JP3977473B2 - Handwritten character recognition method and handwritten character recognition apparatus - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a handwritten character recognition method and apparatus, and more particularly, to a handwritten character recognition method and a handwritten character recognition apparatus that achieve a simplified technique and an improved recognition rate.
[0002]
[Prior art]
Conventionally, as this type of technology, for example, focusing on the fact that the shape of each stroke of a handwritten character differs depending on the character to be recognized and can be classified into basic patterns, What is called stroke matching, in which characters are determined by comparing and determining the shape and the shape of each stroke of a handwritten character to be recognized, and in recent years, various methods that have been proposed in the past Various proposals have been made to introduce a so-called neural network concept to try to improve the recognition rate, and there are also various ones that have been put into practical use.
[0003]
[Problems to be solved by the invention]
Such conventional techniques related to handwritten character recognition, as a whole, are gradually improving the recognition rate with each age, but as the recognition rate is improved, the processing is generally easier. It tends to be complicated.
The present invention has been made in view of the above circumstances, and provides a handwritten character recognition method and a handwritten character recognition apparatus that enable recognition of handwritten characters at a high recognition rate with relatively simple processing.
Another object of the present invention is to provide a handwritten character recognition method and a handwritten character capable of recognizing handwritten characters at a high recognition rate by performing relatively simple identical processing without distinguishing hiragana, katakana, and kanji. To provide a recognition device.
Another object of the present invention is to provide a handwritten character recognizing method and a handwritten character recognizing apparatus capable of recognizing handwritten characters continuously written with relatively simple processing.
[0004]
[Means for Solving the Problems]
The handwritten character recognition method according to the invention of claim 1 is:
For handwritten characters, find the distance from the point at the beginning of writing at each appropriately selected point, and the tangent angle between the tangent at each point and a predetermined reference line,
For a plurality of standard characters to be compared with the handwritten character, the distance and tangent angle at each appropriate point of the standard character acquired in advance in the same manner as the handwritten character are compared with the distance and tangent angle of the handwritten character. Thus, a standard character having distance and tangent angle data that is most approximate to the change in tangent angle with respect to the distance of each point of the handwritten character is used as the recognition result of the handwritten character.
[0005]
This method pays attention to the fact that the distance at an appropriate point from the point at which the character starts writing and the pattern of the tangent angle formed by the tangent at each point and the predetermined reference line are specific to each character. By comparing such a pattern for handwritten characters with such a pattern for standard characters, it is possible to recognize what characters are handwritten characters at a very high recognition rate.
[0010]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to FIGS. 1 to 8.
The members and arrangements described below do not limit the present invention and can be variously modified within the scope of the gist of the present invention.
First, the configuration of a handwritten character recognition apparatus (hereinafter referred to as “the present apparatus”) according to an embodiment of the present invention will be described with reference to FIG.
The apparatus includes an input tablet 1 as an input unit, a digitizer (denoted as “DIG” in FIG. 1) 2, a data processing apparatus (denoted as “DPU” in FIG. 1) 3, and a display drive circuit (see FIG. In FIG. 1, a main component is a display 4 (indicated as “DIR”) 4 and a display device (indicated as “DISP” in FIG. 1) 5.
The input tablet 1 is for inputting handwritten characters with the input pen 1a, and the input pen 1a is moved by tracing the surface of the input window 1b into a desired character shape with the input pen 1a. The position information has a known and well-known configuration that is output as X, Y coordinate data. As an input tablet used for such handwritten character recognition, for example, there are various types such as an electromagnetic induction type and an electrostatic induction type, and any of these types may be used, and it is limited to a specific type. There is no need.
[0011]
The digitizer 2 converts the X and Y coordinate data in the analog signal format output from the input tablet 1 into a digital signal in a predetermined format. The converted signal is input to the data processing device 3 and will be described later. It is used for data processing for simple handwritten character recognition.
