JPH07302305A - Device for recognizing hand-written character - Google Patents

Abstract

PURPOSE:To improve the speed of recognition without reducing the recognition rate. CONSTITUTION:The recognizing device 5 compares and verifies a handwritten character with a character pattern registered in a dictionary 4 and calculates the degree of simularity in each prescribed unit. The degree of simularity is compared with a prescribed threshold, the comparison/verification with the character pattern is stopped in accordance with the compared result and comparison/verification with a succeeding character pattern is started. The prescribed threshold is changed in accordance with the number of strokes of a hand-written character being compared/verified at present.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a handwritten character recognition device for recognizing handwritten characters input by handwriting.

[0002]

2. Description of the Related Art In a conventional handwritten character recognition device,
A comparison test between the input handwritten character and the standard character pattern stored in the dictionary is performed, for example, in a unit of one stroke of the handwritten character, and each time one comparison test is performed,
The degree of similarity of the handwritten character to the character pattern (for example, the closer the handwritten character is to the character pattern, the larger the value) is calculated. Then, the similarity is compared with a predetermined threshold,
When the degree of similarity is smaller than a predetermined threshold value, the process for the character pattern currently undergoing the comparison test is stopped, and the comparison test with the next character pattern is performed.

On the other hand, if the similarity does not become smaller than a predetermined threshold value by the time the comparison test is performed on all strokes of the handwritten character, the character pattern used for the comparison test is the recognition result of the handwritten character. It is designed to be a candidate.

[0004]

By the way, the similarity between a handwritten character and a certain character pattern is, for example, the result of a comparative test up to now (for example, one stroke of a handwritten character is The closer it is to the lines that make up the character pattern corresponding to the stroke,
If the average value of a plurality of specified values is selected from the one with the lowest value among the larger values), the specified number of values must be determined until the comparison test of the number corresponding to the specified value is performed. A predetermined initial value is used as the result of the comparison test that is insufficient.

Since the highest value of the comparison test result is usually used as the predetermined initial value, even if the result of the comparison test for the first stroke of the handwritten character is a small value, the similarity which is the above-mentioned average value. However, there is a problem that it takes time until the comparison and verification with the character pattern that is not so close to the handwritten character is stopped.

As a result, there is a problem in that it takes time to obtain the character recognition result and the user feels troublesome.

Therefore, there is a method of increasing the predetermined threshold value, but if the handwritten character has a characteristic peculiar to the user, for example, the handwritten character is slanted, or the amount of the handwritten character is large, There is a problem that the comparison test for the correct character pattern is easily stopped, and as a result, the recognition rate decreases.

Further, a method of describing in detail the difference from other characters in a dictionary and distinguishing a character pattern that is similar to a handwritten character and a character pattern that is not similar to a handwritten character based on the description However, even in this method, if the handwritten character has a characteristic peculiar to the user, the handwritten character does not match the detailed description in the dictionary, and there is a problem that the recognition rate is lowered.

Further, in this case, there is a problem that enormous amount of time and labor are required to describe the dictionary, the capacity of the dictionary becomes large, and the cost of the apparatus increases.

The present invention has been made in view of such a situation, and it is possible to improve the recognition speed without lowering the recognition rate.

[0011]

A handwritten character recognition apparatus according to claim 1 is a handwritten character recognition apparatus for recognizing a handwritten character input by handwriting. , A dictionary 4 shown in FIG. 1), an extraction unit for extracting the feature amount of the handwritten character (for example, the feature extraction device 2 shown in FIG. 1), and the feature amount extracted by the extraction unit are stored in the storage unit. By comparing and verifying with the standard character pattern that has been calculated, the similarity of the handwritten character to the standard character pattern is calculated,
A verification means (for example, the recognition device 5 shown in FIG. 1) for obtaining character recognition result candidates based on the similarity,
The verification means performs a comparison test for each predetermined unit, calculates a similarity for each predetermined unit, compares the similarity with a predetermined threshold, and then, according to the comparison result, a standard character It is characterized in that the comparison test with the pattern is stopped, the comparison test with the next standard character pattern is started, and the predetermined threshold value is changed in a predetermined case.

A handwritten character recognition device according to a second aspect is
The verification means is characterized in that the predetermined threshold value is changed in accordance with the number of strokes of the handwritten character for which the comparison test is being performed.

A handwritten character recognizing apparatus according to claim 3 is
The verification means is characterized in that the predetermined threshold value is set to a value at which the comparison verification is more likely to be stopped for the stroke at the beginning of writing the handwritten character than for the subsequent strokes.

A handwritten character recognizing device according to a fourth aspect is
The verification means compares the total number of strokes of the handwritten character with a predetermined number of strokes, and according to the comparison result, a predetermined threshold value,
It is characterized in that the comparison test is hard to be stopped over the entire stroke of the handwritten character.

[0015]

In the handwritten character recognition device according to claim 1, the comparison test between the standard character pattern and the handwritten character is performed.
The process is performed for each predetermined unit, and the similarity is calculated for each predetermined unit. Then, the similarity is compared with a predetermined threshold value changed in a predetermined case, and according to the comparison result, the comparison test with the standard character pattern is stopped, and the next standard character pattern is compared. The comparison test is started. Therefore, for example, it becomes possible to quickly stop the comparison test for a character pattern that is not close to the trajectory of the handwriting character at the beginning of writing, and as a result, improve the recognition speed without lowering the recognition rate. You can

In the handwritten character recognition device according to the second aspect of the present invention, the predetermined threshold value is changed corresponding to the number of strokes of the handwritten character for which the comparison test is being performed. Therefore,
For example, it becomes possible to quickly stop the comparison test for a character pattern that does not approximate the stroke at the beginning of handwriting, and as a result, the recognition speed can be improved without lowering the recognition rate. .

In the handwritten character recognition apparatus according to the third aspect, the predetermined threshold value is set to a value at which the comparison verification is more likely to be stopped for the stroke at the beginning of writing the handwritten character than for the subsequent strokes. It Therefore, the comparison and verification for the character pattern that does not approximate the stroke at the beginning of the handwritten character is stopped early, so that the recognition speed can be improved without lowering the recognition rate.

In the handwritten character recognition device according to the present invention, the total number of strokes of the handwritten character is compared with a predetermined stroke number, and a predetermined threshold value is set for all strokes of the handwritten character according to the comparison result. , It is set to a value that makes it difficult to cancel the comparison test. Therefore, for example, it is possible to prevent erroneous recognition due to early termination of comparison and verification for handwritten characters with a small number of strokes.

