JPH0926748A - Handwriting learning method - Google Patents

Abstract

(57) [Abstract] [PROBLEMS] A handwritten character recognition system enables learning of a correct writing order of handwritten characters and a correct character pattern. A learner inputs a character code to be learned from a keyboard. The CPU 50 reads the handwritten character information in the correct writing order corresponding to the character code from the dictionary of the secondary storage 70 and outputs the corresponding reading to the voice output device 20 or the like to prompt the learner to input the handwritten character. . When a learner inputs a handwritten character from the handwritten character input / output shared display device 30, the handwritten character information is compared with the handwritten character information in the correct writing order in the dictionary, and a matching or non-matching message is output from the voice output device 20 or the like. . Further, if they do not match, an incorrect portion such as a writing order, a correct writing order, and a correct character pattern are output to the device 30.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a handwritten character learning method utilizing a handwritten character recognition system for recognizing input handwritten characters and converting them into character codes used in computers and the like.

[0002]

2. Description of the Related Art As a method of recognizing handwritten characters, there is a method of expressing one handwritten character by stroke information and relative position information. That is, the line segment that constitutes a handwritten character is considered as a set of multiple points, the coordinates of each point are expressed with the starting point at the beginning of writing as the origin, and the order of the coordinate information and the set of coordinate information are the length of the line segment. Information, and this is stroke information. Further, the angle formed by each point of the line segment and two consecutive points of the other line segment is obtained for all the points, and the set of the angle information is set as the relative position information.

Usually, for a correct writing order, a dictionary is preliminarily prepared which stores a correspondence table of handwritten character information consisting of this stroke information and relative position information and codes for necessary characters. When finding the character code corresponding to the handwritten character, obtain the handwritten character information of the stroke information and the relative position information for the input handwritten character, collate it with the handwritten character information in the dictionary, and match or compare. The corresponding character is set as the corresponding character, and the corresponding character code is obtained.

[0004]

In order to learn characters, it is necessary to acquire a correct writing order in addition to the character pattern, and it usually takes a lot of time. In addition, it is rare for adults to memorize all characters so that they can be written in a regular order.

On the other hand, in the handwritten character recognition system, in general, if the input handwritten characters are not written in a regular order, it cannot be converted into a corresponding character code.

It is an object of the present invention to provide a handwritten character learning method which efficiently teaches a learner a character pattern and a correct writing order by using this type of handwritten character recognition system.

[0007]

SUMMARY OF THE INVENTION The present invention has a dictionary that stores correspondence between handwritten character information and character codes of individual characters, and collates the input handwritten character information with the handwritten character information in the dictionary. Then, in the handwritten character recognition system that outputs a character code corresponding to a match or a relatively matching one, when the reading of the handwritten character to be learned is output and the learner inputs the handwritten character corresponding to the reading, the input Compare the written character information with the corresponding handwritten character information stored in the dictionary, and if the degree of matching or the degree of matching is high, output that learning was done correctly. , The correct writing order, the correct character pattern, etc. are output.

[0008]

In the dictionary, handwritten character information in a regular stroke order is stored in advance in correspondence with the character code. When a learner inputs a character code to be learned with a keyboard etc., the reading is output to prompt the learner to input handwritten characters, and handwritten character information in the correct writing order corresponding to the character code is read from the dictionary. Hold. By collating the handwritten character information input by the learner with the handwritten character information of the dictionary, not only the error of the character pattern but also the error of the writing order can be found, and the correct writing order and the correct character pattern are learned. It becomes possible to teach people.

[0009]

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of a handwritten character recognition system in which the present invention is implemented. A keyboard input device 10 for inputting character codes and various commands, a handwritten character reading and guidance are output by voice. A voice output device 20 used for the input, a handwritten character input / output shared display device 30 in which an input screen for inputting a handwritten character and a display screen for displaying a handwritten character pattern are configured on the same screen, a learning result, etc. are printed out. 40, a central processing unit that executes recognition processing and learning processing of handwritten characters (C
PU) 50, a memory (RAM) 60 that holds a program to be executed by the CPU 50 and is used as a work memory of the CPU 50, and the correspondence between handwritten character information and character codes in the regular stroke order in advance for each character. The secondary storage 70 holds the stored dictionary.

