KR900005141B1 - Handwritter character recognizing device - Google Patents

Abstract

The handwritten letter recognition system comprises a memory circuit (4) for storing coordinates of stroke on a tablet, a characteristic patern memory circuit (7) stored with standard characteristic pattern, a characteristic point extraction circuit (5a) to decide the number of the charateristic point and extract the position coordinates of the characteristic point, and an identification circuit (5c) to identify the written letter by comparing the position coordinates of the characteristic point with that of the standard characteristic pattern.

Description

Character Recognition Device

1 to 3 are diagrams for explaining a representative method of recognizing a character within the effective time.

4 is a configuration diagram schematically showing a character recognition device according to a first embodiment of the present invention.

FIG. 5 is a diagram showing an example of the configuration of the recognition processing unit in FIG.

6 is a configuration diagram showing an example of the configuration of the identification circuit in FIG.

7 is a view for explaining a control concept of the extracted feature score for the stroke.

8 shows the shape of feature points representing strokes.

9 is a diagram for explaining a problem with pseudo stroke.

10 is a block diagram showing an extraction processing circuit of a local feature point.

11 and 12 show the shapes of the local feature points of the local stroke, respectively.

* Explanation of symbols for main parts of the drawings

1 tablet device 2 coordinate detection circuit

3: previous step processing circuit 4: coordinate sequence memory circuit

5: stroke number detection circuit 5a: feature point extraction circuit

5b: feature point sequence memory circuit 5c: identification circuit

6: recognition processing unit 7: standard feature pattern storage unit

8: Answer determination circuit 11: Head address table

12,13,14: Register 15: Multiplexer

16: adder 17: operation control circuit

18,19: register

[Industrial use]

The present invention relates to a device for recognizing a character or a symbol input through a coordinate input device, and more particularly, to a character recognition device capable of easily and precisely recognizing an input character or symbol at the same time.

[Traditional Technology and Problems]

Character input device (OCR) for inputting character information for a calculator, which conventionally reads alphabetic characters or Japanese Kana characters, etc. described in documents, and compares the character pattern with a standard pattern. Is developed.

However, since there are thousands of kinds of characters constituting Japanese, including Kanji and Kana characters, it is necessary to optically read these various characters using the OCR, and then analyze the character patterns to recognize the characters. A great deal of recognition processing is required, and complicated pattern processing is required to increase the accuracy of recognition. For this reason, it was extremely difficult to adopt the character recognition device using the OCR as the character information input means of a small calculator, such as a personal computer, which is widely used.

On the other hand, in recent years, as the semiconductor technology is advanced, coordinate input devices such as a pressure-sensitive tablet device, etc., which can be obtained at a low price while effectively utilizing the information processing capability of the personal computer and the like which have been greatly improved. Character information input devices have been developed in which handwriting input of character information is used to detect the character information by immediately detecting the character information as coordinate information.

By adopting the character information input method as described above, not only can easily recognize thousands of kinds of character information including Chinese characters, but also can immediately input the character information input means such as the personal computer. As a result, practicality becomes very high.

As described above, the following methods have been proposed as a recognition method used in a character recognition device that recognizes character information input by hand through a coordinate input device.

That is, as one representative recognition method, for example, as shown in FIG. 1, when the letter "英" is handwritten, the shapes of a plurality of strokes forming the letter are determined for each stroke. And a method of identifying the input character from the shape code sequence and the positional information of each stroke.

However, in such a recognition method, in order to improve the recognition performance, it is necessary to prepare about 20-30 types of standard stroke shapes in order to cope with various shapes of strokes that are deformed according to the habits of the writer. In order to precisely recognize the characters by absorbing the deformation of the stroke caused by the deformation noise of the so-called returning stroke or jumping stroke, the stroke-shaped pattern to be prepared is very complicated. The number is inevitably very large.

Therefore, as shown in Fig. 2 (a), the writing stroke of the character to be recognized is sequentially coded using the determined direction nose, for example, the character " 日 " shown in Fig. 2 (b). The writing stroke of the writing stroke is sequentially coded, and the writing stroke is represented by the direction code sequence as shown in Fig. 2C by adding length information and up / down information of the writing instrument to the direction code. A method of simplifying the character recognition process by reducing the load information representing the size is also proposed (however, in Fig. 2, ① to ④ indicate corresponding parts).

