JP3043633B2 - Learning method of handwritten recognition characters - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for learning a handwritten recognition character, and more particularly to a method for learning a handwritten recognition character in a handwriting character recognition system used for a pen input computer or a portable terminal.

[0002]

2. Description of the Related Art In a handwritten character recognition system, a character input by handwriting is compared with a basic pattern of the character possessed by the recognition system, and a plurality of characters are assigned by ranking in descending order of an evaluation value representing the degree of matching. It is extracted as a recognition candidate character group, and the first candidate character having the highest evaluation value is normally displayed on the screen as a recognition result. If the character displayed as the recognition result is different from the handwritten input character intended by the user,
The recognition candidate character group is displayed on the screen, and the user selects an intended character from the group and performs correction. As a method for reducing the occurrence of misrecognition that requires such selection correction, the situation when misrecognition occurs and the selection correction is performed is stored as learning information, and the same handwritten character is input next time. A learning method is used in which the character selected last time is preferentially displayed with reference to the learning information.

Such learning methods include a method of storing and using a pattern of a handwritten input character including a peculiar writing habit as learning information, and a method of gradually correcting a basic pattern by a handwritten input character pattern. (JP-A-61-114387). These have the drawback that they are difficult to apply to simple devices and portable terminals because they require complicated processing and use a large amount of memory because they store character patterns as learning information.

On the other hand, Japanese Patent Application Laid-Open No. Hei 7-175888 proposes a method which does not require storing a character pattern in learning information and can be easily applied to a portable terminal.
In the learning method described in the above publication, when the first candidate of the recognition candidate character group is not the character intended by the user and another character is selected, the first candidate displayed as the recognition result is selected as the first candidate. The difference between the evaluation value and the calculated character is calculated, and only the difference between the evaluation value and the character code are stored as learning information. In other words, the first candidate character that is the recognition result is “misrecognized character” and the selected character is “selected character”, and “misrecognized character” and “selected character”
And the character code of both are "difference in evaluation value"
And memorize it. In the subsequent recognition of handwritten characters, the first candidate in the recognition candidate character group is a “misrecognized character”.
Is satisfied, the evaluation value is corrected based on the "difference in evaluation value" in which the evaluation value of the linked "selected character" is stored, and then reevaluation is performed. This correction is applied to the evaluation value of the "selected character" so that the corrected evaluation value of the "selected character" exceeds the evaluation value of the "misrecognized character" when exactly the same handwritten characters are input. An addition slightly exceeding the “difference in evaluation value” is performed. As a result, after the re-evaluation, the “selected character” becomes the first candidate instead of the “misrecognized character”, and is displayed on the screen as a recognition result.

The basis of the learning method of the prior art is to learn by connecting the first candidate "misrecognized character" as the recognition result and the "selected character" intended by the user, When the “misrecognized character” becomes the first candidate in the recognition, the “selected character” is to be displayed as a recognition result instead of the “misrecognized character” of the first candidate.

[0006]

The problem of the prior art described above is that different characters input by handwriting by a user are different.
When the first candidate of the recognition candidate character group is the same character,
There is a case where learning cannot be performed for both. This is because, in the above-described conventional learning method, the “misrecognized character” and the “selected character” are linked one-to-one, so that a handwritten character that has not yet been learned is input and another learned character has already been learned. If the selected candidate character is the same as the first candidate and the selected candidate character group includes the “selected character” (another character that has been learned), “ This is due to the fact that the "selected character" may be displayed after the evaluation score of the "selected character" is raised and replaced. When the “misrecognized character” and the “selected character” are simply linked in a one-to-one manner, this replacement is inevitable.

More specifically, in the prior art, a recognition candidate character group "T, Z,
It is assumed that “Otsu, 2” is recognized, and “misrecognized character” “te” and “selected character” “2” are learned by the user's selection.

Next, it is assumed that the user inputs "Z", and the recognition candidate character group "te, 2, Z, B" is recognized. At this time, according to the above-described conventional method, the “misrecognized character” “te” and the “selected character” “2” are exchanged based on the learning result, so that the recognition candidate character group “ “2” is presented as “2, Z, B”. When the user selects “Z” because “2” is not the intended character, the conventional learning method described above causes “misrecognized character”
"2" and "selected character""Z" will be learned.

