JPH08147411A - Character recognition device - Google Patents

Abstract

PURPOSE: To provide the character recognition device which can accurately recognize a character even if the character is inputted slantingly. CONSTITUTION: The character recognition device, which extracts plural feature points according to the strokes of the character inputted by writing and compares a feature quantity showing the distribution of segments connecting the feature points on an xy coordinate system with a previously set reference feature quantity to recognize the character inputted by writing, is equipped with a tilt judgement part 7 which judges how much the character is written and inputted slantingly under specific conditions, partial pattern Q value matching (2) 9 which calculates a feature quantity be excluding components generated by the tilt from the distribution of the segments on the xy coordinate system according to the judgement of the tilt judgement part 7.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a character recognition device for calculating a feature amount from a stroke of a character input by handwriting and performing accurate recognition based on the feature amount.

[0002]

2. Description of the Related Art A pattern matching method is generally used to recognize a character written and input by an input means such as a tablet. In this pattern matching method, a feature point is extracted from the coordinate data string of the stroke of a character (pen-on to pen-off) of a character input by handwriting, and the extracted feature point information is extracted in advance by the same method. Matching with a registered reference pattern (hereinafter referred to as a registered pattern), a matching pattern is obtained as a recognition result.

In this pattern matching method, it takes a lot of time to process which stroke of a registered pattern the stroke of a character input by handwriting should be matched with. Further, when the balance of the entire characters is disturbed, for example, when the uneven portion of the character "kawa" is small and the portion of the character is large, there is a disadvantage that an accurate matching result cannot be obtained. Therefore, from the viewpoint of coping with such deformation of the written characters, Japanese Patent Laid-Open No. 229/1987.
An online character recognition device disclosed in Japanese Patent No. 384 is disclosed.

In this on-line character recognition device, large classification is performed according to the number of strokes of a character input by handwriting, and a part to be written as a series of handwriting is defined as a partial pattern.
Then, the middle classification is performed by the vector between the centers of gravity of the partial patterns, and Q as the characteristic parameter of the partial patterns is
Matching is performed using a value that is a value and a value that is a stroke code distribution. As a result, character recognition is performed in response to character deformation and with a small amount of processing.

[0005]

However, such a character recognition device has the following problems. That is, the Q value as the characteristic parameter, which is information for matching, indicates at which position in the xy direction the ± x direction component and the ± y direction component in the xy coordinate system of each segment are distributed. For example, the standard Q value for each stroke of the character "Sat" is + x of each stroke.
Only the direction component and the -y direction component are configured, and when matching is performed, the reference Q value and the Q value of the character written and input are compared, and it is determined whether or not matching is performed based on the difference. are doing. For this reason, when a character input by handwriting is tilted, a direction component that does not exist in the reference Q value is calculated, and accurate matching becomes difficult.

[0006] For example, when the character "Sat" is written upwards, a + y direction component which is not in the standard Q value is generated. Further, when the character "Sat" is written with a right tilt, a component in the -x direction that is not in the reference Q value occurs. As described above, when the characters are written with an inclination, accurate character recognition becomes difficult and the recognition rate is lowered.

[0007]

SUMMARY OF THE INVENTION The present invention is a character recognition device made to solve such problems. That is, according to the present invention, a plurality of feature points are extracted according to strokes of a character that is written and input, and a feature amount indicating a distribution in an xy coordinate system in a segment connecting the feature points and a preset reference value. By comparing with the feature quantity,
It is a character recognition device that recognizes written characters.
An inclination determining unit that determines the degree of inclination when inputting a character to be written and written based on a predetermined condition, and excludes a component caused by the inclination from the distribution in the xy coordinate system in the segment based on the determination by the inclination determining unit. And a feature amount calculation unit that calculates the feature amount.

