JPH09231314A - On-line handwritten character recognizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve a recognition ratio independently of the number of strokes and writing order in the case of a low stroke character. SOLUTION: A tablet 1 extracts a handwritten data string and a preprocessing part 2 executes coordinate string smoothing processing. An on-line type feature extracting part 3 executes rough sorting based upon the number of strokes and executes medium sorting by comparing a vector distance dvec between partial patterns of an input character and a dictionary character with a threshold to calculate the featured value of the input character. An on-line type matching part 4 calculates a distance dBP from a character registered in an on-line type recognition dictionary 5 and the featured value of the input character and calculates a distance from the distances dvec dBP. In the case of a high stroke character, a high stroke character discriminating part 6 outputs a character code in accordance with the distance. A pattern memory conversion part 7 executes conversion into a pattern memory, a contour extracting part 8 executes contour extracting processing and an off-line type feature extracting part 9 calculates an off-line featured value. An off-line type matching part 10 executes matching between an off-line type recognition dictionary 11 and the featured value stored in the pattern memory and a low stroke character discriminating part 12 executes character recognition.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an on-line handwritten character recognition device for identifying a handwritten character in real time.

[0002]

2. Description of the Related Art Conventionally, in online character recognition, there is a pattern matching method as a general character recognition method. In this pattern matching method, the feature point information is extracted in advance from the coordinate data string of the written stroke (writing portion from pen-on to pen-off) by the same method and registered (hereinafter, registered pattern). Pattern recognition) and perform character recognition.
In this pattern matching method, it takes a lot of time to process which stroke of each stroke of the registered pattern each written stroke should be matched with. In addition, when the balance of the whole characters is disturbed, for example, in the case of the character "Kawa", when "Si" is small and "OK" is large, the result that the matching result is not similar is obtained. Weak to.

Therefore, as an on-line handwritten character recognizing device which compensates the above-mentioned drawbacks of the pattern matching method and requires a small amount of processing, for example, Japanese Patent Laid-Open No. 62-22934.
The technology disclosed in Japanese Patent Laid-Open Publication No. H10-26095 has been proposed. In this on-line handwritten character recognition device, large classification is performed according to the number of strokes of a written character, and a portion to be written as a series of writing is defined as a partial pattern. Then, middle classification is performed by the vector between the centers of gravity of the partial patterns, and matching is performed using the Q value and the stroke distribution value as the characteristic parameters of the partial patterns. This makes it possible to perform character recognition that is resistant to character deformation and has a small amount of processing. According to the result of this matching, the candidate characters are ranked and the result is output to a display or the like.

[0004]

However, the conventional online handwritten character recognition device has the following problems. For example, the character "mute" is a partial pattern "mouth" +
Characters are treated as a set of partial patterns, as defined by "A". In this case, the "mouth" is generally
It will be written in the order of "A". As described above, a character having a relatively large number of strokes such as the character "mute" does not cause any problem, but for a character having a small number of strokes such as "+", the character "+" is a partial pattern "-". , "|" Is defined. In this case, when written in the order of "|" and "-", they are erroneously recognized or rejected. As a countermeasure, the character "+" is "-" + "|" and "|"
It is sufficient to define both stroke orders of + and "-", but this causes a problem that the dictionary becomes enormous and processing time is required. Also, characters with a large number of strokes, for example, 14
The letter "accept" in the picture would be defined as "word" + "blade" + "heart". In this way, since the number of partial patterns generally increases and the amount of information increases, even if the matching distance value of one partial pattern is bad, a correct recognition result can be obtained with the matching distance values of other partial patterns. However, the character "+" having a small number of strokes has a problem that there is a high possibility of erroneous recognition because the amount of information is small.

[0005]

In order to solve the above-mentioned problems, the present invention eliminates unnecessary data in a coordinate sequence of handwritten characters obtained by handwriting on a tablet in an online handwritten character recognition device and smoothes it. A pre-processing unit that performs the conversion, an online-system feature extraction unit that extracts a feature amount of the written character from the coordinate data sequence smoothed by the pre-processing unit, and a feature of the written character obtained by the online-system feature extraction unit. When the number of strokes obtained by the stroke number of the written character is equal to or more than a predetermined stroke number, the online method matching unit that calculates the matching distance between the amount and the feature amount of the character to be recognized that is stored in advance in the online method recognition dictionary is It is provided with a high stroke number character identification unit that performs character recognition based on the matching distance output from the online system matching unit.

Then, an area including the entire writing character obtained from the coordinate data string smoothed by the preprocessing unit is divided into a plurality of areas vertically and horizontally, and each of the plurality of areas is one of the writing characters. A pattern memory conversion unit that determines whether or not a part is included and allocates binarized data to each area and stores the binarized data in a pattern memory; a contour extraction unit that reads the pattern memory and extracts contour data; An offline method feature extraction unit that extracts a feature amount of a written character from the contour data obtained by the contour extraction unit, and a feature amount of the written character obtained by the offline method feature extraction unit and an offline method recognition dictionary that are stored in advance. An offline method matching unit that calculates a matching distance with a feature amount of a certain recognition target character, and the number of strokes obtained by the number of strokes of a written character is the predetermined number of strokes. When remote small, on the basis of the matching distance which is output from the off-line mode matching section, and a low number of strokes character identification unit performing character recognition.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION First Embodiment FIG. 1 is a block diagram showing an online handwritten character recognition device according to a first embodiment of the present invention. The difference between the online handwritten character recognition device of the first embodiment and the conventional online handwritten character recognition device is that, when the written characters are low stroke number characters, the offline recognition dictionary has an offline feature amount that does not depend on the writing order and the stroke number. That is, the character recognition is performed by matching with the recognition target character. As shown in FIG. 1, the online handwritten character recognition device according to the first embodiment includes a tablet 1, a preprocessing unit 2, an online system feature extraction unit 3, an online system matching unit 4, an online system recognition dictionary 5, and a high stroke count. The character identification unit 6, the pattern memory conversion unit 7, the contour extraction unit 8, the offline feature extraction unit 9, the offline system matching unit 10, the offline system recognition dictionary 11, the low stroke number character identification unit 12, and the output terminal 13 are provided.

The output side of the tablet 1 is connected to the preprocessing section 2. The coordinate sequence of the output of the preprocessing unit 2 is the online system feature extraction unit 3 and the pattern memory conversion unit 7.
Is connected to the input side of the high stroke number character identifying section 6 and the low stroke number character identifying section 12 and the input side thereof. The output sides of the online method feature extraction unit 3 and the online method recognition dictionary 5 are connected to the input side of the online method matching unit 4. The output side of the online system matching unit 4 is connected to the input side of the high stroke number character identification unit 6. The output side of the pattern memory conversion section 7 is connected to the input side of the contour extraction section 8, and the output side of the contour extraction section 8 is connected to the input side of the off-line method feature extraction section 9. Offline system feature extraction unit 9 and offline system recognition dictionary 1
The output side of 1 is connected to the input side of the offline system matching unit 10. Off-line method matching unit 10
The output side of is connected to the input side of the low stroke number character identification unit 12. High stroke number character identification section 6 and low stroke number character identification section 12
The output side of is connected to the output terminal 13. This on-line handwritten character recognition device is configured by an individual circuit using a semiconductor integrated circuit, program control of a digital signal processor (DSP), or the like.

The tablet 1 is for writing characters. The pre-processing unit 2 performs a smoothing process on the handwritten data and obtains the stroke number of the handwritten character. The online method feature extraction unit 3 roughly classifies the written characters according to the number of strokes, and divides the written characters divided according to the stroke position (hereinafter referred to as the cut position) of the partial pattern of the recognition target character registered in the recognition dictionary. Matching distance d _vec from the partial pattern vector to the character in the recognition dictionary
Is obtained, the middle of the recognition target character is classified, and the Q value which is the feature amount of the partial pattern is further obtained. The online system matching unit 4 uses the online system recognition dictionary 5
The matching distance d _BP between the Q value of the character to be recognized and the Q value of the handwritten character registered in (3) is calculated, and the matching distance is calculated from the distances d _vec and d _BP . The high stroke number character identification unit 6 sorts the matching distance and outputs the corresponding character code when the written character has a high stroke number.

The pattern memory conversion unit 7 has a pattern memory, converts the handwritten character into binary data, and stores it in the pattern memory. The contour extraction unit 8 obtains contour information of the binarized data. The offline system feature extraction unit 9 extracts the offline feature amount of the written character from the contour information. The offline system matching unit 10 obtains the number of strokes of the recognized characters registered in the offline system recognition dictionary 11 and the matching distance between the offline feature amount that does not depend on the writing order and the offline feature amount of the written character. The low stroke number character identification unit 12 sorts the matching distance output from the offline method matching unit 10 and outputs a corresponding character code when the number of written characters is low. Hereinafter, operations (a) to (j) of each unit in FIG. 1 will be described.

(A) Tablet 1 When a character is input by handwriting, the tablet 1 samples the handwritten character at a predetermined sampling frequency, and writes a coordinate data handwriting data string {(x _i , y _i ), i = 1. , 2, ..., n
_j } j (j is a stroke number) is extracted, and the preprocessing unit 2
Send to (B) Pre-Processing Unit 2 FIGS. 2A to 2C are operation explanatory diagrams of the pre-processing unit 2. In the figure, “·” represents a writing data string from the tablet 1 or a feature point. The preprocessing unit 2 has been sent FIG.
For the writing data string shown in (a), noise removal processing,
By performing a moving average process or a smoothing process, the data is smoothed as shown in FIG. 2B, and the number of strokes (the number of strokes) is obtained, and the online feature point extraction unit 3 and the pattern memory conversion are performed. The writing data string after smoothing is output to the unit 7, and the stroke number is output to the high stroke number character identifying unit 6 and the low stroke number character identifying unit 12.

(C) Online Method Feature Extraction Section 3 The offline method feature extraction section 3 extracts data, which is a feature of character recognition, from the output of the preprocessing section 2. Several methods can be considered for this, but one example will be described. As a method of extracting this feature point, a sine sign (positive, negative) between the data of the smoothed data string {(X _i , Y _i ), i = 1, 2, ..., N _j } j. , 0) is calculated by the following equations (1) and (2), and the change point of the sign state is extracted as a feature point. XS _i = Sign (X _i −X _i−1 ) ... (1) YS _i = Sign (Y _i −Y _i−1 ) ... (2) and expressed by +, 0, −. The signs in the x direction and the y direction between the respective data obtained in this way are compared with the signs of the previous data, and if both are the same, they are not registered as feature points. If the coordinates have changed, the coordinates are registered as a feature point for each stroke.

FIG. 2C shows characteristic points to which the start point and the end point of the stroke are added in addition to the points thus obtained. In general, this process may be called linear approximation. These feature points are hereinafter referred to as segments, and are represented by feature points {(X _i , Y _i ), i = 1, 2, ..., I _j } j. FIG. 3 is a diagram showing an example of a character dictionary for calculating a partial pattern vector. Next, the target character is roughly classified according to the number of strokes. for that reason,
Characters that can be the number of strokes are prepared in advance in a character dictionary for each number of strokes (stroke number), as shown in FIG. For example,
It is assumed that the stroke number of the character pattern input by handwriting is 10 strokes. In this case, among the characters stored in the character dictionary, the character "mute" that can be 10 strokes as shown in FIG.
"Dust", "Ai", ... Are selected as candidate characters. next,
Candidate characters obtained by large classification according to the number of strokes as described above are further middle-classified according to inter-pattern vectors described below. Here, the partial pattern refers to a portion of one character to be written as a series in writing, and the partial pattern vector refers to the center of gravity of one partial pattern and the center of gravity of another partial pattern as starting points, Let us say the vector that is the end point.

First, an example of a method of calculating a vector between partial patterns will be described. X, y of each segment in the partial pattern
The component (vector connecting the feature points) is (dx _i = X _i −X
_{i-1, dy i = Y} i -Y i-1) (i = 1,2, ..., n)
Then, the length dl _i of each segment is represented by dl _i = (dx _i ² + dy _i ² ) ^1/2 (3). Further, the center coordinates of each segment normalized by dividing by the character widths HX and HY are (x _i ^* , y
_i ^* ), the normalized barycentric coordinates (X _W , Y _W ) of the partial pattern are

[Equation 1] Is required. The center of gravity of each partial pattern is obtained by the above method, 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. Here, the vector between the partial patterns is x
Consider the direction and the y direction respectively.

FIG. 4 is an explanatory diagram of a vector k between partial patterns, which is the center of gravity of each of the partial patterns "mouth" and "sub" of the character "mah" and the vector B _x (mouth, sub), B _y between the partial patterns.
An example of (mouth, sub) is shown. However, the X mark in FIG. 4 indicates the center of gravity of the mouth and sub, and B indicates the vector. Here, the candidate character selected by the number of strokes is narrowed down by the partial pattern vector to perform the middle classification. Here, as an example, consider the case where "Ai" is written and input. The procedure will be described. Since the written character "O" is 10 strokes, it is 1 in the character dictionary shown in FIG.
Refer to the 0 stroke part. Then, in this character dictionary, the character "mute" is arranged at the first position, and the column "mute"
The partial patterns that make up the partial pattern, the writing order of the partial patterns, the stroke number information of the partial patterns (hereinafter referred to as cut positions), and the inter-partial-pattern vector values calculated in advance from the registered patterns are shown. Below, in order, "dust",
Similar information is lined up for the character "Ai". According to this character order, it is determined whether or not it should be a candidate, and middle classification is performed as follows. First, assuming that the character input by handwriting is "mute", the matching distance d _{vec of the} vector between the partial patterns is calculated as follows.

FIG. 5 is a diagram for explaining the calculation of the matching distance d _vec . As written in the character dictionary shown in FIG. 3, the "mute" has a cut position of (3, 7), that is,
The "mouth" is formed by the first to third strokes,
Although "sub" is formed in the 4th to 7th strokes, the input pattern is "A" in this example, so considering the vector between partial patterns for "A" at this cut position, as shown in FIG. become. In FIG. 5, B _x ^* and By _y ^* are
It is an inter-partial-pattern vector of "a" that is obtained in accordance with the stroke of the "mute" partial pattern.

The inter-partition vector B ^* _x of the partial patterns of the input pattern obtained as shown in FIG.
Nitrous), B ^* _y (mouth, nitrite) and the partial pattern between the vectors B _x (mouth written in the character dictionary, nitrous), B _y (mouth, nitrite) the difference between the is matching distance DVEC, the following equation Calculated in (6). d _vec = | B _x ^* (mouth, nitrous) -B _x (mouth, nitrous) | + | B _y ^* (mouth, nitrous) -B _y (mouth, nitrous) | ··· (6) in general, was written When there are multiple character partial patterns, normalization is performed using the partial pattern number BPN, and the matching distance d
_vec is

[Equation 2] Is calculated according to Set the threshold VECREJ here,
It is determined whether the calculated d _vec is larger than VECREJ.
Then, when d _vec > VECREJ, it is determined that the character is not the referenced character (in this case, “mah”), and the partial pattern vector of the next character is matched. When d _vec ≤ VECREJ,
In order to further check whether or not it is a "mute", the partial pattern Q value is calculated and matched as described below.

