JPS5852266B2 - Real-time character recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a real-time character recognition device having a stroke relative position matrix extraction means and a stroke shape feature extraction means.

Conventional real-time character recognition devices focus on the shape of the strokes that make up a character, encode the strokes into stroke codes, and perform recognition based on that sequence.

According to such a conventional method, although the structure of the entire character remains unchanged, there is a risk that erroneous recognition may occur because the shape of some strokes is deformed.

However, the relative positional relationship of the strokes that make up the characters is
It is stable even if the shape of the stroke changes slightly.

The present invention is based on the above considerations, and recognizes the relative positional relationship of strokes as the main character, and when a character cannot be uniquely determined by the relative positional relationship of strokes,
It is an object of the present invention to provide a real-time character recognition device that further limits candidate characters based on stroke shape characteristics, thereby making it possible to always correctly recognize input characters.

Therefore, the real-time character recognition device of the present invention is a real-time character recognition device that targets real-time input characters expressed in the time system of strokes. degree, the length ratio determined by the total distance of the stroke and the total distance of all the strokes that make up the character, the direction angle that indicates the direction of a straight line from the start point to the end point of the stroke, and the angle determined by three consecutive points on the stroke. stroke shape feature extraction means for extracting the amount of angular change defined by the sum as a stroke shape feature; representative point number determining means for determining the number of representative points for approximating a stroke according to the number of strokes of a human-powered character; and the number of representative points determining means. representative point coordinate determining means for determining representative point coordinates that approximate the stroke of a human-powered character based on the number of representative points determined by the representative point coordinate determining means; a relative position matrix extraction means for extracting a relative position matrix; a position matrix dictionary storing relative position matrices of characters; a position matrix matching means for matching the relative position matrix of human characters with the position matrix dictionary; A shape feature dictionary storing stroke shape features and a shape feature that operates when the character type of the input character is not uniquely determined by the position matrix matching means and matches the stroke shape features of the input character with the shape feature dictionary. It is characterized by having a matching means.

The present invention will be explained below with reference to the present invention.

Fig. 1 is a block diagram of an embodiment of the present invention, Fig. 2 is a diagram showing an example of dividing each stroke of the hiragana ``a'' into three equal parts and using the dividing points as representative points, and Fig. 3 shows an example of the representative points. Figure 4 is a diagram showing an example of serial numbering, Figure 4 is a diagram showing a relative position matrix for the characters in Figure 3, Figure 5 is a diagram explaining the principle of stroke straightness extraction, Figure 6 is a diagram showing the stroke length ratio. A diagram showing the principle of extraction, FIG. 7 is a diagram showing the principle of stroke direction angle extraction,
FIG. 8 is a diagram showing the principle of extraction of the amount of change in angle of a stroke, FIG. 9 is a diagram explaining the sign of the amount of change in angle, and FIG. 10 is a time chart showing the order of processing.

First, in order to facilitate understanding of the present invention, an overview of the present invention will be explained.

In a recognition method that targets real-time input characters expressed as a time series of strokes, the number of strokes and stroke order information can be easily obtained.

It is thought that characters with a large number of strokes, such as kanji, can be classified and identified stably due to the relative vertical and horizontal relationships within the strokes, despite frequent fluctuations in stroke shape.

In real-time manual writing, strokes can be numbered in the order of strokes, so information on relative positions between strokes is extracted as a feature as a stroke position matrix, and characters are classified and identified based on this information.

For characters that are not uniquely determined by the stroke relative position matrix, the character category is determined based on the extracted stroke shape information.

Furthermore, for characters with a small number of strokes such as kana, the number of representative points per stroke is increased to create a stroke relative position matrix.

FIG. 1 is a block diagram of one embodiment of the present invention, in which 1 is a character input device, 2 is a normalization circuit, 3 is a representative point number determination circuit, 4 is a representative point coordinate determination circuit, and 5 is a relative position matrix extraction circuit. circuit, 6 is a position matrix matching circuit, 7 is a position matrix dictionary, 8
1 is a stroke shape feature extraction circuit, 9 is a stroke shape feature matching circuit, 10 is a stroke shape feature dictionary, and 1 is a stroke shape feature extraction circuit.
1 indicates a candidate selection circuit based on the number of strokes.

The character input device 1 is, for example, a tablet.

If human characters are represented by the number of strokes, ak, then the number of strokes, nk, is expressed by the number of points of the first stroke.

The normalization circuit 2 calculates the character width and character height, selects the larger one of the width and height, and normalizes the size so that the larger one is constant.

Set the font size to SXS.

The representative point number determination circuit 3 determines the number of representative points N per l stroke.
is determined as a function of 1 to the number of strokes that intercept a character.

For characters with a large number of strokes, the number of representative points per stroke is decreased, and for characters with a small number of strokes, the number of representative points per stroke is increased.

