JP3066530B2 - Online handwriting recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Overview] An on-line character recognition device that fetches handwritten character data information online and recognizes it in real time. In addition, feature point extraction means for obtaining a feature point and position information of the feature point from the input character pattern information, feature point dictionary means for storing feature points of dictionary characters, feature points of the input character according to the stroke order, Feature point associating means for associating feature points of dictionary characters stored by the feature point dictionary means to obtain candidate characters from the dictionary characters and outputting correspondence information at the time of the association; An identification condition storage means for storing an identification condition defined by using a feature point of the dictionary character, and a candidate character obtained by the feature point association means. The identification condition defined for the input character pattern obtained by searching the identification condition storage means and the characteristic point of the input character pattern and the position information of the characteristic point obtained by the characteristic point extracting means are determined by the characteristic point associating means. Using the obtained correspondence information, the input point is obtained by obtaining a feature point of the input character corresponding to the feature point of the candidate character used in the identification condition and obtaining a line segment between the obtained input character feature points. An identification condition application unit that applies a character to the identification condition obtained by searching the identification condition storage unit and determines whether the identification condition is satisfied, and includes a process of obtaining the candidate character and a process of applying the identification condition. Thus, the recognition process is performed while sharing the information of the feature points.

[Industrial applications]

The present invention relates to a character recognition device, and more particularly, to an online character recognition device that captures and recognizes handwritten character data information online in real time.

[Conventional technology]

The character recognition device reads a pre-written character by, for example, an image scanner or the like, obtains a distance between a feature point of each character and a feature point of dictionary data stored in advance, and determines a character having a closest distance as a candidate character. I have decided. In the above-described method of recognizing a read character with an image scanner, the stroke order cannot be obtained, and the recognition rate is low.

On the other hand, there is a so-called online character recognition device that recognizes, instead of inputting character data with an image scanner, for example, reading information written by a tablet or the like. This on-line character recognition device has a wide application range and is in high demand in that it requires less skill than input devices such as a keyboard. For example, use as a means for directly inputting handwritten characters into a computer in a window business is increasing. Such an online character recognition device has a feature that stroke order information is obtained as compared with the case where the image data is read by a conventional image scanner, and the recognition rate is higher than that of the above-described recognition device.

In recent years, the performance of online character recognition devices has improved. In particular, a technique has been developed that can recognize continuous characters, and its use in inputting at the above-mentioned window has come closer to the practical stage.

However, the performance of identifying characters having similar shapes is not sufficient, and high performance is also required for identifying characters having similar shapes.

In general, an online character recognition device is composed of a large classification that outputs recognition candidates and a detailed similar character identification unit that obtains a correct recognition result by using the recognition candidates as input.

FIG. 5 is a configuration diagram of a conventional online character recognition device. The pattern input 10 is a device for converting handwritten characters into digital information, for example, a digitizer. This digitizer is usually composed of a board and a stylus pen. When the input person writes the board surface with the stylus pen, the digitizer detects the coordinates written with the stylus pen. This detection result is added to the input pattern storage unit 11 and stored. The digital information stored in the input pattern storage unit 11 is added to the feature point extraction unit 12,
The feature point extracting unit 12 extracts a feature point representing a character feature from the stored input pattern, and stores the feature point in the input feature point storage unit 13.

The feature point dictionary unit 14 stores feature points of dictionary characters that are standard character patterns for each dictionary character.

The feature point association unit 15 associates the result of the feature point extraction unit 12 stored in the input feature point storage unit 13 with the feature points of the dictionary characters stored in the feature point dictionary unit 14 in the order of the stroke order. That is, the distance between the input pattern and the dictionary character is obtained from the feature points obtained by the feature point extraction unit 12 in accordance with the stroke order stored in the feature point dictionary unit 14. Further, the feature point associating unit 15 stores, in the recognition code storage unit 16, a character code of a dictionary character that gives a smaller distance from the obtained distance to the dictionary character as a recognition candidate code.

