JPH05342413A - Singular point detecting method for on-line character recognition - Google Patents

Abstract

PURPOSE:To improve recognition by extracting the information of singular points corresponding to various conditions such as the cross or contact of strokes in the case of recognizing online input characters. CONSTITUTION:An online character input equipment 10 reads hand-written characters in the order (online) of strokes. A singular point detection part 11 detects the cross point or contact point of strokes in the online input character as the singular point and even when the coordinates of two points in the two strokes are not completely coincident, however, it is detected as the singular point when those points are positioned within a fixed distance. A curved line structure analyzing part 12 analyzes the curved line structure of strokes and describes the online input character as the rough structure of a character shape regardless of the stroke order corresponding to the structure of the singular point as well. A matching part 13 recognizes the online input character by referring to the standard model of a character shape model part 14.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for detecting singular points in recognition processing of online input characters.

[0002]

2. Description of the Related Art Research on handwritten character recognition technology by online input has been conducted for a long time, but the first factor that makes it difficult to start the technology is the variety of deformation patterns.

For example, "Summary on-line handwritten kanji recognition toward relaxation of writing restrictions such as kanji characters" by Yoko Suminaga
(Nikkei Electronics, 331, pp.115-13
3, December 5, 1983), in the case of online handwritten character recognition, conventionally, various feature quantities describing a shape have been used to cope with a variety of deformation patterns. However, since these characteristics are decided ad hoc, there is a problem that a lot of work is required to create the character shape dictionary. Furthermore, in online recognition, dynamic programming with high computational complexity is used to absorb deformations, although efficient and fast matching is required. In addition, when the structural method is used, a method that largely depends on the writing order is used. Therefore, when the writing order is multi-order and diverse such as Kanji, there is no choice but to enlarge the dictionary. There is.

As one of the solutions to this problem, the applicant has previously proposed in Japanese Patent Application No. 2-333291, and in the process of recognizing an online input character, it has the same coordinates on the stroke but different times. There is a method of describing a character shape feature by using a set of points described in 1. as a singular point, a primitive sequence (a chain of simple curve components) that forms a stroke and their connections, and the structure of the singular point.

[0005]

The above method aims at describing a global structure of a character shape, which is not affected by a flexible and systematic local deformation, for coping with a variety of character deformations. It provides a structure description for online input characters regardless of the stroke writing order. However, it is not easy to accurately extract information about singular points such as stroke intersections and contact points from online input data. In the above method, for example, two singular points as shown in FIG. 14 cannot be identified, so that there is a problem that the ability to identify similar characters is low.

An object of the present invention is to easily extract singular points that can be regarded as stroke intersections and contact points from online input data, enable more precise structure description for online input characters, and make similar character To improve the identification ability of.

[0007]

According to the present invention, when the coordinates of two points of two strokes intersecting or contacting each other match, a set of these two points and a line segment connecting two consecutive points of one stroke. When one point of the other stroke is present inside or close to the other stroke, a set of points on the other stroke and a closer point on the one stroke, and two consecutive points of the two strokes, respectively. When the connecting line segments intersect, the set of points having the smallest distance between the strokes is detected as a singular point.

[0008]

According to the present invention, not only when the coordinates of two points of two strokes written at different times are completely coincident with each other, but even when they are not completely coincident with each other, the distance between them is within a certain range. If any, it is detected as a singular point. This makes it possible to easily extract information about the singularity. As a result, the description of the positional relationship, such as the branching of strokes of online input characters, contact, etc., becomes precise,
More accurate recognition becomes possible.

[0009]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a functional block diagram of an embodiment of the present invention. The online character input device 10 is, for example, a tablet, and reads handwritten characters in the stroke order (online). The singular point detection unit 11 detects singular points such as stroke intersections and contact points from this online input data, and gives them to the curved structure analysis unit 12 together with stroke information of online input characters. The curve structure analysis unit 12
A curved line structure is analyzed based on the stroke information of the online input characters, and the global structure of the character shape is independent of the stroke writing order, etc., due to the primitive series and their connections that make up the stroke, and the structure of singular points. The features described in are extracted. A standard model (dictionary) of character shapes is registered in advance in the character shape model unit 14. The matching unit 13 matches the characteristics of the input character shape obtained by the curve structure analysis unit 12 with the standard model of the character shape model unit 14, and recognizes the online input character. In the following, the singularity detection unit 11, the curved structure analysis unit 12, and the matching unit 13 and the character shape model unit 14 which are related thereto will be described in detail.

