JP3052849B2 - Online character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an on-line character recognition device, and more particularly to a character recognition device for recognizing a character entered while being rotated or inclined in a pen input portable terminal or the like.

[0002]

2. Description of the Related Art Conventionally, this type of on-line character recognition apparatus has been used to rotate (enter from an arbitrary direction with respect to the tablet) or tilt (characters such as vertical and horizontal lines with respect to the tablet) characters written using a tablet. This is used to convert characters into codes by correcting deformations such as writing characters and converting them into characters written straight and performing recognition.

A conventional embodiment will be described in detail with reference to the drawings.

Referring to FIG. 12, in a conventional embodiment, a data input means 10, a correction angle extracting means 23,
It comprises a data conversion unit 24, a classification feature extraction unit 22, a classification unit 34, and a classification standard feature storage unit 35.

[0005] The standard classification feature storage unit 35 stores standard classification features of each category in advance.

[0006] The data input means 10 acquires character data using a tablet or the like. The tablet detects the position of the pen on the tablet at certain sampling intervals. Character data is usually stored as a sequence of sampled coordinate values.

The correction angle extracting means 23 extracts a correction angle for correcting the rotation or inclination of the character from the input character data obtained from the data input means 10.

[0008] The data conversion means 24 is based on the correction angle obtained from the correction angle extraction means 20,
Each coordinate value of the input character data obtained is coordinate-transformed,
Create corrected character data.

The classification feature extraction means 22 extracts classification features used for character shape discrimination from the corrected character data obtained by the data conversion means 21.

The classifying means 34 includes a classifying feature extracting means 22.
The obtained classification feature is compared with the standard classification feature of each category obtained from the classification standard feature storage unit 35 to calculate the degree of difference, and the rank of the recognition candidate is determined based on the degree of difference.

An example of a conventional line graphic processing method for correcting rotation is described in Japanese Patent Application Laid-Open No. 57-189273. In the line figure processing method described in this publication, line figure (character string) data is sequentially rotated and converted, a frequency distribution in the writing direction of each line element is calculated for each of the rotated data, and each frequency distribution is calculated. The data is rotationally transformed by calculating the distance between the pattern and the standard pattern and detecting the rotation angle at which the distance becomes the minimum.

An example of a conventional character recognition device that corrects tilt is described in Japanese Patent Application Laid-Open No. 7-6211. The character recognition device described in this publication detects the direction of a vector connecting the starting point and the ending point of each stroke, and when this direction has a certain inclination, the direction of the Chinese numeral “one” (“→”).
(Direction) or an arithmetic numeral “1” direction (“↓” direction).

[0013]

The first problem is that, in the conventional method of correcting rotation, when performing rotation correction in units of one character, it is impossible to perform stable correction in the front direction. The inability to recognize characters correctly.

The reason is that the frequency distribution in the line element direction of each character is various. In particular, the frequency distribution in the line element direction differs greatly between a character having many vertical strokes and a character having many oblique strokes. Therefore, if the standard pattern is set by averaging the frequency distribution in the line element direction of many characters, the distance between the frequency distribution in the line element direction and the standard pattern when the data is rotated in the front direction is not always minimum. Because there is no.

A second problem is that it is impossible to correct the character if the character is largely rotated or tilted by the conventional method for correcting the tilt.

[0016] The reason is that when a large rotation or inclination is made, it becomes impossible to distinguish between a stroke in the vertical and horizontal directions and a stroke that is originally oblique. Is also converted incorrectly.

It is an object of the present invention to provide an online character recognition apparatus capable of recognizing not only characters having a large number of strokes in the vertical and horizontal directions but also characters having a large number of strokes in an oblique direction.

Another object of the present invention is to provide an on-line character recognition device capable of recognizing characters in which rotation and inclination have occurred simultaneously.

[0019]

A first on-line character recognition apparatus of the present invention converts input character data by recognizing a rotation angle and an inclination angle and recognizes the input character data, and also differs from the standard rotation angle and the standard inclination angle. Recognition is also performed using degrees.