The Q value of the partial pattern is calculated for the candidate characters remaining in the middle classification by the vector between the partial patterns.
Here, the partial pattern Q value is the length of each segment,
A characteristic parameter indicating a direction and a position. In the online character recognition, the movement of the writing pen is the X direction,
The Y direction, the + direction, or the − direction is obtained as important information, and it is this partial pattern Q value that is effectively used. First, a method of calculating the partial pattern Q value will be described. In the following equations (8) to (15), Σ indicates the total stroke, addition regarding all segments, and HX and HY indicate the character width.

[0019]

(Equation 3) Expressions (8) to (15) show the pattern distribution of the partial patterns. For example, the large value shown in Expression (8) means that the pattern written in the right direction in the right part of the written character is It is estimated that there are many distributions.

In the case of the equations (8) to (15), the values are the values of the respective directional positions when the origin is set to the lower left. At this time, the values near the origin become 0 when subjected to multiplication. Therefore, in order to prevent the value from becoming 0, the same applies to the values Q9 to Q16 at each directional position when the origins are exchanged and the origin is set to the upper right. Shall represent the length, direction and position of each stroke segment. The online method feature extraction unit 3 calculates a partial pattern Q value for the candidate characters remaining after the classification by the partial pattern vector, and outputs the Q value to the online method matching unit 4. For example, "Ai"
It is assumed that "dust" remains in the classification by comparing the matching distance d _vec of the partial pattern vector and VECREJ by manually writing. In this case, if the recognition target character is "dust",
As shown in Fig. 3, the cut position of "dust" is (3, 2,
5), and the partial pattern is "ki" + "mu" +
Since it is an "arrow", the input pattern "Ai" is cut at the cut positions (3, 2, 5), and Q1 ^{* to} Q16 ^* respectively ^.
Is calculated and output.

(D) Online system matching unit 4 The online system matching unit 4 receives the output of the online system feature extraction unit 3 and performs matching with an online system recognition dictionary 5 stored in advance to calculate a matching distance value. And outputs it to the high stroke number character identification unit 6. FIG. 6 is a diagram showing an example of the online system recognition dictionary 5. The recognition dictionary 5 is provided with columns for the number of strokes, a partial pattern, and a partial pattern Q value, and information for each partial pattern is stored in each column. In FIG. 6, a plurality of partial patterns are defined for the same character in consideration of the case where the writing order is incorrect (for example, “mouth” is three strokes,
Also defined as one stroke).

The online method matching unit 4 calculates the partial pattern Q value Q1 ^* to be calculated by the online method extracting unit 3.
Q16 ^* is matched with the partial pattern Q value in the online system recognition dictionary 5. In this matching, first, Q1 ^{* to} Q16 ^* calculated for "hi" of "Ai" and Q1 to Q16 of "ki" of the recognition target character "dust" are matched. Next, Q1 ^{* to} Q16 calculated by "D" in "Ai"
^* The recognition object character "挨" of by matching the Q1~Q16 of "no", further "Encounter" of "Tsuji" calculated Q1 ^* ~ in
Q16 ^* and Q1 to Q16 of the recognition target character "dust""arrow"
To match. The sum of the absolute values of the differences in these matchings is the matching distance d _BP .
At this time, the distance d _BP represents how close the input pattern “Ai” is to “Dust”. Generally, the number of strokes BS _j of each partial pattern j is used for weighting as follows,
Let it be the matching distance d _BP .

[0023]

(Equation 4) However, Q _i (j) is the j-th partial pattern Q value of the character dictionary, Q _i ^* (j) is the j-th partial pattern Q value of the input character pattern, and N is the total number of strokes (stroke count). . The distance d _BP obtained as described above and the dvec obtained by matching the vectors between partial patterns obtained by the online system feature extraction unit 3 weighted with α _vec are added to obtain the distance d. d = α _vec · d _vec + d _BP (17) α _{vec in} equation (17) is α _vec = 1/2, 4 to 25 strokes when the number of strokes of the written character is 2 to 3 strokes. Α _vec = 1 for 1 and α _vec = 0 for 1 stroke
And The above process is performed for all candidate characters remaining in the classification by the vector between partial patterns, and d _k {k = 1,2,
, N} is calculated and output to the high stroke number character identifying unit 6. However, n is the number of candidate characters remaining in the middle classification by the vector between the partial patterns.

(E) High stroke number character identification section 6 The high stroke number character identification section 6 is an online system matching section 4
In response to the output d _k and the number of strokes output by the preprocessing unit 2, the input character identification result is output. The high stroke number character identifying unit 6 sorts the output d _k of the online method matching unit 4 in ascending order, and when it is determined that the stroke number is high, outputs the character code corresponding to the sorted order from the output terminal 13.
It is displayed on a display device (not shown). If it cannot be determined that the number of strokes is high, nothing is output. Here, it is considered that the number of strokes is high depending on the characteristic parameter to be extracted, but here it is assumed that characters with 7 strokes or more are high strokes. (F) Pattern memory conversion unit 7 coordinate data string of output data of preprocessing unit 2 {X _i , Y _i }
In response to this, the pattern memory conversion unit 7 causes the two-dimensional array PM
It is converted to (x, y) and output to the contour extraction unit 8. FIG. 7 is a diagram showing pattern memory conversion processing of the pattern memory conversion unit 7. Although various methods can be considered for the pattern memory conversion processing, an example thereof will be described with reference to FIG. 7.

Input pattern is n × m (n ≧ 2, m ≧ 2)
And the output coordinate data string {X _i ,
If Y _i } exists in the divided frame, 1 is set, and if not, 0 is set, and each n × m frame PM is set.
(x, y) {x = 0,1,2,…, n-1, y = 0,1,2,… m-1} is calculated. Hereinafter, a method for obtaining each frame PM (x, y) will be described. here,
As shown in FIG. 7, two adjacent coordinate data strings are represented by {x ₀ ,
y ₀ }, {x ₁ , y ₁ } (provided that y ₀ ≤y ₁ ),
The origin of the frame to be checked is (W _x , W _y ), the widths of the frame in the X and Y directions are W _a and W _b , respectively, the memory area of the frame to be chucked is A, and the result of the frame to be checked is PM _A And At this time, PM _A = 1 when any of the following conditions 1) to 5) is satisfied. Condition 1) A case where two adjacent coordinate points are lined up in the vertical direction, and the memory area A is located on both sides of the two coordinate points, which is expressed by the following equation (18). x ₀ = x ₁ AND W _x ≦ x ₀ <W _x + W _a AND y ₀ ≦ W _y AND y ₁ > W _y + W _b (18) Condition 2) Two adjacent coordinate points are arranged in the horizontal direction. This is a case where the memory area A is located with two coordinates sandwiched therebetween, and is expressed by the following equation (19). y ₀ = y ₁ AND W _y ≦ y ₀ <W _y + W _b AND x ₀ ≦ W _x AND x ₁ > W _x + W _a (19) Condition 3) One of two adjacent coordinate points is This is the case of being included in the memory area A, and is expressed by the following equation (20).

(X ₀ , y ₀ ) A OR (x ₁ , y ₁ ) A (20) Condition 4) A line segment connecting two adjacent coordinate points intersects the memory area A at y = W _y This is the case and is represented by the following equation (21). f (W _x ) <W _y AND f (W _x + W _a ) <W _y AND ((y ₀ <W _y AND y ₁ > W _y ) OR (y ₀ > W _y AND y ₁ <W _y ) ・ ・・ (21) However, f (x) is expressed by equation (25) Condition 5) The line segment connecting two adjacent coordinate points is y = W _y +
This is the case where the memory area A intersects with W _b , and the following equation (22)
It is represented by

F (W _x ) ≧ W _y + W _b AND f (W _y + W _x ) <W _y + W _b AND ((y ₀ <W _y + W _b AND y ₁ > W _y + W _b ) OR (y ₀ >) W _y AND y ₁ <W _y + W _b ) ... (22) where f (x) is expressed by equation (25), and f (x) is as follows: The connecting straight line can be expressed by the following equation (23): y = ax + b (23) When two coordinate sequences (x ₀ , y ₀ ) and (x ₁ , y ₁ ) are substituted, a and b are calculated. ,

(Equation 5) The above processing is performed for all coordinate data strings {X _i , Y _i }, and PM (x, y) is obtained as a result. With this, the binarized data of the input character pattern is created. This PM (x, y) is output to the contour extraction unit 8.

FIGS. 8 (a) and 8 (b) are diagrams showing an example of pattern memory conversion. FIG. 8 (a) is a handwritten input character "a", and FIG. 8 (b) is a converted pattern. It is a figure which shows a memory. (G) Contour extraction unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), the contour of the binarized input character pattern is extracted. FIG. 9 is a diagram for explaining the operation of the contour extracting unit 8, and the arrow in the drawing indicates the direction of contour tracing. A known method for extracting the contour is, for example, starting from the pixel located in the pattern located at the upper end in the pattern memory as a starting point, for example, the first pixel of “1” among the eight neighboring pixels is sequentially arranged. Trace and extract the contour information. For example, when the written character is "A", as shown in FIG.
Three pieces of contour information, that is, the outer contour F1 of "a", the left outline F2, and the right outline F3 are obtained.

(H) Offline Method Feature Extraction Unit 9 The offline method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and calculates the feature amount. Various methods have been considered for extracting a feature amount for character recognition from the contour data in an optical character reader, and an example thereof will be described. First, a contour is input, and a line segment connecting two adjacent coordinate points (pixels) for each contour is classified into four cases as follows. FIG.
FIG. 4 is a diagram showing classification of components of contour data.

As shown in FIG. 10, the left end of the pattern memory is the origin (0, 0), the coordinate value of the target point of the contour data is (x _i , y _i ), and the coordinate value of the next contour data point is (x. _{i + 1} , y _{i + 1} ) and d _x = x _{i + 1} −x _i (26) _dy = y _{i + 1} −y _i (27), then d _y = 0 Then, the horizontal component is obtained if d _x = 0 and the vertical component is obtained if d _x × d _y > 0. −45 ° component If d _x × d _y <0, the + 45 ° component is obtained. Then, the distribution of each component of the input character pattern is obtained for the line segment of the contour having the horizontal component, the vertical component, the −45 ° component, and the + 45 ° component. If the horizontal component (d _y = 0), then q ₀ , q ₁ , q
Only ₈ and ₉ are processed.

[0031]

(Equation 6)

(Equation 7)

(Equation 8) As described above, by performing the calculation of q ₀ to q ₁₅ on the pattern data, it is possible to represent to which position the contour data is distributed. For example, if q ₀ is large, many vertical patterns are distributed to the right, if q ₄ is large, many vertical patterns are distributed to the bottom, and if q ₈ is large, many vertical patterns are distributed to the left. If q ₉ is large, it is estimated that many vertical patterns are distributed above.

Similarly, for q ₂ , q ₃ , q ₁₀ , and q ₁₁ , horizontal patterns q ₄ , q ₅ , q _{12 ,} . It can be estimated how many 45 ° patterns for q _{13 and} −45 ° patterns for q ₆ , q ₇ , q ₁₄ and q ₁₅ are distributed to the right, bottom, left and top, respectively. . (I) Off-line Method Matching Unit 10 The off-line method matching unit 10 calculates from the off-line method recognition dictionary 11 storing the reference feature quantities q ^* _{0 to} q ^* ₁₅ for the characters including the recognition target range and the contour data. By sequentially comparing the feature quantities q _{0 to} q _{15 of} the pattern memory, the difference between the two, that is,
The matching distance dd is calculated from the following equation (44) and output.

[0033]

[Equation 9] However, q ^* _{i, j} and q _j are the feature amount registered in the offline system recognition dictionary 11 and the feature amount of the pattern memory, respectively. nn is the number of feature amounts when a plurality of registered characters are registered for one recognized character (for example, “+”).
Is registered in the off-line method recognition dictionary 11 with the | in the center, the one shifted to the left, and the one shifted to the right).

(J) Low Stroke Number Character Identification Unit 12 When the low stroke number character identification unit 12 receives the stroke number output from the preprocessing unit 2 and determines that this stroke number is low, the low stroke number character identification unit 12 is The character code corresponding to the minimum value of the matching distances dd _i to be output is output from the output terminal 13 and displayed on a display (not shown) or the like. If it cannot be determined that the stroke count is low, nothing is output. For example, “+” is 2 strokes, but if the right end of the written character is “–” and the upper end of “|” are continuous and input in one stroke as a whole, based on the number of strokes. The recognition rate is improved by recognizing by the offline method as compared with the online method for recognizing characters. Also, "+"
For example, when a left-handed person writes, "-" may be written in order from right to left. Even in this case, the recognition rate is improved by adopting the offline method as compared with the online method that depends on the writing order. Is improved.

Further, for example, "+" is replaced by "|", "-"
The recognition rate is improved by the off-line method even in the case where the writing is erroneously input in the order of. Here, it is considered that the number of strokes is determined to be low depending on the characteristic parameters to be extracted. Here, since the high stroke number character identifying unit 6 determines that the number of strokes is 7 or more, the stroke count is 6 strokes. The following characters are low stroke characters. As described above, according to the first embodiment, the pattern memory conversion unit 7, the contour extraction unit 8, the offline system feature extraction unit 9, and the offline system matching unit 1
0, the off-line method recognition dictionary 11 and the low stroke number character identification unit 12 are provided, so that it is possible to provide an online handwritten character recognition device that is not affected by the stroke order and stroke number for low stroke number characters.

Second Embodiment FIG. 11 is a block diagram showing an online handwritten character recognition apparatus according to a second embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. There is. The difference between the online handwritten character recognition device of the second embodiment and the online handwritten character recognition device of the first embodiment is that the output of the online method matching unit 4 and the output of the offline method matching unit 10 are also for high stroke number characters. That is, the matching distance is calculated from and the characters are recognized. As shown in FIG. 11, the online handwritten character recognition device according to the second embodiment includes a tablet 1, a preprocessing unit 2,
Online system feature extraction unit 3, online system matching unit 4, online system recognition dictionary 5, high stroke number character identification unit 26, pattern memory conversion unit 7, contour extraction unit 8, offline system feature extraction unit 9, offline system matching unit 1.
0, offline system recognition dictionary 11, low stroke number character identification unit 1
2 and an output terminal 13.

The output side of the tablet 1 is connected to the preprocessing unit 2. The coordinate sequence of the output of the preprocessing unit 2 is connected to the input side of the online system feature extraction unit 3 and the pattern memory conversion unit 7, and the stroke number is the input side of the high stroke number character identification unit 26 and the low stroke number character identification unit 12. It is connected to the. Online system feature extraction unit 3 and online system recognition dictionary 5
The output side of is connected to the input side of the online method matching unit 4. The output side of the online system matching unit 4 is connected to the input side of the high stroke number character identification unit 26. The output side of the pattern memory conversion unit 7 has a contour extraction unit 8
, And the output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The outputs of the offline system feature extraction unit 9 and the offline system recognition dictionary 11 are output by the offline system matching unit 1.
0 is connected to the input side. The output side of the off-line method matching section 4 is connected to the input sides of the low stroke number character identifying section 12 and the high stroke number character recognizing section 26. The output sides of the high stroke number character identifying section 26 and the low stroke number character identifying section 12 are connected to the output terminal 13. Hereinafter, operations (a) to (j) of each unit in FIG. 11 will be described.