But. In the above recognition method, the writing order of the input characters is changed due to the wrong writing order, or when the stroke is inputted with excessive noise, the sequence information of the stroke representing the input characters is greatly changed. The inconvenience is that you can not recognize the character stably by absorbing the error or noise.

In order to cope with such inconvenience, for example, as shown in FIG. 3, a character recognition method is proposed in which each stroke of a character pattern is represented by a plurality of feature points on the stroke.

In this character recognition method, the number of split points of each stroke is determined according to the number of strokes constituting the character, and the number of strokes of the handwritten characters is divided at equal intervals according to the number of split points. The point and the start point and the end point of the stroke are expressed as the position coordinates of the feature points representing the strokes. Then, character recognition is performed by performing a matching process between the position coordinates of the feature points of these strokes and the position coordinates of the standard pattern feature points prepared in advance.

For example, in the case of Hiragana, which has many curved components and few strokes, the stroke length is divided into five parts, and each stroke is expressed as a feature point position coordinate that approximates six strokes. The strokes are divided into two, and each stroke is expressed by the position coordinates of the feature points approximated by three points. Here, the three-point approximation of the strokes is intended to approximate the strokes reliably because each stroke represented by two straight lines exists relatively in the Chinese character pattern.

However, the majority of the strokes constituting the character may be similarly represented without requiring the information of their even dividing points, in which case each stroke is assigned to a plurality of feature points including the dividing points. Since it is approximated, the number of standard feature patterns is extremely large for the recognition process at this time, and the processing time becomes very long because each of the standard feature patterns needs to be matched with each other. There is a problem.

In addition, since each stroke is represented by the same number of feature points, the matching result obtained from, for example, a short stroke or a linear stroke is likely to be overestimated, whereas from the feature points of the stroke having a lot of information for character identification. The matching result obtained is easily underestimated and therefore not appropriate. Therefore, there are many cases where characters are misrecognized or unrecognized.

In addition, in the extraction of the feature points of the stroke described above, in case of similar characters as shown in FIG. 3, if there is only a local difference in the stroke such as "わ" and "ね" of Hiragana, this is certainly identified. I could not. Therefore, conventionally, a feature extraction means according to a separate method is installed to cope with the above irrationality. However, in this case, the algorithm of recognition processing becomes uselessly complicated and the standard feature pattern to be prepared is increased, thus recognizing the character. There is a problem that the configuration of the apparatus for doing so becomes complicated. For this reason, the above means are not suitable as a character recognition device capable of inputting them at the same time as the creation of the characters (effective time).

[Purpose of invention]

Accordingly, the present invention has been invented to eliminate the above-described drawbacks, and the object of the present invention is to provide a character recognition apparatus having a high recognition rate of handwritten characters by designating a sufficient number of split points (number of feature points) of each stroke of the handwritten characters. have.

[Configuration of Invention]

The present invention for achieving the above object is a coordinate system storage circuit for storing the time series coordinates of the writing stroke of a character or symbol handwritten input through a tablet device, and information on the standard feature pattern of the recognition target category prepared in advance The number of feature points representing each stroke of the written character or symbol is determined based on the number information of the standard feature pattern storage unit to be stored and the feature points representing each stroke determined for each stroke of the standard feature pattern. And a feature point extraction circuit for obtaining the position coordinates of each feature point from the time series coordinates of the stroke stored in the coordinate sequence memory circuit, and a feature point position coordinate of each stroke of the handwritten input character or symbol obtained by the feature point extraction circuit. It has a structure provided with the identification circuit which identifies the said handwritten character or symbol.

[Action]

According to the present invention having the above-described configuration, the minimum required number of feature points is obtained for each stroke of the input character corresponding to the category of the character to be recognized, the number of copies constituting the character, and the like. After the stroke is appropriately expressed, the matching process of the position coordinates of the feature points representing each stroke and the feature information on the standard feature pattern makes it possible to recognize the input characters easily and accurately within the effective time. That is, the accuracy of the approximation of the input stroke is sufficiently increased to obtain a high recognition result.

In addition, since the number of feature points representing the stroke is determined as necessary, the number of feature points can be sufficiently reduced. Therefore, the number of standard feature patterns to be prepared is also reduced, so that the structure of the storage unit can be simplified, and the number of feature points to be contrasted with this is reduced, thereby significantly reducing the recognition processing time.

