JPH0883318A - Method for extracting inflection point, method for roughly sorting dictionary and method and device for recognizing character - Google Patents

Abstract

PURPOSE: To hold always fine character recognition accuracy by preventing the extraction accuracy of an inflection point of a stroke pattern from being deteriorated even when the curvature of the stroke pattern is close to a threshold. CONSTITUTION: An inflection point extracting part 322 finds out the curvature of a stroke pattern inputted to an input stroke buffer 34 and extracts a point on which the curvature is more than the threshold as an inflection point. A pattern checking part 323 matches a dictionary pattern in a reference area obtained by roughly sorting a character recognition dictionary 41 based upon the extracted inflection point with the stroke pattern to recognize a character. At the time, a threshold control part 325 repeats the extraction of an inflection point, the rough sorting of the dictionary and the character recognition of the stroke pattern while changing the threshold set up in a threshold butter 37 and a candidate output part 324 outputs a dictionary pattern having the highest similarity to the stroke pattern, so that always fine character recognition accuracy can be obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention uses an inflection point extraction method for extracting an inflection point from a stroke pattern constituting the handwritten character, which is necessary when recognizing a handwritten character. The present invention relates to a large classification method for a dictionary, a character recognition method using the large classification method for the dictionary, and a character recognition device using each of these methods.

[0002]

2. Description of the Related Art When a character is input with a pen on a transparent tablet (coordinate input device), the handwriting is displayed on a liquid crystal display device laminated below the transparent tablet and input from the transparent tablet. There has been developed a character recognition device that performs character recognition of the recognized characters, displays the recognized characters on the liquid crystal display device, and inputs the characters into the device.
Such a character recognition device is installed in various information devices and enables handwriting of characters and gesture recognition.

In the character recognition method in such a character recognition device, a stroke pattern constituting one character of a handwritten character input from a transparent tablet is vector-divided, the handwriting direction is traced, and the obtained vectors are mutually correlated. The handwritten character is recognized by matching the angle change with a dictionary pattern in the character recognition dictionary.

At this time, in order to improve the efficiency of matching between the angle change between the vectors and the dictionary pattern in the character recognition dictionary and improve the character recognition accuracy, the point where the angle change amount in the stroke pattern is large ( Generally, the points are referred to as inflection points), and the character recognition dictionaries are roughly classified (the reference region of the character recognition dictionary is greatly narrowed) according to the number, positional relationship, and the like.

However, in the dictionary classification method as described above, the threshold value of curvature for extracting the inflection point is set to be constant for all stroke patterns. The point of the stroke pattern indicating the curvature of is originally extracted as an inflection point, but is not extracted, or vice versa, it is extracted as it should not be extracted as an inflection point. Happens. When such a situation occurs, there is a drawback in that the stroke pattern and the dictionary pattern in the character recognition dictionary cannot be matched well and the character recognition accuracy is deteriorated.

[0006]

As described above, in the conventional character recognition device, the character recognition dictionary is roughly classified by extracting the inflection points of the strokes constituting the handwritten character to be recognized, and the character recognition dictionary is classified. Although the efficiency of recognition processing and the accuracy of character recognition have been improved, the threshold value for extracting the inflection point is constant for all stroke patterns. Therefore, in the conventional inflection point extraction method, when the curvature of the stroke pattern is near the boundary of this threshold, there is a drawback that the inflection point extraction accuracy deteriorates. In the large classification method, it may happen that an effective reference dictionary area cannot be obtained, and the conventional character recognition method for character recognition using this reference dictionary area cannot successfully match the stroke pattern and the character recognition dictionary. However, there is a drawback in that the character recognition accuracy may be lowered due to the warp.

In view of the above drawbacks, the present invention has an inflection point extraction method and a dictionary large classification method that can always maintain good character recognition accuracy even when the curvature of the stroke pattern to be recognized is near the threshold value. The object of the present invention is to provide a character recognition method and a character recognition device using these methods.