FIG. 2 shows the concept of the handwritten character learning process according to the present invention and the functional configuration of the CPU 50 related thereto. The learner inputs the character code C _i to be learned from the keyboard input device 10. In CPU50,
When the character code C _i to be learned is input, the handwritten character learning mechanism 51 activates the voice synthesizing mechanism 52, the voice synthesizing mechanism 52 generates a reading of the character corresponding to the character code C _i , and the voice output device 20 outputs a voice. Output. Further, the handwritten character learning mechanism 51 searches the dictionary of the secondary storage 70 using the input character code C _i as a key, and outputs the corresponding handwritten character information I.
(C _i ) is read and held. The learner inputs the corresponding handwritten character I _P (C _i ) ′ from the handwritten character input / output shared display device 30 while listening to the reading read from the voice output device 20. Based on the input handwritten character I _P (C _i ) ′, the handwritten character information creation mechanism 53 creates corresponding handwritten character information I (C _i ) ′ and passes it to the handwritten character learning mechanism 51. The handwritten character learning mechanism 51 compares the learner's handwritten character information I (C _i ) ′ with the handwritten character information I (C _i ) read from the dictionary, and if they match or relatively match, the voice synthesis mechanism 52. Through, the voice output device 20 outputs a message to that effect. On the other hand, if they do not match, a message to that effect is also output,
In addition, the mismatched portion and the portion in which the stroke order is incorrect are output to the handwritten character input / output shared display device 30. Further, the handwritten character learning mechanism 51 activates the handwritten character pattern generation mechanism 54 when they do not match, and the handwritten character pattern generation mechanism 54 corrects the handwritten character based on the handwritten character I (C _i ) of the dictionary. By creating handwritten character patterns I _P (C _i ) in order and outputting them to the handwritten character input / output shared display device 30, the correct writing order and character patterns are taught to the learner.

FIG. 3 shows an example of representation of handwritten character information used in the present invention. I _i (C _i ) is the i-th handwritten character information, which means that it has a character code of C _i .

(S _i-1 , S _i-2 , ..., S _in ) is stroke information, and one line segment (stroke) is a period from when the user starts writing the line segment to when he lifts the pen. For n line segments that constitute a handwritten character, length information (length vector) of each line segment is arranged in order from the first. That is, the element S _ij (j = 1, 2, ... N) is j
The length information of the th line segment is shown. (P _{i-1 ・ 2} , P
_{i−2 · 3} , ... P _{i− (n−1) · n} ) is relative position information, which indicates the relative positional relationship of n line segments that make up one handwritten character.
Position information (angle information) of adjacent line segments, the (j-1) th line segment and the jth line segment, is set as an element P _{i- (j-1) · j} , and the first to (n-1) ) Are arranged in order up to. As described above, the stroke information and the relative position information that constitute the handwritten character information are closely related to the writing order, and even if the same character code has a different writing order, the element array of the corresponding handwritten character information and its The values will be different.

FIG. 4 shows an example of the structure of the dictionary. The dictionary generally associates the handwritten character information I _i (C _i ) in the normal writing order with the character code C _i for many handwritten characters, and is stored in advance in the secondary storage 60 from the first address. There is something. The expression format of the stroke information S _ij and the relative position information P _{i- (j-1) · j} forming the handwritten character information is the same as that in FIG. In the secondary storage, for convenience of retrieval, for example, the handwritten characters are stored in the order of the smallest number of line segments.

Next, an example of the processing flow of handwritten character learning according to the present invention will be described in detail with reference to FIG.

When the learner inputs the learning target character code C _i from the keyboard input device 10 (step 101),
The handwritten character learning mechanism 51 of the CPU 50 activates the voice synthesizing mechanism 52 and passes the input character code C _i . The voice synthesizing mechanism 52 generates a reading of a character corresponding to the character code C _i and outputs the voice from the voice output device 20 (step 102). In addition, the reading may be displayed in a predetermined area on the display surface of the handwritten character input / output shared display device 30 in hiragana and output. On the other hand, the handwriting learning mechanism 51
Reads the handwritten character information I (C _i ) in the regular writing order (basic) corresponding to the input character code C _i from the dictionary of the secondary storage 70 and holds it (step 103).

The learner listens to the reading output from the voice output device 20 or while looking at the hiragana displayed on the handwritten character input / output shared display device 30, the corresponding handwritten character I _P (C _i ) ′. Is input from the handwritten character input / output shared display device 30 (step 104). Based on the coordinate information of the handwritten character I _P (C _i ) ′ input from the handwritten character input / output shared display device 30, the handwritten character information creation mechanism 53 of the CPU 50 determines stroke information that reflects the stroke order of the learner. Handwritten character information I consisting of relative position information
(C _i ) ′ is created and passed to the handwritten character learning mechanism 51 (step 105). Handwriting learning mechanism 51 calculates the distance by comparing the learner handwriting information I (C _i) 'a dictionary of handwritten character information I (C _i) (step 106),
It is determined whether or not they match or close to each other (step 107).
Here, the distance calculation is performed by the learner's handwritten character information I.
Stroke information (S _i-1 , S _i-2 , which composes (C _i ) ′,
S _in ) ′ and relative position information (P _{i− 1,2} , P _i− 2,3, ... P)
_{i- (n-1) n} ) ', and dictionary handwritten character information I (C _i )
Stroke information (S _i-1 , S _i-2 , S _in ) and relative position information (P _{i-1 ・ 2} , P _{i-2 ・ 3} , ... P _{i- (n-1) ・ n} ) For each element of | S _{ij to} S ′ _ij |, | P
_{i- (j-1) · j to} P ′ _{i- (j-1) · j} | are sequentially obtained (j = 1, 2,
... n), and the total sum (referred to as similarity) is calculated.

