JPS6073794A - Character recognition device - Google Patents

Abstract

PURPOSE:To enable the titled device to extract features easily and correctly by correcting the direction division due to the inclination of plural segments, and adding the length of respective segments corresponding to the plural direction compositions. CONSTITUTION:A segment direction correction circuit 33 and a segment length correction circuit 34 are installed. At the circuit 34, in case of the direction division from point 50 to point 51 is theta=6, the division from point 51 to point 52 is theta=5 and the division from 52 to point 53 is theta=6, these can be regarded as the inclination of same segment. Therefore, theta from point 51 to point 52 is corrected from 5 to 6. By executing such correction in order, all divisions can be treated as theta=6 in the range from point 50 to point 54 and the picture feature is correctly and easily extracted. At the circuit 34, the whole length of the desired segments is obtained by adding the length of the segment corresponding to respective divisions including the corrected divisions in the circuit 33. The recognition is executed from the picture feature obtained in such a way.

Description

[Detailed Description of the Invention] ■Technical Field The present invention relates to an online character reading recognition device that writes characters, symbols, etc. (hereinafter referred to as characters, etc.) using a pen on a predetermined tablet and reads the written characters. It is.

[2] Prior art When reading characters, etc. written with a pen on a prescribed tablet online, the information obtained from the paterzo input device for handwritten figures is read using a general-purpose computer connected to the input device. , needed to be recognized by its software.

For this reason, it is necessary to rely on complex character recognition steps, and a long time elapses from input to obtaining recognition results, which is often not suitable for the actual situation of online reading.

Also, as an online device for reading characters written with a pen on a tablet, it is possible to detect changes in the coordinates of the axes (two axes) perpendicular to the tablet surface based on the stroke order of the characters and the length, direction, direction, etc. of each line segment, that is, the stroke of one character. No. 57-185839, and will be explained below using figures.

FIG. 1 shows pen movement angles θ divided into eight types as an example for easily recognizing the direction of line segments of characters, etc. Next, about 3
The numbers in the figure are normalized font sizes. In other words, the direction of the line segment is expressed by the above-mentioned code symbol, and the length of the line segment is determined by dividing the normalized character size and standard dimension C (for example, 10 mm) into 4 equal parts, and dividing 1/4 or less into 0. ,1/2~1
/4I is 1.3/4- to 1/2 is 2.874 or more is 8, and a code sign is assigned to l. The line segment of the first stroke is first in the direction 〆, θ=5. l = 3° then direction →, θ-0, 1 = 3
Therefore, the image feature is set to C04°. The second stroke l is θ-
6. β-2, and the image feature is COI. By combining such image features according to their order of occurrence, characters, etc. can be recognized.

[8] Problems with the Prior Art Although real-time reading devices for characters, symbols, etc. (hereinafter referred to as characters, etc.) based on stroke order recognition are excellent devices, it may be difficult to recognize some characters, etc.

For example, if the inclination θ of the line segment is about 22.5 degrees to the right from the vertical line as shown in Figure 5 (a) and the number "1°" is written, depending on the movement of the child, the slope θ of Figure 5 (b) In other words, the slope θ between points 50 and 51 is a little smaller than 22.5 degrees and belongs to the direction division θ-6, but the slope between points 51 and 52
In between, θ>22.5 degrees and direction division θ-5, and the value of θ also differs in subsequent line segments.

In this way, if different line segment slope divisions are detected even in the same stroke, it becomes impossible to extract the image features.

(4) Purpose of the Invention The present invention solves the above-mentioned problems of the prior art, and allows characters, symbols, etc. written with a pen on a tablet to be recognized without going through the steps of complicated character recognition software for general-purpose computers.
It is an object of the present invention to provide an improved character recognition device that recognizes characters from the image characteristics of line segments based on the order of strokes.

(5) Structure of the Invention The first requirement of the present invention is a means for correcting the directional division based on the slope of a plurality of line segments and clarifying both characteristics, and the second requirement is to correspond to the plurality of directional divisions. This is a means of adding the lengths of each line segment.

The configuration will be explained using FIG. 6.

That is, in the present invention, by providing a line segment direction correction circuit 33 and a line segment length correction circuit 34, the given θ and t are corrected as shown in FIG. 5(b), and the image characteristics are simplified. It is. The details are explained below.

