JPS596420B2 - Online character feature extraction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for extracting characteristics of characters written on a writing board by analyzing the movement of the pen point of the characters in order to identify the characters.

Character identification features in conventional character recognition methods include the following.

Classification of entry area: XY of characters converted to XY coordinate series
Detect the extreme values of both coordinates and use that frame as a frame surrounding the character.
It is divided into regions and characterized by which region the starting point or ending point is located in.

Number of strokes: A stroke is defined as a figure drawn by the pen from the state where the writing pen is "attached" to the paper surface until the state where the writing pen is "removed" from the paper surface, and the shape is written within one character writing area. Count the number of strokes.

Number of X peaks, number of Y peaks: If the coordinate values increase and then decrease over a certain distance, one peak is present.

Matching of two points: A match or mismatch is determined based on whether the distance between two points is within a certain value regarding the starting point, end point, minimum value or maximum value.

X starting point direction: Sample the coordinate value, and if there is no decrease and increase within 3 samples, then +, otherwise -.

Stroke direction: The direction of the stroke is classified into eight directions regarding the start point and end point.

By collecting the above various features and predetermining what value each feature has for standard characters with various shapes and stroke orders,
These were organized into a tree-like structure to identify characters.

However, these characteristics focus on (1) the shape of the start and end points of the stroke, and the characteristics due to the shape of the entire stroke, such as rotation of the stroke, sudden changes in direction, curved shapes and straight lines. (2) If the characters are tilted or small, the X and Y peaks may not be detected properly.

Therefore, in the conventional method, even if a function for detecting various features is provided, the recognition rate does not improve, and the recognition rate cannot be expected to improve with respect to deformation or variation of characters. One side,
There is a method of identifying characters by detecting as a feature a direction signal series generated from changes in the direction of the tangents of each point and the distance of movement of the stroke shape.

This divides the direction of the tangent line by 16 and rotates counterclockwise by 0~15
The algebraic sum of the quantized coefficients within a certain distance is compared with a fixed value, and the direction is increased or decreased (counterclockwise rotation, clockwise rotation), and the output signal with no change is combined with the attachment/detachment signal of the pen on the paper surface. It is characterized by In the embodiment using this method, the part for calculating the tangent direction quantization system and the algebraic sum, the increase/decrease in direction, and the output signals with no change are mechanically generated. Therefore, (1) the speed of signal output is , the accuracy is subject to mechanical constraints, and (2) since only the change in direction is focused, it is not possible to distinguish between 0 and 6, so O must be set as φ. Therefore, an object of the present invention is to improve the above-mentioned drawbacks of the conventional technology, and to approximately classify each stroke constituting a character by a predetermined elementary feature shape consisting of a circle, a curved line, and a straight line shape, The purpose is to extract each stroke as one of the elementary features or a series of them.

For this purpose, in the present invention, three coordinate points are taken out by sequentially shifting one point at a time from the coordinate point series of one stroke, and the product of the direction change of the two vectors given by each three points and the length of one vector is calculated. A process is provided to detect the amount of change. Then, it is divided at coordinate points corresponding to the amount of change larger than a predetermined value, and if there is no such coordinate point and the stroke is not a circular feature, it is divided by a length determined according to the total length of the stroke. A process of dividing and partial stroke is provided. Then, for each partial stroke,
Detect the ratio between the total amount of change and the total length of vectors. Then, by comparing the ratio with a value corresponding to the average length of the vector and detecting a magnitude relationship, it is determined whether the shape is a straight line or a curve. Then, based on information such as the starting and ending point direction of the partial stroke, including this comparison result,
Each partial stroke is made to approximately correspond to a predetermined elementary feature shape such as a circle, a curved line, or a straight line shape. According to the present invention, since partial strokes are set, relatively detailed features can be extracted, and since elementary feature shapes are classified and set in advance, the shapes throughout the entire stroke are relatively limited. It can be extracted using an information code. Next, the detection of the elementary features themselves and the procedure for detecting the elementary features will be explained using the drawings. As shown in FIG. 1, the position information of the figure drawn on the paper on the writing board is converted into XY coordinate information, and after passing through a filter, it is expressed as an XY coordinate point series.