After that, when the user inputs "Z" again and the recognition candidate character group "te, 2, Z, Otsu" is recognized, the first candidate of the recognition candidate character group is "te". , The previous “Z”
The learning result of “misrecognized character” “2” and “selected character” “Z” performed at the time of input is not valid, and “misrecognized character” “te” and “selected character” “2” are not valid. ", And" 2 "is given priority as a recognition result.
It does not become the intended letter "Z".

An object of the present invention is to make it possible to learn a plurality of intended characters in a handwritten character recognition system when input of different characters is the same first candidate and incorrect recognition. An object of the present invention is to provide a method for learning handwritten recognition characters that can reduce the occurrence of recognition.

[0011]

According to the first aspect of the present invention, the input handwritten character is compared with the basic pattern, and the character having the highest evaluation value indicating the degree of matching is displayed as a recognition result. In the handwriting recognition character learning method used in the handwriting recognition system, when a character different from the display character of the recognition result is designated and confirmed by the user, a recognition candidate obtained by comparing the input handwritten character with the basic pattern. The character codes of each character in the character group are arranged in the order of the evaluation value and stored as learning information together with the character code of the character determined by the user.When recognizing handwritten input characters, the input handwritten characters are compared with the basic pattern. When the learning information is searched by the obtained recognition candidate character group, and a plurality of predetermined recognition candidate characters having at least a high evaluation value match. It is characterized by displaying a corresponding confirmation result character as a recognition result.

[0012] According to a second aspect of the present invention, there is provided a method for learning a handwritten recognition character.
2. The method for learning handwritten recognition characters according to claim 1, wherein the determined result character is a selection result character selected by a user from the recognition candidate character group.

[0013] According to a third aspect of the present invention, there is provided a method for learning a handwritten recognition character.
2. The learning method for handwritten recognition characters according to claim 1, wherein the confirmed result character is one of the recognition candidate character group specified by the user, in addition to the selection result character selected by the user from the recognition candidate character group. It is characterized in that it includes additional designated characters that are not included inside.

According to a fourth aspect of the present invention, there is provided a method for learning a handwritten recognition character.
4. The learning method for handwritten recognition characters according to claim 1, wherein the learning information stores characters of the recognition candidate character group starting from a character having a high evaluation value up to a predetermined number of characters. I have.

According to a fifth aspect of the present invention, there is provided a method for learning a handwritten recognition character.
5. The learning method according to claim 1, 2, 3, or 4, wherein when the learning information is searched by a group of recognition candidate characters obtained at the time of recognizing a handwritten input character, the recognition stored in the learning information. It is characterized in that a confirmed result character corresponding to a case where a majority of characters match from the head of the candidate character group is displayed as a recognition result.

[0016]

Next, embodiments of the present invention will be described in detail with reference to the drawings.

The basic technical idea of the method for learning a handwritten recognition character according to the present invention is that when a recognition error occurs and the user selects another character, the recognition obtained for the user's handwritten input character is performed. All the characters in the candidate character group and the character selected by the user are stored in the learning information in association with each other, and when the same handwritten character is subsequently input, the learning information is searched by the obtained recognition candidate character group. Thus, the character selected by the user before is preferentially displayed as a recognition result.

FIG. 2 is a functional block diagram showing the configuration of a first embodiment of a handwritten character recognition system using the present invention. The handwritten character recognition system in FIG. 2 includes an input device 1 such as a tablet, a CPU unit 2 including a handwritten character recognition program and controlling the entire system, and character information 3 including a basic character pattern serving as a reference for handwritten character recognition.
Configuration 1 and ROM3 which stores learning information 41 and temporarily RAM4 to store variables 42 that holds the information, and an output device 5 provided with a liquid crystal screen or the like for displaying the recognition result and the like on the screen Have been. The handwritten character recognition program controlled on the CPU unit 2 includes an input processing unit 21 for distributing an input from the input device 1 to each functional block, and a recognition process for recognizing a user's handwritten character and creating a recognition candidate character group. A unit 22, a learning processing unit 23 for searching, registering, and updating the learning information 41, and a display processing unit 24 for displaying a recognition result and a group of recognition candidate characters on the output device 5 are included.