EXAMPLE

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

4 is a configuration diagram schematically showing a character recognition device according to a first embodiment of the present invention, wherein the information of a character or a symbol to be inputted by using the writing instrument 1a on the tablet device 1 is a tablet device ( Coordinate detection circuit 2 attached to 1) detects time series position coordinate data (time series data) of continuous coordinate positions representing the stroke.

Then, the time-series position coordinate data detected by the coordinate detection circuit 2 is inputted to the previous stage processing circuit 3 so as to detect minute errors such as a detection error, a jumping stroke, a turning stroke, etc. in the tablet device 1. After the noise is removed. It is stored in the coordinate series memory circuit 4 in order. In addition, the preceding step processing circuit 3, for example, the clock position coordinate data stored in the coordinate sequence memory circuit 4 when a passage for one character of the handwritten characters is detected by the coordinate detection circuit 2, for example. The previous step is also performed, such as normalizing the character size.

In this way, the coordinate sequence memory circuit 4 stores time series position coordinate data indicating the handwritten stroke for one character processed as described above as input character data.

The number detecting circuit 5 also detects the number of strokes, i.e., strokes, of the handwritten characters by detecting the up / down of the writing instrument 1a with respect to the surface of the tablet device 1 from the output of the coordinate detecting circuit 2. . Information about the number of strokes detected by the stroke number detection circuit 5 is input to the recognition processing section 6 and the standard feature pattern storage section 7.

The standard feature pattern storage unit 7 stores standard patterns such as all characters and symbols used in accordance with the stroke number information, and standardizes the number of times to be recognized and processed according to the stroke number information from the stroke detection circuit 5. Only feature patterns can be read. In addition, the recognition processing section 6 reads out the standard feature pattern of the corresponding stroke number in order from the standard feature pattern storage unit 7 according to the stroke number information, and at the same time stores the time series position coordinates stored in the coordinate sequence memory circuit 4. After reading the data to obtain the position coordinates of each stroke feature point corresponding to the standard feature pattern, a plurality of characters of the standard feature pattern read out from the standard feature pattern storage unit 7 are sequentially sequenced to each stroke of the input written character pattern. The matching process is executed by comparing the position coordinate data of the feature point indicating and the matching result is output to the answer determination circuit 8. Then, the answer determination circuit 8 identifies and judges each matching processing result, and obtains the category of the standard feature pattern that is most likely to be the recognized result as the identification result of the handwritten characters or symbols. In this case, as the recognition result, a plurality of recognition characters with high similarity may be output as result candidates.

The recognition processing section 6 is composed of a feature point extraction circuit 5a, a feature point sequence storage circuit 5b, and an identification circuit 5c, as shown in FIG. 5, for example. The positional coordinate data in time series is read out from the coordinate sequence memory circuit 4, and the time series data of each stroke is divided at equal intervals according to the maximum number of feature points set in advance for each stroke. As a result, a plurality of feature points including the start point and the end point of each stroke are extracted. For example, if the maximum number of feature points is 9, each stroke of the input character pattern is divided into eight equal parts, seven split points that are eight equal points of the stroke, and nine total points including the start and end points of the stroke. The feature point position coordinates are obtained for each stroke. The feature point extraction circuit 5a obtains the position coordinates of these feature points from the data stored in the coordinate sequence memory circuit 4. Therefore, each stroke is represented by nine feature points, and the position coordinate data of the feature points are stored in the feature point sequence storage circuit 5b.

The feature point detection device is repeated for each stroke until one phrase of the input character is detected. When the feature point position information of each stroke for one character is stored in the feature point sequence storage circuit 5b, the extracted feature point data and the standard feature pattern storage section 7 are stored in the feature point sequence storage circuit 5b. Matching processing with the registered standard feature pattern is executed.

FIG. 6 shows a specific coupling configuration relationship between the identification circuit 5c, the standard feature pattern storage unit 7 and the feature point sequence storage circuit 5b. In FIG. 11, FIG. Represents the head address table.

The head address table 11 registers the number of input characters of the first registered address of the standard feature pattern among the plurality of lists of the standard feature pattern storage unit 7 which classify and register the standard pattern according to the number of letters. So remember. The head address data of the number of strokes detected by the stroke number detection circuit 5 is set in the head address table 11. The head address data from the head address table 11 is applied as a data read address to the standard feature pattern storage section 7 via the register 12.