The identification condition storage unit 17 stores, for each dictionary character, a condition for identifying the dictionary character. The identification condition application unit 18 searches for an identification condition from the candidate candidate codes higher in the recognition candidate code stored in the recognition code storage unit 16 and applies the identification condition to the input pattern stored in the input pattern storage unit 11. Then, as a result of the application, the code of the dictionary character that has passed the identification condition is stored in the recognition result storage unit 19.

The above-mentioned feature point associating unit 15 is for associating from an image, but may be erroneously associated depending on the association. FIG. 6 is a conventional identification condition chart for input characters and dictionary characters corresponding to each other. The input character is "2",
The character “Z” in the feature point dictionary unit 14 is associated with the feature point association unit 15.

In such a case, the identification condition storage unit 17 stores, for example, "Z""the beginning of writing on the first screen has no upwardly curved portion."
And the identification condition applying unit 18 determines that the dictionary character “Z” is not a candidate for the input character. Similarly, if the input character is “3” and the dictionary character “5” is a candidate character, the dictionary character “5” is replaced by the condition that “5” is “two or more strokes”. Remove from candidate characters.

If the katakana "o" is mistaken for "age", it is determined that the candidate is incorrect because of the identification condition of "age""the second and third strokes have a certain degree of intersection". In addition, when "Kan" is associated with "Vague", it is detected that the association is incorrect from the identification condition of "Kan" that "the 9th and 12th strokes do not intersect". I have.

In the case of online handwritten character recognition, it is possible to determine whether or not the candidate character is correct from the above-described strokes, stroke order, and the conditions where the strokes intersect. For this reason, a recognition rate much higher than the recognition read by an image scanner or the like is obtained.

[Problems to be solved by the invention]

In the online handwritten character recognition device as described above, handwritten information can be captured in real time, and it is possible to recognize the writing order in this real time capture, so that the recognition rate is high. ing. However, in the case of online handwritten character input, the input person does not always input consistently with the identification condition, and may input a continued character, that is, a continuous character.

FIG. 7 is an input character chart that cannot be identified under the aforementioned identification conditions. When inputting “3”, if the user once writes between the upper and lower parts of “3” with a brush, the input characters will be two strokes and will pass the identification condition of “5”. Also, if you wrote "o" continuously, that is, you wrote one and two strokes continuously, there would be no third stroke,
The condition for identifying "age" cannot be applied. Similarly, the kanji "kan"
If you continue the part of Sanzui and continue the part after that, the 12th stroke will be lost,
Identification conditions cannot be applied.

Under the above-described conditions, the input person writes the kanji continuously, but the same kanji is used. However, the recognition device changes the number of strokes due to the continuation of the kanji, and the recognition becomes confused.

That is, in the similar character identification in the conventional recognition device,
The identification conditions are directly applied to the stroke order of the input characters.
For this reason, in order to extract a character feature such as a connection feature, an intersection feature, and a curved feature of an image as a condition, it is necessary to obtain a reference point from the input pattern. However, in order to identify a character including a continuous character, it is not possible to specify a location where an identification condition is given by specifying the number of strokes. Further, if a procedure for obtaining a reference point from the input data is added to the identification condition, there is a problem that the storage capacity of a memory for storing the condition is increased and the time for applying the condition is long.

In the identification of "2" in FIG. 6, it is necessary to perform a process of obtaining the writing start part of the first stroke, and in the case of an example such as "3" or "Kan", a character other than the standard stroke number is dealt with. Can not. Further, it is conceivable to store the identification condition one by one for the method of writing a character having a non-standard stroke number. However, there are many methods of writing a character having a non-standard stroke number, and this is not practical.

The present invention aims at online character recognition with a high recognition rate.

[Means for solving the problem]

 FIG. 1 is a functional block diagram of the present invention.

The feature point extracting means 1 adds input character pattern information including the stroke order, and obtains a feature point and position information of the feature point from the input character pattern information. This input character pattern information is character information having a stroke order and a character pattern.

 The feature point dictionary means 2 stores feature points of dictionary characters.

The feature point associating means 3 associates the feature points of the input character with the feature points of the dictionary characters stored in the feature point dictionary means 2 according to the stroke order, obtains candidate characters from the dictionary characters, and performs the association. The corresponding information at the time of is output.