[0019]

1 is a block diagram showing the configuration of an embodiment of a handwritten character recognition device to which the present invention is applied. The input pen 1a is operated when inputting handwritten characters into the handwritten character input device 1. The handwritten character input device 1 is, for example, a tablet, and detects the coordinates of a point sequence forming a trajectory of a handwritten character drawn (written) with the pen 1a on the tablet, and outputs it to the feature extraction device 2. It is done like this. Further, the handwritten character input device 1 detects the pen-down and pen-up of the pen 1 a and outputs it to the feature extraction device 2.

The handwritten character input device 1 is composed of, for example, a tablet made of a transparent member and is integrated with a display device (not shown).

The feature extraction device 2 is a handwritten character input device 1.
From the coordinates of the point sequence forming the trajectory of the handwritten character from, the feature amount of the handwritten character is calculated, for example, by one stroke (pen 1
a is extracted in units of a locus drawn from pen-down to pen-up and is output to the feature buffer 3. In addition, the feature extraction device 2 uses the pen 1
When a is not pen-downed for a predetermined time after pen-up, it is determined that handwriting input of one character is completed, and an end signal indicating the end of inputting one character is output to the recognition device 5. It is designed to do.

The handwritten character input device 1 may be provided with a predetermined number of character input frames, and handwritten characters may be input by the pen 1a into the character input frames. When the character input box in which 1a is pen-down is different from the character input box in which the pen-down is performed last time, it is determined that the handwriting input of one character is completed, and the end signal indicating the end of the input of one character is input to the feature extraction device 2. Can be output.

The feature buffer 3 is adapted to temporarily store the feature amount from the feature extracting device 2. Recognition device 5
When the end signal is received from the feature extraction device 2, reads out the feature amount of the handwritten character stored in the feature buffer 3,
A comparison and verification is performed with the characteristic amount of a standard character pattern registered in the dictionary 4 in advance. Further, the recognition device 5 determines, based on the comparison test result, the similarity of the handwritten input character to each of the character patterns registered in the dictionary 4 (the closer the handwritten character is to the character pattern, the larger the value. Is calculated, a character pattern as a recognition result candidate is obtained from the similarity, and the character pattern is output to the recognition candidate buffer 7. In addition,
The recognition device 5 includes a ROM that stores a recognition program for performing a recognition process described later, a CPU that performs processing according to the program stored in the ROM, and a RAM that temporarily stores data necessary for the operation of the CPU. (None of them are shown).

The comparison result buffer 6 is adapted to temporarily store the intermediate result of the comparison test output from the recognition device 5. The recognition candidate buffer 7 stores the recognition result candidates (character codes described later) output from the recognition device 5 together with the degree of similarity.

Next, the operation will be described. Pen 1
When a is operated and a handwritten character is input to the handwritten character input device 1, the data of the handwritten character (coordinates of the point sequence forming the trajectory of the handwritten character) is stroked in units of the handwritten character input device. 1 to the feature extraction device 2. In the feature extraction device 2, the feature amount of the handwritten character from the handwritten character input device 1 is extracted in units of strokes, output to the feature buffer 3 and stored.

That is, the feature extraction device 2 stores a basic stroke table as shown in FIG. 2, for example, and one stroke of handwritten input from the handwritten character input device 1 is
A ratio (hereinafter referred to as a shape determination value) that can be regarded as each of the four basic strokes as shown in the figure is calculated. It should be noted that basic stroke codes 0 to 3 as unique codes are respectively attached to the four basic strokes, and the shape determination value is output to the feature buffer 3 and stored together with the basic stroke code. Has been done.

Further, the feature extraction device 2 detects the direction of the current stroke with respect to the previous stroke, and
In other words, the quantized one is also extracted as the feature amount of the handwritten character. Specifically, for example, the direction of each of the start point (coordinates at which pen 1a is pen-down) or the end point (coordinates at which pen 1a is pen-down) or the end point (coordinates at which pen 1a is pen-up) relative to the start point of the previous stroke (coordinates at which pen 1a is pen-down). (Hereinafter, each is referred to as a start point-start point, or a start point-end point direction as appropriate), and the direction of each of the start point and the end point of the current stroke with respect to the end point of the previous stroke (hereinafter, respectively, as appropriate, end point-start point, or end point). The direction of the end point) is detected. Then, these four directions (start point-start point, start point-end point, end point-start point,
End direction-direction of each end point) It is detected which one of the directions to which the direction code as a unique code shown in FIG. 3 is closest, and the direction code given to the detected direction is detected. Is output from the feature extraction device 2 to the feature buffer 3 and stored therein.

For example, when handwritten characters "up" are input in the order of numbers as shown in FIG. 4 in the direction of the arrow, the feature amount as shown in FIG. 3 is stored.

In FIG. 5, the number of strokes means
The order of strokes input by handwriting is shown (Fig. 4
Corresponds to the number attached to the stroke shown in). Further, in FIG. 5, the maximum shape determination value is 10
It is set to 0, and is set to a smaller value as the proportion of strokes input by handwriting becomes closer to the basic stroke.

Further, the direction code is, as described above,
Since it corresponds to the direction of this stroke relative to the previous stroke, if the number of strokes is 1,
It is not stored in the feature buffer 3 (the feature extraction device 2 does not extract the direction (direction code) for the first stroke of the handwritten character).

As described above, the feature amount of the handwritten character (the shape determination value and the direction code in this embodiment).
Is extracted in units of one stroke and stored in the feature buffer 3, and thereafter, when the end signal is output from the feature extraction device 2 to the recognition device 5, that is, the feature amount for all strokes of the handwritten character is output to the feature buffer 3. When stored, the recognition device 5 performs the character recognition process of the handwritten character while referring to the dictionary 4.

FIG. 6 shows the structure of the dictionary 4. The dictionary 4 is a unique code added to a standard character pattern (however, as will be described later, the recognition candidate buffer 7
Character code (excluding the code used as the initial value for clearing), the basic stroke code of the basic stroke required to write (draw) that character (Fig. 2), when writing characters in the correct stroke order It is composed of a direction code (FIG. 3) indicating a direction between continuously input basic strokes and an end code. The basic stroke codes are described in the order corresponding to the correct writing order when writing the character, and between the basic stroke codes corresponding to the continuous basic strokes, the four direction codes between the continuous basic strokes ( Start point-start point, start point-end point, end point-start point, end point-end point direction code) are described.

The end code is described after the last basic stroke code of one stroke, and further, the character code is described after that.

The above-described description patterns are repeated in the dictionary 4, and various standard character feature amounts are registered in advance.