In the handwritten character learning mechanism 51, the handwritten character information I (C _i ) 'of the learner is the handwritten character information I of the dictionary.
If it matches or is close to (C _i ), the voice synthesizing mechanism 52 is instructed, and the voice synthesizing mechanism 52 generates a message to the effect and outputs it from the voice output device 20. At the same time, if necessary, the handwritten character pattern generation mechanism 54
Is started, a basic handwritten character pattern is generated by the handwritten character pattern generation mechanism 54 based on the handwritten character information I (C _i ) of the dictionary, and the handwritten character input / output shared display is compared with the handwritten character of the learner. Display on device 30 (step 10)
8). On the other hand, if they do not match, the handwritten character learning mechanism 51 similarly instructs the voice synthesis mechanism 52 to generate a mismatch message and output it from the voice output device 20. Based on this, the strokes determined to be inconsistent (this includes the portion in which the stroke order is incorrect) are emphasized to display the handwritten characters of the learner on the handwritten character input / output shared display device 30 (step 10).
9). Further, the handwritten character learning mechanism 51 activates the handwritten character pattern generation mechanism 54, and in the handwritten character pattern generation mechanism 54, the handwritten character information I of the dictionary.
Create a correct handwritten character pattern based on (C _i ),
The strokes are displayed on the handwritten character input / output shared display device 30 in the correct stroke order.

In the above description, the matching or non-matching message may be displayed on the handwritten character input / output shared display device 30. In learning, the system automatically prompts the learner to input the relevant handwritten character by waiting for the learner's keyboard input device 10 to input the learning target character code, and automatically reading and outputting the learning target character. But it is okay.

Next, an example of a method of creating stroke information and relative position information which constitute handwritten character information will be described with reference to FIGS. 6 to 9.

Generally, handwritten characters are composed of a plurality of line segments. For the sake of convenience, the hiragana "i" is taken as an example here. When the hiragana is “i”, the number of line segments is 2, as shown in FIG. 6, for example. First, in order to create handwritten character information,
The area where each line segment is written is converted to a certain size. Since this is shown in FIG. 6, an example of reduction is shown here, but when the input handwritten character is small, it is enlarged. Then, as shown in FIG. 7, each line segment is represented as a set of a plurality of points. At that time, the total number is fixed, and the number of points is distributed in proportion to the length of the line segment. Stroke information and relative position information are obtained from the point information thus obtained.

FIG. 8 is a diagram showing how to obtain stroke information. First, assuming the XY coordinate system with the starting point of the character as the origin, the coordinates of each point are obtained (FIG. 8A). Next, assuming a vector having two adjacent points forming a line segment as the start point and the end point, a line-up of all the vectors forming the line segment is set as the composite vector of the line segment. This is obtained for each line segment (FIG. 8 (B)).

In FIG. 8 (B), with respect to the left line segment (defined as line segment 1) that constitutes the hiragana “i”, 1 ₁₁ = (x ₁₁ , y ₁₁ ), (x ₁₂ , y ₁₂ ) 1 ₁₂ = (x ₁₂ , y ₁₂ ), (x ₁₃ , y ₁₃ ) _··· 1 _1j = (x _1j-1 , y _1j-1 ), (x _1j , y _1j ) S 1 = { 1 ₁₁ , 1 ₁₂ , ... 1 _1j } are composite vectors. However, (x ₁₁ , y ₁₁ ) = (0,
0). Similarly, for the right side of the line segment (a line segment _{2), S2 = {1 21} , 1 22, ··· 1 2j} is a complex vector.

The stroke information is an array of composite vectors, and is represented by S = (S1, S2) in the case of FIG. Generally, the stroke information of a character having a character code Ci is Si (Ci) = (S _i-1 , S _i-1 , ... S _{is (i)} ) = [S _ij ] (j = 1, ... -It is expressed as S (i)). Here, s (i) is the number of line segments of the character C _i .