Singular point detection unit 11 The stroke information of an online input character is obtained as a sampling point sequence. Focusing on this sampling point, the singularity is detected in the following cases. (1) When the coordinates of two different points are completely the same In FIG. 2A, p ₁ and q ₁ have the same coordinates, but they are different in time. In this case, p ₁ and q ₁ are used as singular points as they are. Note that p ₁ and q ₁ may be end points or inner points of each stroke. (2) In the case where there is another one point inside the line segment connecting two consecutive points (not including the end points) In FIG. 2B, p ₁ , p ₂ and q ₁ have this relationship. In this case, _one of p ₁ and p ₂ that is closer to q ₁ is defined as p ₀ , p ₀ and q ₁ are regarded as the same point (coordinates are different), and they are set as singular points (q ₁ is also the end point of the stroke) Points are also acceptable). In the example of FIG. 2B, p ₁ is p ₀ , and p ₀ (=
Let p ₁ ) and q ₁ be singular points. (3) When a line segment that connects two consecutive two points intersects at a point other than the end points FIG. 2C shows this. In this case, a combination of points (p ₁ , q ₁ ), (p ₁ , q ₂ ), (p ₂ , q ₁ ),
In (p ₂ , q ₂ ), the pair with the smallest distance is selected, and the two points are regarded as the same and set as a singular point. For example,
When (p ₁ , q ₂ ) is selected, p ₁ and q ₂ are singular points. (4) In cases other than (1) to (3), when the distance between the inside of a line segment that connects two consecutive points and a point that is not connected to the two points is less than or equal to a predetermined threshold value d. In d), two points where p ₁ and p ₂ are continuous, q
_1, p ₁ p ₂ the height, and are included in the rectangular width 2d of the center line. In this case, similarly to (2), p
Of ₁ and p _2, the one closer to q ₁ is p _0, and p ₀ (p ₁ or p
₂ ) and q ₁ are singular points. (5) In cases other than (1) to (4), where the distance between the end points of the two strokes is less than or equal to a certain threshold value d ', in this case as well, the singular point is detected in the same manner as (2).

Curve Structure Analysis Unit 12 The curve structure analysis unit 12 receives from the singular point detection unit 11 the stroke information of an online input character together with the singular point.
It is assumed that the output from the online character input device 10 is a curve represented by a segmental line segment (polygonal line) for each stroke as shown in FIG. Where the sample point P
Are serialized by time. That is, the character shown in FIG. 3 is composed of two strokes, (P (0), P (1), ..., P (9)), (P (10), P (1), respectively.
1), P (12)) are written in this order. Furthermore, the two points P (1) and P (10) have the same coordinates but are written at different times. That is,
In FIG. 3, a set of points (P (1), P (10)) is a singular point.

The analysis of the curved structure is performed hierarchically in the order shown in FIG. Step 121 ; Decomposition of Strokes into Primitives An xy monotone curve is x, when moving along that curve.
A curve in which the y-coordinate value is always non-increasing or non-decreasing. The coordinates of two end points of a certain xy monotone curve are P (x ₀ , y ₀ ) and Q (x ₁ , y ₁ ), respectively. However, x ₀ <x _1, and if x ₀ = x _1, then y ₀ <y ₁ . At this time, P is the head of the curve and Q is the curve
Call it the tail. x depending on the positional relationship between the head and tail
The y monotone curve is classified into the following four types.

(1) Horizontal straight line (y ₀ = y ₁ ),
It is written as "-". This is shown in FIG. (2) A curve that rises to the right and descends to the left ((x ₀ −x ₁ ) × (y ₀ −y
₁ )> 0), hereinafter referred to as "/". This is shown in FIG. (3) A vertical straight line (x ₀ = x ₁ ), hereinafter referred to as “|”.
This is shown in FIG. (4) A curve that descends to the right and rises to the left ((x ₀ −x ₁ ) × (y ₀ −y
₁ ) <0), hereinafter referred to as "\". This is shown in FIG.
Shown in. These four types of curves are called curve primitives. That is, a primitive means a simple curve component.