Here, the configuration of the learning information 41 will be described with reference to FIG. As shown in FIG. 1A, a plurality of records indicated by a, b,..., M can be registered in the learning information 41.
Each record has a maximum of n + 1 represented by 0, 1... N
Up to components can be registered. Only the character code is stored in each component of each record, and the component that does not store the character code is left empty. The number of records and the number of components that can be registered are arbitrarily set.

.. M0 at the head of each record of the learning information 41, when erroneous recognition occurs and the user selects another character, the user selects and confirms the “selected character”. ] Is followed by the constituent elements a1 to an,
.., m1 to mn store the character codes of all the characters of the recognition candidate character group in order, and the components that do not store the character codes are left empty.

The user inputs the handwritten character "2", and the recognition candidate character group "te, Z, b, 2, z, shi" is recognized and the first character group is recognized.
After the candidate "te" is displayed, the user performs a [recognition candidate display instruction] to display the recognition candidate character group, and the registration content of the learning information 41 when "2" is selected is shown in FIG. It is shown in b). In the head component a0 of the record a of the learning information 41, the character code of “selected character” “2” is stored,
In the following components a1 to a6, character codes of "te, Z, oto, 2, z, shi" are stored in the arrangement order of the recognition candidate character group, and components a7 to an in which the character codes are not stored are empty. It is left as it is.

FIG. 3 is a flowchart showing the overall operation of the handwritten character recognition system of FIG. Hereinafter, FIG.
The details of the character recognition operation will be described with reference to FIG. 3 while referring to FIG.

The input from the user given to the input device 1 is sorted by the input processing section 21 (steps S1 to S4). If the input is [handwritten characters] (step S1), the input is recognized. ROM 3 by the processing unit 22
The character information 31 is read from (step S5), and compared with the input handwritten character, a recognition candidate character group is created and passed to the learning processing unit 23 (step S6).

The learning processing section 23 stores the passed recognition candidate character group in the variable A of the RAM 3 (step S7), and
The learning information 41 is read from M4 (step S8), and a search is made for a character that matches the recognition candidate character group stored in the variable A (step S9). As a result of the search, if there is a match in the learning information 41, the “selected character” stored at the beginning of the record is entered. Each of the candidate characters is stored in the variable B as a recognition result (step S1).
0). The character stored in the variable B as a recognition result is displayed in the input sentence display column of the output device 5 by the display processing unit 24 (step S11), and then stored in the variable C as a display character (step S12). Thus, the recognition processing for the input handwritten character is completed, and the processing from step S1 is repeated for the next input.

FIG. 4 is a flowchart showing the details of the search processing of the learning information 41 in the above-described steps S8 and S9 performed by the learning processing section 23. First, step S4
In 1, it is determined whether or not there is an unprocessed learning information record in the learning information 41. If there is an unprocessed learning information record, it is read out one record at a time (step S 42). It is checked whether or not the recognition candidate characters of the learning information record are the same as the recognition candidate characters of the constituent elements i1 to in (i is the record number a, b... M) (step S43). Is output as the recognition result (step S44), and if they do not match, the process returns to step S41 and repeats until there is no unprocessed learning information record. Then, the first candidate character at the head of the group of recognition candidate characters stored in the variable A is output as a recognition result (step S45).

Returning to FIG. 3, when the character of the recognition result displayed in the above-described steps S1 to S12 is different from the character intended by the user, the user may input [Recognition candidate display] from input device 1. Instruction] (see FIG. 2). This input is determined by the input processing unit 21 (step S4), and the display processing unit 24 confirms that the recognition candidate character group is stored in the variable A (step S1).
3) A list of recognition candidate characters is displayed in a list in the selection candidate display column of the output device 5 (step S14). If the number of characters in the recognition candidate character group is too large to be displayed at one time, the number of characters that can be displayed from the top is displayed, and the remainder is displayed by operating the next page.