[0008]

In the present invention, the inclination determining section determines the inclination degree when a character to be written is input, and the characteristic amount calculating section calculates the characteristic amount based on the inclination degree.
That is, since the feature amount calculation unit calculates the feature amount by removing the component caused by the inclination in the distribution in the xy coordinate system of the segment of the character to be written and input, the input is performed without inclination. It becomes possible to obtain a feature amount equivalent to the case. This makes it possible to accurately compare the calculated feature amount without including unnecessary components and the preset reference feature amount.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the character recognition device of the present invention will be described below with reference to the drawings. 1 is a block diagram illustrating a character recognition device according to a first embodiment of the present invention, FIG. 2 is a block diagram illustrating a character recognition device according to a second embodiment of the present invention, and FIG. 3 is a third embodiment of the present invention. It is a block diagram explaining a character recognition device. In the character recognition device according to each embodiment, the tablet 1, the preprocessing unit 2, the feature extraction unit 3, the stroke coding unit 4, the large classification unit 5, the middle classification unit 6, the inclination determination unit 7, and the partial pattern Q value matching unit 8 are provided. , The partial pattern Q-value matching unit 9, the partial pattern stroke code matching unit 10, and the output terminal 11 are common, but the internal configurations of the respective inclination determination units 7 are different. The common points will be described in the description of the first embodiment, and the different points will be mainly described in the second and third embodiments.

First, before describing each embodiment, the state of inclination of a character will be described. FIG. 4 is a diagram for explaining the state of inclination of the characters, and (a) to (c) show respective modes in which the character “soil” is taken as an example. The mode shown in FIG. 4A shows each stroke when "Sat" is input in the standard font. In this case, three strokes from the A stroke to the C stroke are shown on the xy coordinate system, and the feature amount (hereinafter referred to as the Q value) is only the + x direction component in the A and B strokes, and the C stroke in the C stroke. -Y direction component only. This is the reference Q value and is preset.

The mode shown in FIG. 4 (b) shows each stroke when "Sat" is input by writing in an upward slope. In this case, compared to the standard case shown in FIG.
It is in the state where the + y direction component is added to the stroke. In addition, the mode shown in FIG. 4C shows each stroke when “Sat” is written with a right inclination, and in this case, the stroke is C stroke compared to the standard case shown in FIG. 4A. It is in a state where the −x direction component is added. In this way, when a character is tilted, a specific direction component is added to the standard font.

In the character recognition device in each of the embodiments, the characters which have been written with inclination as described above are discriminated under a predetermined condition, and the Q value is calculated in a state where the components caused by the inclination are removed, and the reference Q value is obtained. It is characterized in that it is matched with. First, the configuration of the character recognition device in the first embodiment will be described in order. The tablet 1 shown in FIG. 1 is a means for writing and inputting a desired character by touching and moving a pen or the like, and the writing data string {(xi, y
i), i = 1, 2, ... Nj} j are sent to the preprocessing unit 2. FIG. 5 is a diagram for explaining preprocessing / feature point extraction. The writing data string obtained on the tablet 1 is shown in FIG.
It is sent to the preprocessing unit 2 as raw data as shown in (a). The pre-processing unit 2 is configured to perform noise removal processing, moving average processing, or smoothing processing on the writing data string sent from the tablet 1 and send it to the feature extraction unit 3. After smoothing, the data state is as shown in FIG.

The feature extraction unit 3 also extracts feature points using the data sent from the preprocessing unit 2. As a result, the feature points shown in FIG. 5C are extracted.
For example, as a method of extracting this feature point, a smoothed data string {(xi, yi), i = 1, 2, ... Nj}
The sign (positive, negative, sign of 0) in the x and y directions between the data of j is calculated by the equation (1), and the change point of the state of the sign is extracted as the feature point.

[0014]

[Equation 1]

That is, the signs XSi and YSi in the x and y directions between the data are obtained and expressed by +, 0 and-, and if the signs between the respective data are compared with the signs between the previous data, they are characteristic points. Is registered as a feature point, assuming that the state has changed when different. Further, in addition to the feature points thus obtained, points to which a start point and an end point are added are shown as feature points. The feature point information obtained in this way is output to the stroke encoding unit 4 and the middle classification unit 6 shown in FIG.

The stroke coding unit 4 codes each stroke based on the characteristic point information obtained by the characteristic extraction unit 3. Generally, for example, X of each segment,
Classification is performed according to the Y-sign, the angle of the segment, and the rotation angle between the segments, and coding is performed. Figure 6 is a diagram for explaining a stroke coding process, θ1, θ2, (angle between the + x direction) .theta.3 segment angle, θ1 ^-, θ
2 ^- shows the rotation angle between adjacent segments. The stroke data encoded in this way is output to the large classification unit 5 and the partial pattern stroke code matching unit 10.