More specifically, a conversion angle extraction means (20 in FIG. 1) for obtaining a frequency distribution in the direction of each line element from input character data and determining a rotation angle and a tilt angle from the frequency distribution.
Character data conversion means (21 in FIG. 1) for converting coordinate values of input character data based on the rotation angle and the inclination angle obtained by the conversion angle extraction means to create converted character data; Standard angle storage unit (31 in FIG. 1) that stores in advance the standard rotation angle and inclination angle of
And calculating the angle difference from the rotation angle and the inclination angle obtained by the conversion angle extraction means and the standard rotation angle and the standard inclination angle of each category obtained from the standard angle storage unit, and calculating the characteristic difference obtained from the classification feature. And classification means (30 in FIG. 1) for determining the ranking of recognition candidates.

The second online character recognition device of the present invention uses the deviation rotation angle and the deviation inclination angle for calculating the angle difference in the first online character recognition device.

More specifically, deviation angle information input means (40 in FIG. 9) for inputting a deviation rotation angle and a deviation inclination angle representing the tendency of the rotation and inclination of the input character by an operator;
A deviation angle storage unit (4 in FIG. 9) holding the deviation rotation angle and deviation inclination angle obtained from the deviation angle information input means.
1).

The third online character recognition device of the present invention updates deviation angle information in the second online character recognition device by learning.

More specifically, based on the correct category obtained by the result correcting means (50 in FIG. 10), the standard rotation angle and the standard tilt angle of the correct category are obtained from the standard angle storage unit, and the conversion angle extracting means obtains the standard rotation angle and the standard tilt angle. A deviation angle information updating means (FIG. 10) for updating the deviation rotation angle and the deviation inclination angle in the deviation angle storage unit using the obtained rotation angle and inclination angle.
60).

[0025]

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

Referring to FIG. 1, a first embodiment of the present invention comprises a data input unit 10 and a conversion angle extraction unit 20.
, Data conversion means 21, classification feature extraction means 22
, A classifying means 30, a standard angle storage unit 31, and a classification standard feature storage unit 32.

The data input means 10 and the classification feature extracting means 22 are the same as those in the conventional online character recognition device shown in FIG.

The standard angle storage unit 31 stores standard tilt angles and rotation angles of each category in advance. The standard classification feature storage unit 32 stores in advance standard classification features of each category.

The conversion angle extracting means 20 obtains a frequency distribution in the direction of each line element (a line connecting adjacent coordinate values) from the input character data obtained from the data input means 10, and calculates a rotation angle and a tilt angle from the frequency distribution. decide.

The data conversion means 21 performs coordinate conversion of each coordinate value of the input character data obtained from the data input means 10 based on the rotation angle and the inclination angle obtained from the conversion angle extraction means 20 (hereinafter the converted character data). Is called converted character data).

The classification feature extraction means 22 extracts classification features used for character shape discrimination from the converted character data obtained by the data conversion means 21.

The classification unit 30 compares the rotation angle and the inclination angle obtained from the conversion angle extraction unit 20 with the standard rotation angle and the inclination angle of each category obtained from the standard angle storage unit 31, and compares the degrees of difference (angle differences). Degrees), and
The classification feature obtained by the classification feature extraction unit 22 is compared with the standard classification feature of each category obtained from the classification standard feature storage unit 32, and the difference (feature difference) is calculated.
The order of recognition candidates is determined based on the angle difference and the feature difference.

Next, the operation of the first embodiment of the present invention will be described with reference to FIGS.

The input character data obtained from the data input means 10 is converted into a conversion angle extraction means 20 and a data conversion means 21.
Supplied to The conversion angle extraction means 20 obtains a frequency distribution H (θ) in the direction θ of each line element from the input character data (step S200). Next, from the frequency distribution H (θ),
Combination frequency in two directions K (θ, φ) = aH (θ) + bH
(Φ) is calculated (step S201). Where a,
b is a weight coefficient. Further, the synthesis frequency K (θ, φ)
Are obtained (Step S202), and the rotation angle α and the inclination angle β are determined so that the θ and φ coincide with the coordinate axes (Step S203).