(A) Tablet 1 As with the first embodiment, the tablet 1 has a writing data sequence {(x _i , y _i ), i = 1, 2, ..., N _j } of coordinate sequences.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the online system feature extraction unit 3
The writing data string after smoothing is output to the pattern memory converting unit 7, and the stroke number is output to the high stroke number character identifying unit 26 and the low stroke number character identifying unit 12. (C) Online Method Feature Extraction Unit 3 The online method feature extraction unit 3 extracts a feature point from the output of the preprocessing unit 2 by performing, for example, a linear approximation process, as in the first embodiment. Then, the stroke pattern is roughly classified, and the partial pattern of the written character is obtained from the cut position of the partial pattern of the character to be recognized in the character dictionary.
A vector between the partial patterns is calculated, a matching distance d _vec between the vector between the partial patterns of the character dictionary and a vector between the partial patterns of the written character is calculated, and d _vec is compared with VECREJ to perform the middle classification. Further, the Q value of the written character is obtained for each partial pattern in accordance with the partial pattern of the recognition target character remaining in the middle classification.

(D) Online system matching unit 4 The online system matching unit 4 receives the output of the online system feature extraction unit 3 and calculates the matching distance d _BP from the Q value as in the first embodiment. A matching distance value d is calculated from d _vec and d _BP and output to the high stroke count character identification unit 26. (E) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the feature points of the output data of the preprocessing unit 2 and, as in the first embodiment, 2 Convert to the dimension array PM (x, y),
It is output to the contour extraction unit 8. (F) Contour extraction unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), for example, contour tracing is performed to extract contour data, as in the first embodiment. (G) Off-line method feature extraction unit 9 The off-line method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and, for example, as in the first embodiment, the horizontal component, the vertical component, 45 °, for example. The components, q _{0 to} q ₁₅ , of the −45 ° component are obtained.

(H) Offline Method Matching Unit 10 The offline method matching unit 10 stores the reference feature quantities q ₀ ^{* to} q ₁₅ ^* for the characters including the recognition target range, respectively, as in the first embodiment. By sequentially comparing the offline method recognition dictionary 11 and the feature quantities q _{0 to} q _{15 of} the pattern memory calculated from the contour data,
The difference between the two, that is, the matching distance dd is calculated and output. (I) High stroke number character identifying unit 26 The high stroke number character identifying unit 26 calculates a new distance value d from the output d _k of the online method matching unit 4 and the output dd _k of the offline method matching unit 10 according to the following equation (45).
_Search for _highk and _sort in ascending order. When it is determined that the number of strokes is high, the corresponding character code is output from the output terminal 13 and displayed on a display device (not shown) or the like. If it cannot be determined that the number of strokes is high, nothing is output. d _highk = d _k + w · dd _k (45) where w is a weighting coefficient. Here, it is assumed that the number of strokes is high depending on the characteristic parameter to be extracted, but here it is assumed that characters with 7 strokes or more are high strokes.

[0041]

[Outside 1] As shown in Expression (45), the matching distance dd is calculated by the output d _k of the online method matching unit 4 and the output dd _k of the offline method matching unit 10 which is another method that does not depend on the stroke order and the number of strokes. , The recognition rate is improved even in the case of such characters with only a slight difference.

(J) Low Stroke Number Character Identification Unit 12 The low stroke number character identification unit 12 receives the stroke number output from the preprocessing unit 2 and determines that the stroke number is the low stroke number as in the first embodiment. At this time, the character code corresponding to the minimum value of the matching distances dd _i output by the off-line method matching unit 10 is output from the output terminal 13 and displayed on a display device (not shown) or the like. If it cannot be determined that the stroke count is low, nothing is output. Here, it is considered that the determination of the stroke count being the low stroke count depends on the feature parameter to be extracted. Here, since the high stroke count character identification unit 26 determines that the characters of 7 strokes or more are the high stroke count, the stroke count is 6 strokes. The following characters are low stroke characters. As described above, according to the second embodiment,
The low stroke number character has the same advantage as that of the first embodiment, and the high stroke number character is recognized by both the online and offline feature amounts, so that the recognition rate is improved.

Third Embodiment FIG. 12 is a block diagram showing an online handwritten character recognition apparatus according to a third embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. There is. The difference between the online handwritten character recognition device of the third embodiment and the online handwritten character recognition device of the first embodiment is that, for high stroke number characters, the part of the written character according to the cut position of the partial pattern to be recognized. Offline features of the pattern,
This is to perform character recognition by obtaining the matching distance from the matching distance calculated from the offline feature amount of the partial pattern stored in the partial pattern offline method recognition dictionary 35 and the output of the offline method matching unit 4. As shown in FIG. 12, the online handwritten character recognition device according to the third embodiment includes a tablet 1, a preprocessing unit 2, an online system feature extraction unit 3, an online system matching unit 4, an online system recognition dictionary 5, and a high stroke count. Character identification unit 36, partial pattern pattern memory conversion unit 31, partial pattern contour extraction unit 32, partial pattern offline method extraction characteristic unit 33, partial pattern offline method matching unit 34, partial pattern offline method recognition dictionary 3
5, pattern memory conversion unit 7, contour extraction unit 8, offline system feature extraction unit 9, offline system matching unit 1
0, offline system recognition dictionary 11, low stroke number character identification unit 1
2 and an output terminal 13.

The output side of the tablet 1 is connected to the preprocessing unit 2. The coordinate sequence of the output of the preprocessing unit 2 is connected to the input side of the online system feature extraction unit 3 and the pattern memory conversion unit 7, and the stroke count is the input side of the high stroke number character identification unit 36 and the low stroke number character identification unit 12. It is connected to the. Online system feature extraction unit 3 and online system recognition dictionary 5
The output side of is connected to the input side of the online method matching unit 4. The output side of the online system matching unit 4 is connected to the input side of the high stroke number character identification unit 36. The output side of the pattern memory conversion unit 7 has a contour extraction unit 8
, And the output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The output sides of the offline system feature extraction unit 9 and the offline system recognition dictionary 11 are connected to the input side of the offline system matching unit 10. The output side of the off-line method matching unit 4 is connected to the input side of the low stroke number character identifying unit 12.

The output side of the online system feature extraction unit 3 is
It is connected to the online system matching unit 4 and the partial pattern / pattern memory conversion unit 31. The output side of the partial pattern / pattern memory conversion section 31 is connected to the input side of the partial pattern contour extraction section 32. The output side of the partial pattern contour extraction unit 32 is connected to the input side of the partial pattern offline method feature extraction unit 33. The output sides of the partial pattern offline method feature extraction unit 33 and the partial pattern offline method recognition dictionary 35 are connected to the input side of the partial pattern offline method matching unit 34. The output sides of the online method matching unit 4 and the partial pattern offline method matching unit 34 are connected to the input side of the high stroke number character identifying unit 36. The output sides of the high stroke number character identifying unit 36 and the low stroke number character identifying unit 12 are connected to the output terminal 13. The partial pattern pattern memory conversion unit 31 has a partial pattern pattern memory, binarizes the partial pattern of the input character, and stores it in the partial pattern memory.

The partial pattern contour extraction unit 32 extracts contours from the binarized data of the partial pattern of the input character to obtain contour information. The partial pattern offline method feature extraction unit 33 uses the feature information q _{0 to} q _{15 of} horizontal, vertical, 45 °, and −45 ° components from the contour information of the partial pattern.
Is what you want. The partial pattern off-line method matching unit 34 horizontally, vertically, 45 ° of the partial patterns registered in the partial pattern off-line method recognition dictionary 35,
The matching distance between the reference feature quantity q ^* _{0 to} q ^* ₁₅ of the −45 ° component is obtained. High stroke number character identification section 36
Is for recognizing characters by calculating the matching distance from the output of the online method matching unit 4 and the output of the partial pattern offline method matching unit 34. Hereinafter, operations (a) to (n) of the respective units in FIG. 12 will be described.

(A) Tablet 1 As with the first embodiment, the tablet 1 has a coordinate data writing data string {(x _i , y _i ), i = 1, 2, ..., N _j }.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the online system feature extraction unit 3
The smoothed handwritten data string is output to the pattern memory conversion unit 7, and the stroke number is output to the high stroke number character identification unit 36 and the low stroke number character identification unit 12. (C) Online Method Feature Extraction Unit 3 The offline method feature extraction unit 3 extracts a feature point from the output of the preprocessing unit 2 by performing, for example, a linear approximation process, as in the first embodiment. Then, the stroke pattern is roughly classified, and the partial pattern of the written character is obtained from the cut position of the partial pattern of the character to be recognized in the character dictionary.
A vector between the partial patterns is calculated, a matching distance d _vec between the vector between the partial patterns of the character dictionary and a vector between the partial patterns of the written character is calculated, and d _vec is compared with VECREJ to perform the middle classification. Further, the Q value of the written character is obtained for each partial pattern in accordance with the partial pattern of the recognition target character remaining in the middle classification.

Further, the cut position of the partial pattern forming the character and the corresponding coordinate data string are output to the partial pattern pattern memory conversion unit 31. For example, if "A" is written in and the character to be recognized (matching) is "dust", the cut position of "dust" is (3, 2, 5) as shown in FIG. ), So
The coordinate data string of the partial pattern according to the cut position is output to the partial pattern pattern memory conversion unit 31. (D) the online mode matching section 4 online mode matching unit 4 receives the output line topology feature extraction unit 3, as in the first embodiment, calculates a matching distance d _BP from Q value, and d _vec A matching distance value d is calculated from d _BP and output to the high stroke count character identification unit 36.

(E) Partial pattern / pattern memory conversion unit 31 The partial pattern / pattern memory conversion unit 31 outputs, from the online system feature extraction unit 3, a cut position which is stroke information of a partial pattern forming a character and a coordinate data string corresponding thereto. For each partial pattern, similarly to the pattern memory conversion unit 7 of the first embodiment, the partial pattern is vertically and horizontally divided into a plurality of pieces, and the divided areas are adjacent to each other.
It is determined whether or not it intersects with a line segment that connects two coordinate data, and 2
Create the binarized data and store it in the partial pattern memory.
For example, if "A" is written in and the character to be recognized (matching) is "dust", the three partial patterns corresponding to the cut positions of the three partial patterns forming "dust" Each of the coordinate data strings “hisa”, “two”, and “tsuji” is converted into binarized data, and the three converted pattern data are output to the partial pattern contour extraction unit 32.

(F) Partial pattern contour extraction unit 32 The partial pattern contour extraction unit 32 reads out each partial pattern pattern data output from the partial pattern pattern memory conversion unit 31 and uses it as the contour extraction unit 8 of the first embodiment. Similarly, the contour extraction processing is performed, and the contour information is output to the partial pattern offline method feature extraction unit 33. (G) Partial pattern offline method feature extraction unit 33 The partial pattern offline method feature extraction unit 33 receives the contour data for the number of partial patterns output by the partial pattern contour extraction unit 32, and receives the off-line method of the first embodiment. Similar to the feature extraction unit 9, for example, 16 feature quantities q of horizontal, vertical, 45 °, and −45 ° components are extracted from the contour data.
B _{k, 0 to} qB _{k, 15} (k = 1,2, ..., BPN, where BPN is the number of partial patterns) are obtained and output to the partial pattern offline method matching unit 34.

(H) Partial pattern offline method matching unit 34 The partial pattern offline method matching unit 34 outputs the feature quantities qB _{k, 0 to} qB _{k, 15} (k for each partial pattern output from the partial pattern offline method feature extraction unit 33. = 1,2, ..., BP
N) and the feature quantities qB ^* _{k, 0 to} qB ^* _{k, 15} (k = 1,2, ..., BPN) for each matching target partial pattern registered in advance in the partial pattern offline method recognition dictionary 35. , The partial pattern off-line method matching distance value ddb is calculated by the following equation (46) and output.

[0052]

(Equation 10) Here, BPN is the number of partial patterns, N is the number of strokes, and BS _k is the number of strokes of the _kth partial pattern. Where BS
_The reason why the result of dividing _k by N is multiplied is to weight the partial pattern by the number of strokes. For example, if the character you are trying to recognize (match) is "dust" when "A" is written,
The partial pattern offline method recognition dictionary 35 contains "dust".
The three partial patterns that make up "ki", "mu",
Reference feature quantity of "arrow" qB ^* _{k, 0 to} qB ^* _{k, 15} (k = 1,2,3
) Is registered, and the matching distance ddb between the reference feature amount of “ki”, “mu”, and “arrow” and the feature amount of “kyu”, “two”, and “tsuji” is calculated by the equation (47). It is calculated.

(I) High Stroke Number Character Identification Unit 36 The high stroke number character identification unit 36 uses the following equation (47) from the output d _k of the offline method matching unit 34 and the output ddb _{k of the} partial pattern offline method matching unit 34. Then, a new distance value d _highk is obtained and sorted in ascending order, and when it can be determined that the number of strokes is high, the corresponding character code is output from the output terminal 13 and displayed on a not-shown display or the like. If it cannot be determined that the number of strokes is high, nothing is output. d _highk = d _k + w · ddb _k (47) where w is a weighting coefficient. Here, it is considered that the number of strokes is high depending on the characteristic parameter to be extracted, but here it is assumed that characters with 7 strokes or more are high strokes.

[0054]

[Outside 2] Since the matching distance is calculated, the recognition rate is improved even when there is such a slight difference.

(J) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the characteristic points of the output data of the preprocessing unit 2 and is similar to the first embodiment. To a two-dimensional array PM (x, y),
It is output to the contour extraction unit 8. (K) Contour extraction unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), the contour is traced and the contour data is extracted as the contour, as in the first embodiment. (L) Off-line method feature extraction unit 9 The off-line method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and, similarly to the first embodiment, a horizontal component, a vertical component, a 45 ° component, Q ₀ to q of −45 ° component
_Ask for ₁₅ .

(M) Offline Method Matching Unit 10 The offline method matching unit 10 stores the reference feature quantities q ₀ ^{* to} q ₁₅ ^* for the characters including the recognition target range, respectively, as in the first embodiment. By sequentially comparing the offline method recognition dictionary 11 and the feature quantities q _{0 to} q _{15 of} the pattern memory calculated from the contour data,
The difference between the two, that is, the matching distance dd is calculated and output. (N) Low stroke number character identifying section 12 When the low stroke number character identifying section 12 receives the stroke number output by the pre-processing section 2 and determines that the stroke number is low, the matching distance output by the offline method matching section 10 is determined. dd
_The character code corresponding to the minimum value of _i is output terminal 13
And display it on a display device (not shown). If it cannot be determined that the stroke count is low, nothing is output.

Here, it is determined that the stroke number is low
It seems that it depends on the feature parameters to be extracted,
Here, since the high stroke number character identifying unit 36 determines that the characters having 7 strokes or more have a high stroke number, the characters having 6 strokes or less are set as the low stroke number characters.
As described above, according to the third embodiment, the low stroke number character has the same advantages as those of the first embodiment. For a high stroke number character, a partial pattern pattern memory conversion unit 31, a partial pattern contour extraction unit 32, a partial pattern offline method feature extraction unit 33, a partial pattern offline method matching unit 34, and a partial pattern offline method recognition dictionary 35 are provided, and a partial pattern is provided. The recognition rate of the online handwritten character recognition device can be improved because recognition is performed for each of the online and offline feature amounts.