If there is a character intended by the user in the displayed recognition candidate character group, the user designates the character and performs [selection of candidate character] (see FIG. 2). This input is
The character determined by the input processing unit 21 (step S3), and the character selected by the display processing unit 24 is displayed in the input sentence display field of the screen (step S15), and the variable C is displayed as a display character.
(Step S12). When the user confirms the displayed character and performs "determination of the recognition result" (see FIG. 2), the input processing unit 21 makes a determination (step S2), and the learning processing unit 23 stores a variable in which the display character is stored. The contents of C and the variable B storing the recognition result are compared (step S1).
6) If not, it is determined that the user has selected a candidate, and the learning information 41 is updated (step S).
17).

FIG. 5 is a flowchart showing details of the update of the learning information performed in step S17. First, it is confirmed whether a learning information record is registered in the learning information 41 (step S51). If it is registered, it is checked whether or not there is the same recognition candidate character group stored in the variable A (step S52). If there is the same, the corresponding learning information record is deleted. (Step S53). Thereafter, the registered learning information record of the learning information 41 is shifted to empty the first record (step S54), and the display character of the variable C and the group of recognition candidate characters of the variable A are stored in the first record of the learning information 41 (step S54). Step S55). If the same recognition candidate character group has not been registered in the determination in step S52, it is checked whether or not an empty record exists in the learning information 41 (step S5).
6) If there is a vacancy, proceed to step S54 to open a first record and register a new recognition candidate character group. If there is no vacancy, delete the last record of the learning information 41 (step S57), and then go to step S54. Proceed to. If no learning information record is registered in the learning information 41, the process proceeds from step S51 to step S57,
A new recognition candidate character group is registered in the first record.

Here, a series of operations described above will be described using a specific case. First, in the state immediately after the initialization in which nothing is registered in the learning information 41, the user inputs the handwritten character “2” (step S1), and the recognition processing unit 22 recognizes the recognition candidate character group “te, Z, Otsu”. , 2, z,
It is assumed that “shi” has been created (steps S5 and S6). The learning processing unit 23 recognizes the recognition candidate character group “te, Z, oto, 2, z,
Is stored in the variable A (step S7), and the learning information 41 is stored.
(Steps S8 and S9), but since nothing has been registered, the recognition candidate character group “te, Z, second, 2, z,
The first character “te” of “shi” is stored in the variable B as a recognition result (step S10), and the display processing unit 24 sets the “te”
Is displayed on the output device 5 and stored in the variable C (step S12).

Since the recognition result "te" different from the handwritten character "2" intended by the user is displayed, when the user instructs the display of the recognition candidate (step S4), the display processing unit 24 stores the recognition candidate in the variable A. It is confirmed that the character group "te, Z, ot 2, 2, z, shit" is stored (step S13), and the recognition candidate character group "te, Z, ot 2, 2, z, shit" of the variable A is stored. Is displayed (step S14). Since the displayed candidate character group contains the intended character "2", if the user selects "2" (step S3), the corresponding input sentence display field in which the recognition result "te" has been displayed until then. The selected character "2" is displayed at the position (step S15), and "2" is stored in the variable C (step S12). When the user confirms that the intended character "2" has been displayed, and confirms the recognition result (step S2), the variable C storing the display character is displayed.
Is different from “2” of the variable B storing the recognition result (step S16), the learning processing unit 23 determines that the user has selected a candidate and updates the learning information 41 (step S16). Step S17).

At this time, since the learning information 41 is completely empty immediately after the initialization as shown in FIG. 1A, the display character "2" is obtained through the paths of steps S51 and S57 shown in FIG. ”And the recognition candidate character group“ te, Z, Otsu,
2, z, ”is stored in the first record of the learning information 41, and the learning information 41 is in the state shown in FIG.

