JP3411949B2 - How to register stroke order and continuation characters - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a template registration method for on-line handwritten character recognition, and in particular, it is possible to efficiently register a region where a stroke order and strokes are connected to a template to improve recognition accuracy. The present invention relates to a stroke order and continuous character registration method that makes it possible.

[0002]

2. Description of the Related Art In online handwritten character recognition, strokes constituting each character, that is, the stroke of a brush when writing a character, is registered in a template in advance according to the stroke order, and each stroke is input online from a tablet in the order of the stroke. Handwriting characters are recognized by matching the strokes with the strokes of the template. In order to improve the accuracy of character recognition, it is necessary to prepare multiple different templates for one character.

As a conventional registration method for such a template, there is a method for directly registering the time-series information of the writing point of a handwritten character in the template. Further, when registering a template in which the stroke order is changed for the same character, there is a method of registering each stroke by specifying the correspondence with the stroke in the standard template one by one.

[0004]

The conventional on-line character recognition technology is designed to cope with a portion where strokes are connected by various stroke orders and continuous characters depending on individual characteristics (hereinafter, referred to as a connection portion). It is necessary to prepare a large number of templates, which has the drawback of increasing recognition time and dictionary capacity. Also, in order to deal with various stroke orders and connected parts depending on individual characteristics, it is necessary to perform registration processing in some template, but according to the conventional method of registering the relationship with the standard stroke order for each stroke. ， The operation for registration is very complicated.

The accuracy of character recognition is greatly affected by the success or failure of the template registration process, and it is particularly important to be able to deal with various stroke orders and connected parts depending on the individual characteristics. The present invention aims to solve these problems.

[0006]

While the conventional technique generally registers a stroke of a handwritten character as it is in a template, the present invention uses a stroke of the handwritten character and a basic stroke obtained from the template of the character as follows. Regarding the stroke order and the connected portion of the input handwritten characters, the presence or absence of the connected portion is determined, and if there is the connected portion, all the candidates for the stroke order and the connected portion are generated, and the optimum stroke order and the connected portion are specified by calculation. The main feature is to use means for registering information in the template.

In the above processing, when the stroke order and the connected portion to be registered are specified from the generated stroke order and the connected portion candidates, the square of the difference between each stroke length of the template and each stroke length of the input character. By obtaining the one with the smallest sum by calculation, the stroke order to be registered and the connected part can be appropriately specified.

[0008]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of the present invention will be described below. FIG. 1 is a flowchart of a stroke order and continuous character registration method, and FIG. 2 shows an example of a system configuration for implementing the present invention.

The processing device 20 shown in FIG. 2 is a device including a CPU and a memory, and has the following processing means. The input processing unit 21 executes the processes 10 to 12 shown in FIG. 1 and inputs information about handwritten characters to be registered. The candidate generation processing unit 22 processes the processes 13, 14, and 17 shown in FIG.
To generate all the stroke orders obtained from the basic template and, in the case of continuous characters, candidates for connecting parts. The candidate specifying processing unit 23 executes processing 15 and processing 18, and each candidate stroke A candidate registration processing unit 24 that specifies a candidate from the length and the stroke length of each input character.
Is for executing the processing 16 and the processing 19 and registering the stroke order and connection part information of the input handwritten character in the template. The display processing unit 25 performs display processing on the display device 29.

Reference numeral 26 in FIG. 2 is a tablet for inputting strokes of handwritten characters, 27 is a keyboard for inputting character codes and instructions given to the processing device 20, 28 is a template for storing collation information used for character recognition, Two
Reference numeral 9 represents a display device.

FIG. 3 is a diagram for explaining an example of input handwritten characters and strokes of a template, FIG. 4 is a diagram showing an example of stroke order and generated continuous characters (connecting parts), and FIG. 5 is a total stroke length. It is a figure which shows the series of the length of each stroke normalized by.

An embodiment of the present invention will be described below with reference to the flow chart shown in FIG. First, it is assumed that the template 28 stores a character code to be recognized, a maximum stroke number (basic stroke number), a minimum stroke number in the case of continuous characters, and basic stroke detailed information. The basic stroke detailed information is coordinate data of a point sequence such as a start point and an end point of each stroke.