[0008]

According to a first aspect of the present invention, the curvature of a stroke pattern constituting a character input by handwriting is compared with a threshold value to extract an inflection point, and character recognition is performed based on this inflection point. In the inflection point extraction method for obtaining the inflection point in the character recognition device that performs character recognition based on the similarity between the reference area dictionary pattern obtained by roughly classifying the dictionary and the stroke pattern, the character A method is adopted in which the threshold value is changed and the inflection point is extracted again when the recognition result does not satisfy the predetermined condition.

The invention of claim 2 employs a method of changing the threshold value and extracting the inflection point again until the predetermined condition is obtained. According to the invention of claim 3, as the predetermined condition, the dictionary pattern similar to the stroke pattern obtained as the character recognition result this time and the dictionary pattern obtained by the previous character recognition are the same. The method adopted as the condition is adopted.

According to a fourth aspect of the invention, as the predetermined condition,
A method is employed on the condition that the dictionary pattern having a high similarity equal to or higher than a predetermined value with respect to the stroke pattern is obtained as a character recognition result.

According to a fifth aspect of the invention, as the predetermined condition,
The method is employed on the condition that the dictionary pattern having the highest similarity to the stroke pattern is obtained as a character recognition result.

According to a sixth aspect of the present invention, the curvature of a stroke pattern forming a character input by handwriting is compared with a threshold value to extract an inflection point, and the character recognition dictionary is roughly classified based on the inflection point. In the large classification method of the character recognition dictionary in the character recognition device that performs character recognition based on the similarity between the obtained dictionary pattern of the reference area and the stroke pattern, the character recognition result does not satisfy a predetermined condition. In this case, a method is adopted in which the character recognition dictionary is classified again based on the inflection points extracted by changing the threshold.

According to a seventh aspect of the present invention, a method is adopted in which the threshold is changed and the large classification of the character recognition dictionary is repeated until the predetermined condition is obtained. According to the invention of claim 8, as the predetermined condition, a dictionary pattern similar to the stroke pattern obtained as the character recognition result this time and a dictionary pattern obtained by the previous character recognition are the same. The method adopted as the condition is adopted.

According to a ninth aspect of the present invention, as the predetermined condition,
A method is employed on the condition that the dictionary pattern having a high similarity equal to or higher than a predetermined value with respect to the stroke pattern is obtained as a character recognition result.

According to the tenth aspect of the invention, the predetermined condition is that the dictionary pattern having the highest similarity to the stroke pattern is obtained as a character recognition result.

According to an eleventh aspect of the present invention, the curvature of the stroke pattern that constitutes the handwritten character is compared with a threshold value to extract an inflection point, and the character recognition dictionary is roughly classified based on this inflection point. In the character recognition method in the character recognition device that performs character recognition based on the similarity between the obtained reference area dictionary pattern and the stroke pattern, if the character recognition result does not satisfy a predetermined condition, the threshold value is changed. Now, a method of re-recognizing the character of the stroke pattern is adopted.

According to a twelfth aspect of the invention, a method is adopted in which the threshold value is changed and character recognition of the stroke pattern is repeated until the predetermined condition is obtained. According to the invention of claim 13, as the predetermined condition, a dictionary pattern similar to the stroke pattern obtained as the character recognition result this time and a dictionary pattern obtained by the previous character recognition are the same. The method adopted as the condition is adopted.

According to a fourteenth aspect of the present invention, as the predetermined condition, a method is adopted on the condition that the dictionary pattern having a high similarity to the stroke pattern of a predetermined value or more is obtained as a character recognition result. ing.

According to a fifteenth aspect of the present invention, as the predetermined condition, a method is employed on the condition that a dictionary pattern having a highest similarity to the stroke pattern is obtained as a character recognition result.

According to a sixteenth aspect of the present invention, the curvature of the stroke pattern forming the characters input by handwriting is compared with a threshold value to extract an inflection point, and the character recognition dictionary is roughly classified based on the inflection point. In the character recognition device that performs character recognition based on the similarity between the obtained dictionary pattern of the reference area and the stroke pattern, a threshold changing unit that changes the threshold value, and the character recognition result satisfy a predetermined condition. If the predetermined condition is not satisfied by the determining means, the threshold changing means changes the threshold to perform character recognition of the stroke pattern again. And a recognition means.