As a result, the total number of representative points per character is kept approximately constant regardless of the type of character.

The representative point number determination circuit 3 includes, for example, a stroke number counting circuit that counts the number of strokes of a character, and a stroke representative point number conversion circuit that converts the count value of the stroke number counting circuit into a stroke representative number.

The determination of the representative score is described in detail in Japanese Patent Application No. 133863/1983, which was previously filed by the present applicant.

The representative point coordinate determining circuit 4 divides the stroke based on the number of representative points determined by the representative point number determining circuit 3, and determines the coordinates of the divided points as representative points.

If the character approximated by the representative point is P, stroke &, and the representative point is (XF, Y\), then it is expressed as follows.

FIG. 2 shows an example in which each stroke of the hiragana "a" is divided into three equal parts, and the division points are used as representative points.

Representative point coordinate determination circuit 4c Example (f, Stroke length calculation circuit that calculates the total distance of the stroke from the input coordinate point series of the stroke, and calculates the division distance by dividing the stroke length by the number of representative points minus 1) The separation distance calculation circuit includes a separation distance calculation circuit that uses the start point and end point of a stroke as representative points, and a circuit that determines dividing points such that the distance along the stroke between adjacent representative points is equal to the dividing distance.

The division of strokes is described in detail in Japanese Patent Application No. 133864/1983, which was previously filed by the present applicant.

The relative position matrix extraction circuit 5 assigns serial numbers to the representative points in chronological order and extracts a relative position matrix between the representative points.

If the number of representative points of each stroke constituting one character is ni, then the serial number i of the first representative point of the stroke is expressed as follows.

As a result, the input character IP is It is expressed as

In the above formula, the subscript indicates the serial number.

FIG. 3 shows an example of serial numbering of representative points.

The relative position matrix extraction circuit 5 then extracts a relative position matrix between the representative points as follows.

Now, the relative position matrix IR is defined as follows.

In equations (4) and (5), L is the number of levels And, L is It is.

Also, (kij) is a horizontal position matrix, and (Vij) is a vertical position matrix.

kij, Vij are.

FIG. 4 shows the relative position matrix for the characters in FIG. 3, where K1=3. The case where N-3°L=5 is shown.

In Figure 4, for example, the intersection of the 1st row and 3rd column is 'X-2''
This is the point (xt, Yt) and the point (X3.
This shows that the X coordinate values of ¥3) are separated by 12 units.

The relative position matrix extraction circuit 5 includes, for example, a representative point serial numbering circuit that serially numbers representative points, a representative point X coordinate buffer that stores the X coordinate of the representative point, and a representative point Y coordinate buffer that stores the Y coordinate of the representative point. a coordinate buffer, a character frame circuit that calculates the size of a character frame, a horizontal position matrix creation circuit that creates a horizontal position matrix based on the contents of the representative point X coordinate buffer, the size of the character frame, and the number of levels; contents of target Y coordinate buffer,
A vertical position matrix creation circuit that creates a vertical position matrix based on the size of a character frame and the number of levels; a horizontal position matrix buffer that stores a horizontal position matrix; a vertical position matrix buffer that stores a vertical position matrix; It consists of a position matrix synthesis circuit that combines the contents of the matrix buffer and the vertical position matrix buffer.

Note that the extraction of the relative position matrix is described in detail in Japanese Patent Application No. 133872/1987, which was previously filed by the present applicant.

Next, classification and identification based on the number of strokes and relative position matrix will be explained.

First, the candidate selection circuit 11 based on the number of strokes classifies input characters based on the number of strokes, thereby narrowing down candidate categories for the input characters.

The position matrix matching circuit 6 performs matching between the relative position matrix of the input character and the position matrix dictionary 7 within the range of the narrowed down candidate categories, and further narrows down the candidate categories.

The position matrix dictionary 7 stores relative position matrices of characters divided by number of strokes.

Once the number of strokes of the input character is determined, relative position matrices are successively generated from the area determined by the number of strokes of the input character, matching is performed, and similar characters are set as candidate categories.

For characters that cannot be uniquely determined in the relative position line, the character type of the input character is determined using the stroke shape feature.

There are four types of stroke shape characteristics: straightness, length ratio, direction angle, and angle change.

These features are extracted by the stroke shape feature extraction circuit 8.

FIG. 5 explains the principle of stroke straightness extraction.

When the total distance of the stroke is calculated and the distance between the starting point and the ending point is d, the straightness C is defined as follows.

FIG. 6 explains stroke length ratio extraction.

When the total distance of the target stroke is D1 and the total distance of all strokes constituting the character is DT, the length ratio N is defined as follows.

In FIG. 6, it is.

FIG. 7 shows the principle of stroke direction angle extraction, and the direction angle θ is defined as the angle between the straight line from the starting point to the ending point and the X axis.

FIG. 8 explains the extraction of the stroke angle change amount.

