JPH0226267B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for recognizing characters or graphics, and more particularly to a method for recognizing characters or graphics consisting of one stroke handwritten on a tablet.

The processing of characters and graphics handwritten on a tablet with a stylus pen, for example, is called real-time processing or online processing, and is used for various data input and data editing. The reality is that the equipment provided has not yet reached the point of practical application.

The reason for this is that during online input, (1) the use of sequential input to a coordinate point series obtained by sequentially sampling characters or figures, and (2) the sufficient description of a set (stroke) of a coordinate point series, etc. This is probably due to the fact that it is not fully utilized. For example, regarding point (1), the current state of development of online kanji recognition devices is limited to the extent to which strokes are divided depending on the vertical movement of the stylus, and regarding point (2), to the degree of position change between strokes. Most of them are focused, and the composition does not necessarily represent the entire character or figure. Therefore, it is extremely difficult to recognize one-stroke characters, figures, etc., which appear relatively simple to the human eye.

Conventional character recognition mainly involves sampling the coordinate points that make up a stroke and detecting differences in the positional relationship between the sampling points.If the sampling position is good, it can be recognized, but if it is not, it cannot be recognized. It had

Therefore, the present invention improves this drawback and enables recognition regardless of the sampling position by addressing the above two points, namely (1) use of sequential input of coordinate point series, and (2) collection of coordinate point series ( The object of the present invention is to provide a method for recognizing characters or figures consisting of one stroke, which is based on the description of strokes. For this purpose, the character or figure recognition method of the present invention includes a coordinate point sequence extracting means for extracting a coordinate point sequence of a character or figure consisting of one stroke, and a A sub-stroke forming means that forms a stroke and a pair of sub-strokes adjacent to the change point based on the difference between the change points, determine in which direction the pair of sub-strokes are open with respect to the coordinate direction. a surface structure label forming means for obtaining a surface structure label shown in the figure; a first determining means for determining whether a coordinate point sequence constituting a substroke is a straight line or a curved line; By providing a second determining means for excluding the coordinate points and a third determining means for determining whether the coordinate point sequence is a straight line group or a curved line based on the difference between the coordinates of the coordinate point sequence, the sub-stroke is determined whether the sub-stroke is a straight line or not. A feature of the present invention is that local structure labels indicating whether the object has a characteristic or a curve are provided, and recognition is performed based on these labels.

Next, the principle of the present invention will be explained in detail.

As a method for recognizing a character or figure consisting of one stroke, the present invention involves a description of the rough shape of the character or figure to be recognized, that is, global processing, and a combination of what components it is made of. In other words, two types of processing are performed: local processing; global processing embodies point (2) above, and local processing embodies point (1). It is. For example, in character recognition, ``ro'' and ``to'' in hiragana can be distinguished through global processing, but ``ra'' and ``ro'' cannot be distinguished without further local processing.

DESCRIPTION OF THE PREFERRED EMBODIMENTS A method for recognizing characters or figures consisting of one stroke according to the present invention using global processing and local processing will be described in detail below with reference to the drawings.

[] About global processing Global processing is the expression of a stroke as a set of substrokes based on a sequence of coordinate points input from a tablet, and the global graphical structure that can be estimated from adjacent substrokes. Next, this global processing will be explained in detail.

(1) Extraction of substrokes A substroke is a point where the coordinate point series input from the tablet monotonically increases or decreases in the horizontal (X-axis) direction or vertical (Y-axis) direction. Ignoring the interval, sequentially capture the coordinate points that change from increasing to decreasing, or from decreasing to increasing,
Define the space between each coordinate point as one substroke.