According to a seventeenth aspect of the present invention, the re-recognition means repeats character recognition of the stroke pattern by changing the threshold value by the threshold value changing means until the predetermined condition is obtained.

According to the eighteenth aspect of the present invention, as the predetermined condition, the dictionary pattern similar to the stroke pattern obtained as the character recognition result this time and the dictionary pattern obtained by the previous character recognition are the same. It has a structure on the condition that.

The present invention according to claim 19 has a constitution under the condition that the dictionary pattern having a high degree of similarity with respect to the stroke pattern is obtained as a character recognition result.

[0024] The invention of claim 20 has a constitution under the condition that the dictionary pattern having the highest similarity to the stroke pattern is obtained as a character recognition result as the predetermined condition.

[0025]

In the inflection point extraction method of the present invention, when the character recognition result does not satisfy the predetermined condition, the threshold value is changed and the inflection point is extracted again. In the inflection point extraction method according to the second aspect of the present invention, the inflection point is extracted again by changing the threshold value until the predetermined condition is obtained.

In the inflection point extraction method according to the third aspect of the present invention, as the predetermined condition, a dictionary pattern similar to the stroke pattern obtained as the character recognition result this time and a dictionary obtained by the previous character recognition are used. The condition is that the pattern is the same.

In the inflection point extraction method according to the fourth aspect of the present invention, the predetermined condition is that the dictionary pattern having a high similarity of a predetermined value or more with respect to the stroke pattern is obtained as a character recognition result. I am trying.

In the inflection point extracting method of the fifth aspect of the present invention, the predetermined condition is that the dictionary pattern having the highest similarity to the stroke pattern is obtained as a character recognition result.

In the dictionary large classification method according to the sixth aspect of the present invention, when the character recognition result does not satisfy a predetermined condition, the large character recognition dictionary is changed based on the inflection points extracted by changing the threshold value again. Repeat the classification.

In the dictionary large classification method of the seventh aspect of the invention, the large classification of the character recognition dictionary is repeated by changing the threshold value until the predetermined condition is obtained. In the dictionary large classification method of the invention of claim 8, as the predetermined condition, a dictionary pattern similar to the stroke pattern obtained as a result of character recognition this time and a dictionary pattern obtained by previous character recognition are used. The condition is that they are the same.

In the method for classifying a dictionary according to the present invention, the predetermined condition is that the dictionary pattern having a high similarity with the stroke pattern of a predetermined value or more is obtained as a character recognition result. I am trying.

In the dictionary major classification method of the tenth aspect of the present invention, the predetermined condition is that the dictionary pattern having the highest similarity to the stroke pattern is obtained as a character recognition result.

In the character recognition method of the eleventh aspect of the present invention, when the character recognition result does not satisfy a predetermined condition, the threshold is changed and the character recognition of the stroke pattern is performed again.

In the character recognition method of the twelfth aspect of the invention, the character recognition of the stroke pattern is repeated by changing the threshold value until the predetermined condition is obtained. In the character recognition method of the invention of claim 13, as the predetermined condition, a dictionary pattern similar to the stroke pattern obtained as a current character recognition result and a dictionary pattern obtained by previous character recognition are the same. The condition is that

In the character recognition method according to the fourteenth aspect of the present invention, the predetermined condition is that the dictionary pattern having a high degree of similarity with the stroke pattern of a predetermined value or more is obtained as a character recognition result. .

In the character recognition method of the fifteenth aspect of the present invention, the predetermined condition is that the dictionary pattern having the highest similarity to the stroke pattern is obtained as a character recognition result.

In the character recognition apparatus of the sixteenth aspect of the present invention, the threshold value changing means changes the threshold value. The judging means judges whether or not the character recognition result satisfies a predetermined condition. When the determining unit determines that the predetermined condition is not satisfied, the re-recognizing unit changes the threshold value by the threshold value changing unit and performs character recognition of the stroke pattern again.

In the character recognition apparatus of the seventeenth aspect of the present invention, the re-recognition means repeats the character recognition of the stroke pattern by changing the threshold value by the threshold value changing means until the predetermined condition is obtained.

In the character recognition apparatus of the eighteenth aspect of the present invention, as the predetermined condition, a dictionary pattern similar to the stroke pattern obtained as the character recognition result this time, and a dictionary pattern obtained by the previous character recognition. The condition is that they are the same.