Fourth Embodiment FIG. 13 is a block diagram showing an online handwritten character recognition apparatus according to a fourth embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. There is. The difference between the online handwritten character recognition device of the fourth embodiment and the online handwritten character recognition device of the first embodiment is that in the case of low stroke number characters, the output of the offline method matching unit 10 and the output of the online method matching unit 4 are made. From that, the matching distance is calculated. As shown in FIG. 13, the online handwritten character recognition device according to the fourth embodiment includes a tablet 1, a preprocessing unit 2, an online system feature extraction unit 3, an online system matching unit 4, an online system recognition dictionary 5, and a high stroke count. The character identification unit 6, the pattern memory conversion unit 7, the contour extraction unit 8, the offline system feature extraction unit 9, the offline system matching unit 10, the offline system recognition dictionary 11, the low stroke number character identification unit 42, and the output terminal 1.
3 is provided. The output side of the tablet 1 is the preprocessing unit 2
It is connected to the. Of the output of the preprocessing unit 2, the coordinate sequence is
It is connected to the input side of the online method feature extraction unit 3 and the pattern memory conversion unit 7, and the stroke number is connected to the input sides of the high stroke number character identification unit 6 and the low stroke number character identification unit 42.

The output sides of the online system feature extraction unit 3 and the online system recognition dictionary 5 are connected to the input side of the online system matching unit 4. The output side of the online mode matching unit 4 is connected to the input sides of the high stroke number character identifying unit 6 and the low stroke number character identifying unit 42. The output side of the pattern memory conversion unit 7 is connected to the input side of the contour extraction unit 8, and the output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The output sides of the offline system feature extraction unit 9 and the offline system recognition dictionary 11 are connected to the input side of the offline system matching unit 10. The output side of the off-line method matching unit 4 is connected to the input side of the low stroke number character identification unit 42. The output sides of the high stroke number character identifying unit 6 and the low stroke number character identifying unit 42 are connected to the output terminal 13. The low stroke number character identifying unit 42 outputs the output dd of the online system matching unit 4.
_The matching distance d _lowk is calculated from _k and the output d _k of the offline system matching unit 10 to recognize the character. Hereinafter, operations (a) to (j) of each unit in FIG. 13 will be described.

(A) Tablet 1 As with the first embodiment, the tablet 1 has a writing data sequence {(x _i , y _i ), i = 1, 2, ..., N _j } of coordinate sequences.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the online system feature extraction unit 3
The writing data string after smoothing is output to the pattern memory conversion unit 7, and the stroke number is output to the high stroke number character identifying unit 6 and the low stroke number character identifying unit 42. (C) Online Method Feature Extraction Unit 3 The online method feature extraction unit 3 extracts a feature point from the output of the preprocessing unit 2 by performing, for example, a linear approximation process, as in the first embodiment. Then, the stroke pattern is roughly classified, and the partial pattern of the written character is obtained from the cut position of the partial pattern of the character to be recognized in the character dictionary.
A vector between the partial patterns is calculated, a matching distance d _vec between the vector between the partial patterns of the character dictionary and a vector between the partial patterns of the written character is calculated, and d _vec is compared with VECREJ to perform the middle classification. Further, the Q value of the written character is obtained for each partial pattern in accordance with the partial pattern of the recognition target character remaining in the middle classification.

(D) Online system matching unit 4 The online system matching unit 4 receives the output of the online system feature extraction unit 3 and calculates the matching distance d _BP from the Q value, as in the first embodiment. A matching distance value d is calculated from d _vec and d _BP and output to the high stroke count character identification unit 6. (E) High Stroke Number Character Identification Unit 6 The high stroke number character identification unit 6 is similar to that of the first embodiment.
When the output d _k of the online method matching unit 4 is sorted in ascending order and it can be determined that the number of strokes is high,
The corresponding character code is output from the output terminal 13 and displayed on a display device (not shown) or the like. If it cannot be determined that the number of strokes is high, nothing is output. Here, it is considered that the number of strokes is high depending on the characteristic parameter to be extracted, but here it is assumed that characters with 7 strokes or more are high strokes. (F) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the feature points of the output data of the preprocessing unit 2 and, as in the first embodiment, 2 Convert to the dimension array PM (x, y),
It is output to the contour extraction unit 8.

(G) Contour extraction unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), contour tracing is performed and contour information is extracted as a contour, as in the first embodiment. (H) Off-line method feature extraction unit 9 The off-line method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and, similarly to the first embodiment, a horizontal component, a vertical component, a 45 ° component, Q ₀ to q of −45 ° component
_Ask for ₁₅ . (I) Offline Method Matching Unit 10 The offline method matching unit 10 stores the reference feature quantities q ₀ ^{* to} q ₁₅ ^* for the characters including the recognition target range, respectively, as in the first embodiment. By sequentially comparing the recognition dictionary 11 and the feature quantities q _{0 to} q _{15 of} the pattern memory calculated from the contour data,
The difference between the two, that is, the matching distance dd is calculated and output.

(J) Low stroke number character identifying section 42 When the low stroke number character identifying section 42 receives the stroke number output from the preprocessing section 2 and determines that the stroke number is low, the low stroke number character identifying section 42 outputs the online method matching section 5. From d _k and the matching distance value dd _k output by the off-line method matching unit 10, the matching distance value d _lowk of the low stroke number character is calculated by the following formula (48), sorted in ascending order, and is the low stroke number character. If so, the corresponding character code is output, and if not, nothing is output. d _lowk = _{wd k} + dd _k (48) where w is a weighting coefficient. Here, it is considered that the number of strokes is determined to be low depending on the characteristic parameters to be extracted. Here, since the high stroke number character identifying unit 6 determines that the number of strokes is 7 or more, the stroke count is 6 strokes. The following characters are low stroke characters.

For example, when identifying "+" and "na",
With the offline method alone as in the first embodiment, only the big difference can be seen, but with the online method,
"+" Of "|" and "na" of "no" can be collated, and the difference in the fine part of the character can be expressed. Therefore, in the case of low stroke number characters, the offline method and the online method are used together, so that the recognition rate is improved. As described above, according to the fourth embodiment, the pattern memory conversion unit 7, the contour extraction unit 8, the offline system feature extraction unit 9, the offline system matching unit 10, the offline system recognition dictionary 11, and the low stroke number character identification. Since the unit 42 is provided and the low stroke number character whose recognition rate is low due to the small amount of information, the feature quantity of the offline method is added to the feature quantity of the online method, so that the recognition rate of the online handwritten character recognition device is improved.

Fifth Embodiment FIG. 14 is a block diagram showing an online handwritten character recognition apparatus according to a fifth embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. There is. The difference between the online handwritten character recognition device of the fifth embodiment and the online handwritten character recognition device of the first embodiment is that in the case of high stroke number characters, the output of the online system matching unit 4 and the offline system matching unit 10 are different. The matching distance is calculated from the output and the character recognition is performed. In the case of a low stroke number character, the matching distance is calculated from the output of the online method matching unit 4 and the offline method matching 10 to perform the character recognition. That is what I did. As shown in FIG. 14, the online handwritten character recognition device according to the fifth embodiment includes a tablet 1, a preprocessing unit 2,
Online system feature extraction unit 3, online system matching unit 4, online system recognition dictionary 5, high stroke number character identification unit 26, pattern memory conversion unit 7, contour extraction unit 8, offline system feature extraction unit 9, offline system matching unit 1.
0, offline system recognition dictionary 11, low stroke number character identification unit 4
2 and an output terminal 13.

The output side of the tablet 1 is connected to the preprocessing unit 2. The coordinate sequence of the output of the preprocessing unit 2 is connected to the input side of the online system feature extraction unit 3 and the pattern memory conversion unit 7, and the stroke number is the input side of the high stroke number character identification unit 26 and the low stroke number character identification unit 42. It is connected to the. Online system feature extraction unit 3 and online system recognition dictionary 5
The output side of is connected to the input side of the online method matching unit 4. The output side of the online method matching unit 4 has a high stroke number character identifying unit 26 and a low stroke number character identifying unit 42.
Is connected to the input side. The output side of the pattern memory conversion unit 7 is connected to the input side of the contour extraction unit 8, and the output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The output sides of the offline system feature extraction unit 9 and the offline system recognition dictionary 11 are connected to the input side of the offline system matching unit 10. The output side of the online mode matching unit 4 is connected to the input sides of the high stroke number character identifying unit 26 and the low stroke number character identifying unit 42. The output sides of the high stroke number character identifying portion 26 and the low stroke number character identifying portion 42 are connected to the output terminal 13.
Hereinafter, operations (a) to (j) of each unit in FIG. 14 will be described.

(A) Tablet 1 As with the first embodiment, the tablet 1 has a writing data sequence {(x _i , y _i ), i = 1, 2, ..., N _j } of coordinate sequences.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the online system feature extraction unit 3
The writing data string after smoothing is output to the pattern memory conversion unit 7, and the stroke number is output to the high stroke number character identifying unit 26 and the low stroke number character identifying unit 42. (C) Online Method Feature Extraction Unit 3 The offline method feature extraction unit 3 extracts a feature point from the output of the preprocessing unit 2 by performing, for example, a linear approximation process, as in the first embodiment. Then, the stroke pattern is roughly classified, and the partial pattern of the written character is obtained from the cut position of the partial pattern of the character to be recognized in the character dictionary.
A vector between the partial patterns is calculated, a matching distance d _vec between the vector between the partial patterns of the character dictionary and a vector between the partial patterns of the written character is calculated, and d _vec is compared with VECREJ to perform the middle classification. Further, the Q value of the written character is obtained for each partial pattern in accordance with the partial pattern of the recognition target character remaining in the middle classification.

(D) Online system matching unit 4 The online system matching unit 4 receives the output of the online system feature extraction unit 3 and calculates the matching distance d _BP from the Q value as in the first embodiment. The matching distance value d is calculated from d _vec and d _BP , and the high stroke number character identifying unit 26 is calculated.
Output to (E) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the feature points of the output data of the preprocessing unit 2 and, as in the first embodiment, 2 Convert to the dimension array PM (x, y),
It is output to the contour extraction unit 8. (F) Contour extraction unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), for example, contour tracing is performed to extract contour data, as in the first embodiment.

(G) Off-line method feature extraction unit 9 The off-line method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and, for example, as in the first embodiment, the horizontal component and the vertical component, for example. , 45 ° component, −45 ° component q _{0 to} q ₁₅ are obtained. (H) Offline Method Matching Unit 10 The offline method matching unit 10 stores the reference feature quantities q ₀ ^{* to} q ₁₅ ^* for the characters including the recognition target range, respectively, as in the first embodiment. By sequentially comparing the recognition dictionary 11 and the feature quantities q _{0 to} q _{15 of} the pattern memory calculated from the contour data, the difference between the two, that is, the matching distance dd is calculated and output. (I) High stroke number character identifying section 26 The high stroke number character identifying section 26 uses the output d _k of the online method matching unit 4 and the output dd _k of the offline method matching unit 10 to calculate a new distance value d according to Equation (45).
_{When highk} is _calculated and sorted in ascending order, and when it can be determined that the number of strokes is high, the corresponding character code is output from the output terminal 13 and displayed on a display device (not shown) or the like.
If it cannot be determined that the number of strokes is high, nothing is output. Here, it is considered that the number of strokes is high depending on the characteristic parameter to be extracted, but here it is assumed that characters with 7 strokes or more are high strokes.

(J) Low Stroke Number Character Identification Unit 42 When the low stroke number character identification unit 42 receives the stroke number output from the preprocessing unit 2 and determines that the stroke number is low, the low stroke number character identification unit 42 outputs the online method matching unit 4. From d _k and the matching distance value dd _k output by the off-line method matching unit 10, the matching distance value d of the low stroke character is calculated by Expression (49).
_{When lowk} is calculated and sorted in ascending order, if it is determined to be a low stroke number character, the corresponding character code is output, otherwise nothing is output. Here, it is considered that the determination of the stroke count being the low stroke count depends on the feature parameter to be extracted. Here, since the high stroke count character identification unit 26 determines that the characters of 7 strokes or more are the high stroke count, the stroke count is 6 strokes. The following characters are low stroke characters. As described above, according to the fifth embodiment, high stroke number characters have the same advantages as those of the second embodiment, and low stroke number characters have the same advantages.
There are the same advantages as the fourth embodiment.

Sixth Embodiment FIG. 15 is a block diagram showing an online handwritten character recognition apparatus according to a sixth embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. There is. The difference between the online handwritten character recognition device of the sixth embodiment and the online handwritten character recognition device of the first embodiment is that, for high stroke number characters, the output of the online system matching unit 4 and the partial pattern offline system matching unit 34. And the matching distance is calculated for character recognition. For low stroke number characters, the matching distance is calculated from the output of the online method matching unit 4 and the output of the offline method matching unit 10 for character recognition. That is what I did. As shown in FIG. 15, the online handwritten character recognition device according to the sixth embodiment includes a tablet 1,
Preprocessing unit 2, online system feature extraction unit 3, online system matching unit 4, online system recognition dictionary 5, high stroke number character identification unit 36, partial pattern pattern memory conversion unit 31, partial pattern contour extraction unit 32, partial pattern offline system Feature extraction unit 33, partial pattern offline system matching unit 34, partial pattern offline system recognition dictionary 35, pattern memory conversion unit 7, contour extraction unit 8, offline system feature extraction unit 9, offline system matching unit 10, offline system recognition dictionary 11 , A low stroke number character identification section 42, and an output terminal 13.

The output side of the tablet 1 is connected to the input side of the preprocessing unit 2. The output side of the preprocessing unit 2 is connected to the input sides of the online system feature extraction unit 3, the pattern memory conversion unit 7, the high stroke number character identification unit 36, and the low stroke number character identification unit 42. The output sides of the online method feature extraction unit 3 and the online method recognition dictionary 5 are connected to the input side of the online method matching unit 4. The output side of the online mode matching unit 4 is connected to the input sides of the high stroke number character identifying unit 36 and the low stroke number character identifying unit 42. The output side of the pattern memory conversion unit 7 is connected to the input side of the contour extraction unit 8. The output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The output sides of the offline method feature extraction unit 9 and the offline method recognition dictionary 1 are connected to the input side of the offline method matching unit 10. The output side of the offline method matching unit 10 has a low stroke number character identifying unit 42 and a high stroke number character identifying unit 36.
Is connected to the input side. The output sides of the high stroke number character identifying unit 36 and the low stroke number character identifying unit 42 are connected to the output terminal 13.

The output side of the online system feature extraction unit 3 is
It is connected to the online system matching unit 4 and the partial pattern / pattern memory conversion unit 31. The output side of the partial pattern memory conversion unit 31 is connected to the partial pattern contour extraction unit 3
2 is connected to the input side. The output side of the partial pattern contour extraction unit 32 is connected to the input side of the partial pattern offline method feature extraction unit 33. The output sides of the partial pattern offline method feature extraction unit 33 and the partial pattern offline method recognition dictionary 35 are connected to the input side of the partial pattern offline method matching unit 34. The output side of the partial pattern offline method matching unit 34 is
It is connected to the input side of the high stroke number character identification unit 36. Hereinafter, the operations (a) to (n) of each unit in FIG. 15 will be described.