The data conversion means 21 performs coordinate conversion of each coordinate value of the input character data based on the rotation angle α and the inclination angle β (step S210). Classification feature extraction means 22
Extracts the classification feature f (x) (x is an element number) used for character shape discrimination from the converted character data obtained by the data conversion means 21 (step S220).

Next, the classification means 30 determines the rotation angle α and the inclination angle β and the respective categories i obtained from the standard angle storage unit 31.
The standard rotation angle α (i) and the inclination angle β (i) are compared with each other to calculate a difference (angle difference) Da (i),
The classification feature f (x) and the classification standard feature storage unit 32
Standard classification feature g for each category i obtained from
(I; x) are compared to calculate a difference (feature difference) Df (i), and the difference Dn (i) obtained by integrating the angle difference Da (i) and the feature difference Df (i). The order of the recognition candidates is determined based on the recognition candidates (step S300).

Step S300 will be described in detail. In step S300, for example, the classification processing shown in FIG. 3 is performed.

Referring to FIGS. 1 and 3, the classification means 3
In the case of 0, after initialization (step S310), the degree of difference Dn (i) is sequentially obtained for each category i, and K category numbers having a small degree of difference are selected (steps S311 to S315). In step S311, the rotation angle α and the inclination angle β obtained from the conversion angle extraction unit 20 and the standard rotation angle α of each category i obtained from the standard angle storage unit 31
From (i) and the inclination angle β (i), the angle difference Da (i)
Is calculated from the classification feature f (x) obtained from the classification feature extraction unit 22 and the classification standard feature g (i; x) of the category i obtained from the classification standard feature storage unit 32. Df (i) is calculated, and finally the integrated difference Dn (i) is obtained. However, when the angle difference Da (i) is sufficiently large, the process may proceed to step S314 without calculating the feature difference Df (i). In step S312, the degree of difference D (k) of the K-th candidate having the largest degree of difference among the K selection candidates is compared with the degree of difference Dn (i) of the category i. Dn (i) <D
In the case of (k), the category i is converted to the K-th candidate,
It is sorted again in the order of the degree of difference (step S313). In the processing from step S311 to S313, the category number is sequentially increased (step S314), and the number of categories N
(Step S315). Finally, the selected category number is converted into a category code and output (step S316).

Next, the effects of the first embodiment of the present invention will be described. The first embodiment of the present invention does not correct the rotation or inclination of a vertical line or a horizontal line, but converts a character so that a direction having a high frequency of line element directions coincides with a coordinate axis. It is possible to recognize not only characters with a large number of strokes but also characters with a large number of diagonal strokes that are inclined or rotated. Furthermore, since the rotation angle and the inclination angle obtained from the input data are compared with the standard rotation angle and the inclination angle of each category, the rotation and the inclination of the vertical line “|” and the oblique line “/” can be distinguished and recognized.

Further, since both the rotation angle and the inclination angle are detected at the same time, a character in which the rotation and the inclination have occurred simultaneously can be recognized.

[0041]

Next, the operation of the first embodiment of the present invention will be described in detail.

As a first specific example, the data input means 10
It is assumed that character data as shown in FIG. In step S200, the conversion angle extracting means 20 obtains a frequency distribution H (θ) in the direction θ of each line element from the input character data. Point (x _i , y
_i ), the direction θ of the line element connecting the point (x _{i + 1} , y _{i + 1} ) is
For example, it can be calculated by the following equation.

Θ = tan ⁻¹ {(y _{i + 1} −y _i ) / (x _{i + 1} −x _i )} When obtaining the direction frequency distribution H (θ), for example, as shown in FIG. The degrees are divided for each certain angle, and the frequencies of the line element directions within each angle range are totaled (15 in the case of FIG. 5).
Every degree). When totaling the frequencies, an amount proportional to the length of the line element is added as the frequency. The length of the line element is, for example, the Euclidean distance ｛(y _{i + 1} −y _i ) ² + (x _{i + 1} −x
_i ) It can be calculated by ² ｝ ^1/2 . For example, a direction frequency distribution H as shown in FIG.
(Θ) is required.

