JPS6269388A - Method for recognizing handwritten character - Google Patents

Abstract

PURPOSE:To recognize handwritten characters with high accuracy by understanding each stroke of the input characters as a string of feature points consisting of stop points and the start and end points of strokes and recognizing specific characters based on said string of feature points. CONSTITUTION:A stop point detector 2 processes time series coordinate data given from an input device 1 for handwritten characters and detects the start and end points of each stroke and the stop point existing between these start and end points. Then the detector 2 delivers the coordinate values of said points to a stroke feature amount calculating device 3. The device 3 defines the coordinate values of those start and end points and the stop point of each stroke as the feature points of each stroke and calculates the features of strokes. A stroke identification device 4 specifies a standard pattern based on the features of each stroke and the features of each standard stroke are stored previously in a stroke dictionary 41. A character recognizing device 5 contains a character dictionary 51 which stores data showing KANJI (Chinese characters), KANA (Japanese syllaraby), numeric characters, the alphabet, etc. in the form of the combinations of standard strokes.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method for recognizing characters input by hand with a handwritten character input device used as a character input device such as a Japanese word processor.

[Prior art]

In recent years, as an input device for word processors, for example, instead of keyboards that require a certain level of skill to operate, word processors have been configured to LRm characters input by hand with an input pen on a tablet and accept the input of these characters. A handwritten character input device has been developed and put into practical use.

By the way, in conventional devices of this kind, the characteristics of each stroke are expressed based on time-series coordinate data on the tablet of each stroke of a character input by hand as the trajectory of an input pen on a tablet for character input. Coordinate values are selectively detected and used as feature points, each stroke is approximated by a series of these feature points, and the coordinate values of the feature point series approximated to each stroke or the positional relationship of the feature point series are obtained. Characters input by hand with an input pen are recognized based on the stroke characteristics.

Now, as a method for detecting the feature points of each of the above-mentioned strokes, the following two methods have conventionally been mainly adopted.

The first method is to perform character recognition using each point obtained by sequentially dividing input time-series coordinate data at equal distance intervals and the start and end points of a stroke as feature points.

The second method is to detect a point where the time-series movement direction of coordinate values changes from the input time-series coordinate data as a pole,
Character recognition is performed using this and the start and end points of the stroke as feature points.

[Problem that the invention seeks to solve]

However, in the first method described above, although the process of detecting the feature points of the stroke is relatively simple, there is a small possibility that the detected feature points will match the bending point of the stroke, and therefore It is difficult to accurately detect a bending point, and therefore the shape of a stroke having a bending point cannot be accurately detected.

In addition, in the second method, it is necessary to calculate the time-series small movement direction of the coordinate value from the time-series coordinate data of the read stroke each time, which results in an excessive amount of data processing required for input processing. Input processing requires time, and a data processing device with an unnecessarily large capacity is required. Furthermore, this method is susceptible to minute changes in the direction of movement of coordinate values, and many unnecessary feature points may be detected.

[Means for solving problems]

The present invention has been made in view of the circumstances as explained above, and when inputting handwritten characters, the input pen stops at the bending point of the stroke, in other words, the input pen moves on the input surface of the handwritten character input device. By focusing on the fact that when the trajectory of the pen is detected as time-series coordinate data, the same coordinate values are detected continuously over a certain period of time, and by using the stationary point of the input pen as a feature point. It is an object of the present invention to provide a handwritten character recognition method that can accurately detect the bending point of each stroke with a relatively small amount of calculation and recognize characters input by hand with high accuracy.

The present invention provides a handwritten character recognition method that recognizes, as a character, the trajectory of an input pen read as time-series coordinate data representing a position on a character input surface of a character input device, in which the trajectory of one character of the input pen is configured. Detect the starting point, end point, and stationary point of the input pen detected as a coordinate position where the time-series coordinate data repeats the same value a predetermined number or more consecutively, and characterize each of the starting point, end point, and stationary point of each stroke. Based on the coordinate values of a feature point sequence made up of these feature points or the characteristics of each stroke determined from the mutual positional relationship of each feature point, the character input by hand as the trajectory of the input pen is calculated. Characterized by recognition.

[Principle of the invention]

First, the principle of the handwritten character recognition method according to the present invention will be explained.

FIG. 1 shows, as an example, actual measured values of time-series coordinate data when the kanji character ``handled'' is input using a handwritten character input tablet. Note that the measured data has a temporal resolution of 100 points/second and a spatial resolution of 10 points/-1. In the figure, each number indicates the number of times the coordinate value of each position was detected consecutively, and the part marked with an O indicates the part where the above-mentioned number is 4 or more. The underlined numbers each indicate the number of times the same coordinate value overlaps at the position marked with the O mark.