The angle change amount ψ is determined by three consecutive representative points Pi-1, Pi
, Pi+1 is defined as the sum of angles φi.

FIG. 9 explains the sign of the angle change amount. When the stroke is as shown in FIG. 9 A, ψ is negative, and the 9
In cases such as the figure opening, ψ is assumed to be zero, and in cases such as that shown in Figure 9C, ψ is assumed to be positive.

The stroke shape features of the input characters extracted by the stroke shape feature extraction circuit 8 are sent to the stroke shape feature matching circuit 9.

The stroke shape feature matching circuit 9 matches the candidate categories sent from the position matrix matching circuit 6.
The stroke shape features of human characters are matched with the stroke shape feature dictionary 10, and the answer is output.

FIG. 10 is a time chart showing the processing order of the apparatus shown in FIG.

When a character is input by the character input device 1, the normalization circuit 2 then normalizes the character.

After normalization, the representative point number determination circuit 3 determines the representative point of the stroke, and at the same time, the stroke shape feature extraction circuit 8 performs stroke shape feature extraction.

After the number of representative points has been determined, the representative point coordinate determining circuit 4 determines the coordinates of the representative point.

After determining the representative point coordinates, a relative position matrix of the input character is extracted by a relative position matrix extraction circuit.

After extracting the relative position matrix, the candidate selection circuit 11 performs major classification based on the number of strokes, and then the position matrix matching circuit 6 performs matching between the relative position matrix of the input character and the position matrix dictionary 7.

When the character type of human-powered characters cannot be uniquely determined, the extracted stroke shape features and stroke shape patent dictionary 10
Performs matching and outputs the final answer.

As is clear from the above description, according to the present invention, human characters are recognized mainly based on the relative position matrix of representative points of relatively stable strokes, regardless of the character entry person. When the answer to the input character cannot be uniquely determined by the position matrix, recognition processing is performed based on the stroke shape feature, so the character type of the human-powered character can always be determined correctly.

In addition, the stroke shape features including linearity, length ratio, direction angle, and angle change amount capture the global features of the stroke, and are stable against character deformation. Therefore, the present invention According to this method, a high recognition rate can be obtained.

Furthermore, the stroke relative position matrix and the above-mentioned stroke shape feature are complementary features, and by combining both, a stable identification system can be constructed.

[Brief explanation of drawings]

Fig. 1 is a block diagram of one embodiment of the present invention, Fig. 2 is a diagram showing an example of dividing each stroke of the hiragana ``a'' into three equal parts and using the dividing points as representative points, and Fig. 3 is a serial number of the representative points. Figure 4 is a diagram showing the relative position matrix for the characters in Figure 3, Figure 5 is a diagram explaining the principle of stroke straightness extraction, Figure 6 is stroke length ratio extraction. Diagram showing the principle of
FIG. 7 is a diagram showing the principle of stroke direction angle extraction, and FIG. 8 is a diagram showing the principle of extraction of stroke angle change amount.
FIG. 9 is a diagram for explaining the sign of the amount of change in angle, and FIG. 10 is a time chart showing the order of processing. 1...Character input device, 2...Normalization circuit, 3...°°・Representative point number determination circuit, 4...
Representative point coordinate determination circuit, 5... Relative position matrix extraction circuit, 6... Open position matrix matching circuit, 7...
・Position matrix dictionary, 8... Song stroke shape feature extraction circuit, 9... Stroke shape feature matching circuit,
10... Stroke shape feature dictionary, 11...
... Candidate selection circuit based on the number of strokes.

Claims (1)

[Claims] 1. In a real-time character recognition device that targets real-time input characters expressed as a time sequence of strokes, the straightness determined by the total stroke distance and the distance between the start point and the end point, and the total stroke distance and all the characters that make up the character. The length ratio determined by the total distance of the stroke, the direction angle indicating the direction of the straight line from the start point to the end point of the stroke, and the angle change amount defined by the sum of the angles determined by three consecutive points on the stroke are defined as stroke shape characteristics. a stroke shape feature extracting means for extracting a stroke shape feature, a representative point determining means for determining the number of representative points for approximating the stroke according to the number of strokes of the input character, and a representative point determining means for determining the number of representative points for approximating the stroke according to the number of strokes of the input character; representative point coordinate determining means for determining representative point coordinates that approximate the stroke of the representative point; and relative position matrix extraction means for extracting a relative position matrix between the representative points based on the representative point coordinates determined by the representative point coordinate determining means. , a position matrix dictionary storing relative position matrices of characters; a position matrix matching means for matching the relative position matrix of input characters with the position matrix dictionary; and a shape feature dictionary storing stroke shape characteristics of characters. , a shape feature matching means that operates when the character type of the input character is not uniquely determined by the position matrix matching means and matches the stroke shape feature of the input character with the shape feature dictionary. Character recognition device.