JPS6329893A - Hand-written character recognizing system - Google Patents

Abstract

PURPOSE:To attain sure character recognition even if characters with optional size are entered in optional positions on a tablet by automatically segmenting a character or mark written in an optional area of the tablet with optional size by moving an input pen, and executing character recognizing operation. CONSTITUTION:The coordinate position of an input pen 1 is temporarily stored in a position memory 4, a square virtual frame coordinates of the initial pen- down of the input pen which are drawn around the coordinate axes is stored in a virtual frame coordinate memory 5, and when the input pen position is projected from the virtual frame coordinates by moving the input pen 1, the expansion of the virtual frame coordinates is instructed to the memory 5. When the pen position based on a 2nd pen-down is projected from the virtual frame, the end of a 1st hand-written character is instructed to a CPU by an algorithm instruction memory 6. The CPU compares the 1st hand-written character with a reference character previously stored in a dictionary memory 7 to recognize the hand-written character.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a handwritten character recognition method for recognizing handwritten characters written on a tablet.
The present invention relates to a character recognition method that can recognize handwritten characters at any position on a tablet without being affected by their size.

[Technical background of the invention]

A method has been proposed in which characters are recognized by writing on a tablet with an input pen and sequentially detecting the coordinate position of the input pen to read the strokes and shapes. This method is widely used, especially when inputting kanji with many types of characters, because it is sufficient to simply write on the tablet with an input pen. On the other hand, in characters and symbols such as kanji, flat names, katakana, English letters, and numbers, for example, ``dog'', ichidai'', ``u''
,"oh".

There are characters such as ``u'', ``shiJ, riJ, rjJ'' that have a dot or short horizontal bar in the main part of one letter or symbol and in a part slightly away from it.Such characters In the case of , if the dots are too far apart, for example, in the case of ``u'', there is a possibility that it will be identified as ``tsu'' and an apolostrophe. Furthermore, the sentences contain a mixture of muraku, si, semi-voiced marks, punctuation marks, periods, and commas.

In order to identify these various characters, grids are printed integrally on the surface of the tablet, or manuscript paper with grids is placed on the tablet, and each character is written in the grid in order so that it does not overflow. In addition, when writing a plain name, katakana, etc., I use the function keys on the keyboard to switch between them.

However, writing each letter carefully so that it does not protrude from the grid is different from the feeling of writing on a regular piece of paper, and is extremely unnatural and unusable for the operator.

[Purpose of the invention]

An object of the present invention is to provide a handwritten character recognition method that can reliably recognize characters no matter where they are written on a tablet and at what size.

[Structure of the invention]

The handwritten character recognition method of the present invention temporarily stores the coordinate position of the input pen in a position memory, stores the coordinates of a virtual frame of a square centered on the pen-down coordinate axis of the first input pen in the virtual frame coordinate memory, When the input pen deviates from the virtual frame coordinates due to movement, the virtual frame coordinate memory is instructed to enlarge the virtual frame coordinates, and when the pen-down coordinate position of the second input pen deviates from the virtual frame, the first The algorithm instruction memory instructs the central processing unit to finish the first handwritten character, and the central processing unit compares and matches the first handwritten character with standard characters prestored in the dictionary memory, thereby completing the handwritten character. >-z2.

[For creation]

For characters written on the input pen of the tablet in any position and in any size using the input pen, first set a virtual frame centered on the pen-down coordinates of the first stroke, and store the frame coordinates in the virtual frame coordinate memory. Remember. The pen coordinates are also stored in position memory. Next, when the input pen moves and exceeds the virtual frame, the virtual frame is enlarged by the amount of the excess by the algorithm command memory. If the pen-up or the second pen-down violates this virtual frame, the writing before the pen amplifier is considered as one character, and the strokes up to that point are compared with the standard characters stored in the dictionary memory in advance and the central processing unit. Compare and match.

〔Example〕

First, an example of a device implementing the handwritten character recognition method of the present invention will be described with reference to FIG. ing. In this tablet 2, the coordinate position information obtained by the tablet 2 is input to a central processing unit 3 (hereinafter referred to as CPU), which is a main control section and is composed of a microcomputer or the like. Also connected to the CPU 3 are a coordinate position memory 4, a frame coordinate memory 5, an instruction memory 6, and a dictionary memory 7.