FIG. 9 is a diagram showing how to obtain the relative position information. To obtain the relative position information, the point information finally obtained is used as in the case of obtaining the stroke information (FIG. 9 (A)). Using this point information, the angle formed by a line connecting a point on one line segment and two adjacent points on another line segment is obtained (FIG. 9B). This angle is obtained by using the cosine theorem or the like and is stored as a cosine value. For example, FIG. 9 (B)
In, the angle without the starting point 1 of the left line segment (line segment 1) and the first two points of the right line segment (line segment 2) that make up the flat “i” is

[0026]

[Equation 1]

It can be obtained by This is calculated between all the points of the line segment 1 (first line segment) and all the two adjacent points of the line segment 2 (second line segment). The cosine values are arranged with respect to the line segment 1 and the vector P _{1 · 2} = (cos θ ₁₁ , cos θ ₁₂ , ... Cos
θ _1k , cos θ ₂₁ , cos θ ₂₂ , ... Cos θ _jk ) are _created . This is the angle information when line segment 1 to line segment 2 are viewed. Since the number of line segments is 2 for a flat kana "i", the angle information P ₁ .2 is directly used as relative position information.

Generally, assuming that one handwritten character is composed of n line segments, the angle information P ₁ .2 when the line segment 1 to the line segment 2 is first viewed in the writing order. , Angle information P ₂ when line segment 2 to line 3 is viewed, and finally, angle information P when line segment n is viewed from line segment n-1
_{Find n-1 · n} . At this time, the relative position information is P = (P _{1 .2} , P _{2 .3} , ... P _{n-1 .n} ). The relative position information of the character having the character code C _i is P _i (C _i ) = (P _{i−1 · 2} , P _{i−2 · 3} , ... P
It is expressed as _{i- (s (i) -1) ・ s (i)} ). Here, s (i) is the number of line segments of the character C _i .

The relative position information is converted into the first and second positions as described above.
By setting the angle information of the 2nd line segment, the angle information of the 2nd line segment and the angle information of the 3rd line segment, ... As a result, the processing time can be shortened, and the dictionary capacity and work area can be reduced.

The stroke information and the relative position information that constitute the handwritten character information need not be limited to those described above. For example, the relative position information may be obtained as a set of all combinations of relative positional relationships between itself and other line segments for each line segment.

[0031]

As described above, according to the present invention, the learner inputs the handwritten characters and compares them with the correct stroke order and pattern information already registered in the dictionary to compare the error portion, the correct stroke order, and the correct characters. Outputting the pattern enables efficient learning of correct handwritten characters.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an example of a handwritten character recognition system in which the present invention is implemented.

FIG. 2 is a diagram showing a concept of a handwritten character learning process according to the present invention and a functional configuration of a CPU related thereto.

FIG. 3 is a diagram showing a configuration example of handwritten character information.

FIG. 4 is a diagram showing a configuration example of a dictionary.

FIG. 5 is a flowchart for explaining a handwritten character learning process according to the present invention.

FIG. 6 is a diagram showing a specific example of handwritten characters and a reduction example thereof.

FIG. 7 is a diagram showing conversion of a line segment of a handwritten character into a sequence of dots.

FIG. 8 is a diagram illustrating generation of stroke information.

FIG. 9 is a diagram illustrating generation of relative position information.

[Explanation of symbols]

 10 keyboard input device 20 voice output device 30 handwritten character input / output shared display device 40 printing device 50 processing device (CPU) 51 handwritten character learning mechanism 52 voice synthesis mechanism 53 handwritten character information generation mechanism 54 handwritten character pattern generation mechanism 60 memory (RAM ) 70 Secondary storage for dictionary storage

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Noboru Sonehara No. 1-6 Uchisaiwai-cho, Chiyoda-ku, Tokyo Japan Telegraph and Telephone Corporation (72) Go Ito 6-11-11 Shimbashi, Minato-ku, Tokyo NTT PC Communications Co., Ltd.

Claims (2)

[Claims] 1. A dictionary that stores the correspondence between handwritten character information and character codes for each character, and collates the input handwritten character information with the handwritten character information in the dictionary, and matches or relatively matches. A method for learning handwritten characters by a handwritten character recognition system that outputs a character code corresponding to an object, comprising outputting a reading of a handwritten character to be learned and inputting a handwritten character corresponding to the reading when the learner inputs the reading. Compare the handwritten character information that was stored in the dictionary with the corresponding handwritten character information that is stored in the dictionary, and if the match or degree of match is high, output that learning was successful, otherwise output the incorrect part. A learning method for handwritten characters. 2. The handwritten character learning method according to claim 1, wherein when the learning cannot be performed correctly, a correct writing order and a correct character pattern are output based on the handwritten character information in the dictionary.