The line segment direction correction circuit 33 determines whether the directions of the line segment can be considered to be the same from the front and back of the line segment direction division, and if the directions are the same, corrects the direction of the line segment, for example. As shown in Fig. 5(b), the direction division θ-61 from point 50 to point 51, the direction division θ-52 from point 51 to point 52, point 5
In the case of direction segment θ-6 from point 2 to point 53, these can be considered to have the same line segment slope, so point 51
By correcting θ from 5 to 6 to point 52, the value from point 50 to point 53 becomes θ-6. By performing such correction, the first point 50 to the last point 54. can be treated as θ-6, making extraction of both features accurate and easy.

Next, the line segment direction correction circuit 83 and the line segment length correction circuit 34 will be explained using FIG. 6. Line segment direction correction circuit 38
is composed of a register 35 and a line segment direction determination circuit 5-36. The line segment direction classification θ is inputted to the register 35 from the line segment direction register 20, and stores θ that is classified one time before θ stored in the line segment direction register 20.

Furthermore, the line segment direction determination circuit 36 stores θ that is one time before the θ stored in the register 35, so the line segment direction determination circuit 86 stores θ eight times in a row.
Verify the divided θ.

Here, in order to correct θ, only line segment direction divisions having similar θθ values are targeted. For example, if there are 8 categories as shown in Figure 2, the θθ values differ by only one category, such as 0 and 1, 0, 7.1, 0.1, 2.2, 1.2, and 3. It is. Furthermore, only the values that are the same before and after θ to be corrected are corrected. In FIG. 7(a), θ-5 is corrected to θ=6, but in FIG. 7(b), θ=5. Fifth. In (C), θ-6 is not corrected. Therefore, correction is performed on θ only when the following conditions are met.

(1) When the direction division of a line segment differs from the divisions before and after it by one division. and -〇- (2) When the value before and after one half of the line to be corrected is the same.

Next, the line segment circuit length correction circuit 34. As shown in FIG.
It is composed of. Register 37 is distance addition circuit 2
This stores the previous line segment length value sent from 3.
The line segment length corresponding to the line segment direction division θ stored in the register 35 is stored. When correcting θ, as mentioned above, we have explained that the section θ゛ sandwiched between the front and back is corrected, but in Fig. 8(a), rather than correcting θ-6 to θ-5,
It is better to correct the front and rear θ-5 to θ-6 and correct the whole as a straight line with an inclination θ-6. Also, in the example of Fig. 8 (1)), the second θ-5 should be corrected. It is better to correct it to a straight line of θ-6. For this reason, the line segment length determination circuit 38 compares the line segment lengths of the divisions to determine which of the divisions to be corrected is better, and uses the longer one as the division value, and the line segment direction determination circuit 86 send to Next, when correcting the line segment direction division θ, the line segment length addition circuit 39 adds the contents of the distance addition circuit 23, the register 37, and the line segment length addition circuit 39, so that the line including the division to be corrected is The total length is 1 minute.

- When a line segment in a character is not corrected or the end of a - stroke occurs, the line segment length addition circuit 39 sends the data to the line segment length division circuit 26, where the length of the line segment is divided and the stroke ends. At the same time, the segment θ of the line segment direction is sent to the image feature storage device 27 from the line segment direction determination circuit 86. If the line segment direction classification is not corrected, the line segment direction determination circuit 36 and line segment length addition circuit 39 send the line segment direction determination circuit 36 and line segment length addition circuit 39 to the image feature storage device 27 without being corrected. Moreover, if the line segment direction detection circuit 19 divides the direction into 16 divisions as shown in FIG. Be accurate.

(6) Embodiment Using FIG. 10, an improved means for extracting image features of character line segments by correcting multiple line segment direction divisions and adding corresponding line segment lengths, which is a feature of the invention of the present application, is explained. An example of a character recognition device having the following will be described.

Tablet 1, paper 2, stylus pen 3, coordinate reading circuit 4, character 9 separator, detection circuit '57, character coordinate memory 6, XY maximum/minimum detection circuit 7, character size calculation circuit 8, normalization circuit 9.2 coordinate register (t) 10,
Z coordinate register (t-1)il, 2-coordinate differentiation circuit 12,
X coordinate register (t) 13, X coordinate register (t-1)
14, X coordinate difference circuit 15, Y coordinate register (t) 16
, Y coordinate register (t-1) 17, Y coordinate difference and relative position detection circuit 25, line segment length division circuit 26, both feature storage circuit 27, register 28, AND gate 29, ROM 30,
The register 81, the AND gate 82, etc. have a configuration used in a conventional real-time reading device for characters, etc. based on stroke order recognition, and are also used in the invention of Japanese Patent Application No. 57-1.85839, for example. However, this will be explained in detail below.

