JPH0668261A - Pattern recognizing device - Google Patents

Abstract

PURPOSE:To provide a character recognizing device capable of confirming and correcting a recognition result in a short time, easily and with excellent accuracy. CONSTITUTION:A same character string extraction part 39 automatically extracts the character strings of recognition error candidates or the character strings of correction candidates for them which are composed of same or similar category in an inspection information storage part 32 by regarding them as the error of the character in the same category. A correction processing part 36 displays the character string groups of plural same or similar categories which are obtained by the same character string extraction part 39 with input character image on a display part 35 and enables an operator to select plural character strings, convert plural portions into correct characters at one time and input a new correct character.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention automatically recognizes pattern information such as characters, figures and voices, and recognizes information necessary for verifying and confirming and correcting the recognition result, that is, input obtained from a recognition processing system. A recognition information storage unit that stores a character image or voice, a character feature or a voice feature, a recognition result, a candidate character or candidate code that is candidate information included in the recognition result, a distance value, or similarity, and recognition information. Linguistic processing and verification using candidate information are performed, and the recognition error candidate character strings or code strings and the correction candidate character strings or code strings for those character strings or code strings are extracted and stored in the verification information storage unit. The present invention relates to a pattern recognition device that has a verification processing unit and the like and that converts character pattern information or code string information into code data that can be processed by a computer.

[0002]

2. Description of the Related Art FIG. 11 is a block diagram of a conventional example of a pattern (character) recognition device of this type. The image input unit 1 inputs a character to be recognized and converts it into an electric signal. The recognition unit 2 inputs characters converted into electric signals and performs preprocessing for cutting out individual character patterns, feature extraction for expressing parameters used for recognition with a small amount of information, and determines to which category the input characters belong. A recognition process is performed. Correction part 3
A performs post-processing for verifying an ambiguous recognition result using verification processing of recognition result using recognition information, language processing (word knowledge or language rule), and the result is confirmed and corrected by an operator. The output unit 4 outputs the corrected correct character code information to the outside. The control unit 5 has these units 1 to 4
To control.

FIG. 12 is a block diagram of the correction unit 3A in FIG. The recognition information storage unit 31 in the verification processing unit 30 temporarily holds the recognition information sent from the recognition unit 2. The language processing unit 33 performs language processing on the recognition information for verification, and identifies the position of the unnatural character string in the recognition information storage unit 31.
In addition, the unnatural character string and the correction candidate character string are detected and recorded in the verification information storage unit 32. Display processing unit 34
Displays on the display unit 35 the information necessary for confirmation and correction and the result of language processing, and accepts confirmation and correction by the operator.
At this time, the display processing unit 34 displays the contents of the recognition information storage unit 31 and the contents of the verification information storage unit 32 by adjusting the position, size, color, etc. so that the operator can easily perform confirmation and correction processing. It is displayed on the section 35. The operator visually confirms the displayed recognition information and verification information, and if there is a recognition error therein, selects correct information from the candidates using the correction input section 37 or inputs new information. Or complete the data without error. Correction processing unit 36A
Rewrites the contents of the recognition information storage unit 31 with correct code data according to the instruction from the correction input unit 37. The information thus obtained is transferred to the corrected information storage unit 38 as error-free data.

[0004]

In the above-mentioned conventional character recognition device, the recognition information and verification information sent from the recognition unit 2 are displayed on the display unit 35 or output as voice, and the operator visually or hears the information. Had been inspected by. When a recognition error was detected, the correct character was input each time using the keyboard or mouse to create correct data. Therefore, in the recognition correction of the recognition result, the recognition error is detected by the visual check, the cursor is moved to the position of the recognition error, the processing mode is changed from the recognition processing mode to the character input mode, the correct character is input, the correct answer information and the recognition error. Operations such as replacing

Therefore, (a) the operation of the operator is complicated. (B) The data input speed becomes slow due to the large number of correction operations, and (b) it takes time to move the cursor for instructing a recognition error (using the verification information up to an unnatural character string position). There is a problem such as (can be skipped). The voice recognition device had the same problem.

An object of the present invention is to provide a pattern recognition device which can easily and accurately confirm and correct a recognition result in a short time.

