JPH0652363A - On-line handwritten character recognizing method - Google Patents

Abstract

PURPOSE:To improve a recognition rate by analyzing the refusal cause of a recognition refusal character, and performing reprocessing considering character size, stroke size, and distance between strokes. CONSTITUTION:A character is inputted by moving an electronic pen 2 on a tablet 1. An on-line character data input part 3 converts and accepts an input character as a time series data. Character structure pre-processed by a pre- processing part 4, after being analyzed by a structure analysis part 5, is matched with the model of a dictionary 7, and recognition processing is applied to it. A reprocessing part 8 performs the reprocessing on character data for which recognition is refused at a recognition part 6 by using two kinds of noise eliminating methods and three kinds of stroke connecting methods, and the data is delivered to the structure analysis part 5.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an on-line handwritten character recognition method for analyzing the cause of rejection of unrecognizable characters and performing optimum reprocessing.

[0002]

2. Description of the Related Art An on-line handwritten character recognition device recognizes characters in real time by directly using the moving direction information of an electronic pen on a tablet.
In addition, as the recognition method, a method of generating a direction code string of each stroke of the input character and examining its characteristics,
A method is known in which the coordinate position of each stroke, that is, the end point, the inflection point, and the like are characterized. Note that, as a related prior art of this kind, for example, IEEE TRANSACT
IONS ON PATTERN ANALYSIS
AND MACHINE INTELLIGENCE.
VOL. 12, NO. 8 August 1990 p
p787-803 and the like.

[0003]

In the above-described online handwritten character recognition apparatus, character recognition is performed by causing a break (stroke break) in the handwritten character data due to slipping of the electronic pen, irregular movement of the electronic pen, or the like. May be rejected. Conventionally, regarding such unrecognizable character data, using spatial information and temporal information,
The recognition rate was improved by performing various reprocessing. For example, when the distance between two points before and after is less than a threshold value (10 pixels), there is a recognition method in which the recognition rate is improved by connecting these two points with a stroke (Pattern).
Recognition, Vol. 16, No. 5,
pp 447-458, 1983).

However, in the above-mentioned conventional method,
There is a problem that noise is taken in and recognized, or short strokes that should not be removed as noise are removed and recognized because the relativity between the stroke length and the character size is not considered. .

An object of the present invention is to analyze the cause of rejection of a recognition-rejected character and perform online multi-step reprocessing in consideration of the character size, stroke size, and distance between strokes to improve the recognition rate of online handwritten character recognition. To provide a method.

[0006]

In order to achieve the above object, according to the invention of claim 1, in a method of reprocessing a recognition-rejected character in an online handwritten character recognition apparatus, a cause of rejection of character data that has been rejected is recognized. By analyzing the cause of refusal and the table in which the reprocessing procedure already applied to the refusal data is written, it is determined whether or not reprocessing is performed, and when reprocessing is performed, registration is performed. It is characterized in that the plurality of applied reprocessing methods are applied to perform the reprocessing, and the applied reprocessing methods are written in the table.

According to a second aspect of the present invention, a stroke size is one or a plurality of strokes forming a character, and a length of a diagonal line of a rectangle surrounding the stroke is a stroke size, and a length of a diagonal line of a rectangle surrounding the character is a character size. In the reprocessing method, when the ratio between the first stroke size and the size of the character to which the first stroke belongs is smaller than a predetermined threshold value, the first stroke is determined to be noise and the first stroke is determined to be noise. It is characterized in that it is a process of removing the first stroke from the character to which one stroke belongs.

According to the third aspect of the invention, the character is a character having two or more strokes, and a rectangle enclosing the second stroke and a third character.
When there is no overlap with the rectangle surrounding the stroke of,
The shortest distance between the point sequence of the second stroke and the point sequence of the third stroke is defined as the distance between the second and third strokes, and the reprocessing method is the stroke size of the fourth stroke. When the ratio of the distance between strokes is smaller than a predetermined threshold value and the ratio of the distance between strokes and the character size is larger than a predetermined threshold value, the fourth stroke is determined to be noise and the character to which the fourth stroke belongs. Is a process for removing the fourth stroke from.