The following facts can be clearly read from the measured data shown in FIG. That is, at bending points in the middle of the same stroke, the number of duplicates of the same coordinate values is quite high, and at least four or more duplicates occur at the start and end points of each stroke. Therefore, if the position of the coordinate value with the number of repetitions of 4 or more is used as a feature point, not only the bending point of the stroke but also the start point and end point can be accurately detected.

〔Example〕

Hereinafter, the present invention will be explained based on drawings showing embodiments thereof.

FIG. 2 is a block diagram of a handwritten character recognition device used to implement the present invention in accordance with the principles described above.

1 in the figure is a handwritten character input device, and the trajectory of the input ben that moves on the character input surface is set on the input surface.
It is detected as time series coordinate data in the Y coordinate system.

This time-series coordinate data is then given to the stationary point detection device 2.

The stationary point detection device 2 processes the time series coordinate data given from the handwritten character input device 1 to determine the starting point of each stroke,
The end point and the stationary points existing therebetween are detected, and their coordinate values are output to the stroke feature amount calculation device 3. Note that the data processing operation of this stationary point detection device 2 will be described later.

The stroke feature amount calculation device 3 calculates the starting point of each stroke given from the stationary point detection device 2.

The coordinate values of the end point and stationary point are taken as the feature points of each stroke,
The features of the stroke are calculated from the coordinate values of these feature point sequences. Note that the features to be detected for each stroke include the angle of the line segment connecting feature points, the relative length)
, how line segments connecting feature points are connected (clockwise or counterclockwise, etc.).

The characteristics of each of these detected strokes are then provided to the stroke identification device 4.

The stroke identification device 4 classifies the stroke as a specific standard stroke, such as r-J, based on the characteristics of each stroke given from the stroke feature calculation device 3 described above.
A stroke where the characteristic points such as rN, r/J, etc. are only the beginning and end points, or "ko".

r-'J, r<J, r>J, etc. (in the method of the present invention, these are detected as stationary points between the start and end points), respectively. .

In addition, the standard pattern of strokes is created by breaking down each character such as kanji, kana, numbers, alphabets, etc. into strokes according to the order of strokes when handwriting them, and then breaking them down into standard strokes as described above. is stored in advance in the stroke index 1f41 attached to the stroke identification value W4. Then, the stroke feature amount calculation device 3 identifies the stroke characteristics given from the stroke feature amount calculation device 3 as one of the standard strokes by comparing them with the characteristics of the standard strokes stored in the stroke dictionary 41. Then, for example, the code number of the standard stroke is output to the character recognition device 5.

The character recognition device 5 is a device that recognizes a character input by handwriting into the handwritten character input device 1 as a specific character based on a combination of code numbers of standard strokes for one character given from the stroke recognition device 4 described above. This character recognition device 5 has kanji.

A character dictionary 51 is attached that stores data representing each character such as kana, numeral, alpha bent, etc. as a combination of the above-mentioned standard strokes. Each time the aforementioned stroke identification device 4 provides a stroke identification result for one character, the character identification device 5 compares this with the data of stroke combinations of each character stored in the character dictionary 5. , characters input by handwriting from the handwritten character input device 1 are recognized as specific characters.

The method of the present invention is carried out by the apparatus configured as described above, and the operation of detecting a stationary point by the stationary point detection device 2 will be explained in more detail according to the flowchart of FIG.

The stationary point detection device 2 compares each data value of time-series coordinate data of handwritten characters sequentially inputted from the handwritten character input device 1 with the previously inputted data value, and if the values are the same, the number of times of overlap N is determined. 1 and returns to the step of reading the data to read the next data. If the values are not the same, it is determined whether the number of duplications N is 4 or more. If the number of overlaps N is 4 or more (N22), the coordinate value is temporarily stored in the attached memory as a stopping point, the number of overlaps N is reset to 0 (N=O), and the next data is Return to the loading step. Also, the number of duplicates N is not 4 or more (N<4
), the number of repetitions N is reset to 0 (N=0) and the process returns to the step of reading the next data.

Now, the stopping point is basically detected by the process described above, but in the case of handwritten character input, at a bending point in the middle of a stroke, there is only one stopping point where the movement of the input pen stops. This is rather rare, and usually multiple stationary points appear in an extremely narrow range. Therefore, in such cases, the stationary point detection device 2 uses the following principles to detect multiple stationary points once stored in memory. One stationary point is selected from among the stationary points and set as a feature point.