The data processing device 3 processes the data related to the handwritten character input via the digitizer 2 as described later, recognizes the handwritten character, and outputs the result to the display device 5 or an external device (not shown). Thus, for example, a so-called CPU is configured as a main component.
[0012]
The display device 5 is for displaying handwritten characters recognized by the data processing device 3, and is, for example, a known and well-known device using a CRT or a liquid crystal display element. A signal corresponding to the recognition result in the data processing device 3 is input to the display drive circuit 4, where the display device 5 performs processing necessary for display, so that the characters recognized by the display device 5 are displayed in real time. It has come to be.
[0013]
Next, the handwritten character recognition operation by this apparatus will be described with reference to FIGS.
When this apparatus is in an operating state and characters are handwritten on the input tablet 1 using the input pen 1a, the position information of the input pen 1a is input to the data processing apparatus 3 via the digitizer 2. (See step 100 in FIG. 2).
The data input from the digitizer 2 to the data processing device 3 is obtained by expressing position information corresponding to the stroke of the pen tip of the input pen 1a on the input tablet 1 that changes every moment in X and Y coordinates. The data acquired at a predetermined sampling interval by a so-called sampling operation by the digitizer 2.
In addition, information indicating what screen of the handwritten character corresponds to the captured x and y coordinate data is added from the information on the input pen 1a to the input tablet 1. Yes. Therefore, for example, as data at a certain sampling point, (x1, y1, n) and three data are represented as one set. Here, x1 represents the X coordinate value at the sampling point, y1 represents the Y coordinate value at the sampling point, and n represents the portion of the stroke of the handwritten character at the sampling point. That is, a numerical value representing the number of strokes.
[0014]
In this data input process, data is also interpolated between the breaks of each stroke.
That is, for example, if the hiragana “a” as shown in FIG. 3 is taken as an example, the first stroke and the second stroke between the first stroke portion and the second stroke portion. Since the input pen 1a is separated from the input tablet 1 between the end point of the first screen and the start point of the second screen, if the handwritten characters are not continuous characters during There is no so-called actual data input during this period.
In this apparatus, it is assumed that the end point of the first image and the start point of the second image are connected with a straight line, and a predetermined point (for example, in FIG. The x and y coordinate data of “5” along with the square are generated by so-called interpolation. Note that there is no data corresponding to the number of strokes in the normal data described above as to the interpolation data, and it is only possible to identify the interpolation data.
[0015]
In the data processing device 3, the following data conversion is performed on the basis of the x and y coordinate data and the stroke number data of the individual sampling points of the handwritten characters input as described above ( (See step 102 in FIG. 2).
That is, based on the x and y coordinate data, the distance s from the starting point of writing of the handwritten character at each sampling point is obtained, and the tangent angle θ at each sampling point is obtained.
The distance s is from the point at which the first handwritten character to be recognized is first written, for example, in the example of FIG. And the distance including the portion interpolated when the input pen 1a is separated from the input tablet 1 as described above. That is, for example, in FIG. 3, the distance s at the point where “7” is marked with a circle as a sampling point is obtained by connecting the point at the end of writing of the first image and the point of starting writing of the second image with a straight line. As a result, the distance from the starting point of the writing of “A” is calculated. Note that the sampling interval in FIG. 3 is larger than the actual one for easy understanding.
[0016]
The tangent angle θ at the sampling point is calculated as an angle from the X axis of the tangent represented by this equation by calculating a tangent equation from the x and y coordinates at that point.
In this data conversion, the information regarding the number of strokes is stored as it is. Therefore, if the input data at a certain sampling point is (x1, y1, n), this data is converted into (s1, θ1, n) by the data conversion as described above. )
[0017]
After all the input data related to the handwritten characters are converted as described above, normalization is performed with respect to the distance s (see step 104 in FIG. 2). That is, normalization is performed by dividing the value of s at each sampling point obtained by data conversion by the value of s at the end point of writing of the handwritten character. As a result, if the change characteristic of the tangent angle θ with respect to the distance s with respect to a certain handwritten character after normalization is graphed, for example, it becomes as shown in FIG. In this normalization process, the distance s between the individual sample points is a normalized value, but the data relating to the tangent angle and the number of strokes are held as they are.