(A) Tablet 1 As in the first embodiment, the tablet 1 has a coordinate data writing data string {(x _i , y _i ), i = 1, 2, ..., N _j }.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the online system feature extraction unit 3
The writing data string after smoothing is output to the pattern memory conversion unit 7, and the stroke number is output to the high stroke number character identifying unit 36 and the low stroke number character identifying unit 42. (C) Online Method Feature Extraction Unit 3 The offline method feature extraction unit 3 extracts a feature point from the output of the preprocessing unit 2 by performing, for example, a linear approximation process, as in the first embodiment. Then, the stroke pattern is roughly classified, and the partial pattern of the written character is obtained from the cut position of the partial pattern of the character to be recognized in the character dictionary.
A vector between the partial patterns is calculated, a matching distance d _vec s between the vector between the partial patterns of the character dictionary and a vector between the partial patterns of the written characters is calculated, and d _vec and VECREJ are compared to perform the middle classification. Further, the Q value of the written character is obtained for each partial pattern in accordance with the partial pattern of the recognition target character remaining in the middle classification.

(D) Online system matching unit 4 The online system matching unit 4 receives the output of the online system feature extraction unit 3 and calculates the matching distance d _BP from the Q value, as in the first embodiment. The matching distance value d is calculated from d _vec and d _BP , and the high stroke character identifying unit 36
Output to (E) Partial pattern pattern memory conversion unit 31 The partial pattern pattern memory conversion unit 31 receives a cut position of a partial pattern forming a character and a corresponding coordinate data string from the online system feature extraction unit 3, Similar to the embodiment, the conversion process to the pattern memory is performed for each partial pattern. (F) Partial Pattern Contour Extraction Unit 32 The partial pattern contour extraction unit 32 reads out each pattern memory output from the partial pattern pattern memory conversion unit 31, and performs the contour extraction process of each partial pattern as in the third embodiment. And outputs the contour information to the partial pattern offline method feature extraction unit 33.

(G) Partial pattern offline method feature extraction unit 33 The partial pattern offline method feature extraction unit 33 outputs the contour data for the number of partial patterns output by the partial pattern contour extraction unit 32, as in the third embodiment. From the respective contour data, 16 feature quantities qB _{k, 0 to} qB _{k, 15} (k = 1,2, ..., BP) of horizontal, vertical, 45 °, and −45 ° components are received.
(N and BPN are the number of partial patterns) and output to the partial pattern offline method matching unit 34. (H) Partial pattern offline method matching unit 34 The partial pattern offline method matching unit 34 includes a third pattern
Similar to the above embodiment, the feature quantity qB _{k, 0 for} each partial pattern output from the partial pattern offline method feature extraction unit 33.
˜qB _{k, 15} (k = 1,2, ..., BPN) and the feature quantities qB ^* _{k, 0} ˜qB ^* _{k, for} each matching target partial pattern registered in advance in the partial pattern offline method recognition dictionary 35 _{. 15} (k =
1, 2, ..., BPN) and a partial pattern offline method matching distance value ddb is calculated and output.

(I) High stroke number character identifying section 36 The high stroke number character identifying section 36 calculates from the output d _k of the online method matching section 4 and the output ddb _{k of the} partial pattern offline method matching section 34 by the equation (47). The new distance value d _highk is _calculated and sorted in ascending order,
When it is determined that the number of strokes is high, the corresponding character code is output from the output terminal 13 and displayed on a display (not shown) or the like. If it cannot be determined that the number of strokes is high, nothing is output. Here, it is considered that the number of strokes is high depending on the characteristic parameter to be extracted, but here it is assumed that characters with 7 strokes or more are high strokes. (J) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the feature points of the output data of the preprocessing unit 2 and, as in the first embodiment, 2 Convert to the dimension array PM (x, y),
It is output to the contour extraction unit 8.

(K) Contour extraction unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), for example, contour tracing is performed to extract contour data, as in the first embodiment. (L) Off-line method feature extraction unit 9 The off-line method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and, for example, as in the first embodiment, the horizontal component, the vertical component, 45 °, for example. The components, q _{0 to} q ₁₅ , of the −45 ° component are obtained. (M) Offline Method Matching Unit 10 The offline method matching unit 10 stores the reference feature quantities q ₀ ^{* to} q ₁₅ ^* for the characters including the recognition target range, respectively, as in the first embodiment. By sequentially comparing the recognition dictionary 11 and the feature quantities q _{0 to} q _{15 of} the pattern memory calculated from the contour data,
The difference between the two, that is, the matching distance dd is calculated and output.

(N) Low Stroke Number Character Identification Unit 42 When the low stroke number character identification unit 42 receives the stroke number output from the preprocessing unit 2 and determines that the stroke number is low, it is the same as in the third embodiment. In addition, the output d _k of the online method matching unit 4 and the output dd _k of the offline method matching unit 10
In response to this, the matching distance d _lowk for low stroke number characters is calculated from equation (48), sorted in ascending order, and the corresponding character code is output from the output terminal 13 and displayed on a display device (not shown) or the like. If it cannot be determined that the stroke count is low, nothing is output. Here, it is considered that the number of strokes is determined to be low depending on the feature parameter to be extracted. Here, since the high stroke number character identification unit 36 determines that the number of strokes is 7 strokes or more, the stroke count is 6 strokes. The following characters are low stroke characters. As described above, according to the sixth embodiment, the high stroke number character has the same advantages as those of the third embodiment, and the low stroke number character has the same advantages as those of the fourth embodiment. is there.

Seventh Embodiment FIG. 16 is a block diagram showing an online handwritten character recognition apparatus according to a seventh embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. There is. The difference between the online handwritten character recognition device of the seventh embodiment and the online handwritten character recognition device of the first embodiment is that, for low stroke number characters, the output of the partial pattern offline method matching unit 34 and the offline method matching unit. That is, the matching distance is calculated from the output of 10 and character recognition is performed. As shown in FIG. 16, the online handwritten character recognition device according to the seventh embodiment includes a tablet 1, a preprocessing unit 2, an online system feature extraction unit 3, an online system matching unit 4, an online system recognition dictionary 5,
High stroke number character identification unit 6, partial pattern pattern memory conversion unit 31, partial pattern contour extraction unit 32, partial pattern offline system feature extraction unit 33, partial pattern offline system matching unit 34, partial pattern offline system recognition dictionary 35, pattern memory conversion Part 7, contour extraction part 8,
An offline system feature extraction unit 9, an offline system matching unit 10, an offline system recognition dictionary 11, a low stroke number character identification unit 52, and an output terminal 13 are provided.

The output side of the tablet 1 is connected to the input side of the preprocessing section 2. The output side of the preprocessing unit 2 is connected to the input sides of the online system feature extraction unit 3, the pattern memory conversion unit 7, the high stroke number character identification unit 6 and the low stroke number character identification unit 52. The output sides of the online method feature extraction unit 3 and the online method recognition dictionary 5 are connected to the input side of the online method matching unit 4. The output side of the online system matching unit 4 is connected to the input side of the high stroke number character identification unit 6. The output side of the pattern memory conversion unit 7 is connected to the input side of the contour extraction unit 8. The output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The output sides of the offline system feature extraction unit 9 and the offline system recognition dictionary 1 are connected to the input side of the offline system matching group 10. The output side of the off-line method matching unit 10 is connected to the input side of the low stroke number character identification unit 52. The output side of the online system feature extraction unit 3 is connected to the online system matching unit 4 and the partial pattern memory conversion unit 31. The output side of the partial pattern memory conversion unit 31 is connected to the input side of the partial pattern contour extraction unit 32. The output side of the partial pattern contour extraction unit 32 is connected to the input side of the partial pattern offline method feature extraction unit 33.

Partial pattern offline system feature extraction unit 3
3 and the output side of the partial pattern offline method recognition dictionary 35 are connected to the input side of the partial pattern offline method matching unit 34. The output side of the partial pattern offline method matching unit 34 is connected to the input side of the low stroke number character identification unit 52. The output sides of the high stroke number character identifying unit 6 and the low stroke number character identifying unit 52 are connected to the output terminal 13. Low strokes character identifying unit 52, from the output ddb _k partial pattern offline mode matching unit 34, an output dd _k offline mode matching unit 10, and character recognition by calculating a matching distance. Hereinafter, FIG.
The operations (a) to (n) of each unit of 6 will be described.

(A) Tablet 1 As with the first embodiment, the tablet 1 has a writing data sequence {(x _i , y _i ), i = 1, 2, ..., N _j } of coordinate sequences.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the online system feature extraction unit 3
The smoothed writing data string is output to the pattern memory conversion unit 7, and the stroke number is output to the high stroke number character identifying unit 6 and the low stroke number character identifying unit 52. (C) Online Method Feature Extraction Unit 3 The online method feature extraction unit 3 performs, for example, linear approximation processing from the output of the preprocessing unit 2 to extract the feature points, as in the first embodiment. Then, the stroke pattern is roughly classified, and the partial pattern of the written character is obtained from the cut position of the partial pattern of the character to be recognized in the character dictionary.
A vector between the partial patterns is calculated, a matching distance d _vec between the vector between the partial patterns of the character dictionary and a vector between the partial patterns of the written character is calculated, and d _vec is compared with VECREJ to perform middle classification. Further, the Q value of the written character is obtained for each partial pattern in accordance with the partial pattern of the recognition target character remaining in the middle classification.

(D) Online Method Matching Unit 4 The online method matching unit 4 receives the output of the online method feature extraction unit 3 and calculates the matching distance d _BP from the Q value as in the first embodiment. A matching distance value d is calculated from d _vec and d _BP and output to the high stroke count character identification unit 6. (E) High Stroke Number Character Identification Unit 6 As in the first embodiment, the high stroke number character identification unit 6 sorts the output d _k of the online method matching unit 4 in ascending order, and determines that the stroke number is high. At this time, the corresponding character code is output from the output terminal 13 and displayed on a display (not shown) or the like. Here, it is considered that the number of strokes is high depending on the characteristic parameter to be extracted, but here it is assumed that characters with 7 strokes or more are high strokes. (F) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the feature points of the output data of the preprocessing unit 2 and, as in the first embodiment, 2 Convert to the dimension array PM (x, y),
It is output to the contour extraction unit 8.

(G) Contour extraction unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), for example, contour tracing is performed to extract contour data, as in the first embodiment. (H) Offline Method Feature Extraction Unit 9 The offline method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and, for example, as in the first embodiment, for example, the horizontal component, the vertical component, 45 °. The components, q _{0 to} q ₁₅ , of the −45 ° component are obtained. (I) Offline Method Matching Unit 10 The offline method matching unit 10 stores the reference feature quantities q ₀ ^{* to} q ₁₅ ^* for the characters including the recognition target range, respectively, as in the first embodiment. By sequentially comparing the recognition dictionary 11 and the feature quantities q _{0 to} q _{15 of} the pattern memory calculated from the contour data,
The difference between the two, that is, the matching distance dd is calculated and output.

(J) Partial pattern pattern memory conversion unit 31 The partial pattern pattern memory conversion unit 31 receives the cut position of the partial pattern forming the character and the coordinate data string corresponding thereto from the online system feature extraction unit 3, Similar to the third embodiment, the conversion process to the pattern memory is performed for each partial pattern. (K) Partial pattern contour extraction unit 32 The partial pattern contour extraction unit 32 reads out each pattern memory output from the partial pattern pattern memory conversion unit 31 and performs the contour extraction processing of each partial pattern, as in the third embodiment. And outputs the contour information to the partial pattern offline method feature extraction unit 33. (L) Partial pattern offline method feature extraction unit 33 The partial pattern offline method feature extraction unit 33 receives contour data for the number of partial patterns output from the partial pattern contour extraction unit 32, as in the third embodiment. From each contour data, 16 feature quantities of horizontal, vertical, 45 °, and −45 ° components qB _{k, 0 to} qB _{k, 15} (k = 1,2, ..., BP
(N and BPN are the number of partial patterns) and output to the partial pattern offline method matching unit 34.

(M) Partial pattern offline system matching unit 34 The partial pattern offline system matching unit 34
Similar to the above embodiment, the feature quantity qB _{k, 0 for} each partial pattern output from the partial pattern offline method feature extraction unit 33.
˜qB _{k, 15} (k = 1,2, ..., BPN) and the feature quantities qB ^* _{k, 0} ˜qB ^* _{k, for} each matching target partial pattern registered in advance in the partial pattern offline method recognition dictionary 35 _{. 15} (k =
1, 2, ..., BPN) and a partial pattern offline method matching distance value ddb is calculated and output.

[0088] (n) low number of strokes character identifying unit 52 low strokes character identifying unit 52, the output dd _k offline mode matching unit 10, the number of strokes of the output of the partial pattern offline mode output of the matching section 34 ddb _k and the pre-processing unit 2 In response, the matching distance d _lowk of the low stroke number character is calculated by the equation (49), and the matching distance d _lowk is sorted in ascending order. When it is determined that the stroke number is low, the character code corresponding to the sorting result is output, and otherwise, nothing is output. d _lowk = dd _k + w · ddb _k (49) where w is a weighting coefficient. Here, it is considered that the number of strokes is determined to be low depending on the characteristic parameters to be extracted. Here, since the high stroke number character identifying unit 6 determines that the number of strokes is 7 or more, the stroke count is 6 strokes. The following characters are low stroke characters. As described above, according to the seventh embodiment, the high stroke number character has the same advantages as those of the first embodiment. In addition, for low stroke number characters whose recognition rate is low due to the small amount of information, global features are matched by the offline method, and fine character parts are matched by the partial pattern offline method. It is possible to obtain an excellent recognition rate even for characters, and both features are offline systems.
It is possible to provide an online handwritten character recognition device that is not affected by the writing direction.

Eighth Embodiment FIG. 17 is a block diagram showing an online handwritten character recognition device according to an eighth embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. There is. The online handwritten character recognition device of the eighth embodiment differs from the online handwritten character recognition device of the first embodiment in that, for high stroke number characters, the output of the online system matching unit 4 and the offline system matching unit 10 are different. The matching distance is calculated from the output to perform character recognition. For low stroke number characters, the matching distance is calculated from the output of the partial pattern offline method matching unit 34 and the output of the offline method matching unit 10, and the character is calculated. That is to recognize it.

As shown in FIG. 17, the online handwritten character recognition apparatus according to the eighth embodiment includes a tablet 1, a preprocessing unit 2, an online system feature extraction unit 3, an online system matching unit 4, and an online system recognition dictionary 5. , A high stroke number character identification unit 26, a partial pattern pattern memory conversion unit 3
1, partial pattern contour extraction unit 32, partial pattern offline system feature extraction unit 33, partial pattern offline system matching unit 34, partial pattern offline system recognition dictionary 35, pattern memory conversion unit 7, contour extraction unit 8, offline system feature extraction unit 9, an offline system matching unit 10, an offline system recognition dictionary 11, a low stroke number character identifying unit 52, and an output terminal 13. The output side of the tablet 1 is connected to the input side of the preprocessing unit 2. The output side of the preprocessing unit 2 is connected to the input sides of the online system feature extraction unit 3, the pattern memory conversion unit 7, the high stroke number character identification unit 26, and the low stroke number character identification unit 52. The output sides of the online method feature extraction unit 3 and the online method recognition dictionary 5 are connected to the input side of the online method matching unit 4. The output side of the online system matching unit 4 is connected to the input side of the high stroke number character identification unit 26.