According to the fourth aspect of the invention, the last two points of the fifth stroke are defined as p1 and p2, the first two points of the subsequent sixth stroke are defined as p3 and p4, and they are formed by points p1 and p2. And the absolute value of the angular difference between the vector formed by the points p3 and p4 is defined as the pen-up / pen-down angular difference between the fifth stroke and the sixth stroke.
The absolute value of the angular difference between the vector formed by 2 and the vector formed by the points p2, p3 is the connection angle difference between the fifth stroke and the sixth stroke, and the distance between the points p2 and p3 is the fifth stroke. And the distance between the stroke and the character size is smaller than a first threshold, and the angle difference is smaller than a second threshold. When the connection angle difference is smaller than a third threshold value, the process is a process of connecting the two strokes.

According to the fifth aspect of the present invention, in the reprocessing method, the angular difference between two strokes having a front-rear relationship is smaller than a fourth threshold value set to a value equal to or less than the second threshold value, and 2 When the ratio between the distance of one stroke and the character size is less than or equal to a fifth threshold value set to a value equal to or greater than the first threshold value, the process is a process of connecting the two strokes.

According to a sixth aspect of the present invention, in the reprocessing method, the ratio between the distance between two strokes and the character size which are in the context is smaller than the sixth threshold value set to a value equal to or less than the first threshold value. And, when the angle difference is less than or equal to the seventh threshold value set to a value equal to or greater than the second threshold value, the process is a process of connecting the two strokes.

[0012]

[Operation] Characters are input by moving the electronic pen on the tablet. The online character data input unit converts an input character into time series data of coordinate values and takes in the data. The structure of the input character that has undergone smoothing processing and serif removal by the preprocessing unit is analyzed by the structure analysis unit, then input to the recognition unit, and pattern matching is performed with the character pattern model stored in the dictionary for recognition. It is processed. The character data whose recognition is rejected by the recognition unit is reprocessed by the reprocessing unit using two types of noise removal methods and three types of stroke connection methods, and the reprocessed data is passed again to the structure analysis unit. After that, recognition processing is performed and the recognition result is output. This makes it possible to improve the recognition rate of online handwritten characters.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be specifically described below with reference to the drawings. FIG. 1 is a block configuration diagram according to an embodiment of the present invention. In the figure, reference numeral 1 is a coordinate input unit such as a tablet, which is formed integrally with a transparent display board. Two
Is an electronic pen such as a stylus pen. An online character data input unit 3 converts an input character written on the tablet with a pen into time series data (including spatial information and temporal information) of coordinate values and takes in the converted data.

Characters are input by moving the electronic pen 2 while contacting the surface of the tablet 1. The online character data input unit 3 samples the string of coordinate points from the tablet 1 at a constant speed and converts the character shape into a data string of a finite number of points. Further, the pen 2 being lowered onto the surface of the tablet 1 and first contacting with it is called pen down, and the pen 2 being raised from the contact state and separated from the surface of the tablet 1 is called pen up. And the data of the point sequence sampled from pen down to pen up is called stroke, and the character is
It consists of one or more strokes.

Reference numeral 4 denotes a preprocessing unit for performing processing such as serif removal and smoothing on the input raw data of handwritten characters, and 5 is a structure analysis unit for analyzing the structure of the preprocessed handwritten characters. 6 is the structure of the analyzed handwritten character and the dictionary 7
Of the present invention, which will be described later, only for the data for which the recognition is rejected, 7 is a dictionary in which a handwritten character is recognized by performing pattern matching with the model in FIG. A reprocessing unit that reprocesses the processed data and passes the processed data to the structure analysis unit 5 or the result output unit 9.
A result output unit 9 outputs a recognition result to a screen such as a display or a file.

For the handwritten character data that cannot be recognized by the recognition unit 6, the reprocessing unit 8 of the present invention comprehensively manages the analysis of the cause of rejection and the determination of the reprocessing procedure. As the procedure, two types of noise removal methods and three types of stroke connection methods are applied.