A sheet of paper 2 is placed on a tablet 1 and characters are written using a stylus pen 3. Before starting to write one character, the initial position of the pen is detected on three coordinate axes, X, Y, and 2, using the coordinate reading section 9-4.

The character delimiter detection circuit 5 detects the delimitation of one character.
When the character begins to be written, that is, the pen 3 touches the paper 2 and Z=
The coordinate value of the pen from the time it becomes 1 to the time when the pen finishes writing one character and leaves the paper and becomes 2 = 0 is x,
The three axes of y and z are detected and this value is stored in the coordinate memory 6.
to be memorized. Next, the coordinate on either the X or Y axis is Z=
1, that is, when it is detected that the distance on the paper is greater than the value consisting of the previous coordinates, it is assumed that one character has been written. For example, if the character size is 10mm, it will be 15m.
The writing of the character ends when a coordinate with respect to the X-axis or Y-axis that is more than m away is detected.

As described above, the character coordinate memory 6 takes in and stores the coordinates of the X, Y, and Z axes from the beginning to the end of one character. The x, y maximum and minimum detection circuit 7 calculates the maximum and minimum values of the X and Y axes from the coordinates of one character stored in the character coordinate memory 6 and sends them to the character size calculation circuit 8. Character size calculation circuit 10-8 calculates the character size from the maximum value Xmax and minimum value Xm1n on the X axis, and the maximum value Ymax and minimum value Ymin on the Y axis. That is, Xmax-Xmin=A and Ymax
If -Ymin -B is set, A and B are the character sizes in the X-axis direction and the Y-axis direction, respectively.

The normalization circuit 9 performs coordinate transformation to make the size of the written character constant. For example, the size of the character is changed to the size of a square c in the X-axis direction and C in the y-axis direction. When converting to Xt, the coordinates of the character coordinate memory 6 are
Yt, and the corresponding coordinates normalized by the normalization circuit 9 are x, yt, 'x', = (Xt - Xmin
)%, YI/-(YtYmin 4. By performing such processing, even if the sizes of written characters vary, they are converted to a constant size, and the absolute coordinates of each character are Converted to relative coordinates.2 coordinate register 10.X coordinate register 13.Y coordinate register 16
stores the z, x, and y coordinates normalized by the normalization circuit 9. The three registers 10, 18 .
and 16 immediately before storing the normalized coordinates at time t, the contents, that is, the normalized coordinates at time t-1, are stored in 11.14. , and 17.

The two-coordinate machine circuit 12 checks whether the Z-axis has changed from the contents of the Z-coordinate register 10.11. For example, this is a circuit for detecting whether Z=1 changes to 2=0, or from 2=0 to Z=1, or whether Z=0 remains unchanged.

The X coordinate difference circuit 15 calculates the amount of change a in the X coordinate between two sampled points, and the X coordinate register 13
, 14. Letting the contents of Xl and X'L- be expressed by the following equation.

a = X' (Calculate X't-1.

Similarly, the Y coordinate difference circuit 18 is the Y coordinate register 16°17.
Letting the contents of Y'L and Y't-t, the amount of change in the Y-axis direction is calculated by the following equation.

b = y't-y't-1 The line segment direction detection circuit 19 determines the direction of the line segment between the two sampled points, that is, the direction of movement of the pen, and calculates θ from tan θ −− ]. Ru.

Here, in order to simplify the recognition method, it is assumed that the pen movement angle θ is divided into eight types as shown in FIG. 2, for example. At this time, it is possible to classify more accurately, such as 16 types, 32 types, etc., but in the embodiment, a case of 8 types will be described below.

Next, the division value of θ from the line segment direction detection circuit 19 is sent to the line segment direction register 20.

Before the line segment direction detection register 20 receives the division value of θ from the line segment direction detection circuit 19, the contents of the line segment direction detection register 20 up to that point are sent to the direction differentiation circuit 24. You can know the change state of the value.

The line segment length detection circuit 21 calculates the length t of a line segment between two points as follows.

-18 = , = (777 When the length t of the line segment is less than a certain value e (e is a constant determined from the character size C), the comparator circuit 22 determines that it is a minute movement due to hand movement, etc., and ignores it. In this case, there are two coordinate registers 11, X coordinate register 14, and Y coordinate register 17.
Leave the contents as they are and set the normalized coordinates of the next sampling point in the 2-coordinate register 10. X coordinate register 18. Set the Y coordinate register 16.

