the data OO1F stored therein is read. Similarly, the transition is made to the 001Fth record (FIG. 8) of the pointer portion, and the data zero (0) stored therein is read. Zero (0) means that there is no next target catch (final candidate), so the record next to this will be the position registration of the next candidate. This record position 001F is stored in the main storage area of the control unit 5.

Next, the final record of the header part (FIG. 8)
After making a transition, the data (final registration position information) 0109 is read. Since the final registration position information is the final record number in which the data in the pointer part and the data part is registered, a value obtained by adding "1" to this value becomes the next registration position.
However, if the value obtained by adding "1" to the final registration position information exceeds the final record of the file, the registration is interrupted and the registration is not performed. If it does not exceed the final record, the final registration position information of the final record (FIG. 8) in the header part
O1OA obtained by adding 1 to will be the final registration position information after registration.

Next, the 01OA record of the data section (see FIG. 8)
), And the learning information (previous character, rear character, misread character similarity, regular character, regular character rank, regular character similarity) of the character information table 24 is written at this position. When the writing is completed, the process moves to the 001Fth record of the pointer portion (FIG. 8), and the same O1OA as the last registered position information previously obtained is written as the data of this position. Finally, transit to the final record (Fig. 8) in the header section, and enter O1OA as the final registration position information.
To complete the registration.

As a result of the file structure of the learning dictionary 4 configured as described above, a plurality of correct character candidates are sequentially registered for one misread character, and a plurality of registered misread characters are registered. Can be sequentially read out, and appropriate registration and character recognition using the learning dictionary 4 are performed in response to a plurality of cases where erroneous reading has occurred.

[0042]

As described above, according to the learning dictionary updating method in the character recognition device of the first aspect, learning is performed.
Whether to register in the dictionary is automatically determined according to the similarity of the characters related to the correction input in the recognition result,
Registration will be performed only when necessary, and the corrected characters and
The similarity, rank information in candidate characters, and misread characters
A learning dictionary in which the similarity information is registered as one set can be created.

As described above, according to the learning dictionary updating method in the character recognition device of the second aspect, one set
Including the characters before and after the one character to be recognized,
Since it is registered in the learning dictionary, include these characters
One set can be referenced, and misreading when the character string is predetermined
Can be reduced.

As described above, according to the learning dictionary updating method in the character recognition device of the third aspect, one set
Includes the characters before and after the one character that is the target of recognition processing, and the like.
This will be registered in the learning dictionary, including similarity information.
Refer to one set including the surrounding characters and their similarity information.
In this case, misreading when the character string is predetermined can be reduced.

As described above, according to the character recognition device of the fourth aspect, whether or not to register in the learning dictionary is determined.
It is automatically determined according to the similarity of the character related to the correction input in the recognition result, is registered only when necessary, and the corrected input character and its similarity and candidate character are included.
One set of ranking information, misread characters and their similarity information
It is possible to provide a character recognition device having a learning dictionary having the registered contents.

As described above, according to the character recognition device of the fifth aspect, in the learning dictionary, one of the objects to be recognized is recognized.
Register characters in the learning dictionary, including the characters before and after the character
You can refer to one set including these characters and
Misreading when the columns are predetermined can be reduced.

As described above, according to the character recognition device of the sixth aspect, sentences before and after one character to be recognized are recognized.
Since one set including the character and its similarity information is stored , these are referred to as a whole, and misreading when the character string is predetermined can be reduced.

As described above, according to the character recognition device of the seventh aspect, when the recognition result is corrected by referring to the learning dictionary, the recognition result character is registered in the learning dictionary. Since it is required that the information on the preceding and following characters match, misreading in character string units can be reduced.

As described above, according to the character recognition device of the eighth aspect, when the recognition result is corrected by referring to the learning dictionary, all characters registered in the learning dictionary with respect to the normal characters are registered. The condition is that the information in
Then, the correction will be made under strict conditions.