In process 10 shown in FIG. 1, handwritten characters are input using the tablet 26 or the like when the process is started. The number of strokes of this character is Sc. In the example of the handwritten character “I” shown in FIG. 3A, the number of strokes is Sc = 5 because the first person bias is continuously written. The numbers in FIG. 3A represent the order of the input strokes.

In the next processing 11, the character code corresponding to the handwritten character input in processing 10 is input from the keyboard 27 or the like. In process 12, the template 28 is searched based on the character code input in process 11, and the actual stroke number (maximum stroke number) of the corresponding character registered in the template 28 is obtained. The number of strokes of this character, that is, the number of strokes of the basic stroke is Smax. In the example of FIG. 3 (B), Smax = 6. The circled numbers in FIG. 3B indicate stroke numbers of basic strokes.

In process 13, the value of Smax-Sc is checked,
It is determined whether or not the connected portion exists in the handwritten character input by the value of Smax-Sc. In the case of Smax-Sc> 0, since there is a connecting site, the following processing 14
All stroke orders and corresponding connected site candidates are generated.
Specifically, the stroke order is the factorial number of Smax, that is, Smax! =
If there are i candidates, and the candidates for the connected parts are Smax-1 = a and Smax-Sc = b for each stroke order,
_{There are a} C _b = j.

In the example of FIG. 3B, since there are 6 strokes from to, there are 6 factorials in all stroke orders, and 6! = 6 × 5 × 4 × 3 × 2 × 1 = 720 (pieces) of stroke order is generated. FIG. 4A shows an example of the stroke order.

In this example, since the maximum number of strokes Smax is 6 and the number of strokes of the input character is Sc = 5, the candidate of the connecting portion is _6-1 C _6-5 = ₅ C ₁ = 5 ( Exist). If the connection part is represented by a numerical string of 1 when the basic stroke is not connected and 1 when the basic stroke is connected, the connection part when the character of 6 strokes is written in 5 strokes is: "21111", "1211"
1 "," 11211 "," 11121 "," 1111 "
There will be 5 types of 2 ".

Here, the relationship between the stroke order and the connecting portion will be described. FIG. 4B shows an example of the generated consecutive characters. For example, when the stroke order is from to, but when the stroke with the basic stroke of is continuously written, the connected portion is represented as “21111”.

Further, FIG. 4C shows an example of another generated continuous character. If this continuation character,
As shown in (a), in the case of the stroke order of-, (connection) ,, the connection part is represented as "11211". As shown in (b), − (concatenation),
In the case of the stroke order of ,,,, the connecting part is "211
11 ".

In process 15 shown in FIG. 1, a sequence l _ij of the length of each stroke normalized by the total length of the strokes in the continuous state is generated from all the stroke orders and the candidates of the connecting parts generated in the process 14. Then, this is compared with the input character length sequence 1 _c to identify the closest candidate. here,
i in l _ij is the number in the stroke order, and for example, when 6 strokes of characters are input with 5 strokes, 1 ≦ i ≦ 720. Also, j
Is the number of connected parts, and in the example in which 6 strokes of characters are input in 5 strokes, 1 ≦ j ≦ 5.

The identification of this candidate will be described in more detail. For example the length of a sequence of input character When l _c, when the handwriting input as shown in FIG. _{5 (A), l c =} (l c1, l c2, l c3, l c4, l c5) Is expressed as Here, l _c1 to l _c5 are lengths after normalization.

Also, a sequence of stroke lengths l of each candidate
_ij is also normalized by a predetermined length and expressed as l _ij = (l _ij1 , l _ij2 , l _ij3 , l _ij4 , l _ij5 ). FIG. 5 (B) shows an example of a series of stroke lengths in consecutive character candidates as shown in FIG. 4 (B).

When a candidate is specified, the sum of squares of the differences between each element of the stroke length series of the input character and each element of the stroke length series including the connection part calculated from the template for each candidate is minimized. The object is determined to be the optimum stroke order and connection part.

