JP2664624B2 - Feature point detection method for handwritten character recognition - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for inputting handwritten characters. For example, the present invention relates to a method for recognizing a handwritten character input by a handwritten character input device used as a character input device such as a Japanese word processor. 2. Description of the Related Art As an input device of a word processor, instead of a keyboard requiring a certain degree of skill in its operation, for example, a character input by handwriting input on a tablet is recognized and the input of the character is recognized. Handwritten character input devices configured to accept are developed and put into practical use. [0003] This type of device represents the characteristics of strokes based on time-series coordinate data of each stroke of a handwritten character on the tablet as a locus of an input pen on a tablet for character input. Coordinate values are selectively detected to be feature points, each stroke is point-approximated by these feature point sequences, and the stroke is obtained from the coordinate values of the feature point sequences approximated to each stroke or the positional relationship of the feature point sequences. Characters input by handwriting with the input pen are recognized based on the characteristics of the stroke. The following two methods have been mainly employed as methods for detecting the characteristic points of each stroke described above. First, each point obtained by sequentially dividing the input time-series coordinate data at equal distance intervals (equal interval detection points)
And character recognition using the start point and end point of the stroke as feature points. [0005] Second, a point at which the time-series moving direction of the coordinate value changes is detected as a pole from the input time-series coordinate data, and character recognition is performed by using this as a characteristic point using the start point and end point of the stroke. Is what you do. However, in the above-described first method, although the process of detecting the feature point of the stroke is relatively simple, the possibility that the detected feature point and the bending point of the stroke match is small, and therefore, the stroke in the middle of the stroke. It is difficult to accurately detect a bending point, and therefore, it is not possible to accurately detect the shape of a stroke having a bending point. In the second method, it is necessary to calculate the time-series moving direction of the coordinate value from the time-series coordinate data of the read stroke each time. For this reason, the amount of data processing required for the input processing becomes excessive, and it takes time for the input processing, and a data processing device having an unnecessarily large capacity is required. In addition, this method is susceptible to minute changes in the moving direction of the coordinate values and the like, and there is a possibility that many unnecessary feature points will be detected. SUMMARY OF THE INVENTION The present invention is directed to character recognition.
The purpose is to enhance the performance. By the way, when inputting handwritten characters, the input pen stops at the bending point of the stroke. In other words, when the trajectory of the input pen moving on the input surface of the handwritten character input device is detected as time-series coordinate data, the same coordinate value is detected continuously for a certain time or more. By paying attention to this, the stationary point of the input pen is used as a feature point, so that the inflection point of each stroke can be accurately detected with a relatively small amount of computation, and the handwritten character input character can be recognized with high accuracy. Handwritten character recognition methods are conceivable. The present application is thus a stop point
The purpose is to detect By the way, simply,
The more points, the better the recognition accuracy. However, the feature point
If the number is large, the amount of calculation for recognition increases. In addition,
Character recognition accuracy is low when necessary points are adopted as feature points.
I will drop it. This application adopts stop points as feature points
At the same time, in order to improve the character recognition accuracy,
The purpose is to detect a mark. According to the present invention, a starting point of each stroke constituting a locus of an input pen read as time-series coordinate data representing a position on a character input surface of a character input device is provided. An end point and a stop point of the input pen which are detected as coordinate positions where the time-series coordinate data repeats the same value continuously for a predetermined number of times or more are detected, and the start point, the end point, and the stop point are defined as feature points, and Based on the coordinate values of a feature point sequence composed of feature points or the features of each stroke obtained from the positional relationship between each feature point, a handwritten character recognition that recognizes a handwritten input character as a trajectory of an input pen In the feature point detection method, when the stop point exists at a position near the start point or the end point, the stop point among these points
Set the point of the point and start or end point, characterized in that it does not employ other points as feature points. According to the invention of the present application, when there are stationary points near the starting point or the ending point, these points are replaced by one.