[0018]
Subsequently, dictionary selection for selecting standard character data to be compared for determining handwritten characters is performed (see step 106 in FIG. 2).
That is, for standard characters such as hiragana, katakana, and kanji, the distance s, tangent angle θ, and stroke number data at a plurality of appropriate sample points examined in advance are stored in an unillustrated memory (or an external memory such as In other words, it is stored as a dictionary in a so-called hard disk storage device). Basically, all standard character data in the dictionary is collated with the normalized handwritten character data. However, in order to perform collation one by one, in this dictionary selection process, one standard character as described above that is a target to be collated in an appropriate order is selected. .
[0019]
When the standard character to be verified is selected, whether the stroke number data is the same, that is, whether the stroke number of the handwritten character and the stroke number of the standard character appropriately selected from the dictionary are the same If it is determined whether or not they are the same (in the case of YES), the alignment is performed without performing the stroke number matching process (see step 110 in FIG. 2) described later. On the other hand, when it is determined that the number of strokes is not the same (in the case of NO), the stroke number matching process described below is performed (see FIG. 2). Step 110).
[0020]
First, the alignment process is a process of aligning the position of each stroke of the standard character and the handwritten character with respect to the distance s when collating the handwritten character data with the standard character data selected by dictionary selection. is there.
For example, it is assumed that the graph of the change characteristic of the tangent angle θ with respect to the distance s of the selected standard character is as shown in FIG. In the figure, the position of distance s1 is the end position of the first stroke of this standard character, the position of distance s2 is the start position of the second stroke, and the position of distance s3 is the second position. It is assumed that the position of the stroke is the end position, the position of the distance s4 is the start position of the third stroke, and the position of the distance 1 is the end position of the third stroke. Note that the interval between the distances s1 and s2 is a section in which the change in the tangent angle θ with respect to the distance s when the end point of the first image and the start point of the second image are connected by a virtual straight line (Displayed with a two-dot chain line), and between the distances s3 and s4, the tangent angle with respect to the distance s when the end point of the second stroke and the start point of the third stroke are assumed to be connected by a virtual straight line This is a section in which the change of θ is represented (indicated by a two-dot chain line).
[0021]
On the other hand, it is assumed that the graph showing the change characteristics of the distance s and the tangent angle θ after normalization of the handwritten character is as shown in FIG.
Here, the position of the distance sa is the end position of the first stroke of the handwritten character, the position of the distance sb is the start position of the second stroke, and the position of the distance sc is the end position of the second stroke. The position of the distance sd is the start point position of the third image, and the position of the distance 1 is the end point position of the third image. Note that the interval between the distances sa and sb is a section in which the change in the tangent angle θ with respect to the distance s when the end point of the first image and the start point of the second image are connected by a virtual straight line is represented. (Displayed by a two-dot chain line), and between the distances sc and sd, the tangent angle with respect to the distance s when it is assumed that the end point of the second stroke and the start point of the third stroke are connected by a virtual straight line This is a section in which the change of θ is represented (indicated by a two-dot chain line).