[Brief description of drawings]

FIG. 1 is a configuration diagram of a character recognition device according to an embodiment of the present invention.

FIG. 2 is a flowchart for explaining the operation of the character recognition device according to the embodiment of the present invention.

FIG. 3 is a diagram for explaining a learning dictionary updating operation in the character recognition device according to the embodiment of the present invention.

FIG. 4 is a diagram for explaining a character recognition operation using a learning dictionary in the character recognition device according to the embodiment of the present invention.

FIG. 5 is a diagram for explaining a learning dictionary updating operation in the character recognition device according to the embodiment of the present invention.

FIG. 6 is a diagram for explaining a file structure of a learning dictionary in the character recognition device according to the embodiment of the present invention.

FIG. 7 is a diagram for explaining a learning dictionary reference method during a character recognition operation in the character recognition device according to the embodiment of the present invention.

FIG. 8 is a diagram for explaining a learning dictionary updating method during a character recognition operation in the character recognition device according to the embodiment of the present invention.

[Explanation of symbols]

1 Character image data acquisition means 2 Recognition processing unit 3 Fixed dictionary 4 Learning dictionary 5 Control unit 6 Display unit 7 Input section 8 File section 11 reading unit 12 normalization unit 13 Detection cutout part

Continuation of the front page (56) References JP-A 2-171876 (JP, A) JP-A 61-70678 (JP, A) JP-A 63-150782 (JP, A) JP-A 63-27992 (JP , A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G06K 9/00-9/76

Claims (8)

(57) [Claims] 1. A fixed dictionary in which character image data is photoelectrically converted to obtain character image data, and information for performing character recognition processing on the character image data is fixedly stored, and a learning dictionary reflecting a correction input. Performs character recognition processing by referring to it , and if there is correction input made for the recognition result , correction input
Information of the characters and their similarity and rank information in the candidate characters
And the misread character and its similarity information as one set.
According to the similarity of the corrected and input characters in the knowledge
A method for updating a learning dictionary in a character recognition device, wherein the learning dictionary is registered or not registered in the learning dictionary. 2. One of the objects of recognition processing is included in the one set.
Including the front and rear of character, the method of updating the learning dictionary in the character recognition apparatus according to claim 1, wherein the this <br/> to be registered in the learning dictionary. 3. The one set includes characters before and after one character to be recognized and similarity information thereof.
A contract that is characterized in that it is registered in the learning dictionary including
A method for updating a learning dictionary in the character recognition device according to claim 1 . 4. A character image data acquisition means for photoelectrically converting a character image into character image data, a fixed dictionary in which information for performing character recognition processing on the character image data is fixedly stored, and correction input is made. Character and its similarity and
One set of ranking information, misread characters and their similarity information
A learning dictionary to be stored , a recognition processing unit that performs character recognition processing on the character image obtained by the character image acquisition unit by referring to the fixed dictionary and the learning dictionary, and a result of the character recognition processing by the recognition processing unit is displayed. A display unit for inputting information, an input unit for inputting information, and a correction input from the input unit for the recognition result displayed on the display unit.
Similarity, rank information in candidate characters, misread characters, and the like
With similarity information as one set,
A character recognition device, comprising: a control unit that registers or does not register in the learning dictionary according to the similarity of the corrected and input character. 5. The learning dictionary, including characters before and after one character to be recognized , includes the learning dictionary.
The character recognition device according to claim 4, wherein the character recognition device is registered in a learning dictionary . 6. The learning dictionary includes characters before and after a character to be recognized and similarity information thereof.
The contract is characterized in that one set including
The character recognition device according to claim 4 . 7. The recognition processing section, when referring to a learning dictionary and correcting the recognition result,
The character recognition device according to claim 6, characterized in that information on characters before and after the character registered as a recognition result character in the learning dictionary is matched. 8. The recognition processing unit, when referring to the learning dictionary and correcting the recognition result,
All registered in the learning dictionary for authentic characters
The character recognition device according to claim 6, wherein all the pieces of information match each other .