The coordinate position memory 4 is for temporarily storing coordinate information supported by the input pen 1, and is composed of a random access memory (RAM) or the like. The frame coordinate memory 5 stores, for example, a first virtual frame of 511I+11 squares on each side and this first virtual frame centered on the coordinates obtained by the first pen down.
Located concentrically surrounding the virtual frame, each side is 10+nm
It is used to store at least eight coordinates for setting a second square virtual frame, and like the coordinate position memory 4, it is composed of a RAM or the like. The command memory 6 stores a frame coordinate memory 5 so as to enlarge the sizes of the first virtual frame and the second virtual frame when the position of the input pen 1 exceeds the first virtual frame.
If the coordinate position of the second pen-down is located outside the second virtual frame created by the first pen-down, the first stroke ends one character, and the handwriting of the character ends. Read-only memory (R
OM), etc. The dictionary memory 7 stores kanji, hiragana, katakana, numbers, alphabetic characters, symbols, etc., and is composed of ROM, etc., like the instruction memory 6. The tablet 2 is located at a height where signals from the input pen 1 are received. We use a device that can be read from both directions (two axial directions). An example of such a tablet will be explained with reference to FIG.

A plurality of elongated magnetic bodies 21 arranged substantially parallel to each other, and a plurality of excitation lines 21a arranged around the plurality of magnetic bodies 21 at predetermined intervals in a direction orthogonal to the longitudinal direction thereof.
~22h, a plurality of magnetic bodies 21 and a plurality of excitation lines 22a
A position detecting unit 24 including a detection wire 23 wound multiple times over almost the entire circumference of ~22h, a position specifying magnetic generator that generates a steady magnetic field, for example, the input pen 1, and a predetermined cycle. a drive current source 25 that generates an alternating current; a signal selection circuit 26 that sequentially switches and applies an alternating current of a predetermined period to each of the excitation lines 22a to 22h; and a signal selection circuit 26 that sequentially extracts the induced voltage induced in the detection line 23. Take these differences,
It is comprised of a position detection circuit 27 that calculates the position where this differential voltage is equal to a predetermined reference voltage, and from this calculates the height of the input pen 1 with respect to the position detection section 24.

The input pen is provided so that one side of the magnetic field of the bar magnet 11 is exposed at its tip.

If any one of the magnetic lines 22a to 22h is driven! When an alternating current (such as a sine wave) is passed from the II current source 25 and the signal selection circuit 26, a magnetic flux (magnetic field) is generated around the alternating current, and an induced voltage is generated in the detection line 23 due to electromagnetic induction due to this magnetic flux. Since this electromagnetic induction is performed via the magnetic body 21, the larger the magnetic permeability μ of the magnetic body 21, the greater the voltage value of the induced voltage. This is expressed by the formula, induced voltage V
is, V=L(d i/d t)=μ・SN2/Q(d i/
d t) (where, S is the magnetic material, the cross-sectional area of 21, N is the number of turns of the detection wire 23, and a is the length of the magnetic material 21).

By the way, the magnetic permeability μ of the magnetic body 21 changes greatly depending on a steady magnetic field (hereinafter referred to as magnetic bias) applied from the outside. The state of the change varies depending on the composition of the magnetic material, the frequency of the alternating current, or whether the magnetic material is subjected to heat treatment or magnetic field treatment, but here, as shown in Figure 3, when a slight magnetic bias is applied, maximum,
It is assumed that the more magnetic bias is applied, the more the magnetic bias is applied.

Therefore, the tip 1 of the input pen 11 containing the bar magnet 11
When a is located above the magnetic body 21, the magnetic flux emitted from the tip 1a of the input pen 1 crosses the magnetic body 21, as shown in FIG. At this time, the magnetic flux is directly under the tip 1a of the magnetic body 21.
The magnetic material 21 is substantially perpendicular to the magnetic material 21 on both sides thereof. The amount of magnetic bias applied to the magnetic body 21 increases as the angle at which the magnetic flux intersects the magnetic body 21 becomes smaller, so it is smallest immediately below the tip 1a of the input pen 1.
It gradually gets bigger as you move away from here.