[0007]

The pattern recognition device of the present invention has the same character string or code string of recognition error candidates in the verification information storage section, or the same character string or code string of these correction candidates, or a similar string. The same character string or code string extraction unit that automatically considers a character or code in the same category as an incorrect character or code, and a plurality of the same character string or code string extraction unit Character strings or code strings of the same or similar category are displayed on the display together with the input character image or input voice, the operator selects multiple character strings or code strings, and correct characters or correct codes at multiple locations at once. And a correction processing unit that allows input of a new correct character or correct answer code.

[0008]

The present invention detects unnatural recognition results among uncorrected recognition results, automatically extracts recognition information and verification information showing the same tendency, and collects them in the same category. It is designed to allow batch correction operations. As a result, the time required for the operator to make corrections can be reduced, the productivity of data entry can be improved, and the correction operation can be facilitated.

A history information storage unit for accumulating correct characters or correct codes selected or input by an operator, recognition error candidate character strings or code strings, correction candidate character strings or code strings, and recognition information. A correction information storage unit is provided, and the correction unit collects duplicate information in the history information of the history information storage unit to remove other information, and selects information that seems to be the answer to the same category. In order to put together, correction information is calculated and stored in the correction information storage unit, and from the verification information and the recognition information, the same character string or code string of the recognition error candidate as the correction information in the correction information storage unit, or the same correction candidate. Detects the character string or code string of, and the character or code strings of the same character string or code string are the same or similar categories, or the characters or voices of the same category are incorrect By automatically extracting regarded, efficiently using the history information, yet can detect leakage less bulk modify.

The correction unit uses the correct character or the correct code selected or input by the operator to select the same character string or code as a recognition error candidate from a plurality of character strings or code strings displayed on the display unit. A function to rewrite a string, or a character string or a code string of a correction candidate to the correct character or a correct code by the operator, and by repeating this process, for various categories of the character string or code string of the recognition error candidate By having the function of correcting the correct answer character or correct answer code, the operator does not need to input correct answer information, and the productivity of the operator's confirmation and correction work is improved.

The correction unit stores the same character string or code string of the recognition error candidate, or the character string or code string of the correction candidate and the character image or input voice, and the recognition information stored before and after the character information storage unit. By automatically extracting from the above and displaying them on the display unit at the same time, the operator can easily infer the character category of the same character string from the preceding and following characters and identify the same character or similar character. Can support work.

The correction unit determines, from the verification information and the recognition information, the probability of the same character string or code string as the recognition error candidate or the same correction candidate character string or code string as the correction information in the correction information storage unit. A function of detecting the same character string or the same code string that is calculated as the reliability, and the character string or the code string is the same, or the ones made up of similar categories are regarded as the wrong characters or codes in the same category, By having the function of calculating the certainty as reliability and automatically extracting from the order of high reliability, it is possible to efficiently correct the recognition error included in the recognition result.

[0013]

Embodiments of the present invention will now be described with reference to the drawings.

Embodiment 1 FIG. 1 is a basic configuration diagram of a character recognition apparatus according to an embodiment of the present invention. In this embodiment, only the correction unit 3 is different from the conventional character recognition device shown in FIG.

FIG. 2 is a block diagram of the correction unit 3. Correction part 3
Only the modification processing unit 36 is different from the modification unit 3A in FIG. The correction processing unit 36 includes the same character string storage unit 40, and the character strings of recognition error candidates in the verification information storage unit 32 of the verification processing unit 30 or the character strings of these correction candidates are the same or similar categories. The same character string extraction unit 39 that automatically extracts what is formed as an incorrect character in the same category and stores it in the same character string storage unit 40, and a plurality of characters obtained by the same character string extraction unit 39. Of the same
Alternatively, a group of character strings in a similar category is displayed on the display unit 35 through the display processing unit 34 together with the input character image, and the operator selects a plurality of character strings and converts them into correct characters at a plurality of places at once, or creates a new character. The correction processing unit 36 is provided so that a correct answer character can be input.