The output side of the pattern memory conversion section 7 is connected to the input side of the contour extraction section 8. The output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The output sides of the offline method feature extraction unit 9 and the offline method recognition dictionary 1 are connected to the input side of the offline method matching unit 10. The output side of the off-line method matching section 10 is connected to the input sides of the low stroke number character identifying section 52 and the high stroke number character identifying section 26. The output side of the online system feature extraction unit 3 has an online system matching unit 4 and a partial pattern pattern memory conversion unit 31.
It is connected to the. The output side of the partial pattern / pattern memory conversion section 31 is connected to the input side of the partial pattern contour extraction section 32. The output side of the partial pattern contour extraction unit 32 is connected to the input side of the partial pattern offline method feature extraction unit 33. The output sides of the partial pattern offline method feature extraction unit 33 and the partial pattern offline method recognition dictionary 35 are connected to the input side of the partial pattern offline method matching unit 34. The output side of the partial pattern offline method matching unit 34 is connected to the input side of the low stroke number character identification unit 52. The output sides of the high stroke number character identifying unit 26 and the low stroke number character identifying unit 52 are connected to the output terminal 13. Hereinafter, operations (a) to (n) of the respective units in FIG. 17 will be described.

(A) Tablet 1 As in the first embodiment, the tablet 1 has a coordinate data writing data string {(x _i , y _i ), i = 1, 2, ..., N _j }.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the online system feature extraction unit 3
The writing data string after smoothing is output to the pattern memory conversion unit 7, and the stroke number is output to the high stroke number character identifying unit 26 and the low stroke number character identifying unit 52. (C) Online Method Feature Extraction Unit 3 The online method feature extraction unit 3 extracts a feature point from the output of the preprocessing unit 2 by performing, for example, a linear approximation process, as in the first embodiment. Then, the stroke pattern is roughly classified, and the partial pattern of the written character is obtained from the cut position of the partial pattern of the character to be recognized in the character dictionary.
A vector between the partial patterns is calculated, a matching distance d _vec between the vector between the partial patterns of the character dictionary and a vector between the partial patterns of the written character is calculated, and d _vec and VECREJ
Compare with and perform middle classification. Further, the Q value of the written character is obtained for each partial pattern in accordance with the partial pattern of the recognition target character remaining in the middle classification.

(D) Online system matching unit 4 The online system matching unit 4 receives the output of the online system feature extraction unit 3 and calculates the matching distance d _BP from the Q value as in the first embodiment. The matching distance value d is calculated from d _vec and d _BP , and the high stroke number character identifying unit 26 is calculated.
Output to (E) Partial pattern pattern memory conversion unit 31 The partial pattern pattern memory conversion unit 31 receives a cut position of a partial pattern forming a character and a corresponding coordinate data string from the online system feature extraction unit 3, Similar to the embodiment, the conversion process to the pattern memory is performed for each partial pattern.

(F) Partial pattern contour extraction unit 32 The partial pattern contour extraction unit 32 reads each pattern memory output from the partial pattern pattern memory conversion unit 31 and, as in the third embodiment, reads each partial pattern. The contour extraction processing is performed, and the contour information is output to the partial pattern offline method feature extraction unit 33. (G) Partial pattern offline method feature extraction unit 33 The partial pattern offline method feature extraction unit 33 receives the contour data for the number of partial patterns output by the partial pattern online method contour extraction unit 33, as in the third embodiment. Then, from each contour data, horizontal, vertical, 45
16 feature quantities qBk _{, 0 to} qB of-, -45 ° components
_{k, 15} (k = 1,2, ..., BPN, where BPN is the number of partial patterns) is obtained and output to the partial pattern offline method matching unit 34.

(H) Partial pattern offline system matching unit 34 The partial pattern offline system matching unit 34
Similar to the above embodiment, the feature quantity qB _{k, 0 for} each partial pattern output from the partial pattern offline method feature extraction unit 33.
˜qB _{k, 15} (k = 1,2, ..., BPN) and the feature quantities qB ^* _{k, 0} ˜qB ^* _{k, for} each matching target partial pattern registered in advance in the partial pattern offline method recognition dictionary 35 _{. 15} (k =
1, 2, ..., BPN) and a partial pattern offline method matching distance value ddb is calculated and output. (I) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the feature points of the output data of the preprocessing unit 2 and, as in the first embodiment, 2 Convert to the dimension array PM (x, y),
It is output to the contour extraction unit 8.

(J) Contour Extraction Unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), for example, contour tracing is performed to extract contour data, as in the first embodiment. (K) Offline Method Feature Extraction Unit 9 The offline method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and, for example, as in the first embodiment, for example, the horizontal component, the vertical component, 45 °. The components, q _{0 to} q ₁₅ , of the −45 ° component are obtained.

(L) Off-line Method Matching Section 10 The off-line method matching section 10 stores the reference feature quantities q ₀ ^{* to} q ₁₅ ^* for the characters including the recognition target range, as in the first embodiment. By sequentially comparing the offline method recognition dictionary 11 and the feature quantities q _{0 to} q _{15 of} the pattern memory calculated from the contour data,
The difference between the two, that is, the matching distance dd is calculated and output. (M) High stroke number character identifying section 26 The high stroke number character identifying section 26 uses the output d _k of the online method matching unit 4 and the output dd _k of the offline method matching unit 10 to calculate a new distance value d according to equation (45).
_{When highk} is _calculated and sorted in ascending order, and when it can be determined that the number of strokes is high, the corresponding character code is output from the output terminal 13 and displayed on a display device (not shown) or the like.
If it cannot be determined that the number of strokes is high, nothing is output. Here, it is considered that the number of strokes is high depending on the characteristic parameter to be extracted, but here it is assumed that characters with 7 strokes or more are high strokes.

(N) Low Stroke Number Character Identification Unit 52 When the low stroke number character identification unit 52 receives the stroke number output from the preprocessing unit 2 and determines that the stroke number is low, it is the same as in the seventh embodiment. to the output dd _k offline mode matching unit 10 receives an output ddb _k partial pattern offline mode matching unit 34 calculates a matching distance d _{low k} low strokes character from the equation (49), sorted in ascending order Then, the corresponding character code is output from the output terminal 13 and displayed on a display device (not shown) or the like. If it cannot be determined that the stroke count is low, nothing is output. Here, it is considered that the determination of the stroke count being the low stroke count depends on the characteristic parameter to be extracted.
Since the character of 7 strokes or more in 6 has a high stroke count, the character of 6 strokes or less is a low stroke count character. As described above, the eighth
According to the embodiment, the high stroke number character has the same advantages as those of the second embodiment, and the low stroke number character has the same advantages as those of the seventh embodiment.

Ninth Embodiment FIG. 18 is a block diagram showing an online handwritten character recognition device according to a ninth embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. There is. To do. Book 9
The difference between the online handwritten character recognition device of the embodiment of the present invention and the online handwritten character recognition device of the first embodiment is that
For high stroke characters, online matching section 4
Output and partial pattern online method matching unit 3
The matching distance is calculated from the output of No. 4 and character recognition is performed, and the matching distance is calculated from the output of the offline method matching unit 10 and the output of the partial pattern offline method matching unit 34 for low stroke number characters. That is, the character recognition is performed.

As shown in FIG. 18, the online handwritten character recognition apparatus according to the ninth embodiment includes a tablet 1, a preprocessing unit 2, an online system feature extraction unit 3, an online system matching unit 4, and an online system recognition dictionary 5. , High stroke number character identification section 36, partial pattern pattern conversion memory section 3
1, partial pattern contour extraction unit 32, partial pattern offline system feature extraction unit 33, partial pattern offline system matching unit 34, partial pattern offline system recognition dictionary 35, pattern memory conversion unit 7, contour extraction unit 8, offline system feature extraction unit 9, an offline system matching unit 10, an offline system recognition dictionary 11, a low stroke number character identifying unit 52, and an output terminal 13. The output side of the tablet 1 is connected to the input side of the preprocessing unit 2. The output side of the preprocessing unit 2 is connected to the input sides of the online system feature extraction unit 3, the pattern memory conversion unit 7, the high stroke number character identification unit 36, and the low stroke number character identification unit 52.

The output sides of the online system feature extraction unit 3 and the online system recognition dictionary 5 are connected to the input side of the online system matching unit 4. The output side of the online system matching unit 4 is connected to the input side of the high stroke number character identification unit 36. The output side of the pattern memory conversion unit 7 is connected to the input side of the contour extraction unit 8. The output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The output sides of the offline method feature extraction unit 9 and the offline method recognition dictionary 1 are connected to the input side of the offline method matching unit 10. The output side of the offline method matching unit 10 has a low stroke number character identification unit 5
2 is connected to the input side. The output side of the online system feature extraction unit 3 is connected to the online system matching unit 4 and the partial pattern memory conversion unit 31. The output side of the partial pattern / pattern memory conversion section 31 is connected to the input side of the partial pattern contour extraction section 32. The output side of the partial pattern contour extraction unit 32 is connected to the input side of the partial pattern offline method feature extraction unit 33.

Partial pattern offline method feature extraction unit 3
3 and the output side of the partial pattern offline method recognition dictionary 35 are connected to the input side of the partial pattern offline method matching unit 34. The output side of the partial pattern offline method matching unit 34 is connected to the input sides of the high stroke number character identifying unit 36 and the low stroke number character identifying unit 52. The output sides of the high stroke number character identifying unit 36 and the low stroke number character identifying unit 52 are connected to the output terminal 13. Hereinafter, operations (a) to (n) of the respective units in FIG. 18 will be described.

(A) Tablet 1 As in the first embodiment, the tablet 1 has a writing data sequence {(x _i , y _i ), i = 1, 2, ..., N _j } of coordinate sequences.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the online system feature extraction unit 3
The smoothed writing data string is output to the pattern memory conversion unit 7, and the stroke number is output to the high stroke number character identifying unit 36 and the low stroke number character identifying unit 52.

(C) Online Method Feature Extraction Unit 3 The online method feature extraction unit 3 extracts the feature points from the output of the preprocessing unit 2 by performing, for example, linear approximation processing, as in the first embodiment. To do. Then, the stroke pattern is roughly classified, and the partial pattern of the written character is obtained from the cut position of the partial pattern of the character to be recognized in the character dictionary.
A vector between the partial patterns is calculated, a matching distance d _vec between the vector between the partial patterns of the character dictionary and a vector between the partial patterns of the written character is calculated, and d _vec is compared with VECREJ to perform the middle classification. Further, the Q value of the written character is obtained for each partial pattern in accordance with the partial pattern of the recognition target character remaining in the middle classification.

(D) Online system matching unit 4 The online system matching unit 4 receives the output of the online system feature extraction unit 3 and calculates the matching distance d _BP from the Q value, as in the first embodiment. The matching distance value d is calculated from d _vec and d _BP , and the high stroke character identifying unit 36
Output to (E) Partial pattern pattern memory conversion unit 31 The partial pattern pattern memory conversion unit 31 receives a cut position of a partial pattern forming a character and a corresponding coordinate data string from the online system feature extraction unit 3, Similar to the embodiment, the conversion process to the pattern memory is performed for each partial pattern.

(F) Partial pattern contour extraction unit 32 The partial pattern contour extraction unit 32 reads out each pattern memory output from the partial pattern pattern memory conversion unit 31 and, as in the third embodiment, reads out each partial pattern. The contour extraction processing is performed, and the contour information is output to the partial pattern offline method feature extraction unit 33. (G) Partial Pattern Offline Method Feature Extraction Unit 33 The partial pattern offline method feature extraction unit 33 receives contour data for the number of partial patterns output from the partial pattern contour extraction unit 32, as in the third embodiment. From each contour data, 16 feature quantities of horizontal, vertical, 45 °, and −45 ° components qB _{k, 0 to} qB _{k, 15} (k = 1,2, ..., BP
(N and BPN are the number of partial patterns) and output to the partial pattern offline method matching unit 34.

(H) Partial pattern offline system matching unit 34 The partial pattern offline system matching unit 34
Similar to the above embodiment, the feature quantity qB _{k, 0 for} each partial pattern output from the partial pattern offline method feature extraction unit 33.
˜qB _{k, 15} (k = 1,2, ..., BPN) and the feature quantities qB ^* _{k, 0} ˜qB ^* _{k, for} each matching target partial pattern registered in advance in the partial pattern offline method recognition dictionary 35 _{. 15} (k =
1, 2, ..., BPN) and a partial pattern offline method matching distance value ddb is calculated and output. (I) High stroke number character identification unit 36 The high stroke number character identification unit 36 uses the output d _k of the online method matching unit 4 and the output ddb _{k of the} partial pattern offline method matching unit 34 to calculate a new distance value according to Expression (47). d _highk is sorted and sorted in ascending order,
When it is determined that the number of strokes is high, the corresponding character code is output from the output terminal 13 and displayed on a display (not shown) or the like. If it cannot be determined that the number of strokes is high, nothing is output. Here, it is considered that the number of strokes is high depending on the characteristic parameter to be extracted, but here it is assumed that characters with 7 strokes or more are high strokes.

(J) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the feature points of the output data of the preprocessing unit 2 and is the same as in the first embodiment. To a two-dimensional array PM (x, y),
It is output to the contour extraction unit 8. (K) Contour extraction unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), for example, contour tracing is performed to extract contour data, as in the first embodiment. (L) Off-line method feature extraction unit 9 The off-line method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and, for example, as in the first embodiment, the horizontal component, the vertical component, 45 °, for example. The components, q _{0 to} q ₁₅ , of the −45 ° component are obtained.

(M) Off-line Method Matching Section 10 The off-line method matching section 10 stores the reference feature quantities q ₀ ^{* to} q ₁₅ ^* for the characters including the recognition target range, as in the first embodiment. By sequentially comparing the offline method recognition dictionary 11 and the feature quantities q _{0 to} q _{15 of} the pattern memory calculated from the contour data,
The difference between the two, that is, the matching distance dd is calculated and output.

(N) Low Stroke Number Character Identification Unit 52 When the low stroke number character identification unit 52 receives the stroke number output from the preprocessing unit 2 and determines that the stroke number is low, it is the same as in the seventh embodiment. to the output dd _k offline mode matching unit 10 receives an output ddb _k partial pattern offline mode matching unit 34 calculates a matching distance d _{low k} low strokes character from the equation (49), sorted in ascending order Then, the corresponding character code is output from the output terminal 13 and displayed on a display device (not shown) or the like. If it cannot be determined that the stroke count is low, nothing is output.

Here, it is determined that the stroke number is low
It seems that it depends on the feature parameters to be extracted,
Here, since the high stroke number character identifying unit 36 determines that the characters having 7 strokes or more have a high stroke number, the characters having 6 strokes or less are set as the low stroke number characters.
As described above, according to the ninth embodiment, the high stroke number character has the same advantages as those of the third embodiment, and the low stroke number character has the same advantages as those of the seventh embodiment. is there.