In the character recognition apparatus of the nineteenth aspect of the present invention, the predetermined condition is that the dictionary pattern having a high degree of similarity of a predetermined value or more with respect to the stroke pattern is obtained as a character recognition result. .

In the character recognition apparatus of the twentieth aspect of the invention, the predetermined condition is that the dictionary pattern having the highest similarity to the stroke pattern is obtained as a character recognition result.

[0042]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the character recognition device of the present invention. Reference numeral 1 is a coordinate input device, 2 is a display device, 3 is a control device which is the main body of the device, and 4 is an external storage device. The coordinate input device 1 is composed of a transparent tablet 11 and a stylus pen 12, is stacked on the display device 2 composed of a liquid crystal display or the like, and is further integrated with the control device 3.

The transparent tablet 11 and the coordinate position on the display device 2 can be directly designated by the stylus pen 12, and the display information of the display device 2 can be visually recognized through the transparent tablet 11. The display device 2 is a liquid crystal display or the like as described above. Control device 3
Is a transparent tablet 11 specified by the stylus pen 12.
A coordinate input unit 31 for inputting the above coordinate data, a stroke recognition unit 32 for performing character recognition of a stroke for one character input from the coordinate input unit 31, a display for displaying the character recognition result on the display device 2. A processing unit including a unit 33, an input stroke buffer 34 that stores stroke pattern information input from the coordinate input unit 31, a division vector buffer 35 that stores a division vector obtained from the stroke pattern, and a stroke for one character. A similarity buffer 3 that stores the similarity between the pattern and the dictionary pattern in the character recognition dictionary (standard stroke dictionary) 41.
6 and a buffer memory including a threshold value buffer 37 for setting a threshold value for extracting an inflection point from a stroke pattern.

The stroke recognition section 32 is a vector division section 3
21, an inflection point extraction unit 322, a pattern inspection unit 323, a candidate output unit 324, and a threshold value control unit 325.
The external storage device 4 is composed of a magnetic disk device or the like for storing the standard stroke dictionary 41, and may be integrated with the control device 3.

Next, the operation of this embodiment will be described with reference to the flow chart shown in FIG. The user uses the stylus pen 12 of the coordinate input device 1 to make the transparent tablet 11
Handwrite on top. Thereby, the transparent tablet 11 outputs the handwriting of the handwritten character as coordinate point sequence data, and outputs this to the coordinate input unit 31 of the control device 3. In step 201, the coordinate input unit 31 stores the input coordinate point sequence data in the input stroke buffer 34 as a stroke pattern.

When the stroke pattern for one character is stored in the input stroke buffer 34 by the coordinate input unit 31, the stroke recognition unit 32 is activated, and the vector division unit 321 in the input stroke buffer 34 stores it in step 202. Each stroke pattern is divided into vectors indicating the writing direction, and the obtained division vector is stored in the division vector buffer 35. In step 203, the inflection point extraction unit 322 obtains the curvature while tracking the angle change of the division vector in the division vector buffer 35, and extracts the point in the stroke pattern where the curvature is equal to or more than the threshold as the inflection point.

Next, the pattern inspection section 323 executes step 20.
4, the reference dictionary area of the standard stroke dictionary 41 is selected according to the number of extracted inflection points and the positional relationship, and the dictionary is roughly classified, and then the selected reference dictionary area and the division vector buffer 35 are selected. And the similarity between them is calculated, and the calculation result is stored in the similarity buffer 36 in step 205 for each matched dictionary pattern.

In step 206, the pattern inspecting unit 323 determines whether or not the matching of the division vector in the division vector buffer 35 with all the areas of the character recognition dictionary is completed. When the matching is completed, step 211 Proceed to step 207 and proceed to step 207 if all matching is not completed. When the process proceeds to step 211, the candidate output unit 324 selects the dictionary pattern having the highest similarity from the dictionary patterns registered in the similarity buffer 36, and the selected dictionary pattern is displayed on the display unit 33 as the final recognition result. Output.