The operation of the correction unit 3 will be described. First, after the recognition information sent from the recognition unit 2 is verified by the verification processing unit 30, the recognition information is temporarily held in the recognition information storage unit 31. The recognition information in the recognition information storage unit 31 is displayed on the display unit 35 either directly according to an operator's instruction input from the correction input unit 37 or processed by the same character string extraction unit 39. The operator visually confirms the information displayed on the display unit 35, and if there is an erroneously recognized character in the information, the correct input character 37 is used to select the correct character from the candidate characters, or the character input function is selected. Use to enter a new correct character. In the batch correction process, at this time, the recognition information is preliminarily subjected to language processing (for example, to detect a misrecognized character, a process that can be performed by Japanese Patent Laid-Open No. 1-41061: “Japanese sentence erroneous character automatic correction device”) and reliability. For example, Japanese Unexamined Patent Publication No. 62-182982: "Detection of correction candidates based on the degree of character concatenation strength such as" Japanese sentence erroneous character automatic detection method ") is performed to recognize character strings of recognition error candidates and those. The correction candidate character string is detected and recorded in the verification information storage unit 32. Next, the same character string extraction unit 39 separately detects a character string of a recognition error candidate indicating the same category and a character of a correction candidate indicating the same category, and stores them in the same character string storage unit 40. Waits for operator's instruction. The collective correction process is performed by the operator in accordance with a display input from the correction input unit 37, and the recognition error candidate character strings and the correction candidate character strings, which are grouped in the same character string storage unit 40 for each same category, and The character images of the recognition information storage unit 31 corresponding to those character strings are displayed on the display unit 35 based on a request from the display processing unit 34. The operator visually confirms the character string group of the recognition error candidates displayed on the display unit 35, collects the character images belonging to the same category as one group using the keyboard or the mouse, and then selects or inputs the correct answer character. To do. By performing this process until there is no character string of the recognition error candidate in the same character string storage unit 40 (verification information storage unit 32), the recognition error character of the recognition result in the recognition information storage unit 31 can be eliminated. it can. Further, at this time, information (selection information) indicating that the confirmation correction has been performed is recorded in the confirmation information. By adding this selection information, the post-processing of the recognition result can be made more reliable.

FIG. 3 shows the specific contents of the recognition information in the recognition information storage unit 31. The recognition information includes the input character pattern 311, character features 312, and recognition result 313.
A candidate character 314, a distance value or similarity 315, position information 316, and selection information 317 (until the selection information is recorded, it is used as information indicating that the character is a recognition error candidate). The selection information 317 is written in the recognition information when the operator visually judges from the recognition information displayed on the display unit 35 that the same character category is selected, and is, for example, a group of the same character category. , I, i can be represented by a number string. The recognition information storage unit 31 rewrites the content of the recognition result 313 with the correct character according to the instruction from the correction processing unit 36. The rewritten recognition result is replaced with the recognition result before rewriting and displayed on the display unit 35. At this time, the correction information corrected by the operator and the recognition information to be corrected are accumulated as history information in the history information storage unit 41 and are used for subsequent correction of the character string of the recognition error candidate. .

FIG. 4 shows the specific contents of the verification information in the verification information storage unit 32. The verification information includes the character string 321 of the recognition error candidate and the character string 321 of the recognition error candidate.
The storage address 322 of the recognition information in which is present, the character string 323 of the correction candidate, and the reliability 324. Here, the correction candidate character string 323 and the reliability 324 are stored as a pair.

FIG. 5 is a flow chart showing the flow of processing performed by the same character string extraction unit 39 and the operator. The portion surrounded by a solid line shows the flow of processing of the same character string extraction unit 39, and the portion surrounded by a dotted line shows processing by the operator. First, the same character string extraction unit 39 automatically extracts a character string of the same category from the character string of the recognition error candidate transferred from the verification information storage unit 32 and the character string of those correction candidates (step 391). ). The extraction criteria are
For example, there are (1) a method in which recognition error characters detected by verification processing are in the same category, and (2) a method in which correction characters for correcting recognition error characters are in the same category. . Here, the former will be described. Next, the same character string extracting unit 39 selects the input character patterns 311 and the recognition error candidate character strings 321 required for the operator's confirmation and correction, that is, the recognition error characters, from the extracted recognition information, in which the recognition error characters are in the same category. It is detected as an error candidate and displayed on the display unit 35 together with the recognition result 313, and the operator's confirmation and collective correction are accepted (step 392). The same character string extraction unit 39 collects the character categories and displays them on the display unit 35 so that the character categories of the correction candidate character string 324 displayed on the display unit 35 are not displayed in duplicate. The operator visually confirms the input character pattern 311 displayed on the display unit 35, and corrects the input character pattern 311 of the same character category by the correction input unit 37.
To be completed as one group (step 393). At this time, the correction processing unit 36 allocates the selection information 317 to the input character pattern 311 selected by the operator according to the instruction of the correction input unit 37.
The operator corrects the correct characters corresponding to the input character pattern 311 collected as one group by the correction input unit 3
7 is used to select from a correction candidate character string or to input a new character (step 394). The correction processing unit 36 takes the correspondence between the instruction of the correction input unit 37 and the selection information 317 and rewrites the content of the recognition result 313 with the correct character.