One feature point is collected at the stop points. First, the meaning of detecting a stationary point in handwritten character recognition will be described. FIG. 1 shows, as an example, measured values of time-series coordinate data when a Chinese character “handling” is input by a tablet for inputting handwritten characters. The measurement data has a time resolution of 100 points / second and a spatial resolution of 10 points / mm. In the figure, each numeral indicates the number of times that the coordinate value of each position is continuously detected, and the part marked with a circle indicates the part where the number is four or more, and the part marked with a circle The number with an underline adjacent to the symbol indicates the number of times the same coordinate value is duplicated at the position marked with a circle. The following facts can be clearly read from the actual measurement data shown in FIG. That is, at the inflection point in the middle of the same stroke, the overlap of the same coordinate value shows a considerably high numerical value, and at least 4 or more overlaps occur at the start point and the end point of each stroke. . Therefore, if the position of the coordinate value having the number of overlaps of 4 or more is set as the feature point, not only the bending point of the stroke but also the start point and the end point can be accurately detected. Hereinafter, the present invention will be described with reference to the drawings showing embodiments thereof. FIG. 2 is a block diagram of a handwritten character recognition device used to carry out the present invention according to the above-described principle. In FIG. 2, reference numeral (1) denotes a handwritten character input device, and a trajectory of an input pen moving on the character input surface is detected as time-series data in an XY coordinate system set on the input surface. Then, the time-series coordinate data is given to the stop point detecting device (2). The stop point detecting device (2) processes the time-series coordinate data provided from the handwritten character input device (1) to detect the start point and the end point of each stroke and the stop points existing between them, and The coordinate values are output to the stroke feature value calculation device (3). The data processing operation of the stop point detection device (2) will be described later. Stroke feature amount calculation device
(3) uses the coordinate values of the start point, end point, and stop point of each stroke given from the stop point detection device (2) as feature points of each stroke, and calculates the feature of the stroke from the coordinate values of these feature point sequences. Calculate the amount. The features to be detected for each stroke include the angle and length (relative) of the line connecting the feature points, and the connection of the line connecting the feature points.
(Clockwise or counterclockwise, etc.). The features of each of the detected strokes are provided to a stroke identification device (4). The stroke identification device (4) is a stroke feature amount calculation device described above.
Based on the characteristics of each stroke given from (3), that stroke is a specific standard stroke, for example, (a) (b) of FIG.
The characteristic point as shown in (c) is a stroke having only the start and end points, or a bending point as shown in (d), (e), (f) and (g) of FIG. 9 (the stop point between the start and end points in the method of the present invention). Is detected as a standard pattern of a stroke. The standard pattern of strokes is obtained by decomposing each character such as a kanji character, a kana, a number, an alphabet, etc. into each stroke in accordance with the stroke order when handwriting and writing, and decomposing each standard stroke as described above. Are stored in advance in a stroke dictionary (41) attached to the stroke identification device (4). And
The stroke identification device (4) is a stroke feature value calculation device
The feature of the stroke given from (3) is specified as one of the standard strokes by comparing with the feature of the standard stroke stored in the stroke dictionary (41),
For example, the code number of the standard stroke is output to the character identification device (5). The character identification device (5) is a handwritten character recognition device (1) based on a combination of code numbers of standard strokes for one character given from the stroke identification device (4).
Is a device for recognizing a character input by handwriting as a specific character. The character identification device (5) is provided with a character dictionary (51) storing data representing each character such as a kanji, a kana, a number, an alphabet, etc. as a combination of the above-described standard strokes. Each time a stroke result for one character is given from the stroke identification device (4), the character recognition device (5) sends the result to the character dictionary (51).
The character input by handwriting from the handwritten character input device (1) is recognized as a specific character by comparing with the data of the combination of strokes of each character stored in. The present embodiment is implemented by the apparatus configured as described above. The operation of detecting a stop point by the stop point detecting device (2) will be specifically described with reference to the flowchart of FIG. The stop point detection circuit (2) compares each data value of the time-series coordinate data of the handwritten character sequentially input from the handwritten character input device (1) with the previously input data value, and when the data values are the same, Returns to the step of reading the data by adding 1 to the number of duplications N, and reads the next data. If they are not the same value, it judges whether or not the number of duplications N is 4 or more. And the number of duplication N is 4 or more (N
In the case of ≧ 4), the coordinate value is temporarily stored in the attached memory as a stop point, and the number of duplications N is reset to 0 (N = 0). Then, the process returns to the step of reading the next data. If the number N of duplications is not 4 or more (N <4), the number N of duplications is set to 0 (N =
Reset to 0) and return to the step of reading the next data. The stop point is basically detected by the processing described above. However, in the case of handwritten character input, one stop point where the movement of the input pen stops at the inflection point in the middle of the stroke. Location is rather rare, and usually a plurality of stopping points appear in a very narrow range. Therefore, in such a case, the stop point detection device (2)
According to the following principle, one stop point is selected from a plurality of stop points once stored in the memory and set as a feature point. For example, assuming a stroke having an inflection point in the middle as shown in FIG.
