JPH09223186A - On-line character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the recognition rate of the on-line character recognizing device. SOLUTION: A character recognition result in each character block out of characters successively written and inputted to an input part 1 is obtained by a segmentation part 2, a preprocessing part 3 and a recognition part 4 and displayed on an output part 8. At the time of judging the existence of a misrecognized character recognition result due to missegmentation from the display, the number of correct strokes of a written character is inputted to a number-of-strokes setting part 5. A stroke code judging part 6 transfers the misrecognized character recognition result obtained from the recognition part 4 to a stroke pattern dictionary part 7 based upon the inputted number of strokes. The dictionary part 7 corrects the misrecognized character recognition result based upon the number of strokes outputted from the setting part 5. The judging part 6 receives the corrected character recognition result and outputs the received result to the output part 8. The output part 8 displays the corrected character recognition result.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an on-line character recognition device for recognizing handwritten characters in real time.

[0002]

2. Description of the Related Art Conventionally, techniques in such a field include:
For example, there is one described in the following literature. Reference 1: Japanese Patent Laid-Open No. 61-29982 Reference 2: Technical Report of IEICE, 84 [22] (1
984.5), Hiroshi Murase et al., "Recognition of online handwritten character strings by the candidate character lattice method" P.67 Some online character recognition devices cut out characters in free format and perform character recognition. The following methods are available as character cutting methods on the free format. The minimum and maximum values of the x coordinate and the minimum and maximum values of the y coordinate are obtained for each stroke (writing portion from pen-on to pen-off) input by handwriting, and a frame surrounded by them is created. When the frame is in contact with another frame or included in another frame, the entire frame is enclosed by one frame to form a block. Each block is cut out as one character, and each block is matched with the dictionary for character recognition. Also, when the distance between the blocks is short, a plurality of blocks are cut out as a character block representing one character. Character recognition is performed on this character block using a dictionary.

FIG. 2 is an explanatory diagram of a conventional character cutting process. For example, there is a letter "Oki", and when the strokes are enclosed, it is divided into a partial "Sanzui" block and a Tsukuri "Middle" block. The distance α between the blocks is obtained on the x coordinate as shown in FIG. When the distance value α between blocks is smaller than a predetermined value, the character block corresponding to “Oki” is cut out. Other than this, cutting out based on the size of the distance value from pen off to next pen on,
There is a processing method in which a projection pattern of a handwritten character is obtained and then cut out using a distance value for each pattern.

[0004]

However, the conventional online character recognition device has the following problems. Whichever character cutout method is used, the character block cut out as one character differs depending on the distance value α between blocks. For example, when the distance α between the blocks in FIG. 2 is larger than a predetermined value, the uneven “Sanzui” and the created “middle” blocks are cut out as character blocks representing different characters. Therefore, there is a problem that when a character block unit is matched with a dictionary, a recognition result different from what the user originally desires is obtained.

[0005]

In order to solve the above-mentioned problems, the first invention is an online character recognition device,
It has the following configuration. That is, the online character recognition device of the first invention classifies the writing data string into strokes of the writing character, and an input unit for converting the writing characters sequentially input into coordinates to extract the writing data string. A cutout unit that divides the handwriting data string into character blocks corresponding to one character based on the positional relationship between the character blocks, and a recognition unit that performs feature extraction of each of the character blocks and matching of the features, and recognizes the character in each character block Is equipped with.
Furthermore, in this online character recognition device, an output unit for displaying the character recognition result and the corrected character recognition result, and a correct stroke of the written character input when the character recognition result in the recognition unit is incorrect. A stroke number setting unit for setting the number, a dictionary in which a combination of misrecognition result characters caused by the character recognition error, a correction character combined by the combination, and a stroke number of the correction character are stored in association with each other. A stroke pattern dictionary unit that is built-in and that extracts a correction character corresponding to the set number of strokes and a given combination of misrecognition result characters from the dictionary and outputs this as a corrected character recognition result; The combination of the misrecognized result characters is detected on the basis of the number of strokes, and is given to the stroke pattern dictionary unit. A stroke code judging unit which receives the correction character recognition result given to the output section is provided.