That is, (l _c1 −l _ij1 ) ² + (l _c2 −
l _ij2 ) ² + (l _{c3 −} l _ij3 ) ² + (l _{c4 −} l _ij4 ) ²
The candidate with the smallest + (l _c5 −l _ij5 ) ² is determined as the optimal stroke order and connection site to be registered.

The stroke length is normalized as follows. To simplify the description, each stroke is assumed to be a straight line and three strokes. Set the coordinates of the start point of the first stroke to (x
_s1 , y _s1 ) and the coordinates of the end point of the first stroke are (x _e1 , y _e1 ), 2
The coordinates of the start point of the second _stroke are (x _s2 , y _s2 ), the coordinates of the end point of the second _stroke are (x _e2 , y _e2 ), and the coordinates of the start point of the third _stroke are (x _s3 , y
_s3 ) and the coordinates of the end point of the third _stroke are (x _e3 , y _e3 ).

The length of the first stroke is l ₁₁ , the length of the second stroke is l ₂₂ , and the length of the third stroke is l ₃₃ . Further, the length from the end point of the first stroke to the start point of the second stroke is l ₁₂ , and the length from the end point of the second stroke to the start point of the third stroke is l ₂₃ .

L ₁₁ = {(x _s1 −x _e1 ) ² + (y _s1 −y _e1 ) ² } ^1/2 l ₁₂ = {(x _e1 −x _s2 ) ² + (y _e1 −y _s2 ) ² } ^{_{1/2 l 22 = {(x s2}} -x e2) 2 + (y s2 -y e2) 2} 1/2 l 23 = {(x e2 -x s3) 2 + (y e2 -y s3) 2} ^{_{1/2 l 33 = {(x s3}} -x e3) 2 + (y s3 -y e3) 2} ^1/2.

The total length of the stroke in the continuous state is normalized with 255, for example. That is, L = l ₁₁ + l ₁₂ + l
_{If 22} + l ₂₃ + l ₃₃ , the normalization coefficient a is a = 25.
5 / L, and l _c1 = a × l ₁₁ l _c2 = a × l ₂₂ l _c3 = a × l ₃₃ . The l _{c1 to} l _c3 are the stroke lengths of the normalized input characters. Similarly, a candidate stroke length obtained from the template is also normalized by 255.

In process 16 of FIG. 1, the optimum stroke order and connection site specified in process 15 are registered in the template 28, and the process ends. If the maximum number of strokes Smax is equal to the number of strokes of the input character Sc, there is no connected part, so only the stroke order candidates are generated in step 17, and in step 18, the optimum stroke order candidate is calculated by the distance calculation similar to step 15. Is specified, and the stroke order is registered in the template 28 by the process 19.

FIG. 6 shows an example of the data structure of the template used in the present invention. As shown in FIG. 6 (A), the template includes a character code and a maximum stroke number (basic stroke number) Smax for each character to be recognized.
And the minimum stroke number Smin in the case of continuous characters and basic stroke detailed information. The basic stroke detailed information is coordinate data of a sequence of points such as a starting point and an ending point of a stroke for each stroke number (indicated by circled numbers in the figure).

By the stroke order and continuous character registration method according to the present invention, some stroke order information and connection part information are further registered in the template. FIG. 6B shows a specific example thereof.

For example, with respect to the handwritten character input as shown in FIG. 3 (A), the stroke order and the candidate of the connecting part as shown in FIG. 4 (B) are determined to be the optimum by the distance calculation. In this case, stroke order information (a) =
[] And connection site information (b) = [2111
1] and are registered. Although illustration is omitted, if the stroke and the stroke are written in the same stroke order in succession, as the connecting portion information, [121
11] will be registered. The stroke order information (c) and the connection part information (d) in FIG. 6 (B) are (a) in FIG. 4 (C).
7 shows an example of registration in the case where the stroke order and the candidate of the connecting portion shown in are determined to be optimum.

The length of each stroke can be obtained by a simple calculation from the coordinate data of the basic stroke, but in order to speed up the processing, the normalized stroke length is registered in advance in the template. You may do it.