The identification condition storage means 4 stores, for each dictionary character, an identification condition defined by using the characteristic points of the dictionary character.

The identification condition application unit 5 searches the identification condition storage unit 4 for an identification condition defined for each candidate character obtained by the feature point association unit 3, and determines an input condition obtained by the feature point extraction unit 1. Using the feature points of the character pattern and the position information of the feature points and the correspondence information obtained by the feature point associating means 3, the feature of the input character corresponding to the feature point of the candidate character used in the identification condition By finding a point and finding a line segment between the feature points of the input character thus obtained, the input character is applied to the identification condition obtained by searching the identification condition storage means 4, and it is determined whether the identification condition is satisfied. Determine.

That is, the recognition process is performed while the feature point information is shared between the process of obtaining the candidate character and the process of applying the identification condition.

[Action]

When the input pattern information is added, the feature point extracting means 1 obtains a feature point and position information of the feature point from the input character pattern information including the stroke order. The feature point associating means 3 associates the feature points obtained by the feature point extracting means 1 with the feature points of the dictionary characters stored in the feature point dictionary means 2 to obtain correspondence information and to provide candidate characters of the association recognition result. Ask for. Candidate characters are obtained by the feature point association means 3. The identification condition application unit 5 searches the identification condition storage unit 4 from the candidate characters obtained by the feature point association unit 3 to determine the identification condition, and also determines the characteristic points of the input character pattern determined by the characteristic point extraction unit 1 Using the position information of the feature points and the correspondence information obtained by the feature point associating means 3, a feature point of the input character corresponding to the feature point of the candidate character used in the identification condition is obtained. By obtaining a pattern (part of a pattern; for example, a line segment) between feature points, the input character is applied to the identification condition obtained by searching the identification condition storage means 4, and it is determined whether or not the identification condition is satisfied. .

Since the identification condition is determined not by the number of strokes but by the relationship of the feature point position, whether or not the identification condition is satisfied is determined. Therefore, even if there are consecutive characters, the input character can be accurately determined.

〔Example〕

 Hereinafter, the present invention will be described in detail with reference to the drawings.

FIG. 2 is a configuration diagram of the online character recognition device of the embodiment.

The pattern input unit 20 is, for example, a digitizer. When an input person writes an input character on a tablet with a stylus pen, the coordinates at which the pen has moved are read, and the input pattern storage unit 21 stores the input information. This input information is stored in the stroke order written with the stylus pen. That is, the input pattern information is information including the stroke order.

The feature point extraction unit 22 reads out the input pattern information stored in the input pattern storage unit 21 and extracts feature points representing the features of the character from the input pattern information.
Store in 23. The information on the feature points stored in the input feature point storage unit 23 is at least (X, Y) coordinate data of the feature points on a tablet or the like, and other various information conventionally used in the large classification process. . Further, the position of the feature point in the input pattern information (position information of the feature point) is stored in the feature point position storage unit 24. As will be described later, the position information of the feature point is obtained by obtaining the feature point of the input character corresponding to the feature point of the dictionary character by the corresponding data of the feature point in the identification processing, and further inputting the feature point of the input character. This is data for obtaining a point on the character pattern (which point on the input character pattern this characteristic point was originally; the position of the characteristic point on the input character pattern).
Then, the feature point associating unit 25 reads the information of the input feature points stored in the input feature point storage unit 23, and associates the feature points of the dictionary characters described in the feature point dictionary unit 26,
Candidate characters having a short correspondence distance, for example, three candidate characters are extracted from the smallest characters. Then, the extracted result is added to the recognition candidate code storage unit 27, and the corresponding data associated with the feature points is stored in the feature point corresponding data storage unit 28.

The candidate code stored in the recognition candidate code storage unit 27 is
This is a code assigned to one character, and the code is different for each character.