The direction code is represented by 16 kinds of codes of 0 to 15 as shown in FIG. 3, and in the dictionary 4, the direction code is represented by a hexadecimal number. Further, in the dictionary 4, the direction code is described in the place of 16 ⁰ , and when the direction code indicates the direction of start point-start point, start point-end point, end point-start point, or end point-end point, , 1 to 4 are respectively described in the 16 ¹ position thereof.

Therefore, in the dictionary 4, starting point-starting point, starting point-
The direction code of end point, end point-start point, or end point-end point is
10h to 1Fh, 20h to 2Fh, 30h to 3F
h, or a value in the range of 40h to 4Fh, respectively (h represents a hexadecimal number).

Further, the end code is also represented by a hexadecimal number, and its value is, for example, 50h.

Since the user does not always input handwritten characters in the correct stroke order, the dictionary 4 is
In addition to the correct writing order, the basic stroke code and the like are separately described in the order corresponding to the expected incorrect writing order.

The recognition device 5 recognizes handwritten characters from the feature quantity stored in the feature buffer 3 using the dictionary 4 as described above, for example, according to the flowchart shown in FIG.

That is, first of all, in step S1,
Initialization of the recognition candidate buffer 7 and read address for the dictionary 4 (hereinafter, appropriately referred to as dictionary read address)
Character recognition start processing such as initialization of
2, the current character verification start process is performed every time the processes of steps S3 to S6 described later are started for the characters registered in the dictionary 4, and the process proceeds to step S3.

In step S3, the comparison test of the feature amount of the handwritten character stored in the feature buffer 4 with respect to the standard character pattern data stored in the dictionary 4 is performed by, for example, the description unit of the dictionary 4 (shown in FIG. 6). It is performed for each column in which the basic stroke code and the direction code of the dictionary 4 are described).
In step S3, before the above-described processing, the comparison test of handwritten characters is first performed for all the descriptions of the character patterns registered in the dictionary 4, which are the current character patterns (described later). It is determined whether or not it has been broken.

If it is determined in step S3 that the comparison test of the handwritten characters has not been performed for all the descriptions of the above-mentioned character patterns, the process proceeds to step S4 and the comparison test results up to now are used. It is determined whether to continue or stop the comparison test for the character pattern of the dictionary 4 referred to.

When it is determined in step S4 to continue the comparison test for the character pattern of the dictionary 4 which is being referred to, the process returns to step S3 and the processes from step S3 are repeated. When it is determined in step S4 that the comparison test for the character pattern of the dictionary 4 that is being referred to is to be stopped, the process proceeds to step S6.

On the other hand, if it is determined in step S3 that the comparison test of the handwritten characters has been performed for all the above described character pattern descriptions, the process proceeds to step S5, and the rewriting process of the recognition candidate buffer 7 described later is performed. After that, the process proceeds to step S6.

In step S6, it is determined whether or not there is a character pattern in the dictionary 4 that has not been compared and verified with handwritten characters. If it is determined in step S6 that there is a character pattern in the dictionary 4 that has not been compared and verified with the handwritten character, the process returns to step S2, the dictionary read address is updated, and step S is performed again.
The process from 2 is repeated. On the other hand, in step S6, when it is determined that the character pattern that has not been subjected to the comparison test with the handwritten character is not in the dictionary 4, the process ends.

The processing in step S1 or S2 is
This corresponds to the processing of steps S100 and S101 or S102 to S104 of FIG. 8 described later, respectively. Further, the process of step S3 is the process of steps S105 to S109 of FIG. 8 and steps S110 to S116 of FIG. 14 described later, and the process of step S4 is the process of steps S120 to S131 of FIG. 12 described later, and step S5. The process of step S140 of FIG.
Through 155, the process of step S6 corresponds to the process of steps S160 and S161 of FIG.

Next, the details of the processing shown in the flowchart of FIG. 7 will be described with reference to the flowcharts of FIGS. 8, 12 to 14, and 16.

First, in step S100 of FIG. 8, the recognition candidate buffer 7 is initialized.

Here, in the recognition candidate buffer 7, for example, as shown in FIG. 9, a region of a variable B _P for storing a character code of a recognition result candidate of a handwritten character and a similarity of the handwritten character with respect to the character. , Reserved in advance. This variable B _P is an array variable, and the variable B _P [i] has its index (subscript) (subscript) (number) i.
The smaller the number is, the larger the character code having the higher degree of similarity is stored. In FIG. 9, the index (number) i can take a value in the range of 0 to 5, for example. That is, in the variable B _P [i], six character codes can be stored as the recognition result candidates together with the similarity in descending order of similarity. The number of character codes that the variable B _P [i] (recognition candidate buffer 7) can store together with the degree of similarity (in the present embodiment, 6 as described above).
Hereinafter, the number of candidate buffers will be referred to as appropriate.

In step S100, the character code of the variable B _P [0] is set to, for example, 0 as an initial value (0 is assigned to the area for storing the character code of the variable B _P [0]). By doing so, the recognition candidate buffer 7 is initialized, the process proceeds to step S101, the dictionary read address is set to, for example, the first address of the dictionary 4, and the process proceeds to step S102.

In step S102, the comparison result buffer 6 is initialized.

Here, in the comparison result buffer 6, for example, as shown in FIG. 10, an area of a variable B _C for storing a comparison result value obtained as a result of comparison test of handwritten characters is secured in advance. This variable B _{C is} also the above-mentioned variable B
Similar to _P [i], it is an array variable, and the variable B _C [j] stores a smaller comparison result value as the index (number) j increases. Figure 1
At 0, the index j can take values in the range of 0 to 3, for example. That is,
Only four comparison result values can be stored in the variable B _C [j].

The comparison result values stored in the variables B _C [0] to B _C [3] are arithmetically averaged as described later, and the average value obtained as a result is taken as the similarity, Variable B
The number of comparison result values that can be stored in _C [j] (4 in this embodiment as described above) is hereinafter referred to as an averaged number as appropriate.

In step S102, the variable B
_The comparison result buffer 6 is initialized by setting, for example, the highest comparison result value (for example, 100 in this embodiment) to all _C [j] as an initial value.

That is, in step S102, as shown in the flowchart of FIG. 11, first, in step S200, the index i of the variable B _C [j] (hereinafter referred to as a comparison result buffer counter) i is cleared (set to 0). ), The process proceeds to step S201, the initial value (100 as described above) is assigned to the variable [j], and step S2
Go to 02.