Next, in step S201, the conversion angle extracting means 20 converts the direction frequency distribution H (θ) into two directions (θ, θ).
φ) is calculated by the following equation.

K (θ, φ) = a · H (θ) + b · H
(Φ) Here, a and b are weighting factors. FIG. 7 shows an example of the synthesis frequency K (θ, φ) obtained from the direction frequency distribution H (θ) in FIG. 6 (here, a = b = 1). As shown in FIG. 7, the pair of the two directions (θ, φ) may be such that φ-θ is only around 90 ° (the angle returns to 0 ° at 360 °).

Next, the conversion angle extracting means 20 determines in step S202 that θ and φ at which the combined frequency K (θ, φ) is the maximum.
Ask for. In the case of FIG. 7, the maximum is obtained when θ = 210 ° and φ = 315 °. Next, in step S203, the rotation angle α and the inclination angle β are obtained by the following equations.

Α = 360−θ β = φ− (θ + 90) (Set to satisfy −180 ≦ β <180) When θ = 210 and φ = 315 obtained from FIG. 7, α = 1
50, β = 15.

The data conversion means 21 determines in step S210
Then, based on the rotation angle α and the inclination angle β, each coordinate value (x, y) of the input character data is subjected to coordinate conversion by the following equation.

X ′ = (x · cos α−y · sin α) +
(X · sin α + y · cos α) tan β y ′ = (x · sin α + y · cos α) In the case of the input character data of FIG. 4A, the data is as shown in FIG.

Next, in step S220, the classifying feature extracting unit 22 uses the classifying feature f used to determine the character shape from the converted character data obtained by the data converting unit 21.
(X) is extracted. For example, the character data is converted into a character image as shown in FIG. 8A, the direction of the character line portion is detected, and the vertical, upward, horizontal, and downward four as shown in FIG. Create a direction feature pattern for the direction. Next, the direction feature patterns in four directions are summed up for each feature amount for each of 64 regions in each direction as shown in FIG.
A classification feature f (x) of 256 elements is created.

The standard angle storage unit 31 previously stores a standard rotation angle α (i) and inclination angle β (i) of each category i. Standard rotation angle α (i) and standard inclination angle β
(I) is obtained using a character database for creating a dictionary. Using each character data of the same category, a rotation angle α and an inclination angle β are obtained from each character data.
The frequency distribution of the rotation angle and the inclination angle is obtained, and the rotation angle and the inclination angle that are frequently represented are set as the standard rotation angle and the standard inclination angle, respectively. If there are a plurality of frequent groups, a standard rotation angle and a standard inclination angle are obtained for each group (in this case, a multi-template).

The standard classification feature storage unit 32 previously stores standard classification features g (i; x) of each category i. The standard feature g (i; x) for classification is also obtained using a character database for creating a dictionary. For character data belonging to each group for which a standard rotation angle and a standard inclination angle are set, a classification feature is obtained from each character data, and each classification feature is averaged to obtain a classification standard feature g (i; x). I do.

In step S300, the classifying means 30 obtains the difference Dn (i) for each category i, and outputs them as recognition candidates in ascending order of the difference. Difference Dn (i)
Is the angle difference Da (i) and the feature difference Df (i)
For example, it is obtained by the following equation.

Dn (i) = Da (i) + Df (i) The angle dissimilarity Da (i) is obtained from the rotation angle α and the inclination angle β obtained from the input character data and the respective categories i obtained from the standard angle storage unit 31. From the standard rotation angle α (i) and the inclination angle β (i), for example, by the following equation.

Da (i) = sR (α (i) -α) + t
R (β (i) -β) Here, s and t are weighting factors, and R (u) has a relationship as shown in the following equation.