In the large classification unit 5, the stroke coding unit 4
It receives the output from and classifies the target character according to the number of strokes. Therefore, for each stroke number (stroke number), a character that can be the stroke number is prepared in advance in a character dictionary as shown in FIG. For example, if the stroke number of the character pattern input by handwriting is 10 strokes, characters that can be 10 strokes among those stored in the character dictionary (“mush”, “dust”, “a” ...) Select as a candidate character. In addition, the "mute" shown in this embodiment represents the character shown by A in FIG.

Further, the middle classifying section 6 performs a further middle classifying process on the candidate characters obtained by the large classifying section 5 according to the number of strokes, by the partial pattern vectors described below. Here, the partial pattern refers to a portion of one character that is written as a series of characters in writing,
The inter-partial-pattern vector is a vector whose start point and end point are the center of gravity of one partial pattern and the center of gravity of another partial pattern, respectively.

To calculate the inter-partition pattern vector,
First, the length dL1 (dL1 = √ (dx1 ² + dy1 ² )) of each segment when the x and y direction components of each segment in the partial pattern are (dx1, dy1) is obtained. If the center coordinates of each segment normalized by dividing by the character widths HX and HY are (x1 *, y1 *), the center of gravity position (Xw, Yw) of the partial pattern is (2).
Calculate by formula.

[0020]

[Equation 2]

In this way, the center of gravity of each partial pattern is obtained, and the vector between partial patterns is obtained with the center of gravity of one partial pattern and the center of gravity of another partial pattern as the starting point and the ending point, respectively. In addition, here, it is assumed that the vectors between the partial patterns are considered in the x direction and the y direction, respectively.

FIG. 8 is a diagram for explaining vectors between partial patterns. FIG. 8A shows the center of gravity of each of the partial patterns “mouth” and “sub” and the inter-partial-pattern vectors Bx (mouth, sub) and By (mouth, sub) in the case where the character “mute” is taken as an example. . The middle classification unit 6 narrows down the candidate characters selected by the large classification unit 5 based on the vector between the partial patterns to perform middle classification. For example, considering the case where the character "Ai" is written and input, first, since "Ai" has 10 strokes, the candidate character is selected by referring to the 10 strokes in the character dictionary shown in FIG. Here, the partial pattern of the "mute" arranged at the first position, the cut position in the writing order of the partial patterns, and the vector value between partial patterns are read out and compared with the "a" that has been manually input. Similar information is lined up in the character dictionary in the order of “dust”, “a”, ... The middle classifying unit 6 makes comparisons according to this character order to determine whether or not they should be candidates.

For example, when the comparison target for the handwritten character "Ai" is "mute", the partial pattern vectors Bx, By based on the cut position of "mute" written in the character dictionary. The character "A" that is written in
The matching distance dvec, which is the difference between the inter-partial-pattern vectors Bx * and By * (see FIG. 7B) when divided at the same cut position as, is calculated by the equation (3).

[0024]

(Equation 3)

In general, when there are a plurality of partial patterns of characters written and input, the matching distance dvec is calculated by normalizing the partial pattern number BPN. The matching distance dvec in this case is calculated by the equation (4).

[0026]

[Equation 4]

Then, the matching distance dvec calculated here is compared with a predetermined threshold value VECREJ. If the matching distance dvec is larger than the threshold value VECREJ, the handwritten character is the target character (this If
If it is not a "mute"), matching is performed with the partial pattern vector for the next candidate character. If the matching distance dvec is less than or equal to the threshold value VECREJ, it is determined that the character input by handwriting seems to be the target character (“mute”), and calculation and matching of the partial pattern Q value described later are performed. .

After the middle classification of the candidate characters in the middle classification unit 6, the inclination judgment unit 7 performs the processing. In the character recognition device according to the first embodiment, the inclination determination unit 7 includes a character type setting unit 71, a character type Q value setting unit 72, and a QFL.
It is composed of a G determination unit 73. Character type setting unit 71
Then, six types of signals consisting of, for example, "Kanji", "English characters", "Numerals", "Hiragana", "Katakana", and "Symbols" are output according to the type of characters input by handwriting.