Note that PO is the character entry starting point Pn is the ending point. Figure 2 shows the coordinate point P in Figure 1. - An enlarged version of P5, PO-P
When tying 5 as shown, P. - Determine whether the line segment defined by P5 is a straight line or a curved line using the following method. vector F7 de,
The direction change of the vector lower 7F2 for
8), and the length of the F line is 11, the amount of change d1 of P72 is defined as D, = a1 x 11 (a1 is the quantization coefficient), and this is expressed as each vector F Ask about F mouth. Dl, d2d3, d4dl obtained in this way
Based on +D2+D3+D4, D=? , and compare it with a constant value.

', +12+13+14=L, when L≧THO, if 1DI≧THl, it is a curve, and if Dl<THl, it is a straight line. THO is constant and usually l of one stroke
Set the length to /3 to l/4. THl is a constant value determined by the average length l of 1h. The straight lines and curves determined in this way are classified into eight directions by connecting their starting points and ending points as shown in Figure 4, and the straight lines, curves, and circles in each direction are classified into eight directions as shown in Figure 5. Give the code number as shown in .

The straight lines and curved lines indicated by these code numbers are called elementary features. Next, the procedure for detecting elementary features will be explained in detail. (1) Check whether there is a point Pi that satisfies 1diI≧TH3 within d1 to Dn calculated within one stroke of coordinate points PO-Pn. TH3 is a constant value determined by 1h. (
2) If Pi exists, divide the stroke by Pi and PO
Elementary features are detected for each partial stroke -Pi, Pi+&~Pn.

(3) If there is no Pi, check whether the PO-PnOl stroke is a circle and if a curve is detected in 2, check whether it is a circle.

The conditions are THl<TH2≦D, Q≦TH4. Here, TH2 and TH4 are constant values determined in 1 hour. Q is calculated as Q='v-4-, where y and y are the line segment lengths connecting Px and PX+6 in Figure 6.
ρv+. Ki...ki 1v+.

(4) If the curve detected in 2 is not a circle, register it as a curve.

If Pi is not detected, add l from the starting point and go to L
When ≧THO is satisfied, 1D! ≧THl determines whether it is a straight line or a curve, and the elementary feature code is detected and registered based on the direction of the start and end points. (5) Regarding the remaining part of the stroke, the number of coordinate points should be 2 or more, taking into consideration the spring of the character etc.
When O, register the line element feature code.

Based on the principle of the present invention explained above, FIGS.
An embodiment of the character feature extraction method according to the present invention will be described with reference to the drawings.

In FIG. 7, storage circuits 700, 704, 705, and 710 are constituted by RAM (random access memory).

701 is a direction quantization circuit, 702 is a distance calculation circuit, 70
3 is a multiplication circuit, 706 is an addition circuit, 707 is an averaging circuit, 708 is a comparison circuit, 709 is a control circuit that controls the whole, S7OO is a coordinate signal input line, S7Ol is a constant value input line, S7O2 and S7O3 are storage circuits. data output line,
S7O4 is the status signal line of the comparison circuit 708, S7O5, S
7O8 is a start point coordinate signal line, S7O6 and S7O7 are end point coordinate signal lines, S7O9 is an output line of the adding circuit, S71O is an averaging circuit output line, 8711, S712, and S713 are data output lines of the memory circuit 700, and each Point coordinates are output.

S7l4 is a direction quantization output line, S7l5 is a distance value output line, S7l6 is a variation output line, S7l7, Sll8, Sr
l9, S72l, S722 are memory circuit control signal lines, S7
2O is a detection signal line, S723 is an elementary feature division detection signal line,
S724 is an elementary feature determination circuit control signal line, and S725 is a coordinate point signal line. With the above configuration, the coordinate position signal is stored in the storage circuit 700 (RAM) under the control of the control circuit 709.
random access memory).