Tenth Embodiment FIG. 19 is a block diagram showing an online handwritten character recognition apparatus according to the tenth embodiment of the present invention. Elements common to those in FIG. 1 are designated by common reference numerals. There is. The difference between the online handwritten character recognition device according to the tenth embodiment and the online handwritten character recognition device according to the first embodiment is that the output of the offline method matching unit 10 is performed regardless of whether the high stroke number character or the low stroke number character is used. Therefore, the character recognition is performed. As shown in FIG. 19, the tenth book
The online handwritten character recognition device according to the embodiment of the present invention includes a tablet 1, a preprocessing unit 2, a pattern memory conversion unit 7, a contour extraction unit 8, an offline system feature extraction unit 9, an offline system matching unit 10, an offline system dictionary 11, and an identification unit. 6
1, and an output terminal 13.

The output side of the tablet 1 is connected to the input side of the preprocessing section 2. The output side of the preprocessing unit 2 is connected to the input side of the pattern memory conversion unit 7. The output side of the pattern memory conversion unit 7 is connected to the input side of the contour extraction unit 8. The output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The output sides of the offline method feature extraction unit 9 and the offline method recognition dictionary 1 are connected to the input side of the offline method matching unit 10. The output side of the off-line method matching unit 10 is connected to the input side of the identification unit 61. Hereinafter, FIG.
The operations (a) to (g) of each unit of 9 will be described.

(A) Tablet 1 As in the first embodiment, the tablet 1 has a coordinate data writing data string {(x _i , y _i ), i = 1, 2, ..., N _j }.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the online system feature extraction unit 3
The smoothed writing data string is output to the pattern memory conversion unit 7. (C) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the feature points of the output data of the pre-processing unit 2 and, as in the first embodiment, 2 Convert to the dimension array PM (x, y),
It is output to the contour extraction unit 8.

(D) Contour Extraction Unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), contour tracing is performed and contour information is extracted as a contour, as in the first embodiment. (E) Off-line method feature extraction unit 9 The off-line method feature extraction unit 9 obtains the contour data extracted by the contour extraction unit 8 and then, similarly to the first embodiment, a horizontal component, a vertical component, a 45 ° component, Q ₀ to q of −45 ° component
_Ask for ₁₅ . (F) Offline Method Matching Unit 10 The offline method matching unit 10 stores the reference feature quantities q ₀ ^{* to} q ₁₅ ^* for the characters including the recognition target range, respectively, as in the first embodiment. By sequentially comparing the recognition dictionary 11 and the feature quantities q _{0 to} q _{15 of} the pattern memory calculated from the contour data,
The difference between the two, that is, the matching distance dd is calculated and output.

(G) Discriminating Unit 61 The discriminating unit 61 sorts the matching distances output by the off-line method matching unit 10 in ascending order, and outputs the corresponding character codes to the output terminal 13. As described above, according to the tenth embodiment, since the pattern memory conversion unit 7, the contour extraction unit 8, the off-line method feature extraction unit 9, and the identification unit 61 are provided, regarding the high stroke number character and the low stroke number character. , No longer depends on stroke count and stroke order.

Eleventh Embodiment FIG. 20 is a block diagram showing an online handwritten character recognition device according to an eleventh embodiment of the present invention. Elements common to those in FIG. 10 are designated by common reference numerals. There is. The point that the online handwritten character recognition device of the 11th embodiment differs from the online handwritten character recognition device of the 10th embodiment is that
In either case of the high stroke number character or the low stroke number character, the matching distance with the character to be recognized stored in the offline method recognition dictionary 71 having the same stroke number as the input character is calculated. As shown in FIG. 20, the online handwritten character recognition device according to the eleventh embodiment includes a tablet 1, a preprocessing unit 2, a pattern memory conversion unit 7, a contour extraction unit 8, an offline system feature extraction unit 9, and an offline system matching. It includes a unit 70, an offline system recognition dictionary 71, an identification unit 61, and an output terminal 13. The output side of the tablet 1 is connected to the input side of the preprocessing unit 2. The output side of the preprocessing unit 2 is connected to the input sides of the pattern memory conversion unit 7 and the offline system matching unit 70.
The output side of the pattern memory conversion unit 7 is connected to the input side of the contour extraction unit 8.

The output side of the contour extraction unit 8 is connected to the input side of the offline system feature extraction unit 9. The output sides of the offline system feature extraction unit 9 and the offline system recognition dictionary 71 are connected to the input side of the offline system matching unit 70. The output side of the off-line method matching unit 70 is connected to the input side of the identification unit 61. The offline method recognition dictionary 71 stores the offline feature amount (for example, the reference feature amount q ^* _{0 to} q ^* ₁₅ ) of the character pattern to be recognized together with the stroke number. Hereinafter, operations (a) to (g) of each unit of FIG. 20 will be described.

(A) Tablet 1 As with the first embodiment, the tablet 1 has a writing data sequence {(x _i , y _i ), i = 1, 2, ..., N _j } of coordinate sequences.
Is extracted and sent to the preprocessing unit 2. (B) Pre-Processing Unit 2 As in the first embodiment, the pre-processing unit 2 smoothes the writing data of the coordinate sequence, and the pattern memory converting unit 7
The smoothed writing data is output and the number of strokes is output to the off-line method matching unit 70. (C) Pattern memory conversion unit 7 The pattern memory conversion unit 7 receives the coordinate data string {X _i , Y _i } of the feature points of the output data of the pre-processing unit 2 and, as in the first embodiment, 2 Convert to the dimension array PM (x, y),
It is output to the contour extraction unit 8.

(D) Contour extraction unit 8 The contour extraction unit 8 outputs the output PM of the pattern memory conversion unit 7.
Upon receiving (x, y), the contour of the binarized input character pattern is extracted and contour information is output. (E) Off-line method feature extraction unit 9 The off-line method feature extraction unit 9 obtains the contour data {x _ij , i = 1,2, ..., j = 1,2, ...} extracted by the contour extraction unit 8. ,
Similar to the first embodiment, the horizontal component, the vertical component, +45
The feature values of q _{0 to} q ₁₅ are obtained for the ° component and the −45 ° component. (F) Off-line method matching unit 70 The off-line method matching unit 70 receives the feature amount output from the off-line method feature extraction unit 9 and the number of strokes output from the pre-processing unit 2, and matches the off-line method recognition dictionary 71. When performing, the matching process is performed only on the characters that match the number of strokes of the input pattern, and the characters are output to the recognition unit 61.

(G) Discriminating Unit 61 The discriminating unit 61 sorts the matching distances output from the off-line method matching unit 10 in ascending order, and outputs the corresponding character codes to the output terminal 13. As described above, the eleventh embodiment has the same advantages as the tenth embodiment. In addition, the number of strokes is output from the preprocessing unit 2 to the off-line method matching unit 70 so that only the characters that match the number of strokes of the input pattern are matched, so that the processing speed is improved as compared with the tenth embodiment. . The present invention is not limited to the above embodiment,
Various modifications are possible. For example, there are the following modifications (1) to (7).

(1) The processing in the offline system feature extraction unit 9 and the offline system matching unit 10 of the first to eleventh embodiments may be performed by another method used in an optical character recognition device or the like. . (2) The processes of the online method feature extraction unit 3 and the online method matching unit 4 of the first to ninth embodiments are
Other processing may be used. (3) The high stroke number character identifying units 6, 26 and 36 may perform character recognition by further classifying based on the calculated matching distances by a stroke code distribution indicating a stroke distribution. (4) High Strokes character identification unit 6,26,36 from an output d _k of the online mode matching unit 4, an output dd _k offline mode matching unit 10, an output ddb _k partial pattern offline mode matching unit 34 , The weighting coefficient may be multiplied to calculate the matching distance for character recognition.

(5) Low stroke number character identification section 12, 42, 5
2 is the output dd of the off-line method matching unit 10.
and _k, and the output ddb _k partial pattern offline mode matching unit 34, and an output d _k of the online mode matching unit 4 is multiplied by the weighting factor may be the character recognition by calculating a matching distance. (6) In the seventh embodiment, the pattern memory conversion unit 7, the contour extraction unit 8, the offline system feature extraction unit 9, the offline system matching unit 10, and the offline system recognition dictionary 11 are eliminated, and the low stroke number character identification unit 52 is provided. May perform character recognition only from the matching distance output from the partial pattern offline method matching unit 34. (7) In the tenth and eleventh embodiments, the pattern memory conversion unit 7, the contour extraction unit 8, the offline system feature extraction unit 9, the offline system matching unit 10, and the offline system recognition dictionary 11 are eliminated and the identification unit 61 is provided. Is the 7th
Character recognition may be performed based on the matching distance output from the partial pattern offline method matching unit 34 configured in the same manner as in the above embodiment.

[0124]

As described in detail above, first to first
According to the online handwritten character recognition device of the invention of 2, the character recognition is performed for the low stroke number character based on the offline feature amount of the written character, so that it does not depend on the stroke number and stroke order, and the recognition rate is improved. Can be made.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an online handwritten character recognition device according to a first embodiment of the present invention.

FIG. 2 is an operation explanatory diagram of a preprocessing unit.

FIG. 3 is a diagram showing an example of a character dictionary.

FIG. 4 is an explanatory diagram of vectors between partial patterns.

FIG. 5 is an explanatory diagram of calculation of a matching distance d _vec of an inter-partial pattern vector of “a”.

FIG. 6 is a diagram showing an example of an online system recognition dictionary.

FIG. 7 is a diagram showing pattern memory conversion processing.

FIG. 8 is a diagram showing an example of pattern memory conversion.

FIG. 9 is an operation explanatory diagram of a contour extracting unit.

FIG. 10 is a diagram showing component classification of contour data.

FIG. 11 is a configuration diagram of an online handwritten character recognition device according to a second embodiment of the present invention.

FIG. 12 is a configuration diagram of an online handwritten character recognition device according to a third embodiment of the present invention.

FIG. 13 is a configuration diagram of an online handwritten character recognition device according to a fourth embodiment of the present invention.

FIG. 14 is a configuration diagram of an online handwritten character recognition device according to a fifth embodiment of the present invention.

FIG. 15 is a configuration diagram of an online handwritten character recognition device according to a sixth embodiment of the present invention.

FIG. 16 is a configuration diagram of an online handwritten character recognition device according to a seventh embodiment of the present invention.

FIG. 17 is a configuration diagram of an online handwritten character recognition device according to an eighth embodiment of the present invention.

FIG. 18 is a configuration diagram of an online handwritten character recognition device according to a ninth embodiment of the present invention.

FIG. 19 is a configuration diagram of an online handwritten character recognition device according to a tenth embodiment of the present invention.

FIG. 20 is a configuration diagram of an online handwritten character recognition device according to an eleventh embodiment of the present invention.

[Explanation of symbols] 1 tablet 2 pre-processing unit 3 online system feature extraction unit 4 online system matching unit 5 online system recognition dictionary 6, 26, 36 high stroke number character identification unit 7 pattern memory conversion unit 8 contour extraction unit 9 offline system feature Extraction unit 10,70 Offline system matching unit 11,71 Offline system recognition dictionary 12,42,52 Low stroke number character identification unit 13 Output terminal 31 Partial pattern pattern memory conversion unit 32 Partial pattern contour extraction unit 33 Partial pattern offline system feature extraction unit 34 Partial pattern offline system matching unit 35 Partial pattern offline system recognition dictionary 61 Identification unit

Claims (12)