R (u) = | u |, if | u | <180 360 − | u |, otherwise The characteristic difference Df (i) is a classification feature f (x) obtained from the classification feature extraction unit 22. And standard feature storage for classification 3
Standard classification feature g for each category i obtained from
From (i; x), for example, it is calculated by the following equation.

Df (i) = Σx | f (x) −g (i; x) | For example, in the case of input character data as shown in FIG. 4A, the rotation angle is α = 150 and the inclination angle is β = Fifteen. The standard rotation angle of the category “me” (i = 0) is α (0) = 13
5, when the standard inclination angle is β (0) = 0, the angle difference degree is Da (0) = 10 · | 135-1 when s = t = 10
50 | +10. | 0-15 | = 300. The standard rotation angle of the category “ten” (i = 1) is α (1)
= 90 and the display tilt angle β (1) = 0, the degree of angle difference is Da (1) = 750. On the other hand, the category "me"
Figure 8 shows the standard features for the classification of
Since it is similar to (c), the feature difference is, for example, Df
(0) = 500 and Df (1) = 500. Finally, the integrated dissimilarity Dn (i) is Dn (0) = 8
00, Dn (1) = 1250, which is recognized as "me" (the difference from other categories is assumed to be sufficiently large). If the category "me" is not stored in the dictionary as being out of recognition, "ten" is recognized as a handwritten character rotated.

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

Referring to FIG. 9, according to a second embodiment of the present invention, a data input means 10 and a conversion angle extracting means 20 are provided.
, Data conversion means 21, classification feature extraction means 22
, A classifying unit 33, a standard angle storage unit 31, a classification standard feature storage unit 32, a deviation angle information input unit 40, and a deviation angle storage unit 41.

Data input means 10, conversion angle extraction means 20, data conversion means 21, classification feature extraction means 2
2, a standard angle storage unit 31, and a standard feature storage unit for classification 3
2 is the same as that in the online character recognition device of the first embodiment of the present invention shown in FIG.

The deviation angle information input means 40 allows the operator to input deviation angle information using an input device such as a keyboard before starting character recognition. The deviation angle information is information (deviation rotation angle and deviation inclination angle) indicating the tendency of the rotation and inclination of the input character data with respect to the case of writing in the front direction. The deviation angle storage unit 41 holds deviation angle information obtained from the deviation angle information input unit 40.

The classification means 33 is obtained from the rotation angle and inclination angle obtained by the conversion angle extraction means 20 and the standard rotation angle, inclination angle and deviation angle storage section 41 of each category obtained from the standard angle storage section 31. The angle difference is calculated by using the deviation rotation angle and the deviation inclination angle, and the classification feature is compared with the standard classification feature of each category obtained from the classification standard feature storage unit 32 to calculate the feature difference. Then, the order of the recognition candidates is determined based on the angle difference and the feature difference.

Next, the effect of the second embodiment of the present invention will be described. In the second embodiment of the present invention, the degree of angle difference is calculated using the information indicating the tendency of the input character data to rotate and tilt when the character is written in the front direction. Can recognize the written characters.

[0064]

Next, the operation of one example of the second embodiment of the present invention will be described in detail.

The deviation angle storage section 41 previously acquires and holds the deviation rotation angle αp and the deviation inclination angle βp from the deviation angle information input means 40. For example, when writing by always rotating in the clockwise direction by about 15 °, αp = −15, βp
= 0 is stored.

In the classifying means 33, the angle difference Da
(I) is the rotation angle α and the inclination angle β obtained from the input character data, the standard rotation angle α (i) and the inclination angle β (i) of each category i obtained from the standard angle storage unit 31, and the deviation angle From the deviation rotation angle αp and the deviation inclination angle βp obtained from the storage unit 41, for example, it is calculated by the following equation.