In step S202, the comparison result buffer counter j is incremented by 1, and in step S2
In 03, it is determined whether the comparison result buffer counter j is smaller than the averaged number. When it is determined in step S203 that the comparison result buffer counter j is smaller than the averaged number, that is, when the initial values have not been set in all the variables B _C [j], the process returns to step S201 and step S203. In, the processing of steps S201 to S203 is repeated until it is determined that the comparison result buffer counter j is not smaller than the averaged number.

On the other hand, if it is determined in step S203 that the comparison result buffer counter j is not smaller than the averaged number, that is, if the initial values are set in all the variables B _C [j], the processing is terminated. To do.

Returning to FIG. 8, after the comparison result buffer 6 is initialized in step S102 as described above, the process proceeds to step S103, and the read address of the feature buffer 3 (hereinafter referred to as a buffer read address as appropriate). Is set to the address where the basic stroke code and the direction code of the first stroke of the handwritten character are stored (the head address of the data of the line in which the number of strokes is 1 shown in FIG. 5), and the process proceeds to step S104. The current stroke counter (area previously reserved in the RAM configuring the recognition device 5) as a variable for counting the stroke number of the stroke of the handwritten character to be compared and verified is cleared (set to 0). , And proceeds to step S105.

In step S105, data (basic stroke code, direction code, or end code) from the address indicated by the dictionary read address in the dictionary 4 (hereinafter,
As appropriate, these are collectively referred to as a characteristic code), and the process proceeds to step S106 to determine whether the characteristic code read in step S105 is 0Fh or less.

If it is determined in step S106 that the characteristic code is 0Fh or less, that is, if the characteristic code is the basic stroke code, step S1.
In step 07, the shape determination value corresponding to the basic stroke code is read out as the comparison result value from the data of the row of the stroke number indicated by the buffer read address.

That is, for example, the feature quantity shown in FIG. 5 is stored in the feature buffer 3 as the dictionary 4,
When the one shown in FIG. 6 is used, when the process of step S107 is performed for the first time, step S10 is performed.
5, the basic stroke code 1 described at the beginning of the character code “3” is read from the dictionary 4, so that the shape determination value of the row having the number of strokes of 1 and the column having the basic stroke code of 1 is 0. Is read from the feature buffer 3 as the comparison determination value.

After that, in step S107, the shape determination value as the comparison determination value read from the feature buffer 3 is secured in advance in the comparison evaluation register (RAM constituting the recognition device 5) built in the recognition device 5. Area) and the process proceeds to step S108.

In step S108, the buffer read address is set to the address in which the basic stroke code and the direction code of the next stroke are stored (for example, the buffer read address is the row of which the number of strokes is 1 shown in FIG. 5). If it is set to the head address of the data, the number of strokes is set to the head address of the data of the line 2), the process proceeds to step S109, the current stroke counter is incremented by 1, and
Go to step S120 shown in.

In step S120 of FIG. 12, the variable B _C
It is determined whether or not the comparison result value stored in [0] is less than or equal to the comparison result value stored in the comparison evaluation register.

Here, in the variables B _C [0] to B _C [3] secured in the comparison result buffer 6, the comparison result values of handwritten characters are set in descending order. In this embodiment, the closer the stroke of the handwritten character is to the stroke of the character pattern corresponding to the stroke, the larger the comparison result value is. The maximum value of the comparison result value is 10 as described above.
It is set to 0, and its minimum value is set to 0, for example.

Further, in the present embodiment, as a result of the comparison test with respect to the basic stroke code described in the dictionary 4, the shape determination value of each stroke constituting the handwritten character read from the feature buffer 3 and the direction between the strokes are obtained. What is obtained as described below from the direction difference obtained as a result of the comparison test between the code and the direction code stored in the dictionary 4 (hereinafter appropriately referred to as direction difference determination value) is also used as the comparison result value. It is done like this. Recognition device 5
The direction difference determination value obtained in (Step S116 of FIG. 13 described later) and the shape determination value obtained by the feature extraction device 2 as described above are determined to have the same weight when calculating the similarity. However, it is possible to increase the weighting of any of these.

In the recognition device 5, for a certain character pattern registered in the dictionary 4, for example, the arithmetic mean of four small comparison result values obtained from handwritten characters is
The similarity of the handwritten character to the character pattern is set. Therefore, when calculating the similarity of a handwritten character with respect to a certain character pattern registered in the dictionary 4, first, the comparison result value obtained from the handwritten character is the 1st to 4th smallest for that character pattern. You need to ask for things.

Therefore, as described above, step S12
At 0, the variable B is set so that the maximum value of the four comparison result values from the smallest one is set.
The stored value (comparison result value) of _C [0] is compared with the comparison result value stored in the comparison evaluation register.

Then, in step S120, the comparison result value stored in the comparison evaluation register is the variable B.
If it is determined that the comparison result value stored in _C [0] is greater than or equal to the comparison result value, that is, the comparison result value obtained for a certain character pattern has been obtained so far for that character pattern. If it is greater than or equal to the fourth smallest value among the comparison result values, the comparison result value stored in the comparison evaluation register is set to the variable B _C [j] (B _C [0] to B _C [3]). Since it does not need to be set to either), step S1
21 to S131 without performing the processing of S131 of FIG.
Return to 105.

That is, in this case, steps S121 through S
Since the processing of 131 is skipped, the recognition speed can be improved.

On the other hand, in step S120, the comparison result value stored in the comparison evaluation register is the variable B _C [0].
If it is determined that it is not greater than or equal to the comparison result value stored in,
That is, when the comparison result value obtained for a character pattern is smaller than the fourth smallest comparison result value obtained so far for the character pattern, the variable B _C [ 0] to B _C [3], and the rest of the comparison result values stored in the comparison evaluation register excluding the stored value of the variable B _C [0], which is the maximum value, The processing of rearranging in descending order and setting the variables B _C [0] to B _C [3] is performed in steps S121 to S126.

That is, first, in step S121, a comparison result buffer counter (which is an index of the variable B _C [n] (where n = 0, 1, 2, 3) is secured in advance in the RAM constituting the recognition device 5). Area n) is cleared (set to 0), the process proceeds to step S122, and whether or not the comparison result value stored in the variable B _C [n + 1] is less than or equal to the comparison result value stored in the comparison evaluation register. Is determined. In step S122, the variable B _C [n +
1], it is determined that the comparison result value stored in the comparison evaluation register is less than or equal to the comparison result value stored in the comparison evaluation register, that is, the comparison result value stored in the comparison evaluation register is the variable B _C.
If it is greater than or equal to the stored value of [n + 1] and smaller than the stored value of the variable B _C [n], steps S123 to S125 are skipped and the process proceeds to step S126.