A second aspect of the present invention is an online character recognition device having the following configuration. That is, a character type setting unit that presets a first flag that indicates the character type of a writing character, an input unit that performs coordinate conversion of sequentially input writing characters to extract a writing data string, and the writing data string as the writing character string. A stroke that divides the handwriting data string into character blocks corresponding to one character based on the positional relationship of the strokes, feature extraction of each of the character blocks, and matching of the features, and each character block. A recognition unit for recognizing characters in units is provided. Further, this online character recognition device searches a built-in dictionary, sets a second flag indicating a character type of the character recognition result, and sets an erroneous character recognition result of the character recognition result as the first flag. The character type pattern dictionary unit that corrects the character recognition result of the character type indicated by is compared with the second flag and the first flag to determine whether the character recognition result output by the recognition unit is correct or incorrect. When it is determined to be correct, this character recognition result is output, and when it is determined to be incorrect, the incorrect character recognition result is passed to the character type pattern dictionary, and the corrected character recognition result is received from the character type pattern dictionary. A character type code determining unit for outputting and an output unit for displaying the character recognition result output from the character type code determining unit or the corrected character recognition result are provided.

According to the first aspect of the present invention, since the online character recognition device is configured as described above, the input unit converts the sequentially input writing characters into a writing data string, and the cutout unit converts this writing data string. Is divided into character blocks. The recognition unit obtains character recognition for each character block. Here, the output unit once displays the character recognition result. From the display of the character recognition result, it is determined that erroneous recognition has occurred due to a cutting error. If the character recognition result is incorrect, the correct stroke number of the character to be recognized is input and set in the stroke setting unit. This stroke number is given to the stroke pattern dictionary unit and the stroke code determination unit. The stroke code determination unit detects a combination of erroneously recognized result characters from the given number of strokes and passes it to the stroke pattern dictionary unit. The stroke pattern dictionary unit corrects the combination of the misrecognized result characters and supplies the corrected combination to the stroke code determination unit. The corrected character recognition result is supplied from the stroke code determination unit to the output unit,
The output unit displays the corrected character recognition result.

According to the second aspect of the invention, the first flag indicating the character type of the written character is preset by the character type setting unit. The first flag is output to the character type code determination unit and the character type pattern dictionary unit. On the other hand, the handwritten characters sequentially input by the input unit are converted into a handwritten data string, and the cutout unit divides the handwritten data string into character blocks. The recognition unit obtains character recognition for each character block. The character recognition result is supplied to the character type code determination unit. The character recognition result is output from the character type code determination unit to the character type pattern dictionary unit, and the character type pattern dictionary unit sets a second flag indicating the character type of the character of the character recognition result. The character type code determination unit detects whether or not the character recognition result is incorrect by comparing the first and second flags. If the character recognition result obtained from the recognition unit is an incorrect character recognition result, it is passed to the character type pattern dictionary unit. The incorrect character recognition result is corrected by the character type pattern dictionary unit, and the corrected character recognition result is returned to the character type code determination unit. The character type code determination unit outputs the corrected character recognition result to the output unit, and the output unit displays the corrected character recognition result. Therefore, the above problem can be solved.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION First Embodiment FIG. 1 is a functional block diagram of an online character recognition device showing a first embodiment of the present invention. This online Bunyu recognition device has an input unit 1 for inputting written characters such as a tablet. The output side of the input unit 1 is connected to the cutout unit 2, and the output side of the cutout unit 2 is connected to the preprocessing unit 3. The recognition unit 4 is connected to the output side of the preprocessing 3. The processes in the cutout unit 2, the preprocessing unit 3, and the recognition unit 4 are realized by a CPU, a memory, and the like not shown. Furthermore, this online character recognition device is provided with a stroke number setting unit 5, a stroke code determination unit 6, and a stroke pattern dictionary unit 7. The output side of the recognition section 4 is connected to the stroke code determination section 6, and the output side of the stroke number setting section 5 is connected to the stroke code determination section 6 and the stroke pattern dictionary section 7. The stroke pattern dictionary unit 7 is configured to input / output to / from the stroke code determination unit 6. The output side of the stroke code determination unit 6 is connected to the output unit 8 formed of a display or the like.