FIG. 6 shows a display example of the display unit 35 when the operator collectively recognizes and corrects the same character category in the present embodiment, and 351 shows an input character pattern display,
Reference numeral 352 shows a character display of the recognition result, and reference numeral 353 shows a display of a correction candidate character string. Of the recognition information (input character pattern) extracted from the recognition information storage unit 31 by the same character string extraction unit 39, the information necessary for the recognition correction is displayed on the display unit 35.
To display. The operator displays the input character pattern display 35.
1 is confirmed, and the input character pattern group display 354 of the same character category is selected using the correction input unit 37. In order to distinguish the display 354 of the selected input character pattern group and the display 355 of the non-selected input character pattern group, the display unit 35 displays the display unit 35 in different colors and sizes. When the operator completes the selection as one group, he / she selects the correct character corresponding to the character category of this group from the character display 352 of the recognition result or the character string display 353 of the correction candidate, and has a character input function. A new correct character is input using the correction input unit 37 using. The correction processing unit 36 rewrites the content of the recognition result 313 with the correct character according to the instruction of the correction input unit 37. In this way, the correction input unit 37 repeatedly receives operator confirmation and collective correction. In addition, in the case of collectively recognizing and correcting each of the same character categories shown in the present embodiment, the input character pattern display 351 other than the corresponding same character string is displayed on the display unit 35, but the displayed input character pattern display 351 is displayed. For the input character pattern display 351 which is determined by the operator to be corrected, the same character category is collected and the correct character can be repeatedly input.

In the present embodiment, unnatural recognition results are detected among uncorrected recognition results, and recognition information and verification information showing the same tendency are automatically extracted from the uncorrected recognition results, which are summarized for each same category. In this way, the collective correction operation can be performed, which can reduce the time required for the operator to correct the data, improve the productivity of data input, and facilitate the correction operation.

Embodiment 2 FIG. 7 is a diagram showing a batch correction process using the history information 410 in the correction unit 3, where 410 is history information and 411.
Is a correct character input by the operator, 412 is history information (language history information) in which the language processing has worked effectively, 413 is history information in which the operator's correction is valid (input history information), and 420 in FIG. It is a figure which shows the content of the correction information which removed the duplicated information from the information 410. Specifically, in the correction unit 3, when the operator confirms the recognition result in the recognition information storage unit 31 and corrects it, the correct character is selected or input, and at the same time, information related to the correction,
That is, the correct character 411, the character of the recognition error candidate, the character string of the correction candidate, the reliability, and the recognition information (the input character pattern, the characteristic of the character, the candidate information of the recognition result, and the like) are used as the history information 410.
Is recorded in the history information storage unit 41. Next, the correction processing unit 36 creates the correction information 420 at the time when a certain amount of correction processing is performed, for example, when the work of one day or the reading of about 1,000 documents is completed, and the correction information 420 is generated as a result. Is recorded in the correction information storage unit 42. The creation of such correction information 420 is a process of collecting duplicate information from the history information 410 into one and compressing the information, and selecting and collecting information that may be the answer to the same category. It aims at shortening the processing time, reducing the capacity of history information, and improving the reliability of history information. In the same character string extraction unit 39, the information in the verification information storage unit 32 and the recognition information in the recognition information storage unit 31 match the recognition error character in the correction information 420, and the concatenation relationship between the characters also matches. Even if the recognition error character is different, the character string of the correction candidate is automatically extracted and the correct character in the correction information 420 is added,
It is displayed on the display unit 35 together with the character image. By this processing, the operator visually confirms the character pattern 311 displayed on the display unit 35, collects the input character patterns 311 belonging to the same character category as one group using a keyboard or a mouse, and then Select or enter.

In the processing of the present embodiment, if there is the same error tendency as in the past, the character pattern corresponding to the character is detected in the verification information and the recognition information and displayed on the display unit 35 for confirmation. It is an object to be corrected or is put in a correction candidate character string and presented to the operator from the display unit 35, which works effectively for detection of misrecognized characters and is inputted by the operator for correction. The number of times the correct answer is input can be reduced.

