JP3290097B2 - Stroke determination method and determination device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for determining a stroke in a process for recognizing an online input handwritten character, and more particularly to a process for determining whether a stroke is a straight line or a curve in an online input handwritten character recognition process. The present invention relates to a method and a determination device.

[0002]

2. Description of the Related Art A structural analysis method is known as a typical example of conventional online input handwritten character recognition. The structural analysis method analyzes input strokes (handwriting data written from when the pen is turned on until it is turned off), finds basic strokes, matches this basic stroke sequence with the recognition dictionary, and writes characters. Recognize. A typical way to find the basic stroke sequence is to divide the stroke into multiple segments, and change the direction of each segment to 4,
When divided into 8, 16 or more, the direction of each segment is represented by a 4-way code, an 8-way code, a 16-way code, or more.

[0003] This direction code is associated with a basic stroke constituting a kanji. All the input strokes are represented by a basic stroke sequence, and the basic stroke sequence is matched with a recognition dictionary to perform character recognition.

[0004]

However, in the case of Japanese, kanji composed of linear strokes and hiragana, alphabets and numerals composed of both linear strokes and curved strokes are mixed, and the direction of the segment is changed. There was a problem that the number of codes was difficult to determine.

If the number of direction codes is set to a large value, erroneous recognition increases in recognition of kanji, and the recognition tends to become unstable. For example, if the direction of the segment is divided every 30 °, a 12-direction code is obtained, and a problem arises in that the horizontal line of the kanji is recognized as a different line when it is lower than 30 ° due to individual differences. If the number of direction codes is set to be small, the amount of information for hiragana and alphanumeric characters is reduced, causing a problem that the recognition rate of hiragana, alphabetic characters and numerals is reduced.

[0006] If it is known whether the stroke is composed of a linear element or a curved element, for example, the processing of the number of divisions is performed on a linear element, and the processing is performed on the linear element. By applying an effective method to each such as processing the number of divisions,
It is thought to be useful for improving the character recognition rate.

It is an object of the present invention to provide a method and a device for determining whether a stroke of an online input handwritten character is a straight line or a curved line.

[0008]

A stroke determination method according to the present invention comprises a first step of determining a curve length of a stroke from a start point to an end point, and a method of determining a plurality of straight lines approximating to each section of the stroke. A second step of calculating the linear length of the stroke based on the sum of the lengths, and a third step of calculating an error rate for the curve length of the stroke based on a difference between the curve length of the stroke and the linear length of the stroke; A fourth step of sequentially increasing the number of straight lines until the error rate reaches a predetermined first threshold to obtain a number m ₁ of straight lines when the error rate reaches the predetermined first threshold; The number of straight lines is sequentially increased from m ₁ until the rate reaches a second predetermined threshold smaller than the first threshold, and the number m ₂ of straight lines when the error rate reaches the second predetermined threshold is calculated. obtaining a fifth step, the number m ₁ and a straight linear Stroke compared with a threshold value for a predetermined curvature calculates the curvature computed curvature, characterized in that a sixth step determines straight or curve based on the number m ₂ of.

[0009] The stroke determination device according to the present invention comprises:
A curve length calculating means for calculating a curve length of a stroke by a sum of distances between points set during a stroke from a start point to an end point, and a sum of lengths of m straight lines connecting sample points selected from points in the stroke A linear calculating means for calculating a linear length of the stroke based on the error rate; an error rate calculating means for calculating an error rate with respect to the curve length of the stroke based on a difference between the curve length of the stroke and the linear length of the stroke; The number m of straight lines is sequentially increased until a value less than the first threshold value is reached, and the straight line length of the stroke is obtained. The straight line when the error rate based on the straight line length of the stroke reaches a predetermined first threshold value is determined. the stroke determined a first calculation means for calculating the number m _1, the linear length of the stroke by sequentially increasing the number of straight lines from m ₁ to the error rate reaches the second threshold value a predetermined smaller than the first threshold value To find the number m ₂ of straight lines when the error rate based on the length of the straight line reaches a second predetermined threshold value.
Calculating means for calculating the curvature based on the number m ₁ of straight lines and the number m _{2 of} straight lines, and comparing the calculated curvature with a predetermined threshold value of the curvature to determine whether the stroke is a straight line or a curve. Determining means for determining.