The display unit 33 displays the dictionary pattern passed from the candidate output unit 324 on the display device 2 as a recognized character. However, if the candidate output unit 324 cannot select a dictionary pattern whose similarity to the recognition target stroke pattern is equal to or more than a predetermined value from the dictionary patterns registered in the similarity buffer 36, it is determined that the character cannot be recognized, and this is notified. The displayed message is output to the display unit 33, and the message is displayed on the display device 2.

On the other hand, if the process proceeds to step 207, the pattern inspection unit 323 judges whether the similarity between the dictionary pattern registered in the similarity buffer 36 and the stroke pattern to be recognized is high, and the similarity between them is high. The process proceeds to step 211 and the above-described character recognition candidate output process is performed, but if the similarity between the two is low, the process proceeds to step 208.

In step 208, the threshold control unit 325 is activated, and the threshold control unit 325 is activated by the pattern inspection unit 323.
Whether the dictionary pattern registered this time in the similarity buffer 36 at step 205 is the same as the dictionary pattern obtained in the previous processing before the threshold value set in the threshold value buffer 37 is changed. The determination is made by referring to the dictionary pattern at the time of the previous processing registered in the similarity buffer 36. If it is determined that they are the same, the process proceeds to step 209, and if they are not the same, the process proceeds to step 210.

When the process proceeds to step 209, the threshold control unit 325 compares the similarity between the recognition target stroke pattern and the dictionary pattern obtained this time with the similarity between the recognition target stroke pattern and the dictionary pattern obtained last time. If it is determined that the similarity is the same as or lower than that of the previous time, the process proceeds to step 211, the above-described candidate output process is performed, and the dictionary pattern stored in the similarity buffer 36 as the final recognition result. Is output.

When the threshold control unit 325 determines in step 209 that the above-mentioned similarity is higher than the previous time, it proceeds to step 210, and the threshold value set in the threshold buffer 37 is set to the same as the previous time. After changing slightly to the direction,
Returning to step 203, the above processing is repeated.

By the way, while changing the threshold value set in the threshold value buffer 37 by the threshold value control section 325, the above steps 203 to 210 are repeated to obtain the pattern inspection section 323 and store it in the similarity buffer 36. If the type of dictionary pattern is the same (step 208) this time and the last time, and if the above process is further repeated, the similarity degree of this time increases from the previous time in the determination of step 209, and the increase of this similarity degree stops and this time. When is equal to or lower than the last time, it means that the previous similarity is the highest. At this point, similarity buffer 3
By moving to the processing of step 211 in which the candidate output unit 324 outputs the dictionary pattern stored in No. 6,
After all, the character recognition result having the highest similarity, that is, when the inflection point of the stroke pattern is successfully extracted by the inflection point extraction unit 322 and the large classification of the character recognition dictionary 41 is effectively performed, the character recognition result is displayed on the display device. 2 will be output.

FIG. 3A is a diagram showing an example of a stroke pattern to be recognized, and FIG. 3B is a diagram showing an example of a dictionary pattern in the standard stroke dictionary 41. From the recognition target stroke pattern of this example, the inflection point extraction unit 3 shown in FIG. 1 is used by using the threshold value initially set in the threshold value buffer 37 shown in FIG.
When the inflection point is extracted by 22 as shown in FIG. 4A, only the point b is extracted as the inflection point, and the point a which should be originally extracted as the inflection point is not extracted as the inflection point. . Therefore, the pattern inspecting unit 3 can be formed only by the inflection point at the point b.
23 performs a large classification of the standard stroke dictionary 41.
The dictionary pattern example in the character recognition dictionary shown in FIG.
The matching between the recognition target stroke pattern and the dictionary pattern shown in (A) will not be successful.

Therefore, when the threshold control unit 325 changes the threshold set in the threshold buffer 37 to a slightly lower value, the recognition target stroke pattern shown in FIG. 3 (A) is changed to a as shown in FIG. 4 (B). Two inflection points, i.e. and b, are extracted. Therefore, when the pattern inspecting unit 323 roughly classifies the standard stroke dictionary 41 using these two inflection points, the dictionary pattern shown in FIG. Matching of the target stroke pattern will be successful, and highly accurate character recognition will be performed.