[0022]
In order to collate the data of the tangent angle θ with respect to the distance s of the handwritten character and the data of the tangent angle θ with respect to the distance s of the standard character, the position of each stroke number of the handwritten character and the position of each stroke number of the standard character Therefore, for the handwritten character data, in the example shown in FIG. 4B, first, the character writing start point (point with zero distance) and the distance sa (sa For the data up to> s1), the distance coordinate S is converted so as to be data between the starting point of writing and the distance s1, and the coordinates are compressed. Next, for the data between the distances sa to sb, since the distances sa to sb are larger than the distances s1 to s2, coordinate compression is performed in the same manner as described above so that the data is between the distances s1 to s2. . Next, with respect to data between the distances sb to sc, since the distances sb to sc are smaller than the distances s2 to s3, the coordinates are expanded so as to become data between the distances s2 to s3. For the data between the distances sc to sd, since the distances sc to sd are larger than the distances s3 to s4, coordinate compression is performed so that the data is between the distances s3 to s4. As for the data up to the distance sd to 1, since the distance sd to 1 is smaller than the distance s4 to 1, the coordinate expansion is performed so that the data is between the distances s4 to 1. Note that coordinate compression and coordinate expansion itself are not unique to the present invention and are one of general mathematics, and therefore, detailed description thereof will be omitted here.
The graph representing the change characteristic of the tangent angle θ with respect to the distance s of the handwritten character after the alignment is performed as described above, as shown in FIG. The position of each start point and end point of the character matches that of the standard character.
[0023]
On the other hand, the stroke count matching can be said to be processing for changing the stroke count of the standard character selected by the dictionary selection to the stroke count of the handwritten character.
In other words, this process assumes that the handwritten character is not the correct number of strokes because the part of the handwritten character has a so-called continuous writing part. In order to match the two data.
In the following, a specific description will be given. First, as a premise, a handwritten character is input as a three-stroke character, and a graph showing a change characteristic of the tangent angle θ with respect to the distance s is temporarily shown in FIG. FIG. 5 is a graph showing the change characteristic of the tangent angle θ with respect to the distance s. Suppose that
[0024]
Under such a premise, the data of the number of strokes of the standard character is rewritten, assuming that the continuous writing is performed for each portion where the continuous writing is possible. For example, in FIG. 5, the position of distance s1 is the end point of the first stroke, the position of distance s2 is the start point of the second stroke, and the position of distance s3 is the end point of the second stroke. First, assuming that the first picture and the second picture are continuously written, the data between the distances s2 to s3, which is the second picture in FIG. 5, is the first picture. . That is, the data between the distances s2 to s3 before the processing is expressed as, for example, (sn, θn, 2) (“2” in parentheses is an image value indicating the second stroke). ), It is rewritten as (sn, θn, 1) after processing.
As a result, the graph representing the change characteristic of the tangent angle θ with respect to the distance s is a distance s1 represented by a two-dot chain line in FIG. 5 as being between the end point of the first stroke and the start point of the second stroke. The parts s2 to s2 are also assumed to be part of characters, and are represented by solid lines as shown in FIG.
[0025]
Further, for the standard character represented by the characteristic curve shown in FIG. 5, the end point of the second stroke and the start point of the third stroke may be written continuously.
In this case as well, data relating to the number of strokes is rewritten in the same manner as described above, and a graph showing the change characteristic of the tangent angle θ with respect to the distance s is as shown in FIG. That is, in this case, in FIG. 5, the distance s3 to s4 represented by the two-dot chain line is a part of the character, and this part is a characteristic curve represented by the solid line.
Furthermore, with respect to the standard characters represented by the characteristic curve shown in FIG. 5, the end point of the third screen and the start point of the fourth screen may be written continuously. Similarly, the data of the corresponding part is rewritten, and as a result, the change characteristic of the tangent angle θ with respect to the distance s as shown in FIG. 8 is obtained. That is, in this case, in FIG. 5, it is assumed that the portion of the distances s4 to s6 represented by the two-dot chain line is a part of the character, and a characteristic curve is obtained in which this portion is represented by a solid line.
[0026]
In this way, the number of strokes is adjusted by rewriting necessary data on the assumption that the continuous writing has been performed for each portion where there is a possibility of continuous writing. Are temporarily stored in a memory (not shown). In the above description, the case where the number of strokes of the standard character is one more than the number of strokes of the handwritten character has been described as an example. Similarly to the above, the data processing as described above may be performed on the assumption that two or more places are continuously written so as to be the same as the number of strokes of handwritten characters.