For example, if we consider as a specific example the stroke "K" which has a bending point in the middle as shown in Figure 4 fat, first of all,
shows a case where four stationary points are detected near the bending point in the middle of the stroke. That is, four stationary points, R, C, and D, are detected within a predetermined distance from each bending point of the "U"-shaped stroke, and the number of repetitions AN,
8N, CN, DN is AN, 4. BN Tomi 5. CN=
10°ON/6. In such a case, the stationary point detection device 2 selects the stationary point C (CN・10) with the maximum number of repetitions N as a feature point representing this bending point, and sends its coordinate values to the stroke feature amount calculation device 3. Output.

Figure 4 (C) shows the five stationary points A in the same way as in the above case.
- When E is detected, the number of times CN of overlaps between point 0 and point D,
This shows the case where the DNs are both 8 and the same number (CN=8.DN・8). In such a case, the stationary point detection device 2 selects, for example, fl) the stationary point that appears first in time, (2) selects the stationary point that appears last in time, or (3) selects the stationary point that appears last in time. One point is selected as a feature point by any method such as newly setting the intermediate point as a stationary point, and its coordinate value is output to the stroke feature amount calculation device 3.

FIG. 4 fd) shows a case where point C with the same coordinate values is detected twice as a stationary point. That is, at the bending point of the stroke, the time series coordinate data is point A (the number of overlaps AN -4
), move to point B (number of overlaps BN -5), point C (number of overlaps CN, -8), then move once to point C', return to point C again (number of overlaps CN2-3), and then move to point C (number of overlaps CN, -8). When moving in the order of point D (number of overlaps oN -8) and point E (number of overlaps EN -5), the stationary point detection device 2 detects the first number of duplicates CN and the second number of duplicates at the stationary point C. CN2
Let the number of overlaps CNo (-11) obtained by adding the above to be the number of overlaps CN of that point C. Thereafter, the stationary point detection device 2 outputs the coordinate values of the point with the maximum number of repetitions N as a feature point representing this bending point to the stroke feature amount calculation device 3, as in the case of FIG. 4 tc+ described above. .

Therefore, in the example of Fig. 4 fdl, the first number of repetitions CNI of point C and the number of repetitions []N of points are both maximum 8, but the total number of repetitions of the first and second times of point C is 8, which is the maximum. Since the number of times CNo is 11, this point C is selected as a stationary point of this bending point, that is, a feature point representing this bending point, and its coordinate value is output to the stroke feature amount calculation device 3.

Next, detection of the start and end points by the stationary point detection device 2 will be explained.

When inputting handwritten characters, the input pen generally stops moving at the start and end points of each stroke.
Alternatively, since the moving speed becomes somewhat slow, a stopping point will naturally occur if the method of the present invention is applied. Therefore, as shown in Fig. 5(a) and crest), the start point SP and end point EP are detected near the start and end points of the stroke. Needless to say, stopping points A, R, C, E, F, etc. may also be detected within a predetermined distance from the starting point sp and the ending point HP. In such a case, the stationary point detection device 2 does not use the starting point sp and the ending point EP as feature points, but uses the stationary points A, I't.

C and ri, l! , F by the above-described process, and output their coordinate values to the stroke characteristic amount calculation device 3 as feature points representing the start point and end point, respectively.

On the other hand, as shown in FIG. 5 tCi, the starting point SP and the ending point EP
(Figure 5 (C1 shows only the end point EP) If there is no stationary point within a predetermined distance nearby, the stationary point detection device 2 detects the terminal point EP as it is as a feature point representing the terminal point, and its coordinate value is output to the stroke characteristic @ amount calculation device 3.

Furthermore, as shown in FIG. 5(d), if the end portion of the stroke ends with a "spring", the starting portion of the "splash" is a bending point in the middle of the stroke, so a stopping point is detected. However, no stationary point appears at the end of the "splash". Therefore, the end point of "splash", +! 11) The end point of the stroke [There is no stationary point within the predetermined distance of P, and the end point of the "spring" is detected as a feature point representing the end point EP of the stroke, and the "spring" itself also moves from the bending point to the end point. Since it is detected as a part of a stroke having a length between EPs, more accurate stroke recognition is possible.

FIG. 6 is a block diagram showing a second configuration example of an apparatus used to implement the method of the present invention.

In this second device configuration example, similarly to the first device configuration side described above, time-series coordinate data of characters handwritten as a trajectory of an input pen on the input surface of the handwritten character input device 1 is transferred to the stationary point detection device 2. , and the characteristic points of each stroke, that is, the starting point, ending point, and stopping point, are detected. In this device configuration example, the character recognition device 6 compares a part of the coordinate string of the start/end point and the stop point of each stroke for one character detected by the stationary point detection device 2 with a standard character coordinate pattern. , is configured to identify handwritten characters input into the handwritten character input device 1.