(b) shows a case where four stationary points are detected near the bending point in the middle of the stroke. That is, four stop points A, B, C, and D are detected within a predetermined distance at the bending points of this stroke, and the respective numbers of overlaps AN, BN, CN, and DN are represented by A
N = 4, BN = 5, CN = 10, DN = 6. In such a case, the stop point detection device (2) determines the stop point C (CN = 10) where the number of overlaps N is the largest,
The inflection point is selected as a representative feature point, and its coordinate value is output to the stroke feature amount calculation device (3). FIG. 4 (c) shows the number of overlaps CN and D of the point C and the point D when five stop points A to E are detected in the same manner as described above.
The case where N is 8 and the same number (CN = 8, DN = 8) is shown. In such a case, the stop point detection device (2) selects, for example, (A) the stop point that appears first in time. (B)
Select the stop point that appears last in time. (C) One point is selected as a feature point by any method such as newly setting a middle point between the two as a stationary point, and the coordinate value is output to the stroke feature amount calculating device (3). FIG. 4D shows a case where the point C having the same coordinate value is detected twice as a stationary point. That is, at the bending point of the stroke, the time-series coordinate data is point A (duplication count AN =
From 4), move to point B (duplicate number BN = 5), point C (duplicate number CN1 = 8), then move to point C 'once, return to point C again (duplicate number CN2 = 3), Further, point D (duplication count DN = 8), point E (duplication count EN =
In the case of moving in the order of 5), the stop point detection device (2) performs the first overlap number CN1 and the second overlap number CN at the stop point C.
The number of duplications CN0 (= 11) obtained by adding 2 to the number of duplications C at that point C
N is assumed. Then, as in the case of FIG. 4 (c) described above, the stop point detection device (2) sets the point where the number of overlaps N is the maximum as a characteristic point representing this inflection point and uses the coordinate value as the stroke characteristic. Output to the quantity calculation device (3). Therefore, in the example of FIG. 4 (d), the first number of times C1 of the point C and the number of times DN of the point D are both 8 and the maximum, but the sum of the first and second times of the point C is Since the number of overlaps CN0 is 11, this point C is selected as a stationary point of this inflection point, that is, a feature point representing this inflection point, and its coordinate value is output to the stroke feature amount calculating device (3). . Next, detection of the start and end points by the stop point detection device (2) will be described. At the time of inputting handwritten characters, generally, at the start point and end point of each stroke, the movement of the input pen is temporarily stopped or the movement speed is slightly reduced.
A stop point naturally occurs at the start point and the end point. For this reason, for example, as shown in FIGS.5 (a) and 5 (b) (○ represents time-series coordinate data, and ● represents a stop point, respectively), a start point SP and an end point EP are detected near the start point and end point of the stroke. Of course, the stop points A, within a predetermined distance from the start point SP and the end point EP,
B, C and D, E, F, etc. may also be detected. In such a case, the stop point detecting device (2) does not adopt the start point SP and the end point EP as the feature points, and selects one point from each of the stop points A, B, C and D, E, F by the above-described processing. And outputs the coordinate values to the stroke feature amount calculating device (3) as feature points representing the start point and the end point, respectively. On the other hand, as shown in FIG.
If there is no stop point within a predetermined distance near the EP (only the end point EP is shown in FIG. 5 (c)), the stop point detection device (2)
P is detected as it is as a feature point representing the end point, and its coordinate value is output to the stroke feature amount calculating device (3). Furthermore,
As shown in FIG. 5 (d), when the end of the stroke ends with "splash", a stop point is detected because the start of "splash" is a bending point in the middle of the stroke. A stop point does not appear at the end of “Ne” and a stop point does not appear at the end of “Splash”. Therefore, there is no stop point within a predetermined distance of the end point of the "splash", that is, the end point EP of the stroke, and the end point of the "splash" is detected as a feature point representing the end point EP of the stroke. The “ne” itself is also detected as a part of the stroke having the length between the inflection point and the end point EP, so that it is possible to recognize the stroke more accurately. FIG. 6 is a block diagram showing a configuration example of the second embodiment of the present invention. In the second device configuration example, similarly to the first device configuration example, the time-series coordinate data of the character handwritten and input as the trajectory of the input pen on the input surface of the handwritten character input device (1) is detected at the stationary point. Applied to the device (2), characteristic points of each stroke, that is, a start point, an end point, and a stop point are detected. In this embodiment, the coordinate sequence of the start and end points and the stop point of each stroke for one character detected by the stop point detecting device (2) is collectively referred to as a character identification device (6).