In step S122, the variable B
_When it is determined that the comparison result value stored in _C [n + 1] is not less than or equal to the comparison result value stored in the comparison evaluation register, the process proceeds to step S123, and variable B _C [n] is replaced with variable B _C [n + 1]. Stored value is set, and step S12
Go to 4. In step S124, the comparison result buffer counter n is incremented by 1, and step S12
In step 5, it is determined whether the comparison result buffer counter n is smaller than the averaged number (4) -1. Step S
In 125, when it is determined that the comparison result buffer counter n is smaller than the averaged number −1, that is, the variable B _C
If some of the stored values of [0] to B _C [3] are not compared with the stored value of the comparative evaluation register, the process returns to step S122.

On the other hand, when it is determined in step S125 that the comparison result buffer counter n is not smaller than the averaged number -1, that is, the variables B _C [0] to B _C [3].
When the comparison between each of the stored values of No. 1 and the stored value of the comparative evaluation register is completed and the stored value of the comparative evaluation register is smaller than the stored value of variable B _C [3], step S126
In step S127, the variable B _C [n] is set to the comparison result value stored in the comparison evaluation register.

In step S127, the comparison result value stored in the comparison result buffer 6, that is, the variable B _C [0].
To B _C [3], the average value (arithmetic mean value) of the comparison result values stored in each of them is calculated as the degree of similarity, and the process proceeds to step S128, where the total number of strokes of the handwritten character is smaller than the predetermined designated number of strokes. It is determined whether or not.

If it is determined in step S128 that the total number of strokes of the handwritten character is not smaller than the predetermined designated stroke number, the process proceeds to step S129, and it is determined whether or not the stored value of the current stroke counter is smaller than the predetermined designated stroke number. Is determined. If it is determined in step S129 that the stored value of the current stroke counter is not smaller than the predetermined number of strokes, that is, if the stroke of the handwritten character currently undergoing the comparison test is the stroke at the beginning of writing the handwritten character. If not, the process proceeds to step S130.

If it is determined in step S129 that the stored value of the current stroke counter is smaller than the predetermined number of strokes, that is, the stroke of the handwritten character currently undergoing the comparison test is the start of writing the handwritten character. If the stroke is one of the strokes, the process proceeds to step S131, and the similarity calculated in step S127 (for a certain character pattern, one of the first to fourth smallest comparison result values of handwritten characters obtained so far is It is determined whether the average value) is less than or equal to a predetermined threshold value (limit value) B.

In step S131, step S1
When it is determined that the similarity calculated in 27 is less than or equal to the predetermined threshold B, the process proceeds to step S160 shown in FIG. Further, in step S131, step S12
When it is determined that the similarity calculated in 7 is not less than or equal to the predetermined threshold value B, the process returns to step S105 in FIG.

On the other hand, if it is determined in step S128 that the total number of strokes of the handwritten character is smaller than the predetermined number of strokes, that is, if the number of strokes of the handwritten character is small, the process proceeds to step S130 and is calculated in step S127. It is determined whether the similarity is equal to or less than a predetermined threshold value (limit value) A.

In step S130, step S1
When it is determined that the similarity calculated in 27 is not less than or equal to the predetermined threshold value A, the process returns to step S105 in FIG. If it is determined in step S130 that the similarity calculated in step S127 is equal to or less than the predetermined threshold value A, the process proceeds to step S160 shown in FIG.

In step S160 of FIG. 13, it is determined whether or not the handwritten character comparison test has been performed on all the character patterns registered in the dictionary 4. Step S
In 160, when it is determined that the comparison test of the handwritten character has not been performed for all the character patterns registered in the dictionary 4, the process proceeds to step S161, and the dictionary read address is set to the next character pattern. The address is set (updated) to indicate the address, and step S102 in FIG.
Return to.

Therefore, when the similarity calculated in step S127 of FIG. 12 is less than or equal to the predetermined threshold value A or B, the comparison and verification of a certain character pattern with all strokes of the handwritten character is completed. Even if it does not exist, the comparison test is stopped (abort), and the comparison test for the next character pattern is performed.

In step S160, the dictionary 4
When it is determined that the handwritten character comparison test has been performed on all the character patterns registered in step S3, the processing ends.

Here, the threshold value A is, for example, a value similar to that of the conventional apparatus, that is, a value at which the comparison test is hard to be stopped (as described above, when the range of the comparison result value is 0 to 100,
(For example, 50) and the threshold B is equal to the threshold A
It is set to a larger value, that is, a value at which the comparison test is easily stopped (for example, 80 when the comparison result value range is 0 to 100 as described above).

As described above, with respect to a character pattern that is not approximate to a handwritten character, the similarity in the stroke at the beginning of writing the handwritten character does not become so small, but in step S131, the threshold value similar to that in the conventional case is used. If it is compared with a threshold value B larger than A, and the degree of similarity is equal to or less than the threshold value B, the comparison test for the character pattern (hereinafter, appropriately referred to as the current character pattern) that is the target of the comparison test is stopped, and the next The comparison test for the character pattern of is started.

Therefore, the comparison test for a character pattern that is not close to a handwritten character is stopped at the stroke at the beginning of writing the handwritten character, so the number of times the comparison test is executed is reduced, and as a result, the recognition speed is increased. Can be improved.

For the strokes of the handwritten character having a predetermined number of strokes or more, the similarity to the current character pattern is compared with the threshold value A similar to the conventional case in step S130, and the comparison result is determined. , It is determined whether to stop or continue the comparison test with the current character pattern.

Therefore, it is possible to prevent the recognition rate from decreasing.

That is, in this case, as compared with the conventional device,
The recognition speed can be improved without lowering the recognition rate.

Further, as described above, step S12
In 8, the total number of strokes of the handwritten character (the total number of strokes forming the handwritten character) is compared with a predetermined number of strokes, and the total number of strokes (the number of strokes) forming the handwritten character is
When the number of strokes is smaller than the predetermined number of strokes, that is, when the total number of strokes forming a handwritten character is small, the similarity of the handwritten character is compared with the threshold value A similar to the conventional case in step S130. It is also possible to prevent erroneous recognition due to early termination of the comparison test for handwritten characters with few characters.

Returning to FIG. 8, in step S106,
If it is determined that the feature code is not 0 Fh or less, that is, if the feature code is either the direction code or the end code, the process proceeds to step S110 in FIG. 14, and the value of the feature code is 10h to 1Fh. Values in the range, values in the range 20h to 2Fh, values in the range 30h to 3Fh, values in the range 40h to 4Fh, and 50h
Which of the above is determined.