According to the method and the apparatus for determining a stroke according to the present invention, a curve length of a stroke from a start point to an end point is obtained, and the sum of lengths of a plurality of straight lines approximated to respective sections of the stroke is obtained. The linear length of the stroke is determined based on the stroke length, and the error rate with respect to the curve length of the stroke is determined based on the difference between the curve length of the stroke and the linear length of the stroke. The number of straight lines is sequentially increased until the error rate reaches a predetermined first threshold, and the number m ₁ of straight lines when the error rate reaches the predetermined first threshold is obtained. The number of straight lines is sequentially increased from m ₁ until a predetermined second threshold value smaller than the threshold value 1 is reached, and the number m ₂ of straight lines when the error rate reaches the predetermined second threshold value is obtained. . The curvature is calculated based on the obtained number m ₁ of straight lines and the number m _{2 of} straight lines, is compared with a predetermined threshold value for the curvature, and it is determined whether the stroke is a straight line or a curve based on the comparison result.

As described above, according to the stroke determination method and the determination apparatus according to the present invention, by setting two first and second thresholds, a rough straight line approximation to the stroke and a more accurate straight line An approximation can be obtained. As a result, two straight line numbers m ₁ and m ₂ are obtained. When the stroke is composed of straight lines, the number of straight lines does not change much even if the threshold value is slightly changed. If the stroke is composed of a curve, the number of straight lines changes greatly if the threshold is slightly changed. The difference in the number of straight lines can be used to determine whether the stroke is a straight line or a curved line.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, a method and an apparatus for determining a stroke according to the present invention will be described with reference to an embodiment of the apparatus for determining a stroke according to the present invention.

FIG. 1 is a functional block diagram showing a configuration of a stroke determination device according to an embodiment of the present invention.
A stroke determination device according to an embodiment of the present invention includes a storage device (not shown) and a microcomputer (not shown) cooperating with the storage device, and a central processing unit of the microcomputer, as shown in FIG. One stroke read by the online character reading unit 1 that reads online handwritten characters in a stroke order by a device is defined as a point constituting a stroke (hereinafter, referred to as a stroke).
(Also simply referred to as a point) to determine the stroke curve length by summing the lengths of the straight lines connecting the sample points selected from the points in the stroke to determine the stroke straight line length. An error rate (hereinafter, simply referred to as an error rate), which is an approximate linear error rate, is obtained based on a difference between the error rate and a straight line length of a stroke connecting the sample points, and the sample point when the error rate reaches a first threshold value Th1 And the error rate is equal to a second threshold value Th2 (Th1> Th).
A stroke straight line approximation unit 2 for calculating the number of straight lines connecting the sample points when 2) is reached is functionally provided.

The stroke determination device according to one embodiment of the present invention further includes a number of straight lines connecting the sample points when the error rate reaches the first threshold value Th1, and the error rate is set to the second value.
And a stroke curvature calculation unit 3 that calculates the curvature of the stroke based on the number of straight lines connecting the sample points when the threshold Th2 (Th1> Th2) is reached, and determines whether the original stroke is a straight line or a curve based on the calculated curvature. And a stroke type discriminating unit 4 for judging the condition.

The stroke linear approximation unit 2 includes a curve length calculation unit, a linear length calculation unit, an error rate calculation unit, and first and second calculation units. Length calculating means, linear length calculating means, error rate calculating means correspond to first and second calculating means. The stroke curvature calculator 3 corresponds to a curvature calculator, and the stroke type determiner 4 corresponds to a determiner.

Next, the determination operation of the stroke determination device according to one embodiment of the present invention will be described with reference to the flowchart of FIG. 2 and the schematic diagrams of FIGS.

Step 11 in FIG. 2 shows a reading process by the online character reading unit 1.
Steps 21 to 29 in FIG. 2 are steps showing a straight-line approximation process of the stroke in the stroke straight-line approximation unit 2, and Step 31 in FIG. 2 is a step showing the process of the stroke curvature calculation unit 3.
Is a step showing the processing of the stroke type determination unit 4.

FIG. 3 shows an example in the case of a stroke by a straight line, FIG. 4 shows an example in the case of a curved stroke, and in FIGS. 3 and 4, white circles indicate points constituting one stroke, and black circles indicate sample points. Is shown. FIG. 3A and FIG. 4A each show one stroke.

A sequence S of points Pi constituting one stroke read by the online character input device is represented by the following (1).
It is shown in the formula. Here, n is a natural number.

S = {P ₁ , P ₂ ,..., Pn} (1)

A sequence Sm of sample points when one stroke is approximated by m straight lines is shown in the following equation (2). m is a natural number and (m + 1) ≦ n.