Therefore, as shown in FIG.
When the tip 1a of the input pen 11 is placed on the input surface 28 formed on the top of the pen 4, the amount of magnetic bias applied to the magnetic body 21 directly under the tip is set to a slight amount of magnetic bias that maximizes magnetic permeability. When set, as shown in Fig. 5, when an alternating current is applied to the excitation line closest to the position where the input pen 1 is placed (specified position), the voltage generated in the detection line 23 is set to a very thick value, and the alternating current is added. A voltage V that gradually decreases as the position of the excitation line moves away from the specified position. -v7 occurs. In FIG. 5, the horizontal axis is the direction along the magnetic body 11 (
Hereinafter, this will be referred to as the X direction. ), and the vertical axis shows the voltage. Here, the coordinate value X. ~x7 is each voltage V. - Excitation line 22 to which alternating current is applied when v7 occurs
The positions of a to 22h are shown. The position detection circuit 27 detects each voltage V.
. -v7 is taken out, and from this the induced voltage becomes the maximum value
By calculating the coordinate values, the coordinate value Xs of the designated position of the input pen 1 can be determined.

There are various ways to obtain the coordinate value Xs by approximating the waveform near the maximum value in Fig. 5 by an appropriate quadratic function and obtaining the X coordinate value of the maximum value of the function.
(The same applies to the minimum value), focusing on the point that the difference is rOJ, create a difference equation for the function, calculate the chain mark position where the difference is "0", and from this, calculate the coordinate value Xs of the maximum value. You can ask for it.

For each detection voltage v0 to v7, the difference (difference in voltage value), that is, (v1 to V, ), (V, -V), (V
, -V2), . . . (V7- vs), the differential voltage value v0. as shown in FIG. v2. v,
,...v7 is obtained. In FIG. 6, the horizontal axis shows the coordinate position in the X direction, and the vertical axis shows the voltage value. however.

Here, the difference numerator OJ is set to a predetermined reference level ■t having a positive value. In the figure, it can be seen that the voltage value crosses the reference VT level between ■ and ■4, and the difference between them is "0".

Further, even if the input pen 1 is moved along the direction orthogonal to the magnetic body 11, the ratio of the amount of magnetic bias given to each magnetic body in the X direction does not change, so the same coordinate value Xs can be obtained.

Furthermore, by combining the two position detection units 24 so as to be orthogonal to each other, so-called two-dimensional coordinate values in the X and Y directions can be determined.

On the other hand, in order to specify the position, the number of local films ○
Since it is only necessary to apply a magnetic bias of about e, the input pen 1 can be used with the input pen 1 being higher than the magnetic body 21 and somewhat separated from it in the Z) direction.

As shown by the broken line in FIG. 4, when one end 1a of the input pen 1 is separated from the input surface 28 by a height h,
The angle at which the magnetic flux and the magnetic body 21 intersect directly under a does not change, and the amount of magnetic bias applied to the magnetic body 21 also does not change, but on both sides of that, the angle at which the magnetic flux and the magnetic body 21 intersect increases. , the amount of magnetic bias also decreases, and the magnetic permeability, that is, the induced voltage increases.

In FIG. 5, the voltage V, /~v7' is the induced (detected) voltage when one end 20a of the input pen 20 is held at the above-mentioned position at a short distance (for example, about 5 m) in the Z direction. show. Detection voltages v0 to v7 and detection voltage V,
Comparing l to v7', the degree of increase in voltage remains almost unchanged near the specified position, and increases as the distance in the X direction moves away from the specified position, and the curve of the curve is gentle, that is, the difference decreases. I can see that you are doing it.

Detection voltage ■. ' ~ V, / When the differences are calculated in the same way, the differential voltage value v as shown in Fig. 7 is obtained.
1. v,, V3...v7 are obtained. 6th
As can be seen from the figure and FIG. 7, a decrease in the difference appears as a decrease in the slope of the line that crosses the reference level vt.

The two are almost in a proportional relationship. That is, the height h is proportional to the slope of the difference between the detected voltages.

Therefore, the relationship between the height h and the slope of the difference is determined in advance, and each voltage (■) is determined. The height h of the input pen 1 can be calculated by calculating the difference from ~v7 and further calculating the slope.

In addition, when specifying the input position, it is necessary to input a predetermined timing signal to the position detection circuit 27. For example, if the height h of the input pen 1 is used as a parameter,
When the height of the tip 1a of the input surface 28 becomes 0.5nI+ or less, a timing signal can be generated.

Next, the X coordinate value where the difference is actually “O”, for example x9
, the coordinate value xs, and the formula for determining the height h in two directions will be described. First, each detected voltage, for example, V(x), is expressed by a general formula of a quadratic function as shown in the following formula.