In step S110, the feature code is a value in the range of 10h to 1Fh, a value in the range of 20h to 2Fh, a value in the range of 30h to 3Fh, or 40h.
If it is determined that the value is in the range of 4Fh to 4Fh, that is, if the feature code is a start point-start point, start point-end point, end point-start point, or end point-end point direction code, step S1.
11 to S114, respectively, starting point-starting point, starting point-
The direction code of end point, end point-start point, or end point-end point is
The data is read from the data (FIG. 5) of the row having the stroke number indicated by the buffer read address.

That is, for example, when the feature amount shown in FIG. 5 is stored in the feature buffer 3 now, when the buffer read address indicates a line with the stroke number 2, the feature buffer 3 outputs 12, 13, 3, or 2 are read. Further, when the buffer read address indicates a row having the stroke number 3, 10, 13, 9, or 12 is read from the feature buffer 3.

In steps S111 to S114,
After the direction code is read from the feature buffer 3, the process proceeds to step S115, and the direction indicated by the direction code that is the feature code read from the dictionary 4 (hereinafter, appropriately referred to as a standard direction) and steps S111 to S114. In either case, the direction difference from the direction indicated by the direction code read from the feature buffer 3 (hereinafter, appropriately referred to as the input direction) is obtained.

Here, the direction difference is the difference between the direction indicated by the direction code read from the dictionary 4 (standard direction) and the direction indicated by the direction code read from the feature buffer 3 (input direction). The degree is indicated, and in this embodiment, for example, 0
It is adapted to be represented by a value in the range of 8 to 8. That is, when the standard direction and the input direction match, the direction difference is set to 0, and the larger the deviation between the standard direction and the input direction, the larger the direction difference. When the deviation between the standard direction and the input direction is the largest, the direction difference is set to 8.

When the direction difference is obtained, step S116.
Then, for example, the table showing the correspondence between the direction difference and the comparison result value shown in FIG. 15 is referred to, and the direction difference is converted into the comparison result value. According to the table shown in FIG. 15, the direction difference is converted into a larger comparison result value as the value is smaller, that is, the input direction is closer to the standard direction.

When the direction difference is converted into the comparison result value, the result shown in FIG.
The process proceeds to step S120 of 2 and the same process as the above-described process is performed.

On the other hand, when it is determined in step S110 of FIG. 14 that the characteristic code is 50h, that is, the characteristic code read from the dictionary 4 in step S105 of FIG. If the comparison and verification process of all strokes of handwritten characters for the pattern is completed normally without being interrupted,
The character code of the current character pattern is read from the dictionary 4. Then, the process proceeds to step S140 in FIG. 16, and the comparison result value stored in the comparison result buffer 6, that is, the variable B
Among the comparison result values of all strokes of the handwritten character with respect to the current character pattern stored in _C [0] to B _C [3], the average value of the first to fourth smallest values is the current character pattern of the handwritten character. Is calculated as the degree of similarity to, and the process proceeds to step S141.

In step S141, the variable B _P [k]
(However, k = 0, 1, 2, 3, 4, 5) recognition candidate buffer counter (area reserved in advance in the RAM configuring the recognition device 5) k is cleared (set to 0). , Step S142, the variable B
_The character code stored in _P [k] (the storage value of the area reserved for storing the character code of the variable B _P [k]) is
It is determined whether or not it is 0 which is an initial value used when clearing the recognition candidate buffer 7.

Here, the initial value 0 used when clearing the recognition candidate buffer 7 is the variable B, as will be described later.
It is also used as an end code for indicating the end of the character recognition result candidate stored in _P [k].

In step S142, the variable B
_When it is determined that the character code stored in _P [k] is not 0, that is, the character code corresponding to the recognition result candidate is already stored in the variable B _P [k] together with its similarity. In this case, the process proceeds to step S143, and the variable B
_The similarity stored in _P [k] together with the character code (the storage value of the area reserved for storing the similarity of the variable B _P [k]) is the current character pattern calculated in step S140. It is determined whether the degree of similarity is smaller than the similarity. When it is determined in step S143 that the similarity stored in the variable B _P [k] is not lower than the similarity of the current character pattern, the process proceeds to step S144, the candidate buffer counter k is incremented by 1, and the step S1
Return to 42.

In step S143, the variable B
_When it is determined that the similarity stored in _P [k] is smaller than the similarity of the current character pattern, step S145
And the candidate shift counter (R that constitutes the recognition device 5
Variable i as an area secured in advance in AM)
Is set to the minimum value k (recognition candidate buffer counter k) of the index of the variable B _P [k] in which the similarity smaller than the similarity of the current character pattern is stored, and the process proceeds to step S146.

In step S146, the candidate shift counter i is incremented by 1, and the process advances to step S147 to determine whether the character code stored in the variable B _P [i] is 0. When it is determined in step S147 that the character code stored in the variable B _P [i] is not 0, the process returns to step S146 and step S14.
Until it is determined in step 7 that the character code stored in the variable B _P [i] is 0, steps S146 and S147.
The process of is repeated.

If it is determined in step S147 that the character code stored in the variable B _P [i] is 0, that is, the character recognition result candidate stored in the recognition candidate buffer 7 (variable B _P ) is detected. If the end of is found,
In step S148, it is determined whether the candidate shift counter i is smaller than the number of candidate buffers (as described above, the variable B _P (recognition candidate buffer 7) is the number of character codes that can be stored together with the degree of similarity) −1. Is determined.

In step S148, when it is determined that the candidate shift counter i is smaller than the number of candidate buffers−1, that is, when the recognition candidate buffer 7 has a space for storing the character code together with the similarity (variable B
_{For all P} [i] to _BP [(number of candidate buffers-1)],
(If the character code and the similarity are not stored yet), the process proceeds to step S149, and the variable B _P [i + 1]
The stored value of the variable B _P [i] is transferred to and stored in, and the process proceeds to step S150. In step S150, the candidate shift counter i is decremented by 1,
Proceeding to 151, it is determined whether the candidate shift counter i is less than or equal to the recognition candidate buffer counter k.

In step S151, when it is determined that the candidate shift counter i is equal to or larger than the recognition candidate buffer counter k, that is, the variable B whose index i is equal to or larger than k.
The similarity smaller than the similarity of the current character pattern stored in _P [i] and the character code corresponding to the similarity are set in a variable B _P [i] having an index i of k + 1 or more,
If the sequential transfer is not yet completed, the process returns to step S149.