In the first address of the standard feature pattern storage unit 7 designated by the above-mentioned address, a standard feature coordinate sequence for the first stroke of the first characters of the stroke number is stored as a feature pattern. When the character category is j, the first feature point coordinates are represented by (XOj, YOj), the second feature point coordinates are represented by (X1j, Y1j), and the third feature point coordinates are represented by (X2i, Y2j). Will be.

In addition, the header of the first address includes pitch information of the number of feature points for the stroke, the position information of the feature point, the stroke number, and the stroke in relation to the standard feature coordinate system. Information about whether or not a stroke is present, whether such information relates to the final character category in that special block, and whether the standard feature coordinate series is the overall or local character of the character. It is.

Along with the header as described above, the standard feature pattern of each character category for the number of strokes is registered in the standard feature pattern storage unit 7 in order. The header and standard feature pattern of each character category are read out sequentially from the one stored at the head address according to the address from the register 12 which is sequentially output in accordance with the command signal.

The pitch data and stroke number of the number of feature points read out from the standard feature pattern storage unit 7 are transferred to the register 13, and the data stored in the register 14 is cleared at the same time as this transfer timing. The mode of the multiplexer 15 is switched. In other words. The operation mode for recording the feature point information of each stroke for the input character into the feature point sequence storage circuit 5b is switched from the operation mode for reading data of the feature points stored in the feature point sequence memory circuit 5b. According to the output of the register 14, the data read address of the feature point sequence storage circuit 4b is designated. In this case, as described above, the first register 14 is first cleared. The positional coordinate data of the feature point (starting point) is read out from the memory circuit 5b. The position coordinate data of the first feature point (starting point) of the first stroke is input to the identification circuit 5c and read from the standard feature pattern memory 7 to the first feature point of the first stroke of the standard feature pattern ( The distance from the position coordinate data of the starting point) is calculated.

When the coordinate data of one feature point is outputted from the feature point sequence storage circuit 5b through the above process, the data of the feature point pitch stored in the register 14 is stored in the register 14 through the adder 16. The data is added to the data and set in the register 14 again. As a result, the next feature point to be read out from the feature point sequence storage circuit 5b is designated by advancing by the number appearing in the pitch. Similarly, the position coordinate data of the feature point is read out, and the identification circuit 5c calculates the distance of the feature point coordinate of the standard feature pattern. This process is repeatedly executed for each feature point selected sequentially in accordance with the pitch of the feature points, so that the distances obtained for each feature point are sequentially added. When the distance is calculated for all the characteristic points of all the strokes, the addition value of the distance is output as a result of the similarity between the standard feature pattern and the input character, and then from the standard feature pattern storage unit (7). The standard feature pattern of the next category is read out, and the matching process for the next standard feature pattern is executed in the same manner as described above.

The above-described processing is repeatedly executed by the header information until the standard feature pattern being processed is displayed as the last character category of the number of strokes.

7 shows the concept of the number of feature points variably controlled by the feature point pitch described above. As described above, each stroke is stored in the feature point sequence storage circuit 5b, for example, represented by a total of nine feature points including seven divided points obtained by dividing the length of the stroke into eight equal parts and a start point and an end point.

For this feature point sequence, the pitch data designates the extraction pitch of the feature point. For example, when the pitch is given as 1, all the feature points are sequentially extracted from the starting point as shown in FIG. On the other hand, if the pitch is given as 2, the feature points are selected every second as shown in Fig. 7 (c). If the pitch is given as 4, the feature points are selected every fourth as shown in Fig. 7 (b). do. However, when the stroke can be expressed by only one feature point, only one feature point can be selected and supported as shown in (e) of FIG. 7. As an example in which the stroke of one feature point can be expressed, the letter “點” ", The four strokes (" ") of the incident" fire "can be mentioned.

Through the above process, the number of feature points sufficient to represent each stroke is specified according to the pitch information registered in advance in the standard feature pattern storage unit 7, so that the feature points of the input stroke corresponding to the number are selectively extracted. do. That is, the three strokes of the input writing pattern as shown in FIG. 8A, for example, correspond to the first stroke and the feature points representing the respective strokes of the standard feature pattern shown in FIG. 8B. The second stroke is extracted as two feature points (start and end points), and the third stroke is extracted as six feature points.

The distances between the feature points corresponding to the two patterns are calculated, respectively, and the similarity is evaluated.