FIG. 5A is a diagram showing another example of a stroke pattern for recognition, and FIG. 5B is a diagram showing another example of a dictionary pattern in the character recognition dictionary. From the recognition target stroke pattern of this example, the inflection point extraction unit 32 shown in FIG. 1 is used by using the threshold value initially set in the threshold value buffer 37 shown in FIG.
When the inflection point is extracted by 2, the inflection point is extracted at three points a point b point c point as shown in FIG. 6 (A), and the b point that should not be originally extracted as an inflection point It will be extracted as a point. Therefore, when the pattern inspecting unit 323 roughly classifies the standard stroke dictionary 41 at the three inflection points a, b, and c, the standard stroke dictionary shown in FIG. The example of the dictionary pattern in 41 is omitted, and the matching between the stroke pattern to be recognized and the dictionary pattern shown in FIG.

Therefore, when the threshold control unit 325 changes the threshold set in the threshold buffer 37 so as to slightly raise it, the recognition target stroke pattern shown in FIG. 5A becomes a as shown in FIG. 6B. The inflection points are extracted at two points, c and c. Therefore, when the pattern search unit 323 roughly classifies the standard stroke dictionary 41 using the two inflection points, the dictionary pattern shown in FIG. Matching of the target stroke pattern will be successful, and highly accurate character recognition will be performed.

According to this embodiment, the inflection point extracted from the stroke pattern to be recognized can be controlled by changing the threshold value set in the threshold value buffer 37. By controlling, the reference dictionary area obtained by roughly classifying the standard stroke dictionary 41 can be made effective, and by matching the dictionary pattern of this effective reference dictionary area with the stroke pattern, initially, Even if the threshold value is near the boundary of the curvature of the stroke pattern, the stroke pattern can be accurately recognized.

[0060]

As described above, according to the present invention, it is possible to effectively classify the character recognition dictionary by changing the threshold value and controlling the extraction state of the inflection point. Since the character recognition is performed using the classified reference area of the character recognition dictionary, the character recognition accuracy can be always kept good regardless of the curvature of the stroke of the recognition target character.

[Brief description of drawings]

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

FIG. 2 is a flowchart showing a character recognition process of a control unit shown in FIG.

3 is a diagram showing an example of a stroke pattern of a recognition target handled by the apparatus of FIG. 1 and an example of a dictionary pattern in a character recognition dictionary.

FIG. 4 is a diagram showing an example of an inflection point extracting operation by an inflection point extracting unit shown in FIG. 1.

FIG. 5 is a diagram showing another example of a stroke pattern of a recognition target handled by the apparatus of FIG. 1 and another example of a dictionary pattern in a character recognition dictionary.

FIG. 6 is a diagram showing another example of the inflection point extracting operation by the inflection point extracting section shown in FIG. 1;

[Explanation of symbols]

1 ... Coordinate input device 2 ... Display device 3 ... Control device 4 ... External storage device 11 ... Tablet 12 ... Stylus pen 31 ... Coordinate input unit 32 ... Stroke recognition unit 33 ... Display unit 34 ... Input stroke buffer 35 ... Division vector buffer 36 ... Similarity buffer 37 ... Threshold buffer 41 ... Standard stroke dictionary 321 ... Vector division unit 322 ... Inflection point extraction unit 323 ... Pattern inspection unit 324 ... Candidate output unit 325 ... Threshold control unit

Claims (20)