FIG. 1 shows a specific example of a substroke extraction method. Let the sampled coordinate points of the hiragana “so” shown in Figure 1a be 1,
2, 3...10, 11, 12 (Actually, there are many, but I chose 12 to simplify the explanation), each of 1 to 12
A coordinate series (X ₁ , Y ₁ ) consisting of the X and Y coordinates of the point,
(X ₂ , Y ₂ )...(X ₁₂ , Y ₁₂ ) is obtained. The X coordinate increases monotonically between coordinate points 1 and 3, but changes from a monotonous increase to decrease between coordinate points 3 and 4, and
The interval decreases monotonically. It changes from a monotonous decrease to an increase between coordinate points 5 and 6, and monotonically increases between coordinate points 5 to 7. Similarly, in the X coordinate direction, the first
As shown in Figure b, coordinate points 1-3, 5-7, 10-
Between coordinate points 12 and 12, there is a monotonous increase, and between coordinate points 3 and 5 and between 7 and 10, there is a monotonous decrease. And the coordinate points where the X coordinate changes from monotonically increasing to decreasing or from monotonically decreasing to increasing are:
3, 5, 7-10. Therefore, the substroke in the X coordinate direction is the coordinate point 1-3, 3-5, 5-
There are 5 pieces that connect 7, 7-10, and 10-12.

Similarly, the sub-strokes in the Y coordinate direction are as shown in FIG.
There are two pieces connecting the two.

(2) Meaning of substrokes The substrokes extracted by the method described above are expressed as a group of straight lines from a global perspective of characters or figures. This invention utilizes an input stroke, that is, a group of straight lines 1-2, 2-3, 3-4...10-11, 1 connecting each coordinate point in FIG.
As a means of roughly describing the shape of 1-12,
This method uses a method to express the surface structure formed by the sub-strokes extracted by the method described above as a concave shape.

First, each substroke is given a label meaning "+" if it is the same along the input direction, and "-" if it is opposite. This gives each substroke its directionality.

Figure 1 d shows a monotonous increase in the positive direction of the X coordinate,
Directionality is given to each substroke by setting "-" to indicate a monotonous increase in the negative direction.
S _x1 , S _x3 , and S _x5 are "+", and S _x2 and S _x4 are "-".
In addition, the Y coordinate is "+" for a monotonous increase in the negative direction and "-" for a monotonous increase in the positive direction, and as shown in Figure 1e, the substroke S _y1 is "+" and the substroke S _y2 is " −”.

Next, we will introduce the second method of expressing the surface structure of the entire stroke.
This will be explained using figures. Considering adjacent substrokes, when the direction changes from "+" to "-", it changes from monotonous increase to monotonous decrease, so a shape like 〓 is extracted, and from "-" to "+". ”, a 〓-shaped shape is extracted.
Therefore, as shown in Figure 2a, when the directionality of adjacent substrokes in the It is expressed as Also, when "-" → "+", it is extracted that the shape is 〓, and this is labeled as "Dr". Similarly, as shown in Fig. 2b, when the directionality of adjacent substrokes in the Y coordinate direction is "+" → "-", it is concave and is labeled "Du", and "-" → " ＋”
In this case, the label "Dd" is given in the shape of 〓.
The symbols obtained in this way are called surface structure labels, and among these surface structure labels, Dl and Dr are called X-plane labels, and Du and Dd are called Y-plane labels.

(3) Representation and grouping of strokes by global processing Through each of the processes (1) and (2) above, a stroke input as a sequence of coordinate points is represented as a rough shape with a surface structure. The expression method is to arrange the X-plane labels related to the characters or figures in order from the positive to the negative direction of the Y coordinate direction, that is, from the top to the bottom,
This is done by arranging the Y-plane labels from positive to negative in the X-coordinate direction, that is, from left to right.

FIG. 3 shows how to express the hiragana "so" shown in FIG. 1a by global processing. Figure 3a is a reproduction of "so" shown in Figure 1a, and in Figure 3b, the substrokes in the X coordinate direction are S _x1 , S _x2 , S _x3 , S _x4 and S _x5 . If the directions are +, -, +, -, +, respectively,
becomes. Next adjacent substroke (S _x1 ,
S _x2 ), (S _x2 , S _x3 ), (S _x3 , S _x4 ), and (S _x4 , S _x5 ) are labeled on the X side in order from top to bottom.
As shown in the figure, Dl, Dr, Dl, Dr. Similarly, there are two substrokes in the Y coordinate direction, S _y1 and S _y2 , as shown in FIG. 3c, and their Y plane label is Du. Therefore, the representation of ``so'' in hiragana through global processing is as shown in Figure 3d. Hereinafter, stroke expression by this global processing will be referred to as global display.