FIG. 7 is a block diagram showing a third configuration example of an apparatus used to implement the method of the present invention.

In this third device configuration example, as in the first and second device configurations described above, the time-series coordinate data of the characters handwritten as the trajectory of the input pen on the input surface of the handwritten character input device vIll is detected as a stationary point. The feature points of each stroke, that is, the start point, end point, and stationary point, are detected. In this device configuration example, the time-series coordinate data output from the handwritten character input device I includes, in addition to the stationary point detection device 2 described above,
It is also applied to the feature point detection device 7, which detects points equidistantly spaced from each stroke as feature points, as described in the [Junto Technology] section. Then, the coordinate values of the start and end points and the stop points of each stroke detected by the stationary point detection device 2 and the coordinate values of the feature points of each stroke detected by the feature point detection device 7 are determined by the feature point correction device 8. given to.

FIG. 8 is an explanatory diagram of the correction processing by this feature point correction device 8. Now, for example, the stroke "wa" (for example, part of the kanji "ka 1゜", katakana "power", etc.) has been handwritten input into the handwritten character input device 1! In I'3'::, the time-series coordinate data output from the handwritten character input device 1 is in a state as shown in FIG. 8 (al).

The time-series coordinate data output from the handwritten character input device 1 is processed by the stationary point detection device 2, and the detection results of the feature points, that is, the start and end points and stationary points are shown in FIG. The starting and ending points and the bending point are detected as feature points, respectively. .

On the other hand, the time-series coordinate data output from the handwritten character input device 1 is divided at equal distance intervals by the feature point correction device 8 to detect feature points, and the results are as shown in FIG.
The bending point of the stroke is not detected, but instead the curvature of the stroke is represented.

Since the feature point correction device 8 is given both of the above-mentioned data, the feature point correction device 8, as shown in FIG. 8+dl,
Both data are compared, and among the feature points detected by the feature point detection device 7, the feature point closest to each stationary point detected by the stationary point detection device 2 is replaced with the coordinates of each stationary point. Supplementary IF by such feature point correction device 8
This makes up for the drawback that it is difficult to accurately detect bending points in the method commonly used in eastern Japan, which detects positions at equidistant intervals of each stroke as feature points.

Note that the final recognition of handwritten input characters in this third device configuration example is performed by using the same character recognition device 6 as in the second device configuration example described above, by converting the correction result by the feature point correction device 8 into a standard character coordinate pattern. The structure is such that it is compared and identified as a specific character.

〔effect〕

As detailed above, according to the present invention, the same coordinate value continues for a predetermined number of times or more in the time-series coordinate data of a character handwritten as a trajectory of an input pen on the input surface of a handwritten character input device. The repeated coordinate values are detected as the stationary points of the trajectory of the input pen, and each stroke of the input character is captured as a feature point sequence consisting of this stationary point and the start and end points of the stroke, and based on this, the handwritten input is Therefore, when implementing the method of the present invention, the bending point of each stroke of a character input by hand can be recognized with a relatively simple configuration and with a relatively small amount of calculation. This enables reliable detection and highly accurate handwritten character recognition.

[Brief explanation of drawings]

The drawings show embodiments of the present invention, and FIG. 1 is a schematic diagram for explaining the invention in detail, FIG. 2 is a block diagram of an apparatus used to carry out the invention, and FIG. 3 is a stationary Flowchart showing data processing contents of the point detection device, No. 4.5
The figure is an explanatory diagram of the data processing contents of the stationary point detection device, Section 6.
FIG. The block diagram of the third device, FIG. 8, is a schematic diagram for explaining its operation. 1... Handwritten character input device 2... Stationary point detection device 3... Stroke feature amount calculation device 4...
Stroke identification device 5...Character input device patent
Applicant Sanyo Electric Co., Ltd. Representative Patent Attorney Kawano Bunfu Figure 4 @Makinashi-11”) 1゜ 1st Δ < a) P (C) X 5 concave (b (d) 6th Bad (C) Mushroom 8 Figure

Claims (1)

[Scope of Claims] 1. A handwritten character recognition method that recognizes as a character the trajectory of an input pen read as time-series coordinate data representing a position on a character input surface of a character input device, comprising: Detecting a starting point, an ending point, and a stationary point of the input pen, which is detected as a coordinate position where the time-series coordinate data repeats the same value a predetermined number or more consecutively, of each stroke constituting the trajectory; Each point is considered a feature point, and the handwritten input is performed as the trajectory of the input pen based on the coordinate values of a feature point sequence made up of these feature points or the characteristics of each stroke determined from the positional relationship between each feature point. A character recognition method characterized by recognizing characters.