V(x)= a x2+ b x + c
(1) If the interval between each excitation line 22a to 22h is d, the difference equation ΔdV(x) is Δd V(x)=(V(x+ d)−V(x))/d
...12). Therefore, ΔdV(x-d) = (V(x) V(x-d))/d
・・・・・・(3) It becomes. Substituting equation (1) into equation (3) and organizing it.

ΔdV(x-d)=a((2X+b/a)-d)...
...(4) becomes. By the way, X that satisfies Δdv(X-d)=O
From equation (4), the coordinate value xq is xq = −b / 2 a + d / 2
--(5). Although the coordinate value of the vertex of the function is xs, δV(xs)/δX=2axs+b=0,'
-xs=-b/2a
---(6). Substituting equation (6) into equation (5) and rearranging, xs = xq - d / 2
・Old = (7). That is, by finding the X coordinate value xq of the point where ΔdV(x-d) is 0 and subtracting d/2 from it, the coordinate Xs of the vertex of the function can be found.

Next, consider the coordinate value Xq. ΔdV(x-d) is O
The coordinate value xq is between the differential voltage value v3 before and after the reference level Vt and ■ in FIG.
), so from equation (4), y3=a ((2
X3 + b/ a) d) ・・・・・(
”)V4=a((2x4+b/a)-d) ・
-・19), rearranging this, V3-v, = 2a (X3 X4) y 2 a = (V3 V4) / (X) X4
)X, -X, = -d.

2 a = −(V, −V,)/d
--(10). Here, since it indicates the slope of 2a and the difference curve, the height h in the two directions is h
= A I 2 a l −・
= (11). Note that A is a proportionality constant and is determined experimentally. Figure 8 shows a and input pen 1.
This shows the relationship between the height h in two directions of h=0 (m
The change in a with respect to h is shown when a is set to 1 in m).

Also, from the detection voltage equation (6), b / a − d = 2 xq
-(12), so by substituting this into equations (8) and (9).

V3= 2 a (x, + xq)
−(13)V, = 2 a (x, + xq)
...(14). Furthermore, by dividing both sides of equations (13) and (14), vi/vn=(x3-xq)/(xx-xq).

Putting v, / v, = B and rearranging, xq
=(x, -BX4)/(I B) --(1
5). Coordinates x3. Since x4 is known, Xq
can be found from the ratio of ■3 and ■4, and furthermore, (7
) can determine the coordinates XS of the vertex.

Tablet 2 is not limited to the above format;
Other electromagnetic induction type or magnetostrictive type may be used as long as the signal from the input pen can be read from a certain height.

Next, the operating procedure for recognizing handwritten characters will be explained with reference to FIGS. 1 and 9. First, a square first virtual frame 3 parallel to the X-axis and the y-axis is placed at a distance of 2.5 mm above and below, centering on the first pen-down coordinate 31 of the first input pen 1.
2, and a square second virtual frame 33 is similarly set in the frame coordinate memory 5 at a distance of 50 mm above and below the initial pen-down coordinate 31. On the other hand, the first pen-down coordinates 31, ie, X and Y, are temporarily stored in the coordinate position memory 4, and the coordinate values of the first and second virtual frames 32 and 33 are temporarily stored in the frame coordinate memory 5.

From the first pen down to the first pen up, CPU3
The stroke coordinates of the input pen 1 are traced along the trajectory, and if these coordinates are within the first virtual frame 32,
The coordinate position is temporarily stored in the coordinate position memory 4. When the stroke coordinates exceed the first virtual frame 32, the command memory 6 instructs the frame coordinate memory 5 to enlarge the first and second virtual frames 32, 33, enlarges these virtual frames, and stores their coordinate values. It is stored in the coordinate position memory 4. This essentially means that the space for entering characters has become larger, and is completely unaffected by the size of the characters.

If the next pen-down coordinate after pen-up is located outside the above-mentioned final virtual frame coordinates, that is, the second virtual frame 33 that has been expanded to the maximum, it is regarded as one character in the trajectory of the input pen 1 up to that point, The CPU 3 enters an operation of comparing the dictionary in the dictionary memory 6 with the handwritten characters in the coordinate position memory 4.
If there is a match, the end of the 6-character extracted from the dictionary memory 6 will also be determined when the input pen 1 becomes higher than the surface of the tablet 2 in two axial directions, that is, above a certain height, as described above, the tablet 2 It is determined that no coordinate position information has been input, and handwritten characters are recognized based on the trajectory of the input pen 1 up to that point. Therefore, there is no need to give a new instruction to finish writing handwritten characters.