FIG. 2 is a diagram for explaining the first noise removal method of this embodiment, which utilizes the relative relationship between the character size and the stroke size. Here, the stroke size is
A rectangle that exactly encloses a stroke (hereinafter referred to as a circumscribed rectangle)
Is the length of the diagonal line. The character size is the length of the diagonal line of the circumscribed rectangle of the character.

In the first noise removal method, the ratio (stk_) of the size of a stroke (stk_siz) to the size of the character (ch_siz) to which the stroke belongs.
If ch_rt = stk_siz / ch_siz is smaller than a predetermined threshold value (thr_v1 = 0.055) (s
tk_ch_rt <thr_v1), consider the stroke as noise and remove it from the character.

FIG. 3 is a diagram for explaining a second noise removing method of this embodiment, which utilizes the character size, the stroke size, and the distance between strokes. In a character with two or more strokes, the distance (stk_dis) between one stroke (stk) and another stroke (rest_stks)
Is 0 when there is an overlapping portion of the circumscribing rectangle of stk and the circumscribing rectangle of rest_stks, and is the shortest distance between the point sequence of stk and the point sequence of rest_stks when there is no overlapping portion.

In the second noise removal method, for a certain stroke, the ratio of the size of the stroke and the distance between the strokes (stk_dis_rt = stk_siz / st).
k_dis) is a predetermined threshold value (thr_v2 = 0.05)
5) smaller than (stk_dis_rt <thr_v
2) and the ratio of the distance between strokes and the character size (d
is_dim_rt = stk_dis / ch_siz)
Is larger than a predetermined threshold value (thr_v3 = 0.700) (dis_dim_rt> thr_v3), the stroke is regarded as noise and removed from the character.

FIG. 4 is a diagram for explaining the first stroke connection method of the present embodiment using the online input information of characters, the character size, the distance between strokes, and the angle between strokes. Online input information of characters is the pen-down and pen-up information when writing a stroke as described above (that is, the first data in the time series of stroke coordinates indicates the pen-down location, and the last data indicates the pen-up location. (Shown) and stroke order information (the order before and after each stroke) when writing a character.

The last two points of a certain stroke (stk1) are designated as p1 and p2, and the first two points of the subsequent stroke (stk2) are designated as p3 and p4. The pen-up / pen-down angle difference (ang1) between stk1 and stk2 is the absolute value of the angle difference between the vector (p1, p2) and the vector (p3, p4). Further, the connection angle difference between stk1 and stk2 means the vector (p1, p2) and the vector (p2,
It is the absolute value of the angle difference of p3). The distance (stk_dis_1_2) between stk1 and stk2 is the distance between p2 and p3.

In the first stroke connection method of the present embodiment, the ratio of the distance between two strokes and the character size having a front-back relationship (dis_1_2_dim_rt = stk_d).
is_1_2 / ch_siz is a predetermined threshold value (thr_
v4 = 0.2), ang1 is smaller than a predetermined threshold (thr_v5 = 60 degrees), and ang2 is smaller than a predetermined threshold (thr_v6 = 60 degrees),
Connect these two strokes.

FIG. 5 is a diagram for explaining a second stroke connecting method of the present embodiment, which connects strokes in substantially the same traveling direction. Strokes stk1 and s that are related to each other
ang1 of tk2 is a predetermined threshold value (thr_v7 = 15)
2) and dis_1_2_dim_rt is less than or equal to a predetermined threshold (thr_v8 = 0.4), these two strokes are connected. However, the difference between the second stroke connection method and the first stroke connection method is that
stricter conditions for ng1 (thr_v7 <thr_v
5), the condition of dis_1_2_dim_rt is relaxed (thr_v8> thr_v4).

FIG. 6 is a diagram for explaining a third stroke connecting method of the present embodiment for connecting strokes at extremely short distances. Strokes stk1 and stk2 that are related to each other
Dis_1_2_dim_rt is a predetermined threshold value (thr
If less than _v9 = 0.1) and ang1 is less than or equal to a predetermined threshold value (thr_v10 = 90 degrees), these two strokes are connected. However, the difference between the third stroke connection method and the first stroke connection method is dis_
Strict conditions for 1_2_dim_rt (thr_v9
<Thr_v4), the condition of ang1 is relaxed (thr_v4)
_V10> thr_v5).