When the template registered as described above is used for character recognition, it is first determined whether the number of strokes of the input handwritten character is between the maximum stroke number and the minimum stroke number, and within that range. In the case of, the stroke order information and the connection part information corresponding to the number of strokes of the input handwritten character are referred to, and the coordinates of the strokes stored as the basic stroke detailed information are calculated from the arrangement of the stroke numbers of the stroke order information. It is only necessary to obtain the data and, if necessary, connect the strokes according to the connection site information and calculate the distance. At this time, the candidates can be efficiently narrowed down by the series of the normalized stroke lengths.

[0035]

According to the present invention, various stroke orders and connecting parts depending on the characteristics of an individual can be automatically registered in the template simply by inputting the handwritten character to be registered. With the characteristics of, it is possible to recognize with high accuracy regardless of various stroke orders and how to write consecutive characters. In addition, by directly registering the input characters as in the conventional method, the capacity of the dictionary (template) is not significantly increased. Further, unlike other conventional methods, complicated operations and training are not required for dictionary registration, and according to the present invention, it is possible to automatically register the required stroke order and connected parts in the template by a simple operation. .

[Brief description of drawings]

FIG. 1 is a flowchart of a stroke order and continuous character registration method.

FIG. 2 is a diagram showing an example of a system configuration for implementing the present invention.

FIG. 3 is an explanatory diagram of an example of input handwritten characters and strokes of a template.

FIG. 4 is a diagram showing an example of a stroke order and generated consecutive characters.

FIG. 5 is a diagram showing a series of stroke lengths normalized by the total length of strokes (including the length between strokes).

FIG. 6 is a diagram showing an example of a data structure of a template.

[Explanation of symbols]

10, 11, 12 Input processing 13, 14, 17 Candidate generation process 15,18 Candidate identification processing 16, 19 Candidate registration process 20 processing equipment 21 Input processing unit 22 Candidate generation processing unit 23 Candidate Identification Processing Unit 24 Candidate registration processing unit 25 Display processing unit 26 tablets 27 keyboard 28 templates 29 Display

─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Satoshi Suzuki 3-19-2 Nishishinjuku, Shinjuku-ku, Tokyo Nihon Telegraph and Telephone Corporation (72) Inventor Kazuhiko Shinozawa 3-19-3 Nishishinjuku, Shinjuku-ku, Tokyo No. 2 Nihon Telegraph and Telephone Corp. (56) Reference JP-A-8-287185 (JP, A) JP-A-3-62186 (JP, A) JP-A-2-110796 (JP, A) JP-A 60-8991 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G06K 9/00-9/82

Claims (2)

(57) [Claims] 1. A method for registering a stroke order of handwritten characters and a part connecting strokes in a template used for online handwritten character recognition for recognizing the handwritten character from the stroke of the handwritten character, which is the stroke of the handwritten character. The character code of the handwritten character is input, a template is searched from the character code to obtain a stroke registered in the template, and between the stroke of the handwritten character and the stroke of the template, between the strokes of the handwritten character. Whether or not there is a part to be connected is determined, and if there is a part where the strokes are connected from the result of this determination, a candidate for the stroke order, which is the order of strokes, and a candidate for the part for which the strokes are connected according to the stroke order are determined. Is generated, and the stroke order that is closest to the stroke order of the handwritten character from among the generated candidates Distance Starring a site between the nearest stroke at a site between the strokes of handwritten characters is linked to linking
The position specified by calculation is registered in the template, and the part where the stroke order and the stroke are connected is registered in the template. Further, when there is no part where the stroke is connected from the discrimination result, a stroke order candidate is generated and the candidate of the stroke order is generated. The stroke order closest to the stroke order of the handwritten characters is specified by distance calculation from the inside,
A stroke order and continuous character registration method characterized by registering the specified stroke order in a template. 2. The stroke order and continuous character registration method according to claim 1, wherein a stroke order to be registered and a stroke interval to be registered are selected from among the stroke order candidates that have been generated and the candidates of regions where strokes corresponding to the stroke order are connected. The method of specifying the connecting part is characterized by comparing the stroke length series including the connecting part calculated from the template for each candidate with the stroke length series of the input character. How to register stroke order and continuous characters.