For example, a writing area 10 and a selection area 12 are provided on the tablet 1 as shown in FIG. 9, and in this selection area 12, "Kanji", "English characters", "Numerals", and "Hiragana" are provided. , "Katakana", and "symbols" are provided in the six selectors 12a to 12f. To select the character type, one of the six selection units 12a to 12f corresponding to the character to be written and input is selected before writing and input. For example, if it is a kanji character, it is selected by pushing the selection part 12a displayed as "kanji" with a pen. At this time, the coordinate value (x, y) of the position specified by the pen
Is output from the tablet 1 to the character type setting unit 71, and the character type setting unit 71 receives the coordinate values and selects each of the selection units 12a to 12a.
Set the character type setting flag according to the position of 2f by Q depending on the character type.
The value is output to the value setting unit 72.

The character type Q value setting unit 72 receives the character type setting flag, determines whether or not the predetermined component is deleted in the Q value according to the value, and sets the Q value setting flag (QFLG). For example, when a character type setting flag corresponding to a Chinese character is received, the -x direction component and the + y direction component of the Q value are deleted, and at other times (in the case of letters, numbers, hiragana, katakana, and symbols) they are deleted. If not, QFLG = 1; otherwise, QFLG =
Outputs 0.

The QFLG determination unit 73 receives the QFLG from the character type Q value setting unit 72 and outputs the value of this QFLG to the output destination corresponding thereto. That is, QFLG =
When it is 0, this value is used as the partial pattern Q value matching unit 8
When QFLG = 1, this value is output to the partial pattern Q value matching unit 9.

First, the partial pattern Q value matching unit 8
The processing in will be described. The partial pattern Q value matching unit 8 calculates partial pattern Q values for the candidate characters remaining in the intermediate classification by the inter-partition pattern vector in the intermediate classification unit 6, and compares them with the partial pattern Q values in the partial pattern dictionary shown in FIG. Match. The partial pattern Q value in this partial pattern dictionary is created and stored in advance by the registered pattern. Here, the partial pattern Q value refers to a characteristic parameter indicating the length, direction, and position of each segment. The partial pattern Q value is calculated by, for example, equations (5) to (12).

[0033]

(Equation 5)

(Equation 6)

(Equation 7)

(Equation 8)

[Equation 9]

[Equation 10]

[Equation 11]

(Equation 12)

In these equations, Σ is the total stroke, addition for all segments, and HX and HY are character widths. In addition, Q obtained by the equations (5) to (12)
Since 1 to Q8 are values in which the origin is set to the lower left of a rectangular area centered on a character, those near the origin become "0" when subjected to multiplication. Therefore, the origin is replaced with the upper right corner of the quadrangular area, Q9 to Q16 are obtained by the same equations as the equations (5) to (12), and a total of 16 values from Q1 to Q16 are calculated.

For example, when the handwritten character is "O" and there is "dust" remaining in the candidate character, the cut position of the character "dust" is first read from the character dictionary shown in FIG. The cut position in this case is (3, 2, 5), and the partial pattern is in the state shown in FIG. At this time, if the “dust” cut position is divided into the same (3, 2, 5) for the character “Ai”, the state shown in FIG. 11B is obtained. Then, matching is performed with the partial pattern Q value in each of the partial patterns.

In matching, Q1 is set for each partial pattern.
The difference between Q16 and Q16 is calculated, and the sum of the differences is used as the matching distance dBP, and this matching distance dBP is calculated using the equation (13).

[0037]

(Equation 13)

However, Qi (j) in the equation (13) is
j is the Q value of the partial pattern. Next, the obtained matching distance dBP and the matching distance dvec in the inter-partial-pattern vector obtained in the previous process weighted by αvec are added to obtain the distance d1 by the equation (14).

[0039]

[Equation 14]

As αvec in the equation (14), for example, when the number of strokes of the written character is 2-3, αvec = 1/2, and when it is 4-25 strokes, αvec
In the case of = 1 and 1 stroke, αvec = 0. Such an operation is performed for all the candidate characters remaining in the middle classification by the vector between the partial patterns, and sorting is performed at each distance d1.

Next, the partial pattern Q value matching unit 9
The processing in will be described. In the partial pattern Q value matching unit 9, among the Q values calculated by the expressions (5) to (12) performed by the partial pattern Q value matching unit 8 described above,
The Q value is calculated excluding the components (for example, the −x direction component and the + y direction component) generated when the character is inclined, and the matching is performed by this. That is, when excluding the -x direction component and the + y direction component, Q1, Q4, Q5, Q8
And Q9 corresponding to these when the origin is replaced,
A total of eight Q values of Q12, Q13 and Q16 are used.

Therefore, the matching distance dBP obtained by the partial pattern Q-value matching section is calculated using the equation (15).