Next, in order to compare the data for which the number of strokes has been adjusted as described above and the input handwritten character data, the data of each standard character for which the number of strokes has been adjusted as described above is used. As a reference, the handwritten character data is aligned as described above (see step 112 in FIG. 2).
Then, a data distance calculation is performed between the handwritten character data and the standard character data that have been aligned (see step 114 in FIG. 2).
That is, the data distance calculation is to obtain a difference between the tangent angle of the standard character and the tangent angle of the handwritten character, and is specifically calculated by the following arithmetic expression.
[0027]
Data distance ΔΘ = ∫A (s) {θ1 (s) −θ2 (s)} ds
[0028]
The integration interval of this integration is between 0 and 1. In the equation, A (s) is a weighting coefficient, and is set to “1” for actual stroke data and “0.8” for interpolation stroke data, for example. Is. Further, θ1 (s) is a tangent angle with respect to the distance s in the handwritten character, and θ2 (s) is a tangent angle with respect to the distance s in the standard character.
The data distance ΔΘ obtained in this way is temporarily stored in a memory (not shown) of the data processing device 3.
As described above, this data distance calculation is performed for each of the patterns in which a so-called plural pattern is generated in the matching of the number of strokes, and is stored respectively.
[0029]
Subsequently, it is determined whether or not the data distance ΔΘ has been obtained for all the dictionaries in which data for a plurality of standard characters to be compared with handwritten characters are stored by the above-described processing (FIG. 2). If it is determined that the process has not been completed (in the case of NO), the process returns to the previous dictionary selection process (see step 106 in FIG. 2), and the series of processes described above is repeated. Become.
On the other hand, when it is determined that the processing for all of the dictionaries has been completed (in the case of YES), the case where the numerical value is the smallest among the plurality of data distances ΔΘ temporarily stored as described above. Assuming that the standard character is a candidate corresponding to the inputted handwritten character, the data is output from the data processing device 3 to the outside in a predetermined data format, and at the same time, the data is output to the display driving circuit 4, and as a result, the display device 5. The standard character is displayed on (see step 118 in FIG. 2).
[0030]
In the above example, the smallest of the calculated data distances ΔΘ is a candidate for recognition character. For example, the smallest of the data distances ΔΘ is equal to or smaller than a predetermined value. If the object is a recognition character candidate and the data distance ΔΘ is greater than or equal to a predetermined value, even if it is the smallest of the calculated data distances ΔΘ, there is a high possibility of erroneous recognition. In such a case, there may be no recognized character.
[0031]
In the above-described example, the description has been made on the assumption that the program for performing the processing shown in FIG. 2 is stored in a memory (not shown) of the data processing device 3, but it is not always necessary to have such a configuration. Instead, the following configuration may be used.
For example, a necessary program may be stored in a so-called hard disk (not shown) provided separately from the data processing device 3, and the program may be read from the hard disk and executed. In addition, a necessary program is stored in a so-called floppy disk (not shown), while a so-called floppy drive (not shown) is connected to the data processing device 3, and the program of the floppy disk is connected via the floppy drive. May be read and executed.
Further, the program may be executed by other techniques known and well known as this type of program storage technique, such as a method using a magnetic tape, a method using a magnetic disk, a method using a CD-ROM, etc. Of course, this is not a matter of ordinary technical activity for engineers in this technical field.
[0032]
In the example described above, the handwritten character input means is realized by the input tablet 1 and the digitizer 2.
Further, the data conversion means is executed by the data processing apparatus 3 in step 102 shown in FIG. 2, the storage means is executed by the data processing apparatus 3, and the stroke number determination means is executed by the data processing apparatus 3 shown in FIG. By the execution of step 108, the stroke number matching means is executed by the data processing apparatus 3 in step 110 shown in FIG. 2, and the alignment means is executed by the data processing apparatus 3 in step 112 shown in FIG. The integration calculation means is realized by the execution of step 114 shown in FIG. 2 by the data processing device 3, and the determination extraction means is realized by the execution of steps 116 and 118 shown in FIG. 2 by the data processing device 3. Yes.