In addition, when displaying these surface structure labels in binary code, for example, "1" indicates an X-plane label, "0" indicates a Y-plane label, and so on.
Dl is "11", Dr is "10", Du is "01", Dd is "00"
The global representation can be encoded in binary code by this arrangement of four symbols.
For example, the global representation of “so” in Figure 3 (Dl, Dr,
If Dl, Dr: Du) is encoded as (11, 10, 11,
10:01). Note that after "11" there is always "10",
The next ``10'' is always ``11'', the next ``01'' is always ``00'', and the next ``00'' is always ``01'', so use this. Error checking can be performed.

Furthermore, if a global display is used, it can be used for large classifications or groupings in character and graphic recognition without using minute features.

FIG. 4 shows an example of grouping for character or figure recognition. 9 shown in I to I of the same figure
When characters or figures (FIG. 4 d) are globally displayed using surface structure labels, the X surface label is represented as shown in FIG. 4 b, and the Y surface label is represented as shown in FIG. 4 c.
If you look at this Figure 4, in the nine characters and figures shown, the "so" in A can be identified only by the X-side label, but the others cannot be determined only by the X-side label, and are the same. It can be seen that there are four groups with X-plane labels (Ro and Ha, Ni and To, Ho and He, Chi and Li). Furthermore, if you add a Y-plane label,
The three characters A, B, and C in the figure are distinguished by the X- and Y-plane labels, and it can be seen that there are three groups with the same X- and Y-plane labels (Ni, T, C). ho and he, chi and ri). Distinguishing the characters or figures within each of these groups cannot be done by global display, and requires the following local processing.

Next, local processing will be explained.

[] Regarding local processing (1) Local processing (Part 1) In local processing (Part 1), each shape obtained by global processing is separated into constituent components of a straight line, and Extract whether it is a curve, a straight curve, or a straight curve.

FIG. 5 is an example of a sub-stroke whose X-plane label is Dl, where a is straight line + straight line, b is straight line + curved line, and c is curved line + curved line. In FIG. 5, straight lines connecting black circles form substrokes. Furthermore, d
Although it is possible to consider the figures shown in , it is considered that such figures are unlikely to appear as meaningful characters or figures, so they may be excluded from the scope.

Since the images a to c in FIG. 5 belong to the group with the same X-plane label Dl, they cannot be individually distinguished from each other only by the global display. Local processing (part 1) first calculates the distance S _CL of each substroke. Next, the sum S′ _CL of the distances between the coordinate point sequences forming the sub-stroke is calculated. And if S _CL ≒ S′ _CL , it is a straight line,
If this is not the case, it is determined that it is a curve (hereinafter referred to as the determination criterion). The reason for setting S _CL ≒ S' _CL here instead of S _CL = S' _CL is to take into account that when inputting on a tablet with a stylus pen, even if a straight line is drawn, it often curves a little because the input is done by hand. Therefore, it is necessary to statistically decide which range should be considered a straight line, taking into account individual differences. Also, there is a fifth line in the curve.
Of course, things other than those shown in the figures are also included. Therefore, anything that is not S _CL ≒ S′ _CL should be accurately expressed as not being a straight line, but for simplicity, it will be called a curve.

The factors that can be judged as a curve based on the above criteria are: when it is completely curved, when an abnormal coordinate point is detected due to "shaking" during input, and when the substroke has a straight line tilted in two directions. When there is a "discrepancy" due to global shape extraction, etc.

FIG. 6 shows these specific examples, and ~ shown below and ~ above correspond to each other. The area between the black circles indicates a substroke, and the white circle indicates that the portion of each coordinate point is a substroke determined to be a curve. Here, in e of Fig. 6, B ₀ - B ₁ and B ₁ - B ₂ form a substroke, and B ₀ - W, W - B ₁ are substrokes because the X coordinate does not change between W - B ₁ . Care must be taken not to form a stroke.

Next, for each sub-stroke, a line perpendicular to the sub-stroke straight line is drawn from the coordinate point sequence composing that sub-stroke to create a distance histogram. When only a portion where the distance histogram is substantially uniform shows an abnormal value, it is determined that the coordinate point sequence forming the sub-stroke is a straight line (hereinafter referred to as the determination criterion).