The output of the input pattern storage unit 21, the input feature point storage unit 23, the feature point position storage unit 24, the feature point corresponding data storage unit 28, and the output of the recognition code storage unit 27 are added to the identification condition application unit 29,
The feature points of the input characters corresponding to the feature points of the dictionary characters described using the information stored in these storage units 21, 23, 24, 27, and 28 and the identification conditions stored in the identification condition storage unit 30 Is read to determine whether or not the identification condition is satisfied.
That is, the identification condition application unit 29 reads the identification conditions stored in the identification condition storage unit 30, and stores the characteristic points of the input character corresponding to the characteristic points of the dictionary characters used in the identification conditions in the characteristic point data storage. It is obtained from the correspondence data between the characteristic points of the input characters and the characteristic points of the dictionary characters stored in the section 28. Further, a point on the input character pattern for the characteristic point of the input character is obtained from the data stored in the characteristic point position storage unit 24. Then, the identification condition is applied to the input pattern and its characteristic points. In other words, in order to apply, for example, the condition “pattern portion between feature points A and B” in the identification condition as shown in FIG. After obtaining the characteristic points of the input character using the data, it is necessary to obtain the points (positions) of the characteristic points on the input character pattern and obtain the input character pattern information (that is, the pattern portion) between the positions. This means that it is necessary to use the data (position information of the feature points) stored in the feature point position storage unit 24 for that purpose.

That is, it is determined whether or not the candidate character satisfies the obtained identification condition, and the candidate character that satisfies the identification condition is determined as a recognition result. When a plurality of candidate characters satisfy the identification condition, a recognition result having a small distance between feature points is used.

Compared with the conventional technique shown in FIG. 5, the feature point extracting unit 22 stores the input feature points in the input feature point storage unit 23 and also stores the feature point position information in the feature point position storage unit 24. The feature point associating unit 25 stores the recognition candidate code in the recognition code storage unit 27 and stores the feature point corresponding data in the feature point data correspondence storage unit 28.

In the above-described conventional method, the identification condition applying unit 18 determines the identification condition without using the feature point position and the feature point correspondence data, but in the embodiment of the present invention, the feature point position storage position and the Since the discrimination is performed using the feature point correspondence data, it is possible to prevent erroneous discrimination with respect to successive characters. That is, in the related art, the identification condition is used in correspondence with the number of strokes. Seeking letters. Then, the identification condition application unit 29 satisfies the identification condition from the information stored in the feature point position storage unit 24, the input pattern storage unit 21, and further the feature point correspondence data storage unit 28 at the location designated by the identification condition. Seeking what you are.

 Note that this result is stored in the recognition result storage unit 31.

FIG. 3 is a table showing correspondence between feature points and identification conditions in the embodiment. When the input character is “2” and the candidate character is “Z”, the candidate character is determined based on the identification condition of “Z” [corresponding to the characteristic points A and B provided at the upper part which is the characteristic point of “Z”. The pattern part between the characteristic points A and B of the input character does not have a convex upward bend]. It can be determined that the input character is not “Z” by the identification condition that there is no bending. FIG. 3 shows a feature point used for the identification condition, that is, a part of the feature point. As shown in FIG. 3, for example, a start point, an end point, a bending point, and the like of each line segment are characterized on the character pattern. Points.

Conventionally, it is necessary to perform a process of obtaining a writing start portion of the first stroke of “Z”, but in the embodiment, the feature points A,
It is only necessary to find a convex bend for the association of B.

In addition, when the input characters “3”, “O”, and “Kan” in FIG. 6 are input as shown in FIG. 7, the identification conditions of the candidate characters “5”, “age”, and “vague” Previously, the identification conditions passed or could not be applied,
As shown in FIG. 3, if the input character is "3",
[Feature points A and B of the input character corresponding to feature points A and B of "5"
Is on another image], it can be determined that the dictionary character “5” is not a recognition result. Also, input "O"
In the case of, the line segment A, B, C, D for the characteristic points A, B, C, D of the input character corresponding to the characteristic points A, B, C, D of the dictionary character Thus, even if the first stroke and the second stroke of “sai” are written continuously, it can be clearly determined that this character is not “given”. In addition, when the Sanzui part of "Han" and 8 to 10 strokes are written in succession, the 9th and 12th strokes do not intersect under the conventional identification conditions, and the identification condition of "Vague" is satisfied. However, the identification condition in the embodiment is the same as that of the input character corresponding to the characteristic points A, B, C and D of the dictionary character
Line segments A, B, C, D do not intersect for A, B, C, D)
It can be seen that the result is not the recognition result. “A” and “B” of “vague” represent line segments of 9 strokes, and “C” and “D” represent 12 strokes. In the present invention, this line segment is not represented by an image, but is represented by a line segment of a feature point obtained by feature point extraction. Therefore, it is possible to accurately determine whether a candidate character is correct based on an identification condition.