[0033]
【The invention's effect】
As described above, according to the present invention, focusing on the fact that the distances at a plurality of appropriate points from the point at which characters are first written and the tangent angles at each point cause changes specific to each character. By configuring the system so that handwritten characters can be recognized at a high recognition rate with simple processing, the distance from the starting point of writing of handwritten characters to an appropriate sampling point, and data of the tangent angle at each sampling point, Unlike conventional methods, handwritten characters can be recognized without performing special data processing for specific characters, based on the comparison of similar distance and tangent angle data for standard characters. Simple processing is required, and the entire handwritten character and one standard character are matched. Therefore, the recognition rate is higher than the conventional so-called stroke-by-stroke matching. Can character recognition method and handwriting recognition device is to be provided.
Even if handwritten characters are partly written continuously, it can be recognized by changing the stroke number data of the standard characters to be compared. This is unnecessary, and has an effect that a relatively high recognition rate can be obtained only by a relatively simple process.
[Brief description of the drawings]
FIG. 1 is a configuration diagram showing a configuration example of a handwritten character recognition apparatus according to the present invention.
FIG. 2 is a flowchart showing a processing procedure of handwritten character recognition executed by the handwritten character recognition apparatus shown in FIG. 1;
FIG. 3 is a schematic diagram for conceptually explaining sampling points in a handwritten character “A”.
FIG. 4 is a characteristic diagram that contrasts the change characteristic of the tangent angle θ with respect to the distance s for a standard character of a virtual character and the change characteristic of the tangent angle θ with respect to the distance s for a handwritten character. FIG. 4A is a normalized characteristic diagram showing a change characteristic of the tangent angle θ with respect to the distance s for a certain standard character, and FIG. 4B is a tangent angle with respect to the distance s for a certain handwritten character. FIG. 4C is a characteristic diagram before alignment showing the change characteristic of θ. FIG. 4C is a characteristic line after alignment showing the change characteristic of the tangent angle θ with respect to the distance s for a certain handwritten character. FIG.
FIG. 5 is a characteristic diagram showing a change characteristic of a tangent angle θ with respect to a distance s for a virtual standard four-stroke character.
6 is a graph showing a change characteristic of the tangent angle θ with respect to the distance s when it is assumed that the end point of the first stroke and the start point of the second stroke are continuously written for the standard character having the characteristic curve shown in FIG. FIG.
FIG. 7 shows the change characteristic of the tangent angle θ with respect to the distance s when it is assumed that the end point of the second stroke and the start point of the third stroke are continuously written for the standard character having the characteristic curve shown in FIG. FIG.
8 is a graph showing a change characteristic of the tangent angle θ with respect to the distance s when it is assumed that the end point of the third stroke and the start point of the fourth stroke are continuously written for the standard character having the characteristic curve shown in FIG. FIG.
[Explanation of symbols]
1 ... Input tablet
1a ... Pen for input
1b ... Input window
2 ... Digitizer
3. Data processing device
4. Display drive circuit
5. Display device

Claims (1)

For handwritten characters, find the distance from the point at the beginning of writing at each appropriately selected point, and the tangent angle between the tangent at each point and a predetermined reference line,
For a plurality of standard characters to be compared with the handwritten character, the distance and tangent angle at each appropriate point of the standard character acquired in advance in the same manner as the handwritten character are compared with the distance and tangent angle of the handwritten character. A handwritten character recognition method characterized in that a standard character having distance and tangent angle data that most closely approximates a change in tangent angle with respect to the distance of each point of the handwritten character is used as a recognition result of the handwritten character.

Images