If it is determined in step S151 that the candidate shift counter i is not greater than or equal to the recognition candidate buffer counter k, that is, the index i is greater than or equal to k, the current character pattern stored in the variable B _P [i]. The similarity smaller than the similarity and the character code corresponding to the similarity, the variable B whose index i is k + 1 or more.
_When the sequential transfer to _P [i] is completed, the process proceeds to step S153, and the similarity between the current character pattern and the closest similarity variable B is stored.
The similarity of the current character pattern is set in _P [k] together with the character code, and the above-described step S1 of FIG. 13 is performed.
Proceed to 60.

On the other hand, in step S148, when it is determined that the candidate shift counter i is not smaller than the number of candidate buffers−1, that is, when the recognition candidate buffer 7 has no free space for storing the character code together with the similarity. (If the character code and the similarity are already stored in all the variables B _P [i] to B _P [(the number of candidate buffers-1)]), the process proceeds to step S152, and the candidate shift counter i is decremented by 2. , Step S15 described above
Go to 1.

Therefore, in this case, the variable B _P [(the number of candidate buffers-1)] is set to the variable B _P [(the number of candidate buffers-
2)], or the similarity of the current character pattern and its character code are overwritten, and the data stored in the variable B _P [(number of candidate buffers-1)] is erased.

On the other hand, in step S142, the variable B
_If it is determined that the character code stored in _P [k] is 0, that is, the similarity of the current character pattern is smaller than any of the similarities currently stored in the recognition candidate buffer 7, or If they are equal, the process proceeds to step S154,
It is determined whether the recognition candidate buffer counter k is smaller than the number of candidate buffers−1.

In step S154, when it is determined that the recognition candidate buffer counter k is smaller than the number of candidate buffers−1, that is, when the recognition candidate buffer 7 has a space for storing the character code together with the similarity (variable). If the character code and the similarity are not yet stored in all of B _P [0] to B _P [(the number of candidate buffers-1)], the process proceeds to step S155, and the variable B _P [k + 1] is set.
Is set to the end code, and the process proceeds to step S153. In step S153, as described above, the variable B _P
The degree of similarity of the current character pattern is set in [k] together with the character code, and the process proceeds to step S160 in FIG.

On the other hand, when it is determined in step S154 that the recognition candidate buffer counter k is smaller than the number of candidate buffers−1, that is, the variables B _P [0] to BP.
The character code and the similarity are already stored in all [(the number of candidate buffers-1)], and the similarity of the current character pattern is better than any of the similarities stored therein. If less than or equal, then FIG.
3 to step S160. That is, in this case, the current character pattern is not set as the recognition result candidate and is not stored in the recognition result candidate buffer 7.

By the above processing, the similarity degree already stored in the recognition candidate buffer 7 (variable B _P [i]) and the similarity degree of the current character pattern are sorted in ascending order and stored. It is corrected (however, the character code corresponding to the (the number of candidate buffers + 1) th largest similarity is deleted).

In step S160 of FIG. 13, it is determined that all character patterns registered in the dictionary 4 have been handwritten character comparison and verified, and after the processing is completed, the recognition candidate buffer 7 (variable B _P In [i]), among the character patterns registered in the dictionary 4, one having the highest similarity to the handwritten character from 1 to (the number of candidate buffers) is stored. Then, this recognition candidate buffer 7
The character code stored in (variable B _P [i]) is read out as needed, and the character (printing type) corresponding to the character code is displayed on the display device integrated with the handwritten character input device 1. Is displayed.

After that, the user looks at the printed characters displayed on the display device and selects the correct one as the handwritten character recognition result. Then, for example, when the handwritten character is echoed back and displayed on the display device, the selected type character is displayed instead of the handwritten character.

Next, the operation will be further described with reference to FIGS. 17 to 19. Now, for example, when a handwritten character “cape” as shown in FIG. 17 is input to the handwritten character input device 1 by operating the pen 1a, the feature extraction device 2 operates as described above, for example, A feature amount as shown in FIG. 18 is extracted (however, in FIG. 18, the direction code is not shown and only the basic stroke code is shown), and is output to the feature buffer 3 to be stored.

A character pattern as shown in FIG. 19A, for example, is registered in the dictionary 4 (however, in the figure, the direction code is not shown and only the basic stroke code is shown. If the predetermined threshold value A or B described in FIG. 12 is, for example, 50 or 80, the recognition device 5 performs a handwritten character comparison test on the character pattern shown in FIG. For example, it is performed as shown in FIG.

Note that the comparison test is performed using only the basic stroke code and not the direction code for the sake of simplicity.

Further, in FIG. 19 (c) (the same applies to FIG. 19 (b)), among the numbers in the upper and lower two columns described in each column,
The numbers in the upper row indicate the comparison result values (shape determination values) of the basic stroke codes of the dictionary 4 shown in FIG. 19A corresponding to that column, and the numbers in the lower row indicate the comparison result buffer 6.
Shows the degree of similarity which is the average value of the four comparison result values stored in.

That is, for example, since the first stroke of the character pattern "gate", that is, the basic stroke code of the stroke number 1 is 0 (FIG. 19 (a)), the comparison result value (shape determination value) of the handwritten character against it. 18 to 10
It becomes 0.

As described above, the comparison result buffer 6 stores four highest comparison result values of 100 as initial values. To the 4th smallest, but the average value is 1
It becomes 00.

From the above, for example, the column of the first column of the line of the character pattern "gate" is 100 or 100 in the upper or lower row, respectively.
Or 100 is described.

Also, for example, since the basic stroke code of the second stroke of the character pattern "gate", that is, the stroke number 2 is 3 (FIG. 19 (a)), the comparison result value (shape determination value) of the handwritten character against it. ) Is from FIG.
It becomes 0.

As a result of the comparison test for the first stroke of the character pattern "gate", the comparison result buffer 6 stores four highest comparison result values 100. Although 20 is included, the average value of the first to fourth smallest values is 80 (= (20 + 100
+ 100 + 100) / 4).

From the above, for example, in the second column of the row of the character pattern “gate”, 20 or 80 is described in the upper row or the lower row, respectively.

Similarly, the numbers in the respective columns in FIGS. 19 (b) and 19 (c) are assigned in the same manner.

Now, the predetermined designated number of strokes shown in FIG.
For example, when the number is 4, the comparison test up to the number of strokes of the handwritten character is performed using the threshold value B (= 80), and thereafter, the threshold value A (= 50) is used.