By comparing with the standard character coordinate pattern,
The configuration is such that handwritten characters input to the handwritten character input device (1) are identified. FIG. 7 is a block diagram illustrating a configuration example of the third embodiment of the present application. In the third example of the device configuration, the time-series coordinates of the characters input by handwriting as the trajectory of the input pen on the input surface of the handwritten character input device (1) are the same as in the first and second examples of the device configuration. The data is supplied to the stop point detection device (2), and the characteristic points of each stroke, that is, the start point, the end point, and the stop point are detected.
In this embodiment, the time series coordinate data output from the handwritten character input device (1) is
In addition to (2), the present invention is also provided to a feature point detection device (7) for detecting points at equal distance intervals on each stroke as feature points described in the section of the related art. The start and end points of each stroke and the coordinate values of the stop point detected by the stop point detection device (2),
The coordinate value of each stroke detected by the feature point detection device (7) is provided to the feature point correction device (8). FIG. 8 is an explanation of the correction processing by the feature point correction device (8). Now, for example, when a stroke as shown in FIG. 10 (for example, a part of kanji `` power '', katakana `` ka '', etc.) is handwritten to the handwritten character input device (1), the handwritten character input device (1) The output time-series coordinate data is in a state as shown in FIG. The feature points obtained by processing the time-series coordinate data output from the handwritten character input device (1) by the stop point detection device (2), that is, the start and end points and the stop points are shown in FIG. However, ● indicates a feature point detected as a stop point, that is, a start and end point and a bending point, and ○ indicates a feature point which is a start and end point without a stop point, respectively.) It has been detected as a feature point. On the other hand, FIG. 8 (c) shows the result of dividing the time-series coordinate data output from the handwritten character input device (1) at equal distance intervals by the feature point correction device (8) and detecting feature points. As shown in FIG. 7, the curvature point of the stroke is shown instead of detecting the bending point of the stroke. Since both of the above data are given to the feature point correction circuit (8), the feature point correction device (8) compares the two data as shown in FIG. ), The feature points closest to each of the stop points detected by the stop point detection device (2) are replaced with the respective stop points. With such a feature point correction device (8), it is possible to compensate for the disadvantage that it is difficult to accurately detect a bending point in the method of detecting equidistant intervals on each stroke as a feature point conventionally performed in general. Becomes Note that the final recognition of the handwritten input characters in the third apparatus configuration example is performed by the same character identification apparatus (6) as in the above-described second apparatus configuration example, and by the feature point correction apparatus (8). The result is compared as a standard character pattern and identified as a specific character. The writing start time (starting point) and the writing end time
The coordinates at (end point) are generally unstable. Application
In, the stop point exists near the start point or end point.
If one of these stops, one of the above-mentioned stop points will be the starting point.
Or the end point. Therefore, according to the present application, the starting point
Or if the stop point is located near the end point,
Uses a stable stop instead of an unstable start (end)
The accuracy of each feature point is improved, and
Reduces the number of characters, which improves character recognition accuracy and
In addition, the load on the arithmetic processing is reduced.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic diagram for explaining the principle of detecting a stationary point. FIG. 2 is a block diagram of an apparatus according to a first embodiment of the present invention. FIG. 3 is a flowchart showing data processing contents of the stop point detection device. FIG. 4 is an explanatory diagram of data processing contents of the stop point detection device. FIG. 5 is an explanatory diagram of data processing contents of the stop point detection device. FIG. 6 is a block diagram of an apparatus according to a second embodiment of the present invention. FIG. 7 is a block diagram of an apparatus according to a third embodiment of the present invention. FIG. 8 is a schematic diagram for explaining the operation of the third embodiment. FIG. 9 is a diagram illustrating an example of a standard stroke. FIG. 10 is a diagram illustrating an example of an input stroke. [Description of Signs] (1) Handwritten character input device, (2) Stop point detection device, (3) Stroke feature amount calculation device, (4) Stroke identification device, (5) Character identification device.

Claims (1)

(57) [Claims] The start point and end point of each stroke constituting the trajectory of the input pen read as the time-series coordinate data representing the position on the character input surface of the character input device and the time-series coordinate data repeat the same value continuously for a predetermined number of times or more. The start point, the end point, and the stop point are detected as the feature points, and the coordinate values of the feature point sequence constituted by these feature points or the mutual Based on the characteristics of each stroke determined from the positional relationship, in a feature point detection method of handwritten character recognition for recognizing a character handwritten and input as a trajectory of an input pen, If the stationary point is present, one point of the stationary point in these points set as a start point or an end point, wherein the handwriting recognition, characterized in that does not employ other points as feature points Detection method.