FIG. 4 is a configuration diagram of an identification condition application unit. The identification condition application unit 29 includes an identification condition search unit 291, an input character condition creation unit 292,
It comprises an input condition storage unit 293 and a judgment unit 294.

The identification condition search unit 291 reads the recognition candidate codes one by one from the recognition candidate code storage unit 27 in order. An identification condition using the read recognition candidate code as an identification character is searched for and read in the identification condition storage unit 30, and is output to the input character condition creation unit 292.

The input character condition creating unit 292 converts the identification condition described for the dictionary character into the identification condition described for the input character using the data stored in the feature point corresponding data storage unit 28 and the feature point position storage unit 24. Are stored in the input condition storage unit 293. At this time, the determination unit 294 determines the condition stored in the input condition storage unit 293 using the input feature point data stored in the input feature point storage unit 23 and the data stored in the input pattern storage unit 21. Apply to input characters.

If the identification condition is passed, the recognition candidate code is stored in the recognition result storage unit 31. If not, the above operation is repeated for the next candidate code starting from 27.

As described above, according to the present invention, feature points are extracted from the stroke order written sequentially, and are determined using the identification conditions at the feature points. Characters can be identified.

〔The invention's effect〕

As described above, according to the present invention, it is possible to accurately identify a character even if it is a continuous character, thereby reducing the required storage capacity of the identification condition and shortening the execution time.

[Brief description of the drawings]

FIG. 1 is a functional block diagram of the present invention, FIG. 2 is a configuration diagram of an online character recognition device of an embodiment, FIG. 3 is a correspondence table of feature points and a table of identification conditions in one embodiment, and FIG. 5 is a block diagram of a conventional on-line character recognition device, FIG. 6 is a table of corresponding input characters and dictionary characters and a conventional identification condition table, and FIG. 7 is an input character that cannot be identified by the identification conditions. It is a chart. 1. Feature point extracting means 2. Feature point dictionary means 3. Feature point associating means 4. Identification condition storing means 5. Identification condition applying means.

 ──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-58-105388 (JP, A) JP-A-60-89291 (JP, A) JP-A-63-301383 (JP, A)

Claims (1)

(57) [Claims] An input character pattern information including a stroke order is added, and a feature point which is a point representing a feature of the input character is extracted from the input character pattern information, and a position of the feature point on the input character pattern is determined. Feature point extracting means for obtaining position information to be shown; feature point dictionary means for storing feature points of dictionary characters; feature points of the input characters according to the stroke order; and feature points of dictionary characters stored by the feature point dictionary means. A feature point associating means for obtaining candidate characters from the dictionary characters in association with each other and outputting correspondence information at the time of the association; and for each of the dictionary characters, a feature point associating unit is defined using the feature points of the dictionary characters. An identification condition storage unit for storing an identification condition; and an identification condition defined for each candidate character obtained by the feature point association unit, obtained by searching the identification condition storage unit. Using the feature points of the input character pattern obtained by the output means and the correspondence information obtained by the feature point associating means, a feature point of the input character corresponding to the feature point of the candidate character used in the identification condition is obtained. Further determining the position of the feature point of the determined input character on the input character pattern using the position information, applying the input character to the identification condition determined by searching the identification condition storage means, An identification condition application unit that determines whether the identification condition is satisfied is configured to perform recognition processing while sharing feature point information in the processing for obtaining the candidate character and the processing for applying the identification condition. An online handwritten character recognition device characterized by the following.