[0022]

(Equation 1)

The following equation (3) is established between the sequence S of points forming a stroke and the sequence Sm of sample points. The sequence Sm of sample points is a subset of the sequence S of points forming a stroke. is there. Hereinafter, the determination process will be described.

[0024]

(Equation 2)

First, the curve length Lc of one stroke is obtained from the sum of the linear distances between adjacent points constituting the stroke. That is, it is calculated using the following equation (4) (step S21).

[0026]

(Equation 3)

Here, (Pi, Pi + 1) indicates a linear distance of a line connecting the point Pi and the point Pi + 1. Next, the linear length Lλ of one stroke is calculated by the following equation (5) (step S22).

[0028]

(Equation 4)

Here, as described above, m is the number of approximate straight lines. The starting point P ₁ of one stroke is also shown as a point P _1, showing also the end point Pn to point Pn of one stroke. In the equation (5), for example, when the number of straight lines m = 1, the sample point P ₁ ′ = point P ₁ and the sample point Pn ′ =
It is a point Pn, as shown in FIG. 3 (b) with respect to FIG. 3 (a), and as shown in FIG. 4 (b) with respect to FIG. 4 (a).
In the case of approximating the start point of the stroke and P ₁ a curve length Lc of one stroke by connecting the end point Pn with a linear, L
It is approximated by λ = (P ₁ ′, Pn ′).

When the number of straight lines is m = 2, the sample point is the sample point P ₁ ′ = point P ₁ , and the sample point Pia ′ = point P which is the middle point of the stroke and is also the first sample point.
ia and the sample point Pn ′ = point Pn, and the linear distance between the stroke start point P ₁ (= P ₁ ′) and the first intermediate point Pia (= Pia ′) which is also the first sample point, and This is a case where the curve length Lc of one stroke is approximated by the sum of the linear distance between the halfway point Pia (= Pia ′) and the stroke end point Pn (= Pn ′), and the distance Lλ = (P
₁ ′, Pia ′) + (Pia ′, Pn ′). That is, as shown in FIGS. 3C and 4C, a straight line connecting the starting point P1 of _one stroke and the first intermediate point Pia, the first intermediate point Pia and the ending point Pn of one stroke, and Is approximated by the sum of the lengths of two straight lines with the straight line connecting.

Similarly, when the number of straight lines is m = 3, the sample points are: sample point P ₁ ′ = point P ₁ , sample point Pi
a ′ = first intermediate point Pia, sample point Pib ′ = second
Of the stroke and the sample point Pn ′ = point Pn, and the linear distance between the stroke start point P ₁ (= P ₁ ′) and the first middle point Pia (= Pia ′) of the stroke, and the first , A linear distance between the intermediate point Pia (= Pia ′) and the second intermediate point Pib (= Pib ′), which is the second sample point, the second intermediate point Pib (= Pib ′), and the end point of the stroke This is a case where the curve length Lc of one stroke is approximated by the sum of the linear distances with Pn (= Pn ′), and the distance Lλ =
(P ₁ ′, Pia ′) + (Pia ′, Pib ′) + (P
ib ′, Pn ′) to approximate the curve length Lc of one stroke. That is, as shown in FIG. 3D and FIG. 4D, the start point P _{1 of} one stroke and the first intermediate point Pia
And a first intermediate point Pia and a second intermediate point P
The curve length Lc of one stroke is approximated by the sum of the lengths of three straight lines, that is, a straight line connecting ib and the straight line connecting the second intermediate point Pib and the end point Pn of one stroke.

In the following, 1 is obtained by sequentially increasing the number m.
The linear length Lλ of the stroke approximates the curve length Lc of the stroke. FIGS. 4 (e) and (f) show the case where the number m is set to 4 and 5, and the linear length Lλ of one stroke approaches the curve length Lc of the stroke by sequentially increasing the number m. Will be.

Subsequent to step S22, using the calculated one-stroke curve length Lc and one-stroke linear length Lλ, the one-stroke curve length Lc and the one-stroke linear length Lλ with respect to the one-stroke curve length Lc are calculated. An error rate D due to the difference is obtained by the following equation (6) (step S23).

D = (Lc−Lλ) / Lc (6)

Subsequent to step S23, it is checked whether the error rate D is less than a threshold Th1 (step S2).
4). When it is determined in step S24 that the error rate D is equal to or greater than the threshold Th1, a search for a maximum error section is performed subsequent to step S24 (step S27). Here, the section means a section between adjacent sample points obtained at the time of linear approximation.