The input unit 1 transforms coordinates of a handwriting input and sequentially extracts a handwriting data string represented by coordinate values.
The clipping unit 2 takes in the coordinate values of the writing points with respect to the writing data string at regular time intervals, classifies them into strokes, and then automatically writes the writing data string in character units based on the positional relationship between the strokes. Division, that is, division into character blocks. The pre-processing unit 3 performs noise removal processing or smoothing processing on the writing data string in character block units output by the cutout unit 2. The recognition unit 4 performs feature point extraction processing on the writing data string output by the preprocessing unit 3 and performs character recognition by a general pattern matching method. In the pattern matching method, the extracted feature point information is matched with the standard character pattern registered in advance by the same method. Thereby, character recognition is performed. The recognition unit 4 has a function of outputting a shift JIS code of a character code which is the result of character recognition. The stroke setting unit 5 is used when the character recognition result by the recognition unit 4 is incorrect, and sets the correct number of strokes of the character input by the operator. When the stroke code determination unit 6 takes in the character code of the recognition result output by the recognition unit 4 one by one, displays the corresponding character on the output unit 8, and the user determines that the display is erroneous recognition due to a cutting error. The number of strokes obtained from the stroke setting unit 5 is fetched. If it is determined that the recognition is not incorrect, the recognition unit 4
The recognition result output by is displayed again on the output unit 8 as the final character recognition result. When the stroke number set for erroneous recognition is fetched, the stroke code determination unit 6 inputs the shift JIS code of the corrected character from the stroke pattern dictionary unit 7 described below and outputs the character. 8 has the function of displaying. The stroke pattern dictionary unit 7 includes a stroke pattern dictionary.

FIG. 3 is a diagram showing a stroke pattern dictionary in the stroke pattern dictionary unit shown in FIG. In the stroke pattern dictionary, a combination of character codes resulting from erroneous recognition and shift JI of correction characters synthesized by the combination are used.
It has S-codes corresponding to the number of strokes.
The stroke pattern dictionary unit 7 searches the stroke pattern dictionary from the plurality of character codes given from the stroke code determination unit 6 and the set number of strokes, and obtains the shift JIS code of the correction character for the misrecognition result. have. FIG. 4 is a flowchart showing a processing procedure of the online character recognition device of FIG.
The operation of FIG. 1 will be described with reference to FIG. In step S1, the writer inputs characters into the input unit 1 in stroke units. The input unit 1 converts the handwriting input into coordinates and sequentially outputs the handwriting data string {(x _i , y _i ), i = 1, 2, ..., Nj} j to the cutout unit 2. In step S2, the cutout unit 2 causes the writing data string {(x _i , y _i ), i = 1,
From 2, 2, ..., Nj} j, the coordinate values of the writing points at a fixed time interval are fetched and classified for each stroke. Further, the cutout unit 2 automatically divides the writing data string in character units based on the positional relationship between strokes. That is, the cutout process is performed. In step S3, the preprocessing unit 3 performs preprocessing on the writing data string in character units sequentially output from the clipping unit to remove noise or smooth the writing data string. The character-based writing data string output by the preprocessing unit 3 is sequentially transferred to the recognition unit 4. In step S4, the recognition unit 4 extracts feature points from the input writing data string in character units, and performs character recognition by matching using the extracted feature points. The shift JIS code of the character code, which is the result of character recognition, is output to the stroke code determination unit 6.

In step S5, the stroke code determination unit 6 displays the character corresponding to the character code obtained from the recognition unit 4 on the output unit 8 as the first recognition result. In step S6, the writer determines from the display on the output unit 8 whether the character recognition result by the recognition unit 4 is correct or erroneous. In the case of false recognition (N), in step S7,
The writer inputs the correct stroke number of the erroneously recognized character and sets the correct stroke number in the stroke number setting unit 5. Here, if the stroke number is not set, the process returns to step S6. The set number of strokes is output to the stroke code determination unit 6 and the stroke pattern dictionary unit 7. In step S8, the stroke code determination unit 6 outputs to the stroke pattern dictionary unit 7 a character code which is an incorrect character recognition result corresponding to the set number of strokes. The stroke pattern dictionary unit 7 extracts and loads a portion corresponding to the set stroke number in the stroke pattern dictionary of FIG. In step S9, the stroke pattern dictionary unit 7 matches the extracted stroke pattern dictionary portion with the character code given from the stroke code determination unit 6, and obtains the character code of the correction character for erroneous recognition.