[0043]

(Equation 15)

Then, the obtained matching distance dBP and the matching distance dvec in the partial pattern vector obtained in the previous process weighted by αvec are added to obtain the distance d1 by the equation (14). This operation is performed for all candidate characters that are moderately classified by the inter-partition pattern vector, and sorting is performed at each distance d1. As described above, by calculating the Q value by removing the component in the direction generated when the character is tilted, it is possible to obtain the Q value equivalent to that when the character is input without tilting. As a result, it is possible to perform accurate matching between the calculated Q value without including unnecessary components due to the slope and the preset reference Q value.

Next, the processing in the partial pattern stroke code matching unit 10 will be described. The partial pattern stroke code matching unit 10 includes a partial pattern Q value matching unit 8 and a partial pattern Q value matching unit 9.
The partial pattern stroke code distribution is obtained for the candidate characters narrowed down by the middle classification. Then, FIG. 1 created in advance from this distribution and the registered pattern
Matching is performed with the partial pattern stroke code distribution stored in the partial pattern dictionary indicated by 0, and the higher-ranked candidates are ranked.

The range of objects to be ranked is determined, for example, by the ratio with the distance d1 of the first candidate obtained by sorting the distance d1 described above. That is, dj / d1
Ranking is performed with the candidate character of ≦ ZRATE as the target range.

Next, a method of calculating the partial pattern stroke code distribution will be described. As an example, the character written and input is "Ai", and the character selected as the first candidate is "Ai".
The case will be described. First, the cut position (3, 4, 3) of the candidate character "Ai" is read from the character dictionary shown in FIG. As a result, the partial code of the character "Ai" is shown in FIG.
It will be divided into three partial codes as shown in (a) to (c).

Next, the pattern of the character written and input at this position is cut, and the stroke code obtained by the stroke encoding unit 4 is obtained for each of the partial pattern obtained by this and the previously obtained partial pattern. Calculate the distribution of the number of. For example, "Ai" shown in FIG.
When the first partial pattern of the character is viewed, the number of lines corresponding to the partial pattern stroke code distribution “01” in FIG. 10 is 1, the number of lines corresponding to “03” is 1, and the number is “05”.
A stroke code distribution in which the number of lines corresponding to is one is obtained. Then, the partial pattern stroke code distribution obtained in this manner and the registration pattern are calculated and averaged in advance by the same procedure as in FIG.
The partial pattern stroke code distribution indicated by 0 is matched.

In the case of the first partial pattern of the character "Ai" shown in FIG. 12A, the partial pattern stroke code distribution ("" corresponding to the first partial pattern registered in the pattern dictionary shown in FIG. "01" is 0.9, "02"
Is 0.1, "03" is 0.4, and "04" is 0.6
Book, "05" is 1 book). That is,
Find the difference in each partial pattern stroke code distribution,
A matching distance ds obtained by adding each difference is obtained.
0.1 + 0.1 for the first partial pattern of "Ai"
+ 0.6 + 0.6 = 1.4 lines are obtained as the matching distance ds. Similarly, the matching distance ds is calculated for the second partial pattern and the third partial pattern shown in FIGS. 12B and 12C, and these are normalized by the number of partial pattern stroke codes BS to obtain the total. The distance Ds is calculated by the equation (16).

[0050]

[Equation 16]

This distance Ds is calculated for all the candidate characters, and the smallest one is output to the output terminal 11 shown in FIG. 1 as the recognition result.

As described above, in the character recognition apparatus of the first embodiment, it is determined which of the character types is likely to affect the Q value due to the inclination when the handwriting is input. By calculating the Q value excluding the generated unnecessary directional component, it becomes possible to perform accurate matching with the reference Q value. That is, it is possible to accurately recognize even when a character is slanted and is input by handwriting.

Next, a character recognition device according to the second embodiment of the present invention will be described. FIG. 2 is a block diagram illustrating a character recognition device in the second embodiment. In the character recognition device in the second embodiment, the tablet 1, the preprocessing unit 2, the feature extraction unit 3, the stroke coding unit 4, the large classification unit 5, the middle classification unit 6, the partial pattern Q value matching unit 8, the partial pattern Q. The value matching unit 9, the partial pattern stroke code matching unit 10, and the output terminal 11 are the first described above.
The character recognition device is the same as that of the embodiment, but the internal structure of the inclination determination unit 7'is different. Hereinafter, description will be made focusing on the tilt determination unit 7'that is installed.