As an example, the character shown in FIG. 3 is decomposed into primitives. The result is shown in FIG.
That is, the stroke (P (0), P (1)) in FIG. 3 is decomposed into primitives as follows. A ₀ (/): (P (0), P (1), P (2), P (3)), A ₁ (\): (P (3), P (4)), A ₂ (/ ): (P (4), P (5)), A ₃ (\): (P (5), P (6), P (7)), A ₄ (/): (P (7), P (8)), A ₅ (\): (P (8), P (9)), and the stroke (P (10), P (11), P (12)) is one primitive A ₆ ( \): (P (10), P (11), P (12)).

Step 122 : Combining Primitives Next, the primitives are combined. To that end, we introduce a rule for combining four primitives. Now, on a curve that is not an xy monotone curve, we can find two different xy monotone curves that cover the curve, and one of them does not contain the other. Let these two xy monotone curves be a and b, respectively. However, P is included in both a and b, Q and R are arbitrary points in the neighborhood of P, Q is included only in a, and R is included only in b.
coord (P) = (x _p , y _p ), coord (Q) = (x _q ,
y _q ), coord (R) = (x _r , y _r ), the outer product of the vectors (x _q −x _p , y _q −y _p ) and (x _r −x _p , y _r −y _p ). Is positive, that is, (x _q −x _p ) × (y _r −y _p ) − (x _r −x _p ) × (y _q −y _p )> 0.

Under this assumption, the combination of a and b

[Equation 1] Is defined as follows. However, A and B are "/",
Either "\", "|", or "-"
When using either ad or tail, the symbols [A, α, B,
β] is the property of a is A and the property of b is B, and α of a is
And β of b indicate that a and b are bound. (Rule 1) [/, head, \, tail], [|, head, \, tail], [/, head, |, h
[ead], or [/, head, /, head]

[Equation 2] (Rule 2) [\, head, /, head], [\, head,-, head], [-, head,
/, Head] or [\, head, \, head]

[Equation 3] (Rule 3) [/, tail, \, head], [/, tail, |, tail], [|, tail, \, h
ead], or [/, tail, /, tail]

[Equation 4] (Rule 4) [\, tail, /, tail], [\, tail,-, tail], [-, tail,
When /, tail] or [\, tail, \, tail]

[Equation 5]

FIG. 7 is a diagram for explaining the combination of two primitives. That is, the two primitives a and b are

[Equation 6] Bind to form a right-handed system. Here, P is a point taken on the intersection of a and b, Pa and Pb are points on a and b, respectively. Pa corresponds to Q and Pb corresponds to R.

8 to 11 show examples in which the rules 1 to 4 are applied. 8 is a case where rule 1 (equation 2) is applied, FIG. 9 is a case where rule 2 (equation 3) is applied, FIG. 10 is a case where rule 3 (equation 4) is applied, and FIG. 11 is rule 4 Each of the cases where (Equation 5) is applied is shown.

As an example, the combination of the primitives shown in FIG. 6 is calculated as follows.

[Equation 7]

Step 123 ; Construction of Primitive Sequence Next, two primitives are combined.

[Equation 8] By linking, the following series can be generated.

[Equation 9]

For example, for the example of FIG. 6, step 12
When the combination of the primitives obtained in 2 is linked, on the stroke (P (0), P (1), ... P (9)), as shown in FIG.
As shown in 2,

[Equation 10] That is, two primitives can be grouped. Further, the stroke (P (10), P ( 11), P (12)) is a primitive sequence E ₂ comprising a single primitive A _6: constituted by A _6.

Labels <ps, id> are given to the primitive series represented by the equation (1) according to the following rules.

[Equation 11] Here, corner () is a function that takes 1 when the types of the primitives a and b are the same and 0 when they are different.

The meaning of the label <ps, id> is as follows. First, id represents the initial direction of the convex in the primitive series. Next, regarding ps, in the definition expression of ps, one item on the right side

[Equation 12] Is the difference in the convex direction and represents the rotation speed. Next, of the two items on the right side

[Equation 13] At the end of the sequence, FIG. 8 (d), FIG. 9 (d), and FIG.
This is a correction term for the case where the combination shown in (d) and FIG. 11 (d) appears. For example, in the combination of FIG. 8D, it is considered that the downward convex and the convex toward the left appear degenerate at once.

As a result, the labels <1,0> and <2,2> are given to the primitive series E ₀ and E ₁ , respectively. Further, P (5) and P (4) are the h-points of E ₀ and E ₁ , respectively