Subsequently, when the user inputs the same handwritten character "2", the recognition candidate character group "te, Z, oto, 2, z, shit" recognized by the recognition processing section 22 and stored in the variable A. Is searched for the learning information 41 (steps S5 to S9).
At this time, the learning information 41 includes a learning information record previously learned as shown in FIG. 1B, and the constituent elements a1 to a6 of the record a are composed of the recognition candidate character groups “te, Z, oto, 2,
Therefore, the character "2" of the first component a0 of the record a is stored in the variable B as a recognition result, and "2" is displayed in the input sentence display column of the output device 5. (Steps S10 to S12). That is, the character "2" intended by the user is displayed as a recognition result by the effect of learning.

Next, it is assumed that the handwritten character "Z" is inputted and the recognition candidate character group "te, 2, Z, oto, ma" is recognized.
The recognition candidate character group "te, 2, Z, oto, ma" is stored in the variable A and the learning information 41 is searched (steps S5 to S5).
9). Since the learning information 41 includes the learning information record shown in FIG. 1B, the components a1 to a6 of the record a are read in steps S41 and S42 in FIG. 4, and the recognition candidate character group “ , 2, Z, Otsu,
”, But because they do not match, step S
Returning to step 41, the next learning information record is searched. But,
Since there is no next learning information record, step S4
In step 5, the first character "te" of the recognition candidate character group "te, 2, Z, oto, ma" is used as a recognition result, which is stored in the variable B, displayed on the output device 5, and then stored in the variable C (step). S10
To S12).

Since the displayed recognition result "te" is different from the intended character "Z", the user sets the recognition candidate character group "te,
2, Z, Otsu, Ma ”, select“ Z ”and execute the confirmation process. As a result,“ Z ”of the variable C storing the display characters and“ te ”of the variable B storing the recognition result are displayed. The difference is confirmed (steps S13 to S16), and the learning information 41 is updated (step S17). At this time, the learning information 41 is shown in FIG.
In the state (b), the learning information record is registered and the empty record remains, so that the record shift for emptying the first record is performed in step S54 through steps S51, S52, and S56 shown in FIG. Figure 1
The components a0 to a6 of the record a shown in FIG.
As shown in (c), the head is moved to the constituent elements b0 to b6 of the record b, and the first record a becomes empty. The display character "Z" and the recognition candidate character in step S55 of FIG. The group “te, 2, Z, Otsu, Ma” is stored, and the learning information 41 has a plurality of records shown in FIG.

Thereafter, when the user inputs the handwritten character “2” and the recognition candidate character group “te, Z, b, 2, z, shi” is created, the record b in FIG. Component b1
In the case where “2” stored in the constituent element b0 becomes the recognition result because the character string matches b6, and the handwritten character “Z” is input and the recognition candidate character group “te, 2, Z, Otsu, Ma” is created , Since it matches the components a1 to a5 of the record a in FIG. 1D, “Z” stored in the component a0 is the recognition result. That is, even if the first character of the recognition candidate character group is both "te", different handwritten characters "2" and "Z" can be learned.

As described in detail above, the learning information 41
Stores all candidate characters of the group of candidate recognition characters for the user's handwritten input characters and the selected characters selected by the user. When the learning information is searched, all the candidate characters of the group of candidate recognition characters are stored. When the character matches the candidate character, the selected character stored in the learning information is displayed as a recognition result. Therefore, even if the first candidate character at the beginning of the recognition candidate character group is the same,
If there is a difference below the second candidate character, a different selected character can be learned.

Next, a second embodiment of the present invention will be described with reference to FIGS. The difference between the second embodiment and the first embodiment is that the characters stored in the first component of the record of the learning information can be registered even if they are not included in the recognition candidate character group. That is.

FIG. 6 is a functional block diagram showing the configuration of a second embodiment of a handwritten character recognition system using the present invention. The handwritten character recognition system shown in FIG. 6 is different from the functional block diagram shown in FIG. 2 in that a handwritten character recognition program controlled on the CPU 2a includes an input processing unit 21a, a recognition processing unit 22, a learning processing unit 23, In addition to the display processing unit 24, when there is no character intended by the user in the group of candidate characters for recognition of handwritten characters, a candidate addition processing unit 25 for adding and learning the character designated by the user is included. The processing unit 21a is different from the input processing unit 21 in FIG. 2 in that a function of distributing an input to the input device 1 to the candidate addition processing unit 25 is provided.