FIG. 7 shows the configuration of the reprocessing unit 8 of this embodiment, which analyzes the cause of rejection of reject data, determines a multi-step reprocessing procedure, and performs optimum reprocessing. In FIG. 7, the refusal data storage buffer array 26 stores the recognition refusal data before reprocessing. In the reprocessing method table 24,
Various methods are registered for reprocessing. As the reprocessing method, there are the above-described first and second noise removal methods and the first to third stroke connection methods, and other methods can be added. And the reprocessing procedure is
It shows in what order some or all of these reprocessing techniques are applied to the reject data.

The reprocessing history table 23 records the number of the reject data storage buffer array 26 and the reprocessing procedure. The refusal cause analysis unit 21 examines the refusal data for the matching status with the dictionary model, the noise status, and the stroke connection status, and analyzes the cause of the refusal. The reprocessing procedure determination unit 22 refers to the cause of refusal and the reprocessing history table 23 to determine whether or not to reprocess. In the case of reprocessing, the reprocessing procedure is determined, the rejection data is written in the rejection data storage buffer array 26, and the number and the reprocessing procedure are recorded in the reprocessing history table 23. The reprocessing execution unit 25 follows the reprocessing procedure
The refusal data is reprocessed, and the result is the structural analysis unit 5 of FIG.
Hand over.

For example, in the reprocessing method table 24, the first,
It is assumed that the second noise removal method (NE1, NE2) and the first to third stroke connection methods (BC1, BC2, BC3) are registered and the rejection data (ch_rj_0) is input. At this point, nothing is written in the reprocessing history table 23.

The reprocessing procedure determination unit 22 determines a reprocessing method for the rejection data (ch_rj_0) whose rejection cause is analyzed. For example, it is determined to apply in the order of NE1 registered in the table 24 and then BC1. The refusal data (ch_rj_0) is recorded at 0 in the refusal data storage buffer array 26.

The reprocessing execution unit 25 uses the rejection data (ch_
For rj_0), NE1 is applied to remove noise, then BC1 is applied to connect strokes, data is passed to the structural analysis unit, and in the reprocessing history table 23, the reject data storage buffer array 26 No. 0 data is processed in the procedure of NE1-> BC1.
0; NE1-> BC1 is recorded.

Even in this reprocessing, when the recognition is rejected, the reprocessing procedure determination unit 22 refers to the analysis result of the rejection cause and the reprocessing history table 23 (NE1-> B of the 0th data).
(See C1), for example, the procedure of reprocessing is NE
2-> BC3 is determined and reprocessing is executed. Further, in the reprocessing history table 23, 0; NE1-> BC1, NE2-
> BC3 is recorded. Then, the recognition process is performed under different conditions.

[0032]

As described above, according to the invention described in claim 1, with respect to the recognition refusal data of the online handwritten character, the cause of the recognition refusal is analyzed by using the time information and the spatial information, and the optimum decision is made. Decide the reprocessing procedure to remove noise,
Since reprocessing such as stroke connection is performed, the recognition rate of online handwritten characters can be improved.

According to the second aspect of the invention, since the noise removing method using the relative relationship between the character size and the stroke size is adopted, optimum reprocessing can be performed and the recognition rate can be improved.

According to the third aspect of the present invention, since the noise removing method considering the relative relationship between the character size and the stroke size and the distance between the strokes is adopted, it is possible to further improve the recognition rate. Become.

According to the invention of claim 4, since it is a stroke connecting method using online input character information, character size, distance between strokes, and angle between strokes, broken strokes are connected and reprocessed. The character after the stroke is connected is recognized, and as a result, the recognition rate is improved.

According to the invention described in claim 5, since the condition of the connection angle is made stricter than that of the method described in claim 4, and the broken strokes in substantially the same traveling direction are connected, the recognition after the reprocessing is performed. The rate is further improved.

According to the invention of claim 6, the distance condition between strokes is made stricter than that of the method of claim 4,
Since the strokes that are extremely close to each other are connected, the recognition rate after reprocessing is further improved.