In step S27, a search is made for one having the largest error between the straight line length connecting the adjacent sample points and the curve length of the stroke connecting the adjacent sample points. The point of the stroke section k between adjacent sample points is denoted by ｛Pk
₁ , Pk ₂ ,..., Pkh｝. A curve length Lck, a straight line length Lλk, and an error rate Dk as feature quantities based on this point are obtained by the following equations (7), (8), and (9), respectively.

[0037]

(Equation 5)

[0038]

(Equation 6)

[0039]

(Equation 7)

Here, the search for the maximum error section is to search for a section in which the error rate Dk obtained by equation (9) is the maximum.

When the maximum error section is searched in step S27, the maximum error section is subdivided (step S28). The subdivision in step S28 will be described using the stroke section k as an example.

Assuming that a point Pkj constituting the stroke section k exists in the section {Pk ₁ , Pk ₂ ,..., Pkh}
For all j (j = 1, 2, 3,..., H), the linear distance between (Pk ₁ , Pkj) and the linear distance between (Pkj, Pkh) shown in the following equation (10) is calculated. The sum Lλk ′ is obtained, and a point Pkj at which the sum Lλk ′ becomes maximum is set as a division point (sample point).

[0043]

(Equation 8)

Subsequent to the division in step S28, for each of the two newly divided sections, the calculation of the characteristic amounts of the curve length Lck, the linear length Lλk, and the error rate Dk is performed using the points of the section (7). It is calculated by the equations (8) and (9) (step 29), and the processing from step 22 is repeated. The error rate D is reduced by this repetition. If it is determined in the check in step S24 that the error rate D is less than the threshold Th1, the number m of straight lines at that time is determined based on the number of sample points following step S24. ₁ is stored (step S2
5). In FIG. 2, it is indicated as approximate sample point storage. Subsequent to step S25, the error rate D becomes equal to the threshold Th2 (T
It is checked whether h2 <Th1) or less (step S26).

In step S26, the error rate D is equal to the threshold T.
If it is h2 or more, the process is repeatedly executed from step S27 following step S26, and the search for the maximum error section is repeated. With this execution, the error rate D decreases. In step S26, the error rate D is equal to the threshold T.
When it is determined that it is less than h2, the number m ₂ of straight lines at that time is stored based on the number of sample points in step S25 immediately before. When it is determined in step S26 that the linear approximation error rate D is less than the threshold Th2, the linear approximation is terminated, and the curvature C of one stroke is calculated (step 31).

The curvature C of the stroke is calculated using the number m _{1 of} straight lines satisfying the threshold value Th1 and the number m ₂ of straight lines satisfying the threshold value Th2. Assuming that the number of straight lines when the error rate D is less than the threshold value Th1 is m ₁ and the number of straight lines when the error rate D is less than the threshold value Th2 is m ₂ , the curvature C is given by the following equation (11). Desired.

[0047]

(Equation 9)

The starting point of a stroke composed of linear elements,
If the end point and the corner point are set to be extracted as the sample points of the linear approximation, the threshold value Th2 becomes the threshold value Th1
Even if it is set smaller, m ₁ ≒ m _{2. In} this case, the curvature C is close to 1. If the stroke is composed of curved elements, then m ₁ ≪m ₂ and the curvature C is greater than one.

Subsequent to step S31, the threshold value C of the curvature C
By setting t, for example, Ct = 1.5, and when the stroke curvature C is larger than Ct, the stroke is a curved stroke. Otherwise, the stroke is determined to be a straight stroke, and a straight line and a curved line are determined (step 41).

FIG. 3 is a diagram showing a process near a straight line of a stroke constituted by linear elements. FIG. 4 is a diagram showing a process near a straight line of a stroke constituted by curved elements. FIG. 3D shows a result of the linear approximation when the threshold values Th1 and Th2 are satisfied. FIG. 4D shows a linear approximation result when the threshold value Th1 is satisfied, and FIG. 4F shows a linear approximation result when the threshold value Th2 is satisfied.

[0051]

As described above, according to the stroke judging method and the judging device according to the present invention, a rough straight line approximation and a more accurate straight line approximation can be obtained for the stroke. One of the obtained number m ₁ and m ₂ of the straight line, if the stroke is constituted by a straight line, even if the threshold is slightly changed, the number of straight lines does not change much, if the stroke is configured by curves, threshold Even if it changes slightly, the number of straight lines greatly changes, and an effect is obtained in which it is possible to determine whether the stroke is a straight line or a curved line based on the difference in the number of straight lines.

[Brief description of the drawings]

FIG. 1 is a functional block diagram showing a configuration of a stroke determination device according to an embodiment of the present invention.