FIG. 7 is a flowchart showing the overall operation of the handwritten character recognition system of FIG. Compared to the flowchart of FIG. 3, steps S38 to S40 related to the processing of the candidate addition processing unit 25 are added. Steps S21 to S37 correspond to steps S1 to S17 in FIG. 3, respectively, and are the same as those in FIG. 3, so that only the processing of candidate addition processing unit 25 will be described below. Candidate addition processing unit 2
5 is activated by the input processing unit 21a when the user performs [additional designation to a recognition candidate] when there is no character intended by the user in the recognition candidate character group for the handwritten input character (step S38). ), A character not included in the recognition candidate character group is designated as an additional character by a method other than handwriting input (step S39), and this additional character is treated in the same manner as a selected character from the recognition candidate character group (step S4).
0), display on the output device 5, and store it in the variable C (step S35, step S32). As a method of specifying the additional character, the input from the keyboard may be used in the case of a device that also uses a keyboard, and in the case of the device without a keyboard, the virtual keyboard is divided and displayed in the selection candidate display area of the screen, and the input pen And a method of inputting by a method of selecting and instructing.

Here, the user inputs the handwritten character “5”, and the recognition processing unit 22 shown in FIG. 6 generates a recognition candidate character group “ra, S, J, j, ri, shi” not including “5”. A specific description will be given by taking the case of creation as an example. Steps S21 and S in FIG.
25, the recognition candidate character group “ra, S,
“J, j, ri, shi” is stored in the variable A in the learning processing unit 23 (step S27) and compared with the learning information 41 (steps S28 and S29).
Assuming that all of (a) is empty, the first character "ra" of the recognition candidate character group is stored in the variable B as a recognition result (step S30), and is stored in the variable C after display (steps S31 and S32). .

Since the displayed "la" is not a character intended by the user, when "recognition candidate display instruction" is performed, the recognition candidate character group "ra, S, J, j, ri, "
Are displayed in a list (steps S23, S33, S3
4). Since the displayed character group of the recognition candidate does not include the character “5” intended by the user, the user performs “designation of addition to the recognition candidate” (step S38), and sets “5” as the additional character. When specified (step S39), the additional character "5" is treated as a selected character (step S40), and is displayed on the output device 5 and stored in the variable C (step S3).
5, S32).

Subsequently, when [determination of the recognition result] is performed (step S22), the learning information 41 is compared after comparing the variable B “la” with the variable C “5” (step S36).
Is updated (step S37), and the learning information 41 includes, as shown in FIG.
The display character “5” is stored in 0, and the recognition candidate character group “ra, S, J, j, ri, shi” is stored in the constituent elements a1 to a6.

When the user again inputs the handwritten character "5" and the recognition candidate character group "ra, S, J, j, ri, shi" is generated (steps S21, S25, S2).
6), the recognition candidate character group “ra, S, J, j, ri, shi” and the character group “ra, S,
J, j, ri, shi ", the learned" 5 "becomes the recognition result (steps S27 to S30).

With the above operation, according to the second embodiment, even if the intended character is not included in the recognition candidate character group, learning is performed by candidate addition processing, and the intended character is recognized from the subsequent handwritten character recognition. It becomes possible to recognize.
As an extended use of this method, it is possible to input handwritten abbreviations not registered in the system and display registered real characters.

In the description of the first and second embodiments, a specific example in which all the character codes of the recognition candidate character group are stored in the learning information has been described. Generally, a recognition candidate character group is composed of characters whose evaluation value exceeds a certain standard, and the number of characters included varies. If the number of components (number of characters) that can be registered in the learning information is set sufficiently large, all characters in the recognition candidate character group can be registered. However, if the number of components is limited in consideration of the memory capacity, the recognition If the number of characters in the candidate character group exceeds n, the character code of the first n characters is stored, and the following is discarded.