FIG. 5 is a diagram showing an example of erroneously recognized characters.
For example, even if a character of “state” is input to the input unit 1, the recognition unit 4 causes “ri”, “ri”, “ri” due to misrecognition.
Character codes of "82e2", "82e2", "82e"
2 "is output. The writer sets the correct stroke number of the character" state "to 6, and the stroke code determination unit 6
Is the character code “82e2” for which the number of strokes is 2,
“82e2” and “82e2” are passed to the stroke pattern dictionary unit 7. The stroke pattern dictionary unit 7 has 6 strokes and the passed character code group is “82e2”.
Since they are "82e2" and "82e2", the correction character corresponding to them is selected. As a result, the character code “8f42” of the character “state” is selected. The character code “8f42” of “state” is output to the stroke code determination unit 6 as the corrected character recognition result. Step S10
In, the stroke code determination unit 6 replaces the character code from the recognition unit 4 with the character code of the corrected character given from the stroke pattern dictionary unit 7. Step S1
In 1, the stroke code determination unit 6 outputs the final character recognition result to the output unit 8. As a result, when the recognition result is determined to be correct in step S6, the character corresponding to the character code from the recognition unit 4 is output as it is, and in the case of erroneous recognition, the correction obtained in step S10. The character corresponding to the character code is displayed. As described above, in the first embodiment, the stroke number setting unit 5
A stroke code determination unit 6 and a stroke pattern dictionary unit 7 are provided, a correct stroke number is set when the first character recognition result is erroneous recognition, and the character code generated by the erroneous recognition and the set stroke number are corrected. It is configured to output the character code of the character. Therefore, the erroneous recognition caused by the clipping error is corrected, and the character recognition rate of the online character recognition device is improved.

Second Embodiment FIG. 6 is a functional block diagram of an online character recognition device showing a second embodiment of the present invention. This on-line sentence recognition device has an input unit 11 for inputting written characters such as a tablet. The output side of the input unit 11 is the cutout unit 12
The output side of the cutout unit 12 is connected to the preprocessing unit 13. The recognition unit 14 is connected to the output side of the preprocessing 13. Cutout unit 12, preprocessing unit 13, and recognition unit 1
The process in 4 is realized by a CPU, a memory, and the like (not shown). Furthermore, this online character recognition device is provided with a character type setting unit 15, a character type code determination unit 16, and a character type pattern dictionary unit 17. The output side of the recognition unit 14 is connected to the character type code determination unit 16. The character type setting unit 15 is connected to the input unit 11, and the character type setting unit 15
The output side of is connected to the character type code determination unit 16 and the character type pattern dictionary unit 17. Character type pattern dictionary unit 17
The character type code determination unit 16 and the character type code determination unit 16 input and output each other. The output side of the character type code determination unit 16 is connected to the output unit 18 formed of a display or the like.

Input unit 11, cutout unit 12, and preprocessing unit 13
The recognition unit 14 has the same function as the input unit 1, the cutout unit 2, the preprocessing unit 3, and the recognition unit 4 in the first embodiment. The character type setting unit 15 is a character type that the writer wants to input. (AN
K character, Kanji character, etc.) is set in advance as the first flag. The character type code determination unit 16 includes the recognition unit 14
The function of determining whether the character type of the recognition result matches the flag set by the character type setting unit 15 by referring to the character code flag dictionary described later by capturing the character code of the character recognition result output by Have Then, when the character type of the recognition result matches the preset flag, the character type code determination unit 16 outputs the recognition result to the output unit 18.
When the character type of the recognition result does not match the preset flag, the shift JIS code of the corrected character is input from the character type pattern dictionary unit 17, which will be described later, and is output to the output unit 1.
It is supposed to be displayed on 8. Character type pattern dictionary 1
7 includes a character type code flag dictionary and a character type pattern dictionary.