The written coordinate position information is read out three points at a time under the control of the control circuit 709, and is entered into the direction quantization circuit 701 and the distance calculation circuit 702 for calculation. The direction quantized value and the vector length are output to output lines S7l4 and S7l5, respectively, and input to a multiplication circuit 703, where the amount of change d is calculated. The length of the vector is also stored in the storage circuit 705 by the output line S7l5.
The signal is input to an adding circuit 706 and an averaging circuit 707.

The storage circuit 705 stores the length of the vector under the control of the control circuit 709. The addition circuit 706 calculates the addition of vector lengths, that is, the total length, and the average value circuit 707 calculates the average 1h of distance values. The amount of change d is the signal line S7l6
At the same time, the signal is outputted to the memory circuit (RAM) 704 and is compared with the constant value TH3 input by the constant value input line S7Ol in the comparator 708. As a result, Di≧T
If it is H3, the state is output to the state signal line S7O4.
At the same time, the detection signal is output to the signal line S72O to notify the control cape of the detection, and the memory circuit (RAM)
Starting point and ending point at 710. Temporarily store coordinates. Although not shown in the diagram, the control circuit 709 is capable of detecting coordinate points in the memory circuit 700, and transfers start and end point signals regarding one stroke via signal lines S7O7 and S7O8. However, D
If i≧TH3 is detected, it means that there has been a sudden change during the stroke at this point. Therefore, in order to divide the stroke into partial strokes at this point, the control circuit 709 determines the start point and end point coordinates of this partial stroke. The data is stored in the storage circuit 710 via the. In FIG. 8, 801 and 802 are adder circuits, 803 is a division circuit, 804 is a 4-input, 2-output switch, 805 is a comparison circuit, 806 is an AND gate, and 807
808 is an inverter, 808 is a constant value generation circuit, 809 is an elementary feature determination circuit, 801 and 811 are storage circuits, and 812 is a discriminative feature determination circuit. S8OO, OOn chant'-F pleasure QQnri QOnri is the start and end point coordinate output line,
S8O4, S8O5 are constant value output lines, S8O6 is constant value input lines, S8O7 is elementary feature code output lines, S8O8, S
8O9 is a memory circuit data output line, S81O is a feature presence/absence output line, and S8ll is a detection signal line. With the above configuration, the operation of this embodiment following FIG. 7 will be described below.

The amount of change d stored in the storage circuit 704 is added to the addition circuit 801.
is added.

The lengths of the vectors stored in the storage circuit 705 are added in the addition circuit 802, and the added output value is input to the comparator 805.
1/n of the total length L of the stroke generated in the dividing circuit 803
is compared with the value THO, and its output is outputted to the signal line S8ll as an elementary feature division detection signal. In order to switch between the start and end point coordinates of one stroke and the start and end point coordinates of a stroke, the switch 804 outputs the start and end point coordinates stored in the storage circuit 710 and the control circuit 709 under the control of the status signal line S7O4. Switch the start and end point coordinates. A constant value generation circuit 808 generates constant values THl and TH2 based on the average length 1h of the distance and coordinate point number information, and outputs constant values THl and TH2 to the elementary feature determination circuit 80 through constant value output lines S8O4 and S8O5.
Enter into 9. AND gate 806 connects the detection signal line S8l
The control gate is used to gate the detection signal of l, and the control gate is opened or closed depending on the operation result of the comparison circuit 708. The elementary feature determination circuit 809 uses addition circuit output lines S8OO (additional change amount) and S8Ol (additional distance), start and end point coordinate output line S8O2
(Start point coordinates and S8O3 (end point coordinates), constant value output line S8O4 (one step value THl) and S8O5 (constant value TH
2) The constant input line S8O6 (constant value TH4) is input, a determination operation is performed by activation of the control signal line S724 of the control unit 709, and the elementary feature code is output to the output line S8O7. The storage circuit 810 stores the elementary feature code via the elementary feature code output line S8O7. Reference numeral 811 denotes a storage circuit in which identification features are stored, and the storage contents are read out under the control of the control unit, and elementary feature codes are read out from the storage circuit 810 according to the contents, and are compared and determined in the identification feature determination circuit 812 to determine the identification characteristics. Existence (1) and absence (0) are characteristic presence/absence output line S8l
Output to O.