The inclination determination unit 7'of the character recognition apparatus in the second embodiment comprises a stroke number setting unit 74, a stroke number Q value setting unit 75, and a QFLG determination unit 73. In the stroke number setting unit 74, the stroke number is calculated from the data for one character output from the preprocessing unit 2, the stroke number is substituted into the variable Strsu, and the Q value depending on the stroke number setting unit 75 is calculated.
Output to.

The number-of-strokes Q-value setting unit 75 determines the degree of inclination of the character in accordance with the value of the variable Strsu, and determines whether the Q-value is calculated by removing the component caused by the inclination. Set the indicated QFLG. For example, Strsu
When ≧ 7 (the character is 7 strokes or more), the Q value of −
QFL as deleting x-direction component and + y-direction component
If G = 1 is set and Strutu ≦ 6 (characters are 6 strokes or less), it is not deleted and QFLG = 0 is set.
Output to the QFLG determination unit 73. Here, 1 ≦ Str
Let su ≦ 25.

The QFLG judging unit 73 receives the QFLG from the Q value setting unit 75 based on the number of strokes and outputs the value of this QFLG to the output destination corresponding thereto. That is, QFLG = 0
In the case of, the partial pattern Q value matching unit 8
When QFLG = 1, this value is output to the partial pattern Q value matching unit 9. As a result, the processing by the partial pattern Q value matching unit and the processing by the partial pattern Q value matching unit 9 described above are performed, and further the processing by the partial pattern stroke code matching unit 10 is performed to output an accurate recognition result. Output from the terminal 11.

In the character recognition apparatus according to the second embodiment, it is noted that when the number of strokes is larger than the predetermined stroke number, the influence of the Q value due to the inclination of the character becomes large. The Q value is calculated by excluding such components. By setting the number of strokes to a value larger than the maximum number of strokes for letters, numbers, hiragana, katakana, and symbols, it is automatically determined that a kanji character that is easily affected by the Q value due to the character inclination has been input, and is unnecessary. It becomes possible to accurately calculate the Q value excluding these components. As a result, it becomes possible to accurately recognize the characters even if the characters are inclined and input.

Next, a character recognition device according to the third embodiment of the present invention will be described. FIG. 3 is a block diagram illustrating a character recognition device in the third embodiment. In the character recognition device in the third embodiment, the tablet 1, the preprocessing unit 2, the feature extraction unit 3, the stroke coding unit 4, the large classification unit 5, the middle classification unit 6, the partial pattern Q value matching unit 8, the partial pattern Q. The value matching unit 9, the partial pattern stroke code matching unit 10, and the output terminal 11 are the first described above.
This is the same as the character recognition device in the embodiment and the second embodiment, but the internal configuration of the inclination determination unit 7 ″ is different. Hereinafter, the description will be centered on the different inclination determination unit 7 ″.

The inclination judging section 7 '' of the character recognition apparatus in the third embodiment is composed of a stroke code setting section 76, a stroke code Q value setting section 77, and a QFLG judging section 7.
And 3. The stroke code setting unit 76 uses information about strokes including the −x direction component and the + y direction component, which are pre-classified stroke codes for the number of characters output from the stroke encoding unit 4, and strokes not including the stroke components. , -X direction component and + y
The number of strokes including the direction component is SY, the number of strokes not including it is SN, and these are output to the Q value setting unit 77 by the stroke code.

In the stroke value Q value setting unit 77, a predetermined value is determined according to a ratio indicating how much the -x direction component and the + y direction component are included in the total number of strokes according to the determination formula shown in the formula (17). QFLG of is output.

[0061]

[Equation 17]

That is, the threshold value I in equation (17)
(However, 0 ≦ I ≦ 1) is set, and (SY / (SY + S
N)) is less than or equal to the threshold value I, QFLG = 1 is output to calculate the Q value excluding the −x direction component and the + y direction component, and QFLG = 0 otherwise.
Is output.

The QFLG judging section 73 receives the QFLG from the Q value setting section 77 based on the stroke code, and outputs the value of this QFLG to the output destination corresponding to this. That is,
When QFLG = 0, this value is output to the partial pattern Q value matching unit 8, and when QFLG = 1, this value is output to the partial pattern Q value matching unit 9. As a result, the processing by the partial pattern Q value matching unit and the processing by the partial pattern Q value matching unit 9 described above are performed, and further the processing by the partial pattern stroke code matching unit 10 is performed to output an accurate recognition result. Output from the terminal 11.