[Claims] 1. An inflection point is extracted by comparing the curvature of a stroke pattern constituting a character input by handwriting with a threshold value,
An inflection for obtaining the inflection point in a character recognition device that performs character recognition based on the similarity between the dictionary pattern of the reference area obtained by roughly classifying the character recognition dictionary based on the inflection point and the stroke pattern. A point extraction method, wherein when the character recognition result does not satisfy a predetermined condition, the threshold value is changed and the inflection point is extracted again. 2. The inflection point extraction method according to claim 1, wherein the inflection point is extracted again by changing the threshold value until the predetermined condition is obtained. 3. The predetermined condition is that a dictionary pattern similar to the stroke pattern obtained as a result of character recognition this time and a dictionary pattern obtained by previous character recognition are the same. The inflection point extraction method according to claim 1 or 2, characterized in that. 4. The predetermined condition is a condition that the dictionary pattern having a high degree of similarity equal to or higher than a predetermined value with respect to the stroke pattern is obtained as a character recognition result. Inflection point extraction method described. 5. The inflection point according to claim 1, wherein the predetermined condition is a condition that a dictionary pattern having a highest similarity to the stroke pattern is obtained as a character recognition result. Extraction method. 6. An inflection point is extracted by comparing the curvature of a stroke pattern forming a character input by handwriting with a threshold value,
Large classification method of the character recognition dictionary in the character recognition device that performs character recognition based on the similarity between the reference area dictionary pattern obtained by roughly classifying the character recognition dictionary based on the inflection point and the stroke pattern Then, when the character recognition result does not satisfy a predetermined condition, the character recognition dictionary is re-classified based on the inflection point extracted by changing the threshold value again. Classification method. 7. The dictionary major classification method according to claim 6, wherein the major classification of the character recognition dictionary is repeated by changing the threshold value until the predetermined condition is obtained. 8. The predetermined condition is that a dictionary pattern similar to the stroke pattern obtained as a result of character recognition this time and a dictionary pattern obtained by previous character recognition are the same. The method for classifying a dictionary according to claim 6 or 7, characterized in that. 9. The predetermined condition is a condition that the dictionary pattern having a high degree of similarity of a predetermined value or more with respect to the stroke pattern is obtained as a character recognition result. Major classification method of the listed dictionary. 10. The large dictionary according to claim 6, wherein the predetermined condition is a condition that a dictionary pattern having a highest similarity to the stroke pattern is obtained as a character recognition result. Classification method. 11. An inflection point is extracted by comparing the curvature of a stroke pattern constituting a handwritten input character with a threshold value, and a character recognition dictionary is roughly classified based on this inflection point to obtain a reference area. In the character recognition method in the character recognition device that performs character recognition based on the similarity between the dictionary pattern and the stroke pattern, when the character recognition result does not satisfy a predetermined condition, the threshold is changed to change the stroke pattern. A character recognition method characterized by performing character recognition again. 12. The character recognition method according to claim 11, wherein the character recognition of the stroke pattern is repeated by changing the threshold value until the predetermined condition is obtained. 13. The predetermined condition is a condition that a dictionary pattern similar to the stroke pattern obtained as a result of character recognition this time and a dictionary pattern obtained by previous character recognition are the same. The character recognition method according to claim 11 or 12, characterized in that. 14. The predetermined condition is a condition that the dictionary pattern having a high degree of similarity of a predetermined value or more with respect to the stroke pattern is obtained as a character recognition result. Character recognition method described. 15. The character recognition method according to claim 11, wherein the predetermined condition is a condition that a dictionary pattern having a highest similarity to the stroke pattern is obtained as a character recognition result. . 16. A reference area obtained by roughly classifying a character recognition dictionary based on the inflection points by extracting the inflection points by comparing the curvature of a stroke pattern forming a handwritten input character with a threshold value. In a character recognition device that performs character recognition based on the similarity between the dictionary pattern and the stroke pattern, a threshold value changing unit that changes the value of the threshold value, and determine whether the character recognition result satisfies a predetermined condition. And a re-recognition means for changing the threshold value by the threshold value changing means and recognizing the character of the stroke pattern again when the predetermined condition is judged not to be satisfied. A character recognition device characterized in that 17. The character recognition device according to claim 16, wherein the re-recognition unit repeatedly performs the character recognition of the stroke pattern by changing the threshold value by the threshold value changing unit until the predetermined condition is obtained. . 18. The predetermined condition is that a dictionary pattern similar to the stroke pattern obtained as a result of character recognition this time and a dictionary pattern obtained by previous character recognition are the same. The character recognition device according to claim 16 or 17, characterized in that. 19. The predetermined condition is a condition that the dictionary pattern having a high degree of similarity of a predetermined value or more with respect to the stroke pattern is obtained as a character recognition result. Character recognition device described. 20. The character recognition device according to claim 16, wherein the predetermined condition is a condition that a dictionary pattern having a highest similarity to the stroke pattern is obtained as a character recognition result. .