If the first pen-down of the second stroke is within the second virtual frame 33, it is assumed that the - character is in the middle of being written, and the handwriting is further traced.

This is the case, for example, with dakuten and handakuten in flat names and katakana, or with the English letters "i" and "j." In other words, if the main part of the character falls within the first virtual frame 32, and "〃" or "." falls within the second virtual frame 33 outside of it, this point is also determined to be part of one character. Therefore, there will be no misunderstanding.

Furthermore, if the first dot is small, as in the case of ``u'' in a flat name, the virtual frame may not be enlarged, or even if it is slightly enlarged, if the second stroke is brought from a slightly distant place, the first dot may not be enlarged. It may end up recognizing only the dot, and eventually recognize it as ",tsu" and not being able to recognize two consecutive "u" strokes. Such misrecognition is prevented by the second virtual frame 33.

The procedure of the above-mentioned operation will become clearer if you refer to the flowchart in FIG. Since there is no point in setting them, the two virtual frames may be integrated once they reach a certain point.

Further, in the above embodiment, the first and second virtual frames 32
．． Although the length of each side of 33 was set to 5 and 10 mm, it is needless to say that the length is not limited to these values.

〔Effect of the invention〕

As described above, the handwritten character recognition method of the present invention automatically cuts out characters and symbols drawn in any size on a tablet in a free area according to the movement of the input pen, and enters the character recognition operation. Freed from the unnatural constraint of having to write in a predetermined size in a predetermined position as in the past, it is now possible to recognize handwritten characters with the natural feeling of writing with a pen on paper. can.

In addition to alphanumeric characters, it is also possible to accurately recognize characters in which flat names and katakana dots form part of the character. Furthermore,
By using a dictionary that performs matching using vector components of characters, which requires a relatively small amount of memory, the entire device can be constructed with just a few chips, making it extremely compact, inexpensive, and easy to manufacture. A character recognition device can be configured.

[Brief explanation of the drawing]

FIG. 1 is a block diagram schematically showing an embodiment of the present invention, FIG. 2 is an explanatory diagram illustrating an example of the tablet,
Figure 3 is a characteristic diagram of magnetic bias versus magnetic permeability, Figure 4 is a diagram showing the state of magnetic flux applied to a magnetic material from an input pen, and Figure 5 is a diagram showing the magnetic flux applied to a magnetic material from an input pen.
The figure is a graph showing an example of the induced voltage generated in the detection line, Fig. 6 is a graph showing an example of the differential voltage value, Fig. 7 is a graph showing another example of the differential voltage value, and Fig. 8 is the differential voltage value. Graph showing an example of the relationship between and the height of the input pen in the Z direction, No. 9
The figure is an explanatory diagram showing the relationship between the first and second virtual frames;
Figure 0 is a flowchart for explaining the operating procedure of this method, in which 1 is an input pen, 2 is a tablet, and 3 is a C
PU, 4 is a coordinate position memory, 5 is a frame coordinate memory, 6 is an instruction memory, and 7 is a dictionary memory.

Claims (4)

[Claims] (1) A handwritten character recognition method that recognizes handwritten characters written on a tablet with an input pen by detecting the coordinate position of the tablet, and the coordinate position of the input pen is temporarily stored in a position memory. The virtual frame coordinates of a square centered on the pen-down coordinate axis of the first input pen are stored in the virtual frame coordinate memory, and when the input pen deviates from the virtual frame coordinates due to the movement of the input pen, the virtual frame coordinates When the second pen-down coordinate position of the input pen deviates from the virtual frame, the end of the first handwritten character is centrally calculated using the algorithm instruction memory. A handwritten character recognition method in which a handwritten character is recognized by instructing a unit to compare and match the first handwritten character with standard characters stored in advance in a dictionary memory in this central processing unit. (2) In claim 1, the virtual frame coordinates are composed of first and second virtual frame coordinates made up of two large and small squares, and these two first and second virtual frame coordinates are A handwritten character recognition method which is expandable by the algorithm command memory and which recognizes handwritten characters by identifying characters such as alphabets, flat names, katakana, symbols, dots and voiced marks using the first and second virtual frame coordinates. (3) In claim 1, the tablet recognizes the end of the first handwritten character as the pen-up when the input pen is positioned at a predetermined height from the surface of the tablet. Handwritten character recognition method. (4) The handwritten character recognition method according to claim 1, wherein the two large and small squares have an area ratio of 4:1.