In this embodiment, the recognition information and the verification information showing the same tendency as the history processing information corrected by the operator are automatically extracted from the uncorrected recognition information, and the correction information is collectively collected for each same category. With this, the history information corrected by the operator can be used to efficiently make a batch correction with less omission of detection.

Embodiment 3 In FIG. 9, the correction processing unit 36 uses the correct character information selected or input by the operator as the history information 410 recorded in the history information storage unit 41 to facilitate the subsequent correction processing. 7 is a flowchart of an embodiment that can be performed. This is because the information on the correct character selected or input by the operator is held in the history information storage unit 41, and the same recognition error candidate is selected from among the various character strings displayed on the display unit 35. By selecting a character string by the operator, the erroneously recognized character in the recognition information storage unit 31 is rewritten to the correct character of the history information 410 input immediately before. Further, by repeating this process, it is possible to rewrite various categories of character strings of recognition error candidates into correct characters. First, a “recognition result correction mode” indicating whether the correct category is selected / input before or after selecting the same category is given as a default value of the system that can be changed by the operator. In this embodiment, in the case of inputting the correct character before selection, the process proceeds in the following steps. In this figure, the solid line part is automatic processing,
The broken line shows the processing by the operator. First, the recognition result correction mode is input (step 501). The operator inputs the correct character to the character string of the recognition error candidate displayed on the display screen (step 501). Next, a plurality of recognition error candidate character string images in the same category as the correct character are selected (step 503). Next, at the time when the selection is completed, that is, when the next correct answer character is input to perform correction work for another category on the same screen display, at the time when the correction work of the misrecognized character in one correction information is finished. The selected recognition error candidate character string (recognition result) is rewritten to the correct answer character (steps 504 and 50).
5). Next, it is determined whether or not the correction work for another category is performed on the same screen display (step 506). Finally, it is determined whether there are no recognition error candidate characters in the same character string storage unit 40 collected using one piece of verification information or correction information (step 507).

In the present embodiment, the correct character input by the operator is used to rewrite the same character string of the recognition error candidate or the character string of the correction candidate to the correct character. Since it is not necessary to input information, the productivity of operator confirmation and correction work is improved.

Fourth Embodiment FIG. 10 is a diagram showing a display screen of the display unit 35 for explaining the fourth embodiment. In this case, the correction unit 3 has the same configuration as that of FIG. When the same character string extraction unit 39 displays the content of the same character string storage unit 40 on the display unit 35 according to the instruction of the correction processing unit 36, the same character string extraction unit 39 also recognizes the recognition information (character pattern) stored before and after the recognition error character. Information storage unit 31
Are automatically extracted and displayed on the display unit 35 at the same time. However, if there is no recognition information before or after the similar recognition information, the recognition information to be corrected, that is, the digit of the recognition information is aligned so that it is positioned at the center character of the input character pattern display 351, and the display unit 35 is displayed. To display. Further, when the display unit 35 cannot display all the recognition information of the same category on one screen, the screen switching mode is displayed on the display unit 35, and the operator performs the recognition and collective correction processing while switching the screens. . The operator confirms the character type of the recognition information of the correction target displayed on the display unit 35 with reference to the recognition information displayed before and after the correction information, and uses the correction input unit 37 to combine the same character categories into one group to perform a correction process. I do.

In the present embodiment, the extracted same character string and the characters before and after the same character string are simultaneously displayed on the display unit 35, so that the operator can change the character category of the same character string from the preceding and succeeding characters to the isomorphic characters and similar characters. The identification can be easily guessed, and the operator's confirmation work can be supported.

Fifth Embodiment In the fifth embodiment, the correction unit 3 calculates the certainty of the verification process result at the time of the verification process as the reliability 323, and records the result in the verification information storage unit 32. The probability of the same character string extraction result when extracting the same character string using the correction information is calculated as the reliability 323, and the subsequent same character string extraction unit 39 uses the same character string or a similar category. Is regarded as an incorrect character in the same category, its certainty is calculated as the reliability 323, and the same character string extraction unit 3
In No. 9, the information is automatically extracted from the highest reliability, and the same category can be collectively displayed by using this information.