- When the length t of the line segment is 0 or more, the length t of the line segment is sent to the distance addition circuit 23. The direction differentiation circuit 24 calculates the line segment based on the segment direction θ divisions sent from the line segment direction detection circuit 19 and the segment direction register 20 which is the division value of the line segment direction θ obtained at the previous sampling point. Determine whether the directions are the same or different. If the division values of θ are the same, a signal is sent to the distance addition circuit 23, and the length t of the line segment sent from the comparison circuit 22 is added. If the division values of θ are different, a different signal is sent to the distance addition circuit 23, and the added value of the line segment 14-length t added to the distance addition circuit 23 is sent to the line segment length division circuit 26. With,
The length t of the line segment sent from the comparison circuit 22 is newly set in the distance addition circuit 22.

As a result, when the same division value of θ continues,
Since the total length of the line segments is obtained, it indicates the length of the line segments until the segment value of θ changes. The line segment length dividing circuit 26 divides the added value of the line segment length t sent from the distance adding circuit 23 into a plurality of values.

For example, the normalized character size C is 4 and 4 degrees. to simplify subsequent processing.

Next, image feature extraction will be explained.

One car is from Z=1 to z-0, and consists of a read-only storage device 27 that extracts features for each stroke and a register 28, and stores the contents of the register 28, the contents of the line segment direction register 20, and the line segment length. The content output from the division circuit 26 is set to an address (
The address is read from the read-only storage device 27 and set in the register 28. A read-only memory (ROM) 27 is generated at the timing of the output signal from the AND gate 29 between the output signal of the direction differentiating circuit 24 and the timing signal CKI, which is output when the direction of the line segment changes.
Since it is read from and set to register 28,
It is read when the direction of the line segment changes. Further, since the contents of the register 28 are also read out from the read-only storage device 27 as an address, the state in which the line segment direction has changed is set in the register 28. Also z-
When the value is O, the contents of the register 28 are sent to recognition, and then the contents of the register 28 are cleared.

For example, the case where the number 114.11 is written as shown in FIG. 8 will be explained.

The direction of the line segment is divided into 8 types in Fig. 2 and indicated by θ, and the length of the line segment is determined by dividing the character size C into 4 equal parts: 1 11 3
1 3 - - - - - 1 - - - - - 2.7 or more - 4 2
4, denoted by t as 4 2 3.

Fig. 4 shows an explanation of this case, in which the first stroke becomes 2=0 once, and the second stroke becomes 1.

In the first stroke, the line is 1 (direction/θ-5, length t=3)
Then, it becomes -(direction → θ-O, Z=a) and one stroke is completed.

The feature classification of this L is 004.

In the second stroke, there is only 1 (~-6, t-2), and similarly C
oIO image features.

Furthermore, recognition is performed based on both features.

Recognition consists of a read-only storage device 30 and a register 81, and is performed at the timing of the output of the AND gate 32 between the output signal of the Z coordinate differentiating circuit 12 and the clock signal CK2, which is output when Z=1 changes to 2=0. Read-only storage device 30
The read register 31 is set from. In the example of FIG. 3, both features CO4 of the first image are input to the read-only storage device 30, read out, and set in the register 31. Next, from the image feature COI of the second image and the content read out for the first image set in the register 31, further reading is performed, and COI becomes C04.
The result of the combination with I is read out, and this becomes the recognition barrier.

The features of the present invention are a line segment direction correction circuit 88 and a line segment length division correction circuit 34, the former consisting of a register 35 and a line segment direction determination circuit 36, and the latter consisting of a register 37 and a line segment length determination circuit 38. It consists of 17- By adopting such a configuration, for example, the fifth
By correcting θ and t, which are divided into a plurality of sections as shown in FIG. 3B, image features of character line segments can be easily extracted, and recognition accuracy can be improved by reducing noise.