In the first and second embodiments described above,
In order to make the basic idea easier to understand, if the recognition candidate character group obtained by the recognition processing unit for the handwritten input characters and the recognition candidate character group stored in the learning information all match, the learning The description has been made so that the confirmation result character (selection confirmation character or additional designation character) stored at the head of the information record is displayed. However, since handwritten characters have a peculiar habit and also have a variation at every input, the effect of learning may not be reflected on condition that all the characters match. To remedy this, it is conceivable that processing is performed not based on matching of all characters but on the condition of matching a plurality of characters with high evaluation values.
Considering that the number of characters in the recognition candidate character group itself varies,
It is a practical method to display a confirmed result character as a recognition result when the character matches a majority character from the head of the recognition candidate character group stored in the learning information.

[0047]

As described above, in the method for learning handwritten recognition characters according to the present invention, a plurality of characters of a group of recognition candidate characters for handwritten input characters and the user are selected or added to the learning information and fixed. Since the determined result characters are stored in association with each other, even if the user inputs different handwritten characters and the first candidate of the recognition candidate character group is the same character and is not the intended character, the second and subsequent candidates It can be identified by the difference between the recognition candidate characters. That is, the first
Even if the candidates have the same character, they can be learned separately and reflected in the subsequent processing, which has the effect of reducing the occurrence of erroneous recognition.

Also, learning is performed for characters that are not included in the recognition candidate character group, and subsequent handwritten characters can be displayed as a recognition result from the input. Therefore, learning can be performed for handwritten characters having a large peculiar peculiarity. Thus, the effect of reducing erroneous recognition and improving input efficiency can be expected.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram showing a data configuration of learning information used in the present invention.

FIG. 2 is a functional block diagram showing a configuration of a first embodiment of a handwritten character recognition system using the present invention.

FIG. 3 is a flowchart showing an operation of the first embodiment of FIG. 2;

FIG. 4 is a flowchart illustrating an operation of a search process of learning information.

FIG. 5 is a flowchart illustrating an operation of a learning information update process.

FIG. 6 is a functional block diagram showing a configuration of a second embodiment of a handwritten character recognition system using the present invention.

FIG. 7 is a flowchart showing the operation of the second embodiment of FIG. 6;

FIG. 8 is an explanatory diagram showing contents of learning information according to the second embodiment of FIG. 6;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input device 2, 2a CPU part 3 ROM4 RAM 5 Output device 21, 21a Input processing part 22 Recognition processing part 23 Learning processing part 24 Display processing part 25 Candidate addition processing part 31 Character information 41 Learning information 42 Variable

Claims (5)

(57) [Claims] 1. A method for learning a handwritten recognition character used in a handwritten character recognition system for comparing an input handwritten character with a basic pattern and displaying a character having the highest evaluation value representing the degree of matching as a recognition result. When a character different from the displayed character is confirmed by the user,
The input handwritten characters are compared with the basic pattern, and the character codes of each character of the group of recognition candidate characters obtained are arranged in the order of the evaluation value, and stored together with the character code of the character determined by the user as learning information, and handwritten. enter at the time of recognition of the input character
The learning information is searched by a group of recognized candidate characters obtained by comparing the input handwritten character with the basic pattern, and a determination result corresponding to a case where a plurality of predetermined recognized candidate characters having at least a high evaluation value match. A method for learning a handwritten recognition character, characterized by displaying a character as a recognition result. 2. The learning method for handwritten recognition characters according to claim 1, wherein the confirmed result characters are selection result characters selected by a user from the recognition candidate character group. 3. The confirmed result character is, in addition to a selection result character selected by the user from the recognition candidate character group, an additional designated character not included in the recognition candidate character group specified by the user. 2. The learning method for handwritten recognition characters according to claim 1, further comprising: 4. The handwriting recognition according to claim 1, wherein the learning information stores characters of the recognition candidate character group in descending order of evaluation value up to a predetermined number of characters. How to learn letters. 5. When the learning information is searched by using a group of candidate characters obtained at the time of recognizing a handwritten input character, when a majority of characters from the head of the group of candidate characters stored in the learning information match, 4. A corresponding confirmed result character is displayed as a recognition result.
Or a method for learning handwritten recognition characters according to 4.