On the other hand, there is no guarantee that the order (stroke number) of the strokes stored in the standard feature pattern storage unit 7 and the order of the strokes inputted by hand are always matched. You can cope with it by contrast. By providing such flexibility, it is possible to cope with the recognition process of Chinese characters with a large number of strokes. That is, when a part of the collation result deviates greatly from the predetermined evaluation value in a range of strokes, it becomes useful to improve the processing efficiency by stopping the subsequent recognition process.

According to the above-described apparatus, each stroke of the character pattern is represented by a necessary and sufficient number of feature points, that is, the input stroke is represented by a number of feature points set according to the feature amount of the stroke of the recognition target category. By performing the recognition process, the distance between the specific points obtained from a simple stroke such as a straight line is not excessively evaluated, and an overall straightforward evaluation of the input character pattern can be obtained. In other words, the number of feature points is reduced for simple strokes such as straight lines, and the number of feature points necessary for expressing the strokes is given to strokes such as curves, so that the approximate accuracy of the strokes by the feature points is increased. It is possible to increase the precision of. Further, as described above, each stroke is represented by an appropriate number of feature points, so that the number of feature points of the entire character can be reduced, thereby reducing the amount of computation and thereby speeding up the recognition process. Therefore, it is possible to maximize the practical advantage by removing the irrationality seen in the conventional character recognition method.

However, as shown in FIG. 9, when the letter "WA" is handwritten, it may not be possible to obtain a sufficiently reliable recognition result only by the above-described character recognition process. That is, when the character pattern is displayed as shown in Fig. 9 (a), the standard characteristic patterns of the letters "わ" and "ね" are given as shown in Figs. 9 (b) and 9 (c). There is little difference in the distance between the feature points. In other words, in the above case, since there is only a wrong place near the end point of the third stroke, the difference between the approximated strokes according to the feature points becomes very small. For this reason, it is expected that the recognition evaluation result of the input characters for the two standard feature patterns will be about the same.

In order to cope with such irrationalities, the apparatus extracts the local feature points of the stroke and reevaluates the recognition process.

Fig. 10 shows the configuration of a control circuit for solving the above irrationality, which is activated when sufficient identification cannot be made for the plurality of character recognition candidates obtained by the above-described recognition process.

The standard feature pattern storage section 7 registers the coordinate data of the local feature points so as to be distinguished by the information indicating whether or not the above-described local feature is identified in order to identify such a local stroke feature. In this case, the above-mentioned identification information is read out from the standard feature pattern storage unit 7 and applied to the operation control circuit 17. The information indicating the number of feature points and the data of the feature point positions are stored in the register 18. Set to (19). According to the information set in this register 18, the coordinate data of the feature point of the local partial stroke is read out as a new feature point from the coordinate sequence memory circuit 4, and the address by the data set in the register 19 is read. Is stored in the feature point sequence storage circuit 5a under the control of?. Also in this case, as in the selective extraction process of the feature point described above, the local partial stroke is designated as, for example, between adjacent feature points among the feature points obtained first, and the partial stroke is equally divided to extract a predetermined number of local feature points. It is controlled as if The distance from the feature points of the standard feature pattern is calculated by selecting the required number of feature points by designating the pitch of the feature points in the above-described manner.

In other words, the recognition processing according to the local feature points of the local stroke portion is performed similarly to the recognition processing by the whole feature point extraction described above.

By the above processing method, for example, as shown in Fig. 11A, feature points representing respective strokes of the input characters are extracted, and as shown in Figs. 11B and 11B. The standard feature patterns of the two similar character patterns with the coordinate data of the local feature points are contrasted with each feature point. Therefore, the result of the recognition processing including the local feature points is greatly different with respect to the two standard patterns so that the identification can be performed. The same is true of the Chinese characters. For example, as shown in FIG. 12, when the identification between the characters "閔" and "問" is not sufficiently high, localization is performed on the stroke described in "門". The recognition may be performed by extracting the feature points.

Thus, by using the information of the local feature point appropriately it is possible to dramatically improve the accuracy of the character recognition. Furthermore, the extraction of the local feature points can be performed in the same manner as the extraction of the feature points for the entire stroke as described above, so that the structure of the device is not complicated and the recognition processing algorithm can be refined. Therefore, not only the device configuration is easy but also the production of a standard feature pattern can be easily achieved.