Since the method of summing up the identification feature presence/absence signals and outputting a character code from the summing result is a conventional method, the description thereof will be omitted. Table 1 shows examples of standard Lai characters written using elementary feature codes.

In Table 1, / represents a combination of elementary features, and / represents a connection with the next stroke. FIG. 9 shows one type of character identification feature format that detects whether a feature is present (1) or absent (0) from the combination and connection relationship of elementary feature codes specified in advance for each character for detected and registered elementary features.

In FIG. 9, 900 is a feature number, 901, 902, and 903 are designated pits, and 904 and 905 are elementary feature codes. 9
00 is a feature number used for classifying features.

901 specifies in what stroke number an elementary feature code such as 904 or 905 appears.

A field 902 specifies the number in which the elementary feature should appear or should not appear in the specified stroke.

903 specifies the relationship between the elementary features specified in 901 and 902 and the elementary feature to appear next.

The relationships are: (1) the elementary feature specified below must appear next; (2) it may or may not appear; (3) it must not appear next; the last in the stroke. It is possible that

(XX). +(XX)1+... is the designation of elementary features, +
is (XX). represents a logical 0R combination that requires only (XX)1.・represents a combination of the specified elementary feature with the elementary feature that should appear next. + is (XX)′o or (X
Represents a logical 0R combination that only requires either Xyl or Xyl. In addition, for example, if XXO detects as an elementary feature of the X1st partial stroke of the XOth stroke in an unknown character, then as the next partial stroke (XX
)'o is detected, it is compared with the character identification feature of each character, and in the case of the character whose character identification feature includes the feature number in Figure 9, a C comparison judgment is made in the identification feature of the feature number FXXX. , in this case, there is a match with the feature code shown by 904, and the combination of the next elementary feature also satisfies the combination condition with the feature code 905, so for that character, (XXO
) are output as having elementary features, and characters are identified by aggregating them. In the embodiment described above, the direction of direction change is equally divided into 16 as shown in FIG. 3, but it is also possible to divide the frequency into 20 to 32 by non-uniform division depending on the type of character figure. .

Furthermore, the quantization coefficients may not be determined as shown in FIG. 3, but may be set appropriately. The directional frequency division value, quantization coefficient value, etc. should be selected depending on the character type and size to be identified. In any case, the curvature can be approximated and classified into straight lines and curved lines. In the example, the coordinate point series is 3 points P
F[] for the vector G due to O, Pl, P2 This direction was determined, but the direction of the vector was determined in advance as shown in Figure 3, and from F'7 lower ゛1 (d-F77 A similar effect is produced by determining the directional quantization value.

Here, α and β are vectors P7l, F and this direction (exist)
~7, -1~-8), γ indicates the amount of direction change with respect to the vector F (IO)α. In the example, 8 types of straight lines, 16 types of curves, and 1 type of circle are detected from the directions divided into 8 directions as shown in Fig. 4. However, depending on the character, the straight line portions may be divided into even smaller parts. It is possible to increase the number of straight line elementary features, or to introduce two types of long and short divisions into the length of a straight line, and use them as elementary features.On the other hand, it is possible to configure discriminating features by combining curved elementary features that are roughly approximated.

In any case, by time-sequential combination of elementary features of straight lines, curves, and circles, identification features can be constructed and characters can be identified.O As explained above, according to the present invention, 3
Since the curvature is approximated from the product of the change in the direction of a point and the distance, and a curve is distinguished from a straight line, the conventional method calculates the curvature from the calculation of a slight change in the tangent angle direction of the figure and a slight change in the distance, and distinguishes between a curve and a straight line. Unlike , curves and straight lines can be distinguished in a relatively simple way, and features such as the stroke shape direction of the start and end points and the rotation of the stroke can be used as distinguishing features of characters by time-series combination of elementary features. By absorbing character changes and fluctuations using the logical sum of elemental feature combinations, the identification rate is improved, and further features such as the number of strokes, start/end point stroke shape direction, and Stokker rotation can be handled uniformly as identification features. By proposing a detection form, it is only necessary to configure one type of feature detection circuit. Therefore, according to the present invention, it is possible to obtain an economically effective recognition device.