FIG. 7 shows the character type pattern dictionary unit 17 in FIG.
FIG. 8 is a diagram showing a structure of a character type code flag dictionary incorporated in FIG. 8, and FIG. 8 is a diagram showing a structure of a character type pattern dictionary incorporated in the character type pattern dictionary unit 17 in FIG. In the character type code flag dictionary, each character type, a range of character codes of that character type (shift JIS range), and a flag of each character type are stored in a corresponding form. For example, the range of character codes 889F to 9872 is given to Kanji,
Correspondingly, a flag "0000" representing a Chinese character is stored. The character type pattern dictionary of FIG. 8 has a combination of character codes resulting from erroneous recognition and a shift JIS code of a corrected character synthesized by the combination in association with each other. The character type pattern dictionary unit 17 includes the character type code determination unit 1
It has a function of obtaining a shift JIS code of a corrected character from a plurality of character codes given from No. 6 and the character type pattern dictionary of FIG. FIG. 9 is a flowchart for explaining the processing procedure of the online character recognition device of FIG. 6, and the operation of FIG. 6 will be described with reference to this figure. Step S21
In the above, the writer previously selects the character type of the handwritten input character, and the character type is set to the character type setting unit 15 via the input unit 11.
To instruct. The character type setting unit 15 sets a flag corresponding to the character type of the handwritten input character. In step S22, the writer inputs characters to the input unit 21 in stroke units. The input unit 11 converts a handwriting input into coordinates and writes a handwriting data string {(x _i , y _i ), i = 1, 2, ..., N.
j} j is sequentially output to the cutout unit 12.

In step S23, the clipping unit 12 causes the writing data string {(x _i , y _i ), i = 1, 2, ..., N.
From j} j, the coordinate values of the writing points at fixed time intervals are fetched and classified for each stroke. Further, the cutting portion 12
Automatically divides the writing data string into character units based on the positional relationship between strokes. That is, the cutout process is performed. In step S24, the preprocessing unit 13 performs preprocessing on the writing data string in character units sequentially output from the cutout unit to remove noise or smooth the writing data string. The character-based writing data string output by the preprocessing unit 13 is sequentially transferred to the recognition unit 14. In step S25, the recognition unit 14 extracts feature points from the input writing data string in character block units, and performs character recognition using the extracted feature points. The shift JIS code of the character code, which is the result of character recognition, is output to the character type code determination unit 16.

In step S26, the character type code determination unit 16 outputs the character recognition result from the recognition unit 14 to the character type pattern dictionary unit 17. The character type pattern dictionary unit 17
The second flag corresponding to the character type of the input character recognition result is set by referring to the character type code flag dictionary. The character type code determination unit 16 compares the second flag with the first flag. For example, the character type pattern dictionary unit 17 sets a flag "0000" if the character recognition result is a Kanji character, and sets a flag "0001" if the character recognition result is an English capital letter, and the character type code determination unit 16 sets the second flag and the first flag. It is determined whether it matches the flag of. If they do not match, that is, if erroneous recognition occurs, the process proceeds to step S27. In step S27, the character type code determination unit 17 passes the character code group of the erroneous recognition result to the character type pattern dictionary unit 17.
The character type pattern dictionary unit 17 obtains the character code of the corrected character by using the character type pattern dictionary shown in FIG.

FIG. 10 is a diagram showing an example of erroneously recognized characters. Even if the writer inputs "A", "B", and "C" in capital letters, the recognition unit 14 causes "A",
Character codes corresponding to "1", "3", and "C" may be output. In this case, "B" is erroneously recognized as "1" and "3", and the character types are different. The character type pattern dictionary unit 17 searches for the character code “8271” of the corrected character from the character codes “8250” and “8252” corresponding to the respective “1” and “3”. Character code "826
1 "corresponds to the character" B ". The character code of this modified character is output to the character type code determination unit 17. In step S28, the character type code determination unit 16 determines the character recognition result from the recognition unit 14. Among them, the character code of the erroneous recognition is replaced with the character code of the corrected character, and this is output to the output unit 18 as the corrected character recognition result.
In 9, if the character recognition result from the recognition unit 14 is determined to be correct in the determination in step S26, the output unit 18 displays the character recognition result. If there is an erroneous recognition, the corrected character recognition result obtained in step S28 is displayed. As described above, in the second embodiment, the character type setting unit 15 and the character type code determining unit 1
6 and the character type pattern dictionary unit 17 are provided, the character type is limited in advance, and when a character recognition result other than the limited character type is obtained, the character recognition result corrected by the character type pattern dictionary unit 17 from the erroneous character recognition result. Is configured to obtain. As a whole, the recognition rate of the online character recognition device is improved.
Note that the present invention is not limited to the above embodiment, and various modifications are possible. For example, as a character recognition result, shift JI
Although the S code is used, the present invention can improve the recognition rate even in a device using another character code.