Next, the procedures for extracting and recognizing both features will be described, and there will be no problem in applying the conventional techniques. In other words, one stroke is from Z=1 to 2=0, and it consists of a read-only storage device 27 that extracts features for each stroke, and a register 28, and the contents of the register 28, the contents of the line segment direction register 20, and the line segment The content output from the long division circuit 26 is an address (address).
As such, the data is read from the read-only storage device 27 and set in the register 28. The output signal from the direction differentiating circuit 24, which is output when the direction of the line segment changes, and the output signal from the AND gate 29 with the timing signal CKI are read out from the read-only memory (ROM) 27 and stored in the register 28. Since the line segment is twisted, it is read when the direction of the line segment changes. Further, since the contents of the register 28 are read from the read-only storage device 27 using the address 18-, the state in which the line segment direction has changed is set in the register 28. When z-0, the contents of the register 28 are sent to recognition, and then the contents of the register 28 are cleared.

For example, as shown in FIG. 3, the number ++4. The case where ++ is written will be explained.

The direction of the line segment is divided into 8 types as shown in Figure 2 and indicated by θ, and the length of the line segment is determined by dividing the character size C into 4 equal parts, and 0 for less than 1.
, -~- is 1, main~↓ is 2.1 or more is 1 11 4 2 4, 4. 2 4°3 and denoted by l.

FIG. 4 shows an explanation of this case, where the first stroke is 1 degree z-0 and the second stroke is 1 degree.

In the first stroke, the line is first] (direction θ-5, length t-3)
Then, -(direction→θ-0, t=8) is set, one stroke is completed, and the feature classification of this L is ended as CO4.

In the second stroke, there is only l(θ-6, t-2), and similarly C
This is an image characteristic of OI.

Furthermore, recognition is performed based on both features.

Recognition consists of a read-only storage device 30 and a register 31, and is performed at the timing of the output of the AND gate 82 between the output signal of the Z coordinate differentiating circuit 12 and the clock signal CK2, which is output when Z=1 changes to 2=0. Read-only storage device 30
and set it in the read register 31. In the example of FIG. 3, both features CO4 of the first image are input to the read-only storage device 30, read out, and set in the register 31. Next, from the image feature COI of the second image and the content read out for the first image set in the register 81, further reading is performed, resulting in COI, COI, and COI.
The result of the combination with I is read out, and this becomes the recognition result.

■Effects When writing characters on a tablet and recognizing online characters, it was previously very difficult to draw features because the pen moved up, down, left and right due to hand shake, etc., but the correction method of the present invention has made it easier. Accurate feature delineation becomes possible, and an inexpensive and high-performance character recognition device can be implemented.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the configuration of a conventional character recognition device. FIG. 2 is a diagram showing directional divisions of line segments. FIG. 3 is a diagram illustrating characters written on the tablet. FIG. 4 (a) is an explanatory diagram of image feature extraction in the first image, (1)) is an explanatory diagram of image feature extraction in the second image, and FIG. 5 is a diagram showing problems in the conventional technique. FIG. 6 is a diagram showing an embodiment of the character recognition device of the invention, showing the configuration of a line segment direction correction circuit and a line segment length division correction circuit of the invention. 1. Tablet 2, Paper 3, Stylus pen 4, Coordinate reading circuit 5, Character delimiter detection circuit 6, Character coordinate memory 7, X, Y maximum/minimum detection circuit 8, Character size calculation circuit 9, Normalization circuit 10, Z coordinate register (1) 21- 11.2 Coordinate register (1-1,) 12, 2-coordinate machine division circuit 13, X-coordinate register (1) 14, X-coordinate register (t-1) 15, X Coordinate difference circuit 16, Y coordinate register (1) 17, Y coordinate register (t-1) 18, Y coordinate difference circuit 19, line segment direction detection circuit 20, line segment direction register 21, line segment length detection circuit 22, comparison circuit 23, distance addition circuit 24, direction differentiation circuit 25, relative position detection circuit 26, line segment length classification circuit 27, image feature storage circuit 28, register 29, AND gate 30, read-only storage devices 22-31, Shi Ji Staro 2. AND gate 33, line segment direction correction circuit 34, line segment length correction circuit 35, register 86, line segment direction determination circuit 87, register 38, line segment length determination circuit 39, line segment length addition circuit 23 - 527 1 year old

Claims (1)

[Claims] (1) Capture the XYz coordinates of the pen at regular intervals, detect the direction and length of the line segment from the XY coordinates, divide each line segment into multiple parts, and depict the characteristics from the order of occurrence of the direction and length divisions of the line segment. and a means for correcting the direction division of the line segment into one line segment direction division representing the direction division of the plurality of line segments from before and after the direction division of the line segment in the character recognition device; A character recognition device characterized by comprising means for adding the lengths of corresponding line segments.