In addition, the present invention uses numbers (0 to
9), alphanumeric characters α to Z), and some other symbols can be used for real-time recognition.

[Brief explanation of the drawing]

Figure 1 is a diagram showing a character figure expressed in an XY coordinate series,
Figure 2 shows the coordinate point P in Figure 1. - An enlarged view of P5, Fig. 3 is an explanatory diagram of the directional quantization coefficient, Fig. 4 is an explanatory diagram of the direction of the elementary feature code, Fig. 5 is an explanatory diagram of the elementary feature code and the elementary feature shape, Fig. 6 is an explanatory diagram for determining whether a curve is a circle or not; FIGS. 7 and 8 are block diagrams showing an embodiment of the online character feature extraction method according to the present invention; and FIG. 9 is an explanatory diagram of character identification features. It is. 700, 704, 705, 710: Memory circuit, 70
1: Direction quantization circuit, 702: Distance calculation circuit, 703:
Multiplication circuit, 706: Addition circuit, 707: Averaging circuit, 7
08: Comparison circuit, 709: Control circuit, S7OO: Coordinate signal input line, S7Ol: - Fixed value input signal line, S7O2,S
7O3: Memory circuit data output line, S7O4: Status signal line, S7O5, S7O8: Start point coordinate output line, SlO6, S
7O7: End point coordinate output line, S7O9: Addition circuit output line,
S7lO: Averaging circuit output line, S7ll, S7l2, S
7l3: point coordinate output line, S7l4: direction quantization output line,
S7l5: Distance output line, S7l6: Change amount output line, S7
l7, S7l8, S7l9, S72l, S722, S7
24: Memory circuit control signal line, S72O: Detection signal line, S
723: Elementary feature division detection signal line, S725: Coordinate point number output line, 801, 802: Addition circuit, 803: Division circuit,
804: Switching circuit, 805: Comparator, 806: AND circuit, 807: Inverter, 808: Fixed value generation circuit, 8
09. Elementary feature determination circuit, 810, 811: Memory circuit, 8
12: Discrimination feature determination circuit, S8OO, S8Ol: Addition circuit output line, S8O2, S8O3: Start and end point coordinate signals, S
8O4, S8O5, S8O6: - constant signal line, S8O7:
Elementary feature code output line, S8O8, S8O9: memory circuit data output line, S8lO: feature presence/absence output line.

Claims (1)

[Claims] 1 The pen point movement of writing characters on the writing board is detected for each stroke as a coordinate point series of XY coordinates, and one stroke or a partial stroke obtained by dividing it is divided into predetermined elementary characteristic shapes of circles, diameters, and curves. Approximate to
In an online character feature extraction method that extracts one elementary feature or a series of multiple elementary features for each stroke, (a) one stroke is sequentially extracted from the coordinate point series of one stroke.
A process in which the coordinate points are taken out three by three by shifting the points, and the product of the direction change (a) of the two vectors given by each three points and the length (l) of one vector is detected as the amount of change (d). and (b) the amount of change larger than the predetermined value (
Divide each stroke at the coordinate point corresponding to TH_3) and set partial strokes, and if each stroke does not have a change amount larger than the predetermined value (TH_3) and it is not a circle element feature, The process of setting partial strokes by dividing each stroke by a length determined according to the total length of the stroke, and (c) the ratio of the sum of the amount of change of each partial stroke to the sum of the length of the vector (D) (d) a process of comparing and detecting the magnitude relationship between a value determined according to the average vector length (ln) in the partial stroke and the ratio (D) in the partial stroke; e) Extracting one elementary feature or a series thereof from each stroke by making each partial stroke approximately correspond to a predetermined elementary feature shape of a circle, a curve, and a straight line, based on information including the comparison results. An online character feature extraction method featuring