[0020]

As described above in detail, according to the first invention, the stroke number setting unit for setting the stroke number, the stroke code determination unit, and the stroke pattern dictionary unit are provided, and the first character recognition result is In the case of erroneous recognition, the correct stroke number is set, and the erroneous character recognition result is corrected based on the erroneous character recognition result and the set stroke number. Therefore, the erroneous recognition caused by the clipping error is corrected, and the character recognition rate of the online character recognition device can be improved. According to the second invention, a character type setting unit, a character type code determination unit, and a character type pattern dictionary unit are provided, flags corresponding to the character types are set in advance, and when a character recognition result other than the set character type is obtained, Since the corrected character recognition result is obtained from the erroneous character recognition result, the erroneous recognition caused by the clipping error is corrected, and the character recognition rate of the online character recognition device can be improved.

[Brief description of drawings]

FIG. 1 is a functional block diagram of an online character recognition device showing a first embodiment of the present invention.

FIG. 2 is an explanatory diagram of a conventional character cutout process.

3 is a diagram showing a stroke pattern dictionary in a stroke pattern dictionary unit in FIG.

FIG. 4 is a flowchart showing a processing procedure of the online character recognition device of FIG.

FIG. 5 is a diagram showing an example of erroneously recognized characters.

FIG. 6 is a functional block diagram of an online character recognition device showing a second embodiment of the present invention.

7 is a diagram showing the structure of a character type code flag dictionary built in the character type pattern dictionary unit in FIG.

8 is a diagram showing the structure of a character type pattern dictionary built in the character type pattern dictionary unit in FIG. 6;

9 is a flowchart illustrating a processing procedure of the online character recognition device in FIG.

FIG. 10 is a diagram showing an example of erroneously recognized characters.

[Explanation of symbols]

 1, 11 Input unit 2, 12 Cutout unit 3, 13 Pre-processing unit 4, 14 Recognition unit 5 Stroke setting unit 6 Stroke code determination unit 7 Stroke pattern dictionary unit 8, 18 Output unit 15 Character type setting unit 16 Character type code determination unit 17 Character type pattern dictionary

Claims (2)

[Claims] 1. An input unit for converting a sequentially input handwritten character into coordinates to extract a handwritten data string, classifying the handwritten data string into strokes of the handwritten character, and writing the handwritten data string from a positional relationship of the strokes. Is divided into character blocks corresponding to one character, a recognition unit that performs feature extraction of each of the character blocks and matching of the features, and recognizes characters in each of the character blocks; An output unit that displays the character recognition result, a stroke number setting unit that sets the correct stroke number of the written character that is input when the character recognition result in the recognition unit is incorrect, and the character recognition error occurs. A built-in dictionary in which a combination of misrecognition-result characters, a correction character synthesized by the combination, and a stroke number of the correction character are stored in association with each other, A stroke pattern dictionary unit that extracts a correction character corresponding to a predetermined number of strokes and a given combination of misrecognition result characters from the dictionary, and outputs this as a corrected character recognition result, and the set stroke number. A stroke code determination unit which detects a combination of the erroneously recognized result characters based on the above, gives the stroke pattern dictionary unit, receives the corrected character recognition result from the stroke pattern dictionary unit, and gives it to the output unit. An online character recognition device. 2. A character type setting unit for presetting a first flag indicating a character type of a written character, an input unit for coordinate-converting sequentially input written characters to extract a written data string, and the written data string. A cutout unit that classifies the strokes of the written characters and divides the written data string into character blocks corresponding to one character based on the positional relationship of the strokes, performs feature extraction of each of the character blocks, and performs matching to the features, A character recognition unit for each character block and a built-in dictionary are searched, a second flag indicating the character type of the character recognition result is set, and an erroneous character recognition result out of the character recognition result is detected as the first character. The character type pattern dictionary unit that corrects the character recognition result of the character type indicated by the flag 1 and the character recognition output by the recognition unit by comparing the second flag and the first flag. It is determined whether the result is correct or incorrect, and when it is determined to be correct, this character recognition result is output, and when it is determined to be incorrect, the incorrect character recognition result is passed to the character type pattern dictionary, and the character type pattern dictionary is output. A character type code determination unit that receives and outputs a corrected character recognition result from the character type code determination unit; and an output unit that displays the character recognition result output from the character type code determination unit or the corrected character recognition result. Online character recognition device.