FIG. 7 shows an example of the determination criteria.
This figure shows one sub-stroke P ₁ - P ₂ shown in figure a and the coordinate point sequence Q ₁ , Q ₂ , ...Q ₅ ... that constitutes it, and the coordinate Q ₅ is greatly deviated from the others. There is. Figure b shows a distance histogram of each coordinate point, showing that only coordinate point _Q5 is an abnormal point. Therefore, based on the determination criteria, it is determined that the coordinate point sequence forming the sub-strokes P ₁ to P ₂ is a straight line, and the coordinate point Q ₅ is excluded. In this way, the coordinate point sequences forming each sub-stroke in FIG. 6f are determined to be straight lines.

Note that the level of abnormal values in the determination criteria is determined statistically, and the number of abnormal values is also determined by the interval between coordinate point sequences. This is because when the sampling interval is narrow, several abnormal parts may be sampled.

Each sub-stroke extracted from a character or figure is further given meaning by assigning a label "S" to those determined to be straight lines and "C" to those determined to be curved lines according to the determination criteria. It will be done. Hereinafter, this label will be called a local structure label, the local structure label S will be called a straight line label, and the local structure label C will be called a curved label. By combining this local structure label with the above-mentioned surface structure label, the shape of a character or figure can be expressed as code information.

FIG. 8 shows an example of identifying characters or figures using surface structure labels and local structure labels. In FIG. 8, a indicates five characters or figures A to E, b indicates a global display, c indicates the shape of each substroke, d indicates an X-plane label, and e indicates a local structure label.

By adding this criterion, among the factors that are determined to be curved according to the criterion, those determined to be curved even though the actual shape is straight are correctly determined to be straight. The other cases (Fig. 5a and Fig. 6) are excluded from the scope of the present invention because they rarely occur in practice. The remaining cases are determined by the following local processing (Part 2).

(2) Local processing (Part 2) Local processing (Part 2) generates a list of differences in the coordinate values of the coordinate point sequences that make up the sub-stroke, for a sub-stroke that is determined to be a curve according to the criteria. When the difference is a large value, that is, when the difference in the X or Y coordinate of each coordinate point sequence exceeds the set threshold, it is determined that it is a straight line group,
If the difference does not become a large value, it is determined that it is a curve (hereinafter referred to as the determination criterion). According to this criterion, the line in FIG. 6 is determined to be a straight line group.

Note that the threshold value for determining whether the difference is a large value is determined depending on the size of the characters or figures to be written in the tuplet.

In addition, there are many types of figures that consist of one stroke, but in order to identify characters or figures that are used in daily life and consist of one stroke, which this invention targets for identification, it is practical to add the above-mentioned criteria ~. The above is sufficient.

It is easily understood that when displaying the surface structure label and the local structure label as code information using binary codes, they can be displayed as a series of 4-bit codes.

FIG. 9 shows one embodiment of global processing and local processing according to the present invention.

In the figure, 10 is a tablet, 11 is a stylus pen for writing characters or figures, 12 is a coordinate point string storage memory that stores the X and Y coordinates of the coordinate point string of one character or one figure, and 13 is the X coordinate and Y coordinate point string. A change point detection circuit for detecting a change point in a coordinate direction, 14
is a change point storage memory that stores the coordinates of this change point. Here, the change point storage memory 14 stores the coordinates of two adjacent sub-strokes related to the change point (for example, 1, 3,
5 coordinates of each coordinate point) are stored. 15 is an inter-change point difference circuit that calculates the difference between change points; 16 is a comparison circuit; 17 is a conversion circuit that converts into surface structure labels; 18 is a description buffer that performs global display;
Reference numeral 9 denotes a distance extraction circuit that integrates the distance between coordinate point sequences forming each sub-stroke, 20 a histogram storage memory that creates and stores a distance histogram, and 21 a determination circuit that determines whether each sub-stroke is a straight line or a curved line. , 22 is a description buffer that gives local structure labels of straight lines and curves to each sub-stroke, 23 is a difference extraction circuit that extracts the coordinate difference of each coordinate point sequence, and 24 determines whether it is a group of straight lines or a curve according to the determination criteria. 25 is a comparison circuit that compares the results of the description buffer 22 and the determination circuit 24;
26 is a description buffer that adds a local structure label to a substroke.