In the character recognition apparatus according to the third embodiment, if the -x direction component and the + y direction component are included in a certain number or more with respect to the total number of strokes, the character originally includes an oblique segment. It is judged as it is and Q
When the value is calculated and the -x direction component and the + y direction component are less than a certain value with respect to the total number of strokes, it is determined that the character originally having a small number of diagonal segments is inclined, and the -x direction component and the + y direction component The feature is that the Q value is calculated excluding the components. This makes it possible to accurately calculate the Q value excluding unnecessary components caused by the inclination and improve the recognition rate.

In the present embodiment, the -x direction component and the + y direction component are described as examples of the components excluded by the partial pattern Q value matching unit 9 in the Q value calculation, but the present invention is not limited to this. It is only necessary to appropriately exclude components that are often generated by the inclination. For example, the components to be excluded may be set arbitrarily, and may be changed according to the characteristics (individual difference or the like) of the input character.

[0066]

As described above, the character recognition device of the present invention has the following effects. That is, in the character recognition device of the present invention, since the feature amount calculation excluding the component generated when the character is inclined due to the degree of inclination of the character is calculated, even if the character to be written and input is inclined, the reference feature amount It becomes possible to perform accurate matching with. As a result, it is possible to obtain a high recognition rate even for character deformation such as character inclination.

[Brief description of drawings]

FIG. 1 is a block diagram illustrating a character recognition device according to a first embodiment of the present invention.

FIG. 2 is a block diagram illustrating a character recognition device according to a second embodiment of the present invention.

FIG. 3 is a block diagram illustrating a character recognition device according to a third embodiment of the present invention.

FIG. 4 is a diagram for explaining a state of inclination of a character, where (a) shows a standard, (b) shows an upward slope, and (c) shows a right inclination.

5A and 5B are diagrams illustrating preprocessing and feature point extraction, in which FIG. 5A shows raw data, FIG. 5B shows smoothed data, and FIG. 5C shows feature extraction.

FIG. 6 is a diagram illustrating stroke coding processing.

FIG. 7 is a diagram showing a configuration example of a character dictionary.

FIG. 8 is a diagram for explaining vectors between partial patterns,
(A) shows a vector between partial patterns, and (b) shows an example of "a".

FIG. 9 is a diagram showing an example of a tablet.

FIG. 10 is a diagram showing a configuration example of a partial pattern dictionary.

FIG. 11 is a diagram illustrating calculation of a partial pattern Q value.

FIG. 12 is a diagram showing partial patterns of “a” in (a) to (c).

[Explanation of symbols]

 1 tablet 2 pre-processing unit 3 feature extraction unit 4 stroke coding unit 5 large classification unit 6 medium classification unit 7 inclination determination unit 8 partial pattern Q value matching unit 9 partial pattern Q value matching unit 10 partial pattern stroke code matching unit 11 output Terminal

Claims (5)

[Claims] 1. A feature quantity indicating a distribution in an xy coordinate system in a segment connecting a plurality of feature points in accordance with strokes of a character input by handwriting, and a preset reference feature. A character recognition device for recognizing a character input by handwriting by comparing with an amount, and an inclination determination unit that determines the degree of inclination at the time of input of a character input by handwriting based on a predetermined condition, A character amount calculating unit that calculates the characteristic amount by excluding the component generated by the inclination in the distribution in the xy coordinate system in the segment based on the determination by the inclination determining unit. apparatus. 2. The character recognition device according to claim 1, wherein the inclination determination unit determines the degree of inclination according to the type of the character input by handwriting. 3. The character recognition device according to claim 1, wherein the inclination determination unit determines the degree of inclination according to the number of strokes of the character input by handwriting. 4. The inclination determining unit determines the inclination depending on how much of the distribution in the xy coordinate system in the segment, which is caused by the inclination, is included in the total stroke number of the character to be written and input. The character recognition device according to claim 1, wherein a condition is judged. 5. The feature quantity calculation unit calculates the feature quantity by excluding the −x direction component and the + y direction component when the lower left corner of the character is the origin in the distribution in the xy coordinate system of the segment. The character recognition device according to claim 1, wherein the character recognition device is a character recognition device.