The reliability 323 is determined by Japanese Patent Laid-Open No. 62-18298.
No. 2: As shown in "Japanese typographical error automatic detection method" and the like, there are cases where the one required as the degree of connection strength of the concatenated character strings is used and where correction information is used. In the case of the former, the same character string extraction unit 39 selects from the same character string storage unit 40 in accordance with the instruction of the correction processing unit 36, the one in the same category and the one having a high degree of connection, and displays it on the display unit 35. When the number of times of use 421 in the correction information is used as the reliability 323, when the correction information is valid by displaying the correct character as described (correct use 422)
When the other characters do not work effectively and are selected or input (negative use 423), the comparison value is used, and the one with the highest comparison value is selected and displayed on the display unit 35.

In the present embodiment, a character string of a recognition error candidate or a character string of a correction candidate showing the same category is extracted in descending order of reliability. As a result, the recognition included in the recognition result is performed. Errors can be corrected efficiently.

Embodiment 6 This embodiment shows a case where a recognition target pattern is voice. In the correction unit 3 in the case of voice recognition, the same voice extraction unit (corresponding to the same character string extraction unit 39) collects candidates of misrecognized voices that indicate the same category, and adds them to the voice (category name) for the label of the voice. Subsequently, the collected original voices are output in the order of high reliability. At this time, if the original voice is in the same category as the voice of the label, an instruction indicating the same category is given using a terminal such as a tablet or a mouse. By converting the erroneously recognized voice candidate corresponding to the designated original voice to the code information indicated by the category name of the voice output as a label, the erroneously recognized voice is corrected to the correct voice, and the voice recognition confirmation and correction are performed. Can be done efficiently.

In the present embodiment, the case of voice recognition is shown, in which candidates of erroneously recognized voices are collectively output as voices, only the same voices are selectively extracted, and then correction instructions are collectively issued to make a correct answer. By converting it to a code, you can check and correct it efficiently.

[0034]

As described above, the present invention has the following effects. (1) The invention of claim 1 detects an unnatural recognition result among uncorrected recognition results, and automatically extracts recognition information and verification information showing the same tendency, and extracts the recognition information and the verification information from each unrecognized recognition result for each same category. By collectively performing the batch correction operation, the time required for the correction by the operator can be reduced, the productivity of data input can be improved, and the correction operation can be facilitated. (2) The invention of claim 2 automatically extracts the recognition information and the verification information showing the same tendency as the history information corrected by the operator from the uncorrected recognition results, and collectively collects them in the same category. By making the correction operation possible, it is possible to make a batch correction efficiently using history information and with little detection error. (3) The invention according to claim 3 uses the correct character or the correct code selected or input by the operator to convert the same recognition error candidate character string or code string or correction candidate character string or code string to the correct character. Alternatively, by rewriting the correct answer code, the operator does not need to input the correct answer information, and the productivity of the operator's confirmation and correction work is improved. (4) According to the invention of claim 4, by displaying the extracted same character string or the same code string and the characters or codes before and after the same character string on the display unit at the same time, the operator can display the same character string or the character category of the same code string. It is possible to easily infer the homomorphic character or the homomorphic code or the similar character or the similar code from the preceding or following characters or codes, and it is possible to support the operator's confirmation work. (5) The invention according to claim 5 is to automatically extract the recognition error candidate character strings or code strings or the correction candidate character strings or code strings in the same category in descending order of reliability. As a result, it is possible to efficiently correct the recognition error included in the recognition result.

[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 configuration diagram of a correction unit 3 in FIG.

FIG. 3 is a diagram showing the contents of recognition information in a recognition information storage unit 31 in FIG.

4 is a diagram showing the contents of verification information in a verification information storage unit 32 in FIG.

FIG. 5 is a flowchart showing processing performed by the same character string extraction unit 39 and an operator in FIG.

FIG. 6 is a diagram showing a display of a display unit 35 when an operator collectively modifies the same character category.

7 is a diagram showing a display of a recognition result on the display unit 35. FIG.

8 is a diagram showing correction information 420 obtained by removing redundant information from the history information 410 of FIG.

FIG. 9 is a flow chart of a third embodiment in which information on correct characters selected or input by an operator is used to facilitate subsequent correction processing.

FIG. 10 is a diagram for explaining the contents of the display unit 35 in the fourth embodiment.