[Claims] 1. A pre-processing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by writing on a tablet, and the handwriting from the coordinate data sequence smoothed by the pre-processing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online method matching unit that calculates a matching distance, and a high stroke number character that performs character recognition based on the matching distance output from the online method matching unit when the stroke number obtained by the number of strokes of the written character is a predetermined stroke number or more. An identification unit and a region including the entire written character obtained from the coordinate data string smoothed by the pre-processing unit are divided into a plurality of regions vertically and horizontally. A pattern memory conversion unit that determines whether or not each of the plurality of regions includes a part of the writing character, allocates binarized data to each region, and stores the binarized data in a pattern memory. A contour extraction unit that reads the pattern memory and extracts contour data, an offline method feature extraction unit that extracts the feature amount of the written character from the contour data obtained by the contour extraction unit, and an offline method feature extraction unit The offline method matching unit that calculates the matching distance between the feature amount of the written character and the feature amount of the recognition target character that is stored in advance in the offline method recognition dictionary, and the number of strokes obtained by the stroke number of the written character is When the number of strokes is smaller than the predetermined number of strokes, character recognition is performed based on the matching distance output from the offline matching unit. Online handwriting recognition apparatus characterized by a low number of strokes character identifying unit, comprising the performing. 2. A preprocessing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by handwriting on a tablet, and the handwriting from the coordinate data sequence smoothed by the preprocessing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online matching unit that calculates a matching distance, and a region including the entire written character obtained from the coordinate data string smoothed by the preprocessing unit is divided into a plurality of regions vertically and horizontally, and each of the plurality of regions is divided. A pattern memory conversion unit that determines whether or not a region includes a part of the written character, allocates binarized data to each region, and stores the binarized data in a pattern memory; A contour extraction unit for reading the memory and extracting contour data, an offline method feature extraction unit for extracting the feature amount of the written character from the contour data obtained by the contour extraction unit, and an offline method feature extraction unit. An offline method matching unit that calculates a matching distance between the feature amount of the written character and the feature amount of the recognition target character that is stored in advance in the offline method recognition dictionary, and the number of strokes obtained by the stroke number of the written character is predetermined. When the number of strokes is greater than or equal to the number of strokes, based on the matching distance output from the online method matching unit and the offline method matching unit, a high stroke number character identification unit that performs character recognition, and the number of strokes obtained by the stroke number of the written character is Output from the offline matching unit when the number of strokes is smaller than a predetermined number. Based on the serial matching distance, and a low number of strokes character identification unit performing character recognition, on-line handwritten character recognition apparatus characterized by comprising. 3. A preprocessing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by writing on a tablet, and the writing from the coordinate data sequence smoothed by the preprocessing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online matching unit that calculates a matching distance, and a division based on the coordinate data string according to the stroke position of a partial pattern that indicates a portion to be written as a series in writing when a character to be recognized is written A region including each partial pattern of the written character is vertically and horizontally divided into a plurality of regions, and each of the plurality of regions includes a part of the partial pattern of the written character. It is determined whether or not each of the areas is assigned binarized data, and the partial pattern pattern memory conversion unit that stores the partial pattern pattern memory in the partial pattern pattern memory, and the partial pattern pattern memories are read and A partial pattern contour extraction unit that extracts the contour data of the partial pattern; and a partial pattern offline method feature extraction unit that extracts the feature amount of the partial pattern of the written character from the contour data of the partial pattern obtained by the partial pattern contour extraction unit And calculating a matching distance between the feature amount of the written character obtained by the partial pattern offline type feature extraction unit and the feature amount of the partial pattern of the recognition target character stored in advance in the partial pattern offline type recognition dictionary. Partial pattern off-line matching section, and the writing character straw When the number of strokes obtained by the number of strokes is equal to or greater than a predetermined number of strokes, based on the matching distance output from the online method matching unit and the partial pattern offline method matching unit, a high stroke number character identification unit that performs character recognition, Whether the area including the entire written character obtained from the coordinate data string smoothed by the preprocessing unit is vertically and horizontally divided into a plurality of areas, and each of the plurality of areas includes a part of the written character. It is determined whether or not each area is assigned binarized data and stored in a pattern memory, a pattern memory conversion unit that reads the pattern memory and extracts contour data, and a contour extraction unit that obtains the contour data. An offline method feature extraction unit that extracts the feature amount of the written character from the contour data that has been written; and the brush obtained by the offline method feature extraction unit. An offline method matching unit that calculates a matching distance between the character amount of the written character and the character amount of the character to be recognized that is stored in advance in the offline method recognition dictionary, and the number of strokes obtained by the stroke number of the written character is the predetermined number. An online handwritten character recognition device, comprising: a low stroke number character identification unit that performs character recognition based on the matching distance output from the offline type matching unit when the number of strokes is smaller than the number of strokes. 4. A preprocessing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by writing on a tablet, and the handwriting from the coordinate data sequence smoothed by the preprocessing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online method matching unit that calculates a matching distance, and a high stroke number character that performs character recognition based on the matching distance output from the online method matching unit when the stroke number obtained by the number of strokes of the written character is a predetermined stroke number or more. An identification unit and a region including the entire written character obtained from the coordinate data string smoothed by the pre-processing unit are divided into a plurality of regions vertically and horizontally. A pattern memory conversion unit that determines whether or not each of the plurality of regions includes a part of the writing character, allocates binarized data to each region, and stores the binarized data in a pattern memory. A contour extraction unit that reads the pattern memory and extracts contour data, an offline method feature extraction unit that extracts the feature amount of the written character from the contour data obtained by the contour extraction unit, and an offline method feature extraction unit An offline method matching unit that calculates a matching distance between the characteristic amount of the written character and the characteristic amount of the character to be recognized that is stored in advance in the offline method recognition dictionary, and the number of strokes obtained by the stroke number of the written character is When the number of strokes is smaller than the predetermined number of strokes, the output from the online method matching unit and the offline method matching unit is performed. Based on the etching distance, and a low number of strokes character identification unit performing character recognition, on-line handwritten character recognition apparatus characterized by comprising. 5. A pre-processing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by writing input on a tablet, and the handwriting from the coordinate data sequence smoothed by the pre-processing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online matching unit that calculates a matching distance, and a region including the entire written character obtained from the coordinate data string smoothed by the preprocessing unit is divided into a plurality of regions vertically and horizontally, and each of the plurality of regions is divided. A pattern memory conversion unit that determines whether or not a region includes a part of the written character, allocates binarized data to each region, and stores the binarized data in a pattern memory; A contour extraction unit for reading the memory and extracting contour data, an offline method feature extraction unit for extracting the feature amount of the written character from the contour data obtained by the contour extraction unit, and an offline method feature extraction unit. An offline method matching unit that calculates a matching distance between the feature amount of the written character and the feature amount of the character to be recognized, which is stored in advance in the offline method recognition dictionary, and the number of strokes obtained by the stroke number of the written character is predetermined. When the number of strokes is greater than or equal to the number of strokes, based on the matching distance output from the online method matching unit and the offline method matching unit, a high stroke number character identification unit that performs character recognition, and the number of strokes obtained by the stroke number of the written character is When the number of strokes is smaller than a predetermined number, the online method matching unit and the offline Based on the matching distance which is output from the mode matching unit, and a low number of strokes character identification unit performing character recognition, on-line handwritten character recognition apparatus characterized by comprising. 6. A preprocessing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by writing on a tablet, and the handwriting from the coordinate data sequence smoothed by the preprocessing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online matching unit that calculates a matching distance, and a division based on the coordinate data string according to the stroke position of a partial pattern that indicates a portion to be written as a series in writing when a character to be recognized is written A region including each partial pattern of the written character is vertically and horizontally divided into a plurality of regions, and each of the plurality of regions includes a part of the partial pattern of the written character. It is determined whether or not each of the areas is assigned binarized data, and the partial pattern pattern memory conversion unit that stores the partial pattern pattern memory in the partial pattern pattern memory, and the partial pattern pattern memories are read and A partial pattern contour extraction unit that extracts the contour data of the partial pattern; and a partial pattern offline method feature extraction unit that extracts the feature amount of the partial pattern of the written character from the contour data of the partial pattern obtained by the partial pattern contour extraction unit And calculating a matching distance between the feature amount of the written character obtained by the partial pattern offline type feature extraction unit and the feature amount of the partial pattern of the recognition target character stored in advance in the partial pattern offline type recognition dictionary. Partial pattern off-line matching section, and the writing character straw When the number of strokes obtained by the number of strokes is equal to or greater than a predetermined number of strokes, based on the matching distance output from the online method matching unit and the partial pattern offline method matching unit, a high stroke number character identification unit that performs character recognition, Whether the area including the entire written character obtained from the coordinate data string smoothed by the preprocessing unit is vertically and horizontally divided into a plurality of areas, and each of the plurality of areas includes a part of the written character. A pattern memory conversion unit that determines whether or not each region is assigned binarized data and stores the binarized data in a pattern memory; a contour extraction unit that reads the pattern memory and extracts contour data; An offline method feature extraction unit that extracts the feature amount of the written character from the obtained contour data, and the offline method feature extraction unit An offline method matching unit that calculates a matching distance between the feature amount of the written character and the feature amount of the character to be recognized that is stored in advance in the offline method recognition dictionary, and the number of strokes obtained by the stroke number of the written character is the predetermined number. An online handwriting, comprising: a low stroke number character identification unit that performs character recognition based on the matching distance output from the offline method matching unit and the online method matching unit when the number of strokes is smaller than the number of strokes. Character recognizer. 7. A preprocessing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by writing on a tablet, and the handwriting from the coordinate data sequence smoothed by the preprocessing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online method matching unit that calculates a matching distance, and a high number of strokes for character recognition based on the matching distance output from the online method matching unit when the stroke number obtained by the stroke number of the written character is a predetermined stroke number or more. The stroke of the character identification part and the partial pattern that indicates the part to be written as a series in writing when the recognition target character is written. A region including each partial pattern of the written character divided based on the coordinate data string according to the position is divided into a plurality of regions vertically and horizontally, and each of the plurality of regions is a partial pattern of the written character. It is determined whether or not a part is included, binarized data is allocated to each area, and a partial pattern pattern memory conversion unit that stores the partial pattern pattern memory in each partial pattern, and the partial pattern pattern memories A partial pattern contour extraction unit for reading and extracting contour data of each of the partial patterns, and a partial pattern offline for extracting a feature amount of the partial pattern of the written character from the contour data of the partial pattern obtained by the partial pattern contour extraction unit A system feature extraction unit, and a feature amount and a partial pattern of the written character obtained by the partial pattern offline system feature extraction unit. A partial pattern offline method matching unit that calculates a matching distance with the feature amount of the partial pattern of the character to be recognized, which is stored in advance in the online offline recognition dictionary, and a coordinate data string smoothed by the preprocessing unit. The obtained region including the entire written character is vertically and horizontally divided into a plurality of regions, it is determined whether or not each of the plurality of regions includes a part of the written character, and each region has a binary value. Pattern memory conversion unit that allocates the converted data and stores it in a pattern memory, a contour extraction unit that reads the pattern memory and extracts contour data, and a feature amount of the written character from the contour data obtained by the contour extraction unit. An offline method feature extraction unit for extracting, a feature amount of the written character obtained by the offline method feature extraction unit, and an offline method recognition dictionary in advance. An offline method matching unit that calculates a matching distance with the stored characteristic amount of the recognition target character, and an offline method matching unit when the stroke count obtained by the stroke number of the written character is smaller than the predetermined stroke count. An on-line handwritten character recognition device comprising: a low stroke number character identification unit that performs character recognition based on the matching distance output from the partial pattern offline type matching unit. 8. A preprocessing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by writing input on a tablet, and the handwriting from the coordinate data sequence smoothed by the preprocessing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online matching unit that calculates a matching distance, and a division based on the coordinate data string according to the stroke position of a partial pattern that indicates a portion to be written as a series in writing when a character to be recognized is written A region including each partial pattern of the written character is vertically and horizontally divided into a plurality of regions, and each of the plurality of regions includes a part of the partial pattern of the written character. It is determined whether or not each of the areas is assigned binarized data, and the partial pattern pattern memory conversion unit that stores the partial pattern pattern memory in the partial pattern pattern memory, and the partial pattern pattern memories are read and A partial pattern contour extraction unit that extracts the contour data of the partial pattern; and a partial pattern offline method feature extraction unit that extracts the feature amount of the partial pattern of the written character from the contour data of the partial pattern obtained by the partial pattern contour extraction unit And a matching distance between the feature amount of the written character obtained by the partial pattern offline type feature extraction unit and the feature amount of the partial pattern of the recognition target character stored in advance in the partial pattern offline type recognition dictionary is calculated. The partial pattern offline method matching unit and the pre-processing unit smooth the image. The region including the entire written character obtained from the coordinate data string is divided into a plurality of regions vertically and horizontally, and it is determined whether each of the plurality of regions includes a part of the written character, A pattern memory conversion unit that allocates binarized data to each area and stores it in a pattern memory, a contour extraction unit that reads the pattern memory and extracts contour data, and the contour data obtained by the contour extraction unit from the contour data An offline method feature extraction unit that extracts a written character feature amount, a feature amount of the written character obtained by the offline method feature extraction unit, and a feature amount of a recognition target character that is stored in advance in an offline method recognition dictionary. And an off-line matching unit that calculates the matching distance of the writing method, and when the number of strokes obtained by the number of strokes of the written character is equal to or greater than a predetermined stroke count, A high stroke number character identifying portion that performs character recognition based on the matching distance output from the touching portion and the offline type matching portion, and when the stroke number obtained by the stroke number of the written character is smaller than the predetermined stroke number, An online handwritten character recognition device comprising: a low stroke number character identification unit that performs character recognition based on the matching distance output from the offline system matching unit and the partial pattern offline system matching unit. 9. A preprocessing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by handwriting input on a tablet, and the handwriting from the coordinate data sequence smoothed by the preprocessing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online matching unit that calculates a matching distance, and a division based on the coordinate data string according to the stroke position of a partial pattern that indicates a portion to be written as a series in writing when a character to be recognized is written A region including each partial pattern of the written character is vertically and horizontally divided into a plurality of regions, and each of the plurality of regions includes a part of the partial pattern of the written character. It is determined whether or not each of the areas is assigned binarized data, and the partial pattern pattern memory conversion unit that stores the partial pattern pattern memory in the partial pattern pattern memory, and the partial pattern pattern memories are read and A partial pattern contour extraction unit that extracts the contour data of the partial pattern; and a partial pattern offline method feature extraction unit that extracts the feature amount of the partial pattern of the written character from the contour data of the partial pattern obtained by the partial pattern contour extraction unit And calculating a matching distance between the feature amount of the written character obtained by the partial pattern offline type feature extraction unit and the feature amount of the partial pattern of the recognition target character stored in advance in the partial pattern offline type recognition dictionary. Partial pattern off-line matching unit and smoothing by the pre-processing unit The region including the entire written character obtained from the converted coordinate data string is divided into a plurality of regions vertically and horizontally, and it is determined whether or not each of the plurality of regions includes a part of the written character. , A pattern memory conversion unit that allocates binarized data to each area and stores it in a pattern memory, a contour extraction unit that reads the pattern memory and extracts contour data, and from the contour data obtained by the contour extraction unit An offline method feature extraction unit that extracts the written character feature amount, the written character feature amount obtained by the offline method feature extraction unit, and the feature amount of the recognition target character that is stored in advance in the offline method recognition dictionary. An off-line matching unit that calculates a matching distance between the online character and a stroke number obtained by the stroke number of the written character is equal to or more than a predetermined stroke number. Based on the matching distance output from the expression matching unit and the partial pattern offline method matching unit, a high stroke number character identifying unit for performing character recognition, and the stroke number obtained by the stroke number of the written character is more than the predetermined stroke number. An online handwritten character characterized by comprising a low stroke number character identification unit that performs character recognition based on the matching distance output from the offline type matching unit and the partial pattern offline type matching unit when the size is small. Recognition device. 10. A preprocessing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by writing in a tablet, and a coordinate data sequence smoothed by the preprocessing unit. An area including the entire written character is vertically and horizontally divided into a plurality of areas, it is determined whether or not each of the plurality of areas includes a part of the written character, and binary data is stored in each area. A pattern memory conversion unit for allocating and storing in a pattern memory, a contour extraction unit for reading the pattern memory and extracting contour data, and an offline for extracting a feature amount of a handwritten character from the contour data obtained by the contour extraction unit. A system feature extraction unit, and matching between the feature amount of the written character obtained by the offline system feature extraction unit and the feature amount of the recognition target character stored in advance in the offline system recognition dictionary. And offline mode matching unit for calculating the distance, based on the matching distance which is output from the off-line mode matching unit, and an identification unit for performing character recognition, on-line handwritten character recognition apparatus characterized by comprising. 11. A pre-processing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by writing on a tablet, and a coordinate data sequence smoothed by the pre-processing unit. An area including the entire written character is vertically and horizontally divided into a plurality of areas, it is determined whether or not each of the plurality of areas includes a part of the written character, and binary data is stored in each area. A pattern memory conversion unit for allocating and storing in the pattern memory, a contour extraction unit for reading the pattern memory and extracting contour data, and a feature amount of the written character is extracted from the contour data obtained by the contour extraction unit. An offline method feature extraction unit, a feature amount of the written character obtained by the offline method feature extraction unit, and an offline method recognition dictionary that are stored in advance and have the same number of strokes as the written character. An off-line method matching unit that calculates a matching distance with the feature amount of the recognition target character, and an identifying unit that performs character recognition based on the matching distance output from the off-line method matching unit are provided. A unique online handwriting recognition device. 12. A preprocessing unit for smoothing by removing unnecessary data of a coordinate sequence of handwritten characters obtained by handwriting on a tablet, and the handwriting from the coordinate data sequence smoothed by the preprocessing unit. An online method feature extraction unit for extracting a character feature amount; a feature amount of the written character obtained by the online method feature extraction unit; and a feature amount of a recognition target character stored in advance in an online method recognition dictionary. An online method matching unit that calculates a matching distance, and a character recognition unit that performs character recognition based on the matching distance output from the online method matching unit when the number of strokes obtained by the stroke number of the written character is equal to or more than a predetermined stroke number. The stroke number character identification part and the stroke of the partial pattern indicating the part to be written as a series in writing when the character to be recognized is written. A region including each partial pattern of the handwritten character divided based on the coordinate data string according to the position is divided vertically and horizontally into a plurality of regions, and each of the plurality of regions is a partial pattern of the handwritten character. A partial pattern pattern memory conversion unit that determines whether or not a partial pattern is included, allocates binarized data to each area, and stores the binary data in the partial pattern pattern memory for each partial pattern; And a partial pattern contour extraction unit that extracts contour data of each of the partial patterns, and a partial pattern that extracts the feature amount of the partial pattern of the written character from the contour data of the partial pattern obtained by the partial pattern contour extraction unit. An offline method feature extraction unit, the partial pattern offline method feature extraction unit, and the feature amount and partial pattern of the written character. A partial pattern offline method matching unit that calculates a matching distance with a feature amount of a partial pattern of the character to be recognized, which is stored in advance in a turn-offline method recognition dictionary, and the number of strokes obtained by the stroke number of the written character is the predetermined number. When the number of strokes is smaller than the number of strokes, the on-line handwritten character recognition device is provided with a low stroke number character recognition unit that performs character recognition based on the matching distance output from the partial pattern offline method matching unit.