Da (i) = sR (α (i) -α-α
p) + tR (β (i) −β−βp) For example, using the above-described specific example, in the case of input character data as shown in FIG. 4A, the rotation angle is α = 150, and the inclination angle is β = 15. Assuming that the standard rotation angle of the category “me” (i = 0) is α (0) = 135 and the standard inclination angle is β (0) = 0, the angle difference is Da (0) = 150 (s = t = 10).
). Finally, the integrated difference is Dn
(0) = 650, and there is no deviation angle information (Dn
(0) = 800), it is more easily recognized as “me”.

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

Referring to FIG. 10, according to a third embodiment of the present invention, a data input means 10 and a conversion angle extracting means 2 are provided.
0, data conversion means 21, classification feature extraction means 22
, Classification means 33, standard angle storage section 31, classification standard feature storage section 32, deviation angle storage section 41, result correction means 50, display means 51, and correction information input means 52
And a deviation angle information updating means 60.

Data input means 10, conversion angle extraction means 20, data conversion means 21, classification feature extraction means 2
2, a standard angle storage unit 31, and a standard feature storage unit for classification 3
2, the classification means 33, and the deviation angle storage unit 41 are the same as those in the online character recognition device according to the second embodiment of the present invention shown in FIG.

The result correction means 50 displays the recognition candidates obtained from the classification means 33 on the display using the display means 51. If the recognition result is erroneously read, the operator uses the correction information input means 52 to recognize the recognition candidate. Correct the result.

The deviation angle information updating means 42 stores the standard rotation angle α (ic) of the correct category from the standard angle storage 31 based on the correct category ic obtained by the result correcting means 50.
And the standard inclination angle β (ic), and using the rotation angle α and the inclination angle β obtained from the conversion angle extraction means 20,
The deviation rotation angle αp and the deviation inclination angle βp in the deviation angle storage unit 41 are updated.

Next, the operation of the third embodiment of the present invention will be described with reference to FIGS.

Steps S200 to S300 are the same as those in the operation of the first embodiment shown in FIG.

The result correction means 50 displays the recognition candidates obtained by the classification means 33 on the display by the display means 51, and when the recognition result is erroneously read, the correction information input means 52 allows the operator to use a keyboard or the like. The recognition result is corrected (Step S500).

The deviation angle information updating means 42 stores the standard rotation angle α (ic) of the correct category from the standard angle storage 31 based on the correct category ic obtained by the result correcting means 50.
And the standard inclination angle β (ic), and using the rotation angle α and the inclination angle β obtained from the conversion angle extraction means 20,
The deviation rotation angle αp and the deviation inclination angle βp in the deviation angle storage unit 41 are updated (step S600).

Next, effects of the third embodiment of the present invention will be described. The third embodiment of the present invention automatically learns the information indicating the tendency of the input character data to rotate and tilt with respect to the case of writing in the front direction. Characters can be adaptively recognized.

[0078]

Next, the operation of one example of the third embodiment of the present invention will be described in detail.

The deviation angle information updating means 42 stores the standard rotation angle α (ic) of the correct category from the standard angle storage 31 based on the correct category ic obtained by the result correcting means 50.
And the standard inclination angle β (ic), and using the rotation angle α and the inclination angle β obtained from the conversion angle extraction means 20,
The deviation rotation angle αp and the deviation inclination angle βp in the deviation angle storage unit 41 are updated by, for example, the following equation.

Αp ′ = ｛n · αp + Q (α (ic) −α
−αp｝ / (n + 1) βp ′ = {n · βp + Q (β (ic) −β−βp} /
(N + 1) Here, n is a weight coefficient for determining the degree of update, and Q (v) is a function as shown in the following equation.

Q (v) = v, if | v | ≦ 180 v−360, if v> 180 v + 360, other For example, using the above specific example, the case of input character data as shown in FIG. , The rotation angle is α = 150, and the inclination angle is β = 15. It is also assumed that the deviation angle storage unit stores the deviation rotation angle αp = −15 and the deviation inclination angle βp = 0. When the correct answer category is “ten”, the standard rotation angle is α (1) = 90 and the standard inclination angle is β (1) = 0, so αp and βp are updated as follows (n = 4).