FIG. 11 is a configuration diagram of a conventional example of a character recognition device.

12 is a configuration diagram of a correction unit 3A of the character recognition device of FIG.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 Input section 2 Recognition section 3, 3A Correction section 4 Output section 5 Control section 30 Verification processing section 31 Recognition information storage section 32 Verification information storage section 33 Language processing section 34 Display processing section 35 Display section 36, 36A Correction processing section 37 Correction Input unit 38 Corrected information storage unit 39 Same character string extraction unit 40 Same character string storage unit 41 History information storage unit 42 Correction information storage unit 311 Input character pattern 312 Character characteristics 313 Recognition result 314 Candidate character 315 Distance value 316 Position information 317 Selection information 321 Character string of recognition error candidate 322 Storage address of recognition information in which character string of recognition error candidate exists 323 Reliability 324 Character string of correction candidate 351 Input character pattern display 352 Character display of recognition result 353 Character of correction candidate Column display 354 Selected input character pattern group display 355 No selected input character pattern Turn group display 391 to 394 steps 410 History information 411 Correct characters 412 Language history information 413 Input history information 420 Correction information 421 Number of uses 422 Positive number of uses 423 Negative number of uses 501 to 507 steps

Claims (5)

[Claims] 1. Recognition information necessary for automatically recognizing pattern information such as characters, figures, voices, and verifying and confirming and correcting the recognition result, that is, an input character image or voice obtained from a recognition processing system, A recognition information storage unit that stores a character feature or a voice feature, a recognition result, a candidate character or a candidate code that is candidate information included in the recognition result, a distance value, or similarity, and a language for the recognition information. Verification that performs verification by processing and candidate information, extracts character strings or code strings of recognition error candidates, and character strings or code strings of correction candidates for those character strings or code strings and stores them in the verification information storage unit In a pattern recognition device having a processing unit and the like, including a correction unit for converting character pattern information or code string information into code data that can be processed by a computer, A character string or code string of a recognition error candidate in the verification information storage unit, or a character string or code string of those correction candidates that are the same or similar to each other has a wrong character or code in the same category. Input the same character string or code string extraction unit that is automatically regarded as one and a plurality of same or similar category character string groups or code string groups obtained by the same character string or code string extraction unit It is displayed on the display along with the character image or input voice, the operator selects multiple character strings or code strings, converts multiple places at once into correct characters or correct voices, or inputs a new correct character or correct code. A pattern recognition device, comprising: 2. A history information storage unit for accumulating correct characters or correct codes selected or input by an operator, character strings or code strings of recognition error candidates, character strings or code strings of correction candidates, and recognition information. , A correction information storage unit, wherein the correction unit collects duplicate information in the history information of the history information storage unit into one and removes other information, The correction information for selecting / combining is calculated and stored in the correction information storage unit, and a character string or a code string of the same recognition error candidate as the correction information in the correction information storage unit from the verification information and the recognition information, or The same correction candidate character string or code string is detected, and if the character string or code string has the same or similar category, the character or voice of the same category is detected. Pattern recognition apparatus according to claim 1, wherein automatically extracting it is assumed that Tsu. 3. The correction unit uses the correct character or the correct code selected or input by the operator to identify the same recognition error candidate from a plurality of character strings or code strings displayed on the display unit. A function to rewrite a character string or code string, or a correction candidate character string or code string to the correct character or correct code selected by the operator,
3. The pattern recognition device according to claim 2, which has a function of correcting various categories of character strings or code strings of recognition error candidates into correct characters or correct codes by repeating this process. 4. The correction unit stores the same character string or code string of a recognition error candidate, a character string or code string of a correction candidate, a character image or an input voice, and recognition information stored before and after the character information. 3. The pattern recognition device according to claim 1, wherein the pattern recognition device automatically extracts from the recognition information storage unit and simultaneously displays them on the display unit. 5. The correction unit includes a character string or a code string of a recognition error candidate that is the same as the correction information in the correction information storage unit from the verification information and the recognition information, or a character string or a code string of the same correction candidate. The same character string or the same code string detection function that calculates the probability of reliability as the reliability, and the character string or the code string that is the same or similar category is the same character or code is wrong 3. The pattern recognition apparatus according to claim 1, wherein the pattern recognition apparatus has a function of calculating the reliability as a reliability and automatically extracting the reliability from the highest reliability.