FIG. 2 is a flowchart for explaining a determination operation in the stroke determination device according to the embodiment of the present invention; This will be described with reference to schematic diagrams of FIGS.

FIG. 3 is a schematic diagram for explaining a determination operation in the stroke determination device according to the embodiment of the present invention;

FIG. 4 is a schematic diagram for explaining a determination operation in the stroke determination device according to the embodiment of the present invention.

[Explanation of symbols]

 1 online character reading section 2 stroke linear approximation section 3 stroke curvature calculation section 4 stroke type determination section

Continuation of the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06K 9/46-9/62 G06T 7/60 JICST file (JOIS)

Claims (5)

(57) [Claims] 1. A first step of obtaining a curve length of a stroke from a start point to an end point, and a step of obtaining a straight line length of a stroke based on a sum of lengths of a plurality of straight lines approximated to respective segments of the stroke. Step 2, a third step of calculating an error rate for the stroke curve length based on a difference between the stroke curve length and the stroke linear length, and a straight line until the error rate reaches a predetermined first threshold value. A fourth step of sequentially increasing the number of the error rates to obtain the number m ₁ of straight lines when the error rate reaches a predetermined first threshold, and a second predetermined step in which the error rate is smaller than the first threshold. A fifth process of sequentially increasing the number of straight lines from m ₁ until the threshold is reached to determine the number m ₂ of straight lines when the error rate reaches a second predetermined threshold, and the number m ₁ of straight lines and the straight line The curvature is calculated based on the number m ₂ and the calculated curvature is determined in advance. A sixth step of determining whether the stroke is a straight line or a curve by comparing the stroke with a threshold value for the calculated curvature. 2. A first step of obtaining a curve length of a stroke based on a sum of distances between points set during a stroke from a start point to an end point, and m straight lines connecting sample points selected from the points in the stroke. A second step of obtaining a linear length of the stroke based on the sum of the lengths of the strokes, and a third step of calculating an error rate for the curve length of the stroke based on the difference between the curved length of the stroke and the linear length of the stroke. A third step of sequentially increasing the number m of straight lines until the error rate reaches a predetermined first threshold to obtain a straight line length of the stroke;
A fourth step of obtaining the number m ₁ of straight lines when the error rate based on the straight line length of the stroke obtained in the step has reached a predetermined first threshold; A fifth step of sequentially increasing the number of straight lines from m ₁ until the second threshold value is reached is determined, and an error rate based on the straight line length of the stroke determined in the fifth step is determined as A sixth step of obtaining the number m ₂ of straight lines when the predetermined second threshold value is reached, the number m ₁ of straight lines and the number m of straight lines
7. A seventh step of calculating the curvature based on the _{second step,} and an eighth step of comparing the calculated curvature with a threshold value for a predetermined curvature to determine whether the stroke is a straight line or a curve. The method for determining the stroke to be performed. 3. The method for determining a stroke according to claim 2, wherein the number of straight lines in the third and fifth steps is increased by selecting a sample point so as to maximize the straight line length of the stroke. A stroke determination method characterized by increasing the number of strokes. 4. A curve length calculating means for determining a curve length of a stroke by a sum of distances between points set during a stroke from a start point to an end point, and m straight lines connecting sample points selected from points in the stroke. Linear length calculating means for calculating a linear length of the stroke based on the sum of the lengths of the strokes; and error rate calculating means for calculating an error rate for the curve length of the stroke based on a difference between the curve length of the stroke and the linear length of the stroke. The number m of the straight lines is sequentially increased until the error rate reaches a value less than a predetermined first threshold value, and the straight length of the stroke is obtained.
A first calculating means for calculating the number m ₁ of straight lines when the threshold value is reached, and a second predetermined value having an error rate smaller than the first threshold value.
The number of straight lines is sequentially increased from m ₁ until the threshold value is reached, the straight line length of the stroke is obtained, and the number m ₂ of straight lines when the error rate based on the straight line length of the stroke reaches a second predetermined threshold value is calculated. A second calculating means for calculating, a curvature calculating means for calculating a curvature based on the number m ₁ of straight lines and the number m _{2 of} straight lines, and comparing the calculated curvature with a predetermined threshold value of the curvature to determine whether the stroke is a straight line. A stroke judging device comprising: a judging means for judging whether or not the stroke is a curve. 5. The stroke judging device according to claim 4, wherein the increase in the number of straight lines in the first and second calculating means is performed by selecting a sample point so that the straight line length of the stroke is maximized. A stroke determination device characterized by increasing the number.