As a result, as described above, for example, the character patterns “gate” and “fix” that do not approximate handwritten characters are used.
Is a relatively large value (for example, the first or second stroke) in the stroke at the beginning of handwriting.
Stroke is 100 or 80 respectively (Fig. 19).
(C)), but with the second stroke, the threshold value B (= 8)
0) or less, where the comparison test for these character patterns is stopped.

On the other hand, according to the conventional device, the threshold value A (= 5) is set regardless of the number of strokes of the handwritten character for comparison and verification.
Since 0) is used, as shown in FIG. 19B, the comparison test is performed up to the fourth or fifth stroke for the character pattern “mon” or “fix” that is not approximate to the handwritten character. Be seen.

Therefore, according to the handwritten character recognition apparatus of FIG. 1, the comparison and verification for the character pattern which is not approximate to the handwritten character is stopped earlier than in the conventional case, so that the recognition speed is improved. be able to.

Further, since the threshold value A similar to that in the conventional case is used for the strokes after the predetermined designated number of strokes, the reduction of the recognition rate can be prevented.

Although the handwritten character recognition device to which the present invention is applied has been described above, the present invention is also applicable to, for example, a portable computer or an electronic notebook device including a block for recognizing handwritten characters. .

In this embodiment, two thresholds, A and B, are used as thresholds for deciding whether to stop or continue the comparison test. However, in addition to this, three or more thresholds are used. You can

That is, for example, three threshold values A to C (however,
When A <B <C is used, two predetermined designated stroke numbers a and b (provided that a <b) are provided. Then, if the stored value of the current stroke counter (the number of strokes of the handwritten character currently undergoing the comparison test) is smaller than the specified stroke number a, the threshold value C is set. The threshold value B may be used, and the threshold value A may be used when the number of strokes is equal to or more than the designated stroke number b.

Further, in the present embodiment, the threshold to be used is changed corresponding to the number of strokes of the handwritten character which is currently undergoing the comparison test. It is also possible to change the stroke number corresponding to the subtraction value obtained by subtracting the number of strokes for which the comparison test is being performed. However, in this case, it is necessary to use a large threshold value when the subtraction value is large and a small threshold value when the subtraction value is small.

Further, the method of calculating the feature amount and the comparison result value extracted from the handwritten character is not limited to the above.

In this embodiment, the dictionary 4 is not hierarchically structured (FIG. 6), but the dictionary 4 has a hierarchical structure as shown in FIG. 20 (a), for example. It is possible to use a structured dictionary (having a tree structure) (the details of the dictionary having such a structure are disclosed in, for example, Japanese Patent Laid-Open No. 02-062683).

In this case, in the conventional case, for example, as shown in FIG.
As shown in (b), a comparison test is performed on character patterns that do not approximate handwritten characters up to the third stroke. According to the present invention, the comparison test is shown in FIG. 20 (c), for example. Thus, the stroke is stopped at the first stroke (note that in FIG. 20 (b) and FIG. 20 (c),
X or y of (x, y) indicates the comparison result value or the similarity at each time point). Therefore, when the dictionary 4 is hierarchically constructed, the recognition speed can be further improved.

[0139]

As described above, according to the present invention, the recognition speed can be improved without lowering the recognition rate.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of a handwritten character recognition device to which the present invention is applied.

FIG. 2 is a diagram showing a correspondence relationship between a basic stroke and a basic stroke code attached to the basic stroke.

FIG. 3 is a diagram showing a correspondence relationship between a quantized direction and a direction code attached to it.

FIG. 4 is a diagram showing handwritten characters.

FIG. 5 is a diagram showing a feature amount extracted from the handwritten character of FIG.

FIG. 6 is a diagram illustrating a configuration of a dictionary 4.

FIG. 7 is a flowchart illustrating an operation of the recognition device 5.

FIG. 8 is a flowchart illustrating a more detailed operation of the recognition device 5.

FIG. 9 is a diagram illustrating a recognition candidate buffer 7 in which character codes and similarity of handwritten character recognition result candidates are stored.

FIG. 10 is a comparison result buffer 6 in which comparison result values are stored.
It is a figure explaining.

FIG. 11 is a flowchart illustrating more details of the process of step S102 of FIG.

FIG. 12 is a flowchart illustrating a more detailed operation of the recognition device 5.

FIG. 13 is a flowchart illustrating a more detailed operation of the recognition device 5.

FIG. 14 is a flowchart illustrating a more detailed operation of the recognition device 5.

FIG. 15 is a diagram showing a correspondence relationship between direction differences and comparison result values.

FIG. 16 is a flowchart illustrating a more detailed operation of the recognition device 5.

FIG. 17 is a diagram showing handwritten characters.

FIG. 18 is a diagram showing a feature amount extracted from the handwritten character of FIG. 17.

FIG. 19 is a diagram for explaining how the handwritten characters of FIG. 17 are compared and verified.

FIG. 20 is a diagram showing another configuration example of the dictionary 4.

[Explanation of symbols]

 1 Handwritten character input device 1a Pen (input pen) 2 Feature extraction device 3 Feature buffer 4 Dictionary 5 Recognition device 6 Comparison result buffer 7 Recognition candidate buffer

Claims (4)

[Claims] 1. A handwriting character recognition device for recognizing a handwritten character input by handwriting, a storage means for storing a standard character pattern, an extracting means for extracting a feature amount of the handwritten character, and the extracting means. By comparing and testing the feature amount extracted by the standard character pattern stored in the storage means, the similarity of the handwritten character to the standard character pattern is calculated, and based on the similarity. A character recognition result candidate, and the verification means performs the comparison test for each predetermined unit, calculates the similarity for each predetermined unit, and calculates the similarity and the predetermined value. Compare with the threshold value, according to the comparison result, the comparison test with the standard character pattern is stopped, the comparison test with the next standard character pattern is started, and in a predetermined case, Handwritten character recognition apparatus characterized by changing the serial predetermined threshold. 2. The handwritten character recognition device according to claim 1, wherein the verification unit changes the predetermined threshold value in accordance with the number of strokes of the handwritten character which is being compared and verified. . 3. The verification means sets the predetermined threshold value to a value at which the comparison verification is more likely to be stopped for a stroke at the beginning of writing the handwritten character than for subsequent strokes. The handwritten character recognition device according to claim 2. 4. The verification means compares the total number of strokes of the handwritten character with a predetermined number of strokes, and according to the comparison result, the predetermined threshold value is applied to the comparison of the total strokes of the handwritten character. The handwritten character recognition device according to claim 2 or 3, wherein the value is set such that the test is difficult to stop.