Αp ′ = {4 · (−15) + Q (90−150 − (− 15)} / (4 + 1) = − 21 βp ′ = {4.0 + Q (0−15−0)} / (4 + 1) = -3 The angle difference before updating category “ten” is Da (1) = 6
00, the updated angle difference is Da
(1) = 510, which makes it easier to recognize “ten”.

[0083]

The first effect is that not only characters having many strokes in the vertical and horizontal directions but also characters having many strokes in the oblique direction can be recognized or rotated.

The reason is not to correct the rotation or inclination of a vertical line or a horizontal line, but to convert a character so that a direction having a high frequency of line element directions coincides with a coordinate axis.

The second effect is that the vertical line “|” and the diagonal line “/”
Is to be able to recognize the rotation and inclination of the object.

The reason is to compare the rotation angle and the inclination angle obtained from the input data with the standard rotation angle and the inclination angle of each category.

A third effect is that a character in which rotation and inclination occur simultaneously can be recognized.

The reason is that both the rotation angle and the inclination angle are detected simultaneously.

The fourth effect is that characters written while being rotated or inclined with respect to the frontal direction can always be adaptively recognized.

The reason for this is that information indicating the tendency of the input character data to rotate and tilt with respect to the case of entry in the front direction is used.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a first embodiment of the present invention.

FIG. 2 is a flowchart for explaining the operation of the first embodiment of the present invention;

FIG. 3 is a flowchart illustrating a classification process.

4A and 4B show specific examples of character data. FIG. 4A shows input character data, and FIG. 4B shows converted character data.

FIG. 5 is a diagram showing division in a line element direction;

FIG. 6 is a diagram showing a specific example of a frequency distribution in a line element direction;

FIG. 7 is a diagram showing a specific example of a synthesis frequency.

FIGS. 8A and 8B are diagrams showing specific examples of classification features. FIG. 8A shows a character image, FIG. 8B shows a direction feature, and FIG.

FIG. 9 is a block diagram showing a second embodiment of the invention;

FIG. 10 is a block diagram showing a third embodiment of the invention;

FIG. 11 is a flowchart for explaining the operation of the third embodiment of the present invention;

FIG. 12 is a block diagram showing a conventional embodiment.

Claims (6)