Now, in FIG. 9, when one character or figure consisting of one stroke is input to the tablet 10 with the stylus pen 11, the coordinates of the coordinate point sequence of the sampled character or figure are stored in the coordinate point sequence storage memory 12. . The change point detection circuit 13 detects a change point where the polarity of the increasing point of the X coordinate and Y coordinate of the coordinate point sequence changes, that is, the coordinate point of each substroke. Then, the change point storage memory 14 stores the coordinate points of substrokes adjacent to each change point in pairs. The inter-changing point difference circuit 15 calculates the difference in the coordinates of each changing point and transmits it to the comparing circuit 16.
The comparison circuit 16 compares this difference with a positive or negative reference voltage to give directionality to each sub-stroke. The conversion circuit 17 applies a surface structure label to each adjacent pair of substrokes from the oriented substrokes. And description battle 1
8 outputs surface structure labels as code information for global display.

Each time the distance extraction circuit 19 is supplied with the coordinate output of each sub-stroke from the conversion point storage memory 14,
Based on the coordinates of the coordinate point sequence transmitted from the coordinate point sequence storage memory 12, the distance between the coordinate point sequence forming the substroke and the distance to the substroke are calculated. The histogram storage memory 20 creates a distance histogram from the distance to the sub-stroke, and eliminates abnormal points based on the determination criteria. The determination circuit 21 determines whether the coordinate string constituting each sub-stroke is a straight line or a curved line based on the determination criteria. This determination result is then held in the description buffer 22.

On the other hand, the difference extraction circuit 23 uses the change point storage memory 1
4, each time a sub-stroke coordinate input is supplied, a difference between each coordinate of a sequence of coordinate points constituting the sub-stroke is extracted, and this is transmitted to the determination circuit 24. The determination circuit 24 determines whether the coordinate point sequence constituting the sub-stroke is a straight line group or a curved line according to the determination criteria, and transmits this to the comparison circuit 25. As a result, the comparison circuit 25 converts the description buffer 22
Among the curves held in , those that correspond to a group of straight lines are corrected as a group of straight lines and are transmitted to the description buffer 26 .
The description buffer 26 gives a straight line label S and a curved line label C to each substroke. In this way, a character or figure consisting of one stroke can be identified by the local structure label output from the description buffer 26 and the surface structure label output from the description buffer 18.

As explained above, according to the present invention, it is possible to process and recognize characters or figures online by combining global processing and local processing, and also by changing the sampling position of characters or figures. Accurate identification can be performed without causing any difference in identification results.

[Brief explanation of drawings]

Figure 1 is an illustration of substroke extraction and orientation, Figure 2 is an illustration of surface structure labels, Figure 3 is an illustration of global processing, and Figure 4 is an illustration of grouping for character or figure recognition. , FIG. 5 to FIG. 7 are explanatory diagrams of local processing, FIG. 8 is an explanatory diagram of local structure labels, and FIG. 9 is a configuration diagram of an embodiment of the present invention. In the figure, 10 is a tablet, 11 is a stylus pen, 12 is a coordinate point sequence storage memory, 13 is a change point detection circuit, 14 is a change point storage memory, 15 is a difference circuit between change points, 16 is a comparison circuit, and 17 is a screen. 18 is a description buffer; 19 is a distance extraction circuit; 20 is a histogram storage memory; 2
1 is a judgment circuit, 22 is a description buffer, 23 is a difference extraction circuit, 24 is a judgment circuit, 25 is a comparison circuit, 2
6 indicates the description buffer.

Claims (1)

[Claims] 1. Coordinate point sequence extraction means for extracting a coordinate point sequence of a character or figure consisting of a stroke; substroke forming means for forming a substroke from a change point in the coordinates of the coordinate point sequence in one coordinate direction; A surface structure label forming means for obtaining a surface structure label indicating in which direction the pair of substrokes is open with respect to the coordinate direction from the pair of substrokes adjacent to the change point by the difference; a first determining means for determining whether a coordinate point sequence constituting a stroke is a straight line or a curve; a second determining means for excluding abnormal coordinate points from a distance histogram between a sub-stroke and a coordinate point sequence; By providing a third determining means for determining whether a coordinate point sequence is a straight line group or a curved line based on the difference in each coordinate, a substroke has a local structure that indicates whether the substroke has linearity or curvedness. A character or figure recognition method characterized by assigning labels and performing recognition based on these labels.