[Brief description of drawings]

FIG. 1 is a block diagram according to an embodiment of the present invention.

FIG. 2 is a diagram illustrating a first noise removal method of this embodiment.

FIG. 3 is a diagram illustrating a second noise removal method of the present embodiment.

FIG. 4 is a diagram illustrating a first stroke connection method of the present embodiment.

FIG. 5 is a diagram illustrating a second stroke connection method of the present embodiment.

FIG. 6 is a diagram illustrating a third stroke connection method of the present embodiment.

FIG. 7 is a diagram illustrating a configuration of a reprocessing unit according to the present exemplary embodiment.

[Explanation of symbols]

 1 tablet 2 electronic pen 3 online character data input unit 4 preprocessing unit 5 structural analysis unit 6 recognition unit 7 dictionary 8 reprocessing unit 9 result output unit

Claims (6)

[Claims] 1. A method of reprocessing a recognition rejected character in an online handwritten character recognition device, analyzing the cause of rejection of character data of which recognition is rejected, and the cause of rejection and the reprocessing already applied to the rejected data. Refer to the table in which the procedure is written and decide whether or not to reprocess, and when reprocessing, apply the registered reprocessing methods, execute the reprocessing, and apply. An online handwritten character recognition method, characterized in that the reprocessing method is written in the table. 2. The reprocessing, comprising one or a plurality of strokes forming a character, wherein a length of a diagonal line of a rectangle surrounding the stroke is a stroke size, and a length of a diagonal line of a rectangle surrounding the character is a character size. In the method, when the ratio of the first stroke size to the size of the character to which the first stroke belongs is smaller than a predetermined threshold, the first stroke is determined to be noise, and the character to which the first stroke belongs. The online handwritten character recognition method according to claim 1, wherein the first stroke is removed from the first stroke. 3. A character string having two or more strokes, and when there is no overlapping portion of a rectangle surrounding the second stroke and a rectangle surrounding the third stroke, the point sequence of the second stroke and the third stroke. The shortest distance between the point sequences is set as the distance between the second and third strokes, and in the reprocessing method, the ratio of the stroke size of the fourth stroke to the stroke distance is smaller than a predetermined threshold value, and When the ratio of the distance between strokes and the character size is larger than a predetermined threshold value, the fourth stroke is determined to be noise, and the fourth stroke is removed from the character to which the fourth stroke belongs. The online handwritten character recognition method according to claim 1. 4. The last two points of the fifth stroke are p1,
p2, and the first two points of the subsequent sixth stroke are p
3, p4, the absolute value of the angular difference between the vector formed by the points p1 and p2 and the vector formed by the points p3 and p4 is
The pen-up / pen-down angle difference between the fifth stroke and the sixth stroke is defined as the absolute value of the angle difference between the vector formed by the points p1 and p2 and the vector formed by the points p2 and p3. The connection angle difference with 6 strokes, the distance between points p2 and p3 is the distance between the 5th stroke and the 6th stroke, and in the reprocessing method, the distance between the two strokes in the front-back relationship and the character size ratio are First
2. When the angle difference is smaller than the threshold value, the angle difference is smaller than the second threshold value, and the connection angle difference is smaller than the third threshold value, the process is a process of connecting the two strokes. Online handwriting recognition method. 5. The reprocessing method is such that an angular difference between two strokes having a front-back relationship is smaller than a fourth threshold value set to a value equal to or less than the second threshold value, and a distance and a character between the two strokes. The online handwritten character recognition method according to claim 4, wherein when the size ratio is less than or equal to a fifth threshold value set to a value greater than or equal to the first threshold value, the process is a process of connecting the two strokes. . 6. The reprocessing method, wherein the ratio of the distance and the character size of two strokes having a front-rear relationship is smaller than a sixth threshold value set to a value equal to or less than the first threshold value, and an angle difference is 5. The online handwritten character recognition method according to claim 4, wherein when the value is equal to or less than a seventh threshold value set to a value equal to or higher than the second threshold value, the process is a process of connecting the two strokes.