In addition, the present invention is not limited to the above embodiments, and for example, a local feature point sequence is created by setting a feature point over a plurality of strokes instead of a single stroke, or for a plurality of partial strokes of one stroke. Local feature points may be selected and created. In addition, although the recognition process by a local feature point may be performed for every standard feature pattern, the recognition process by a local feature point may be performed only about the thing with high similarity after complete | finishing the whole stroke recognition with respect to all the standard feature patterns. It is also useful to extract local feature points by joining adjacent strokes, and may be modified and carried out without departing from the gist of the present invention.

[Effects of the Invention]

As described above, according to the present invention, not only the character is recognized from the characteristic point of the stroke of the entire character, but also the local feature point extraction of the stroke is efficiently carried out using the same means, and the evaluation result is used for the character recognition. By doing so, highly accurate recognition results can be obtained. Therefore, for example, it is possible to realize a large number of practical effects, such that it is possible to recognize characters input by handwriting through a decompression tablet device with high accuracy within the effective time and input them to a personal computer.

Claims (7)

Coordinate system memory circuit 4 for storing time-series coordinates of handwritten strokes of letters or symbols handwritten through the tablet device 1, and standards for storing information on standard feature patterns of categories to be recognized in advance. The feature pattern storage unit 7 determines the number of feature points representing each stroke of the written character or symbol based on the number information of the feature points representing each stroke determined for each stroke of the standard feature pattern. And a feature point extracting circuit 5a which obtains the position coordinates of each feature point from the time series coordinates of the stroke stored in the coordinate sequence memory circuit 4, and the handwritten input characters or symbols obtained by the feature point extracting circuit 5a. An identification circuit 5c for identifying the handwritten input characters or symbols by matching the feature point position coordinates of the respective strokes with the feature point position coordinates of the respective strokes of the standard feature pattern. Character recognition device characterized in that provided. 2. The feature point extraction circuit (5a) according to claim 1, wherein the feature point extraction circuit (5a) obtains the position coordinates for the feature points of the handwritten characters or symbols, and equally divides the handwritten strokes according to the number of feature points to obtain the position coordinates of the stroke positions. Character recognition device characterized in that. The character recognition apparatus according to claim 1, wherein the number of feature points representing the stroke is determined for each stroke according to the characteristics of the stroke. Coordinate system memory circuit 4 for storing time series coordinates of handwritten strokes relating to characters and symbols that are handwritten and input through the tablet device 1, and information on the standard feature patterns of categories to be recognized in advance and localized. A standard feature pattern storage unit (7) for storing identification information representing a feature, and each of the letters or symbols inputted by the handwriting based on the number information of the feature points representing each stroke determined for each stroke of the standard feature pattern A feature point extraction circuit 5a which determines the number of feature points representing the stroke and obtains the positional information of each feature point from the time series coordinates of the stroke stored in the coordinate sequence memory circuit 4, and this feature point extraction circuit ( A feature point sequence storage circuit 5b for storing feature point position coordinates relating to each stroke of the handwritten input character or symbol obtained by 5a), and this feature An identification circuit 5c for identifying the handwritten input characters or symbols by matching the feature point coordinates from the point sequence memory circuit 5b with the feature point position coordinates for the respective strokes of the standard feature pattern, and the identification circuit When the reliability of identification by (5c) is low, the arithmetic control circuit 17 in the feature point extraction circuit 5a generates feature point position information by locally extracting the feature points of the stroke according to the identification information. And the handwritten input character or symbol is identified again in the identification circuit 5c by comparing the locally extracted feature point position coordinates with the feature point position coordinates set locally in the standard feature pattern. Character recognition device. The feature point extraction circuit (53) according to claim 4, wherein the feature point extraction circuit (53) obtains the position coordinates of the feature points with respect to the handwritten characters or symbols, and obtains the position coordinates of the stroke positions determined by equally dividing the writing strokes according to the number of feature points. Character recognition device characterized in that the. The character recognition apparatus according to claim 4, wherein the number of feature points representing the stroke is determined for each stroke according to the characteristics of the stroke. 5. The method of claim 4, wherein the feature points of the strokes that are extracted locally are increased by increasing the number of extraction feature points for the strokes or by subdividing partial strokes between the feature points set by dividing the strokes according to the number of feature points. Character recognition device characterized in that.

Images