(57) [Claims] 1. A method according to claim 1, wherein each of the coordinate values of the input character data is coordinate-converted based on a rotation angle and a tilt angle obtained from a frequency distribution in a direction of each line element (a line connecting adjacent coordinate values) of the input character data. Extracting the classification features used for character shape discrimination from the data, standard rotation angles, standard inclination angles, classification standard features of each category stored in advance, and the rotation angles, the inclination angles, the classification features, An online character recognition device that determines the ranking of recognition candidates based on the degree of difference. 2. Coordinate conversion of each coordinate value of the input character data based on a rotation angle and a tilt angle obtained from a frequency distribution in a direction of each line element (a line connecting adjacent coordinate values) of the input character data; Extracting the classification features used for character shape discrimination from the data, standard rotation angles, standard inclination angles, classification standard features of each category stored in advance, and the rotation angles, the inclination angles, the classification features, An on-line character recognition device that determines the order of recognition candidates based on the degree of difference between a deviation rotation angle and a deviation inclination angle indicating the tendency of rotation and inclination of an input character. 3. The method according to claim 1, wherein the deviation rotation angle and the deviation inclination angle are updated using a standard rotation angle and a standard inclination angle obtained from a correct answer category input by an operator, and the rotation angle and the inclination angle. The online character recognition device according to claim 2, wherein 4. A data input means for acquiring input character data, and a frequency distribution in a direction of each line element (a line connecting adjacent coordinate values) is obtained from the input character data obtained from the data input means. Conversion angle extraction means for determining a rotation angle and a tilt angle; and, based on the rotation angle and the tilt angle obtained from the conversion angle extraction means, coordinate conversion of each coordinate value of the input character data obtained from the data input means, and conversion. Data conversion means for creating character data; classification feature extraction means for extracting classification features used for character shape discrimination from the converted character data obtained by the data conversion means; standard rotation angle and inclination of each category A standard angle storage unit that stores angles in advance; a standard classification storage unit that stores standard classification characteristics of each category in advance; The angle difference is obtained from the rotation angle and the inclination angle obtained from the output unit and the standard rotation angle and the standard inclination angle of each category obtained from the standard angle storage unit, and the classification feature obtained by the classification feature extraction unit is obtained. Classifying means for determining a feature difference of each category from the classification standard feature of each category in the classification standard feature storage unit, and determining a ranking of recognition candidates based on the angle difference and the feature difference. Online character recognition device. 5. A data input means for acquiring input character data, and a frequency distribution in a direction of each line element (a line connecting adjacent coordinate values) is obtained from the input character data obtained from the data input means. Conversion angle extraction means for determining a rotation angle and a tilt angle; and, based on the rotation angle and the tilt angle obtained from the conversion angle extraction means, coordinate conversion of each coordinate value of the input character data obtained from the data input means, and conversion. Data conversion means for creating character data; classification feature extraction means for extracting classification features used for character shape discrimination from the converted character data obtained by the data conversion means; standard rotation angle and inclination of each category A standard angle storage unit that pre-stores angles, a classification standard feature storage unit that pre-stores standard classification features of each category, Deviation angle information input means for inputting a deviation rotation angle and a deviation inclination angle representing a tendency of inclination and inclination, and a deviation angle storage unit holding the deviation rotation angle and the deviation inclination angle obtained from the deviation angle information input means. A rotation angle and a tilt angle obtained from the conversion angle extraction means, a standard rotation angle and a standard tilt angle of each category obtained from the standard angle storage unit, and a deviation rotation angle and a deviation tilt angle obtained from the deviation angle storage unit. From the angle difference,
The characteristic difference of each category is obtained from the classification feature obtained by the classification feature extraction means and the classification standard feature of each category in the classification standard feature storage unit, based on the angle difference and the characteristic difference. An online character recognition device, comprising: classification means for determining the ranking of recognition candidates. 6. A data input means for acquiring input character data, and a frequency distribution in a direction of each line element (a line connecting adjacent coordinate values) is obtained from the input character data obtained from the data input means, and the frequency distribution is obtained from the frequency distribution. Conversion angle extraction means for determining a rotation angle and a tilt angle; and, based on the rotation angle and the tilt angle obtained from the conversion angle extraction means, coordinate conversion of each coordinate value of the input character data obtained from the data input means, and conversion. Data conversion means for creating character data; classification feature extraction means for extracting classification features used for character shape discrimination from the converted character data obtained by the data conversion means; standard rotation angle and inclination of each category A standard angle storage unit that pre-stores angles, a classification standard feature storage unit that pre-stores standard classification features of each category, And a deviation angle storage unit that holds a deviation rotation angle and a deviation inclination angle representing a tendency of inclination, and a rotation angle and an inclination angle obtained by the conversion angle extraction unit and a standard of each category obtained from the standard angle storage unit. Determine the angle difference from the rotation angle and the standard inclination angle and the deviation rotation angle and the deviation inclination angle obtained from the deviation angle storage unit,
The characteristic difference of each category is obtained from the classification feature obtained by the classification feature extraction means and the classification standard feature of each category in the classification standard feature storage unit, based on the angle difference and the characteristic difference. Classification means for determining the ranking of recognition candidates; display means for displaying recognition candidates; correction information input means for specifying a correct answer category; and recognition candidates obtained by the classification means being displayed by the display means and misread. If there is, the correct information category is obtained from the operator by the correction information input means, a result correcting means for correcting the recognition result, and based on the correct category obtained from the result correcting means,
The standard rotation angle and the standard inclination angle of the correct answer category are obtained from the standard angle storage unit, and the deviation rotation angle and the deviation inclination angle in the deviation angle storage unit are updated using the rotation angle and the inclination angle obtained by the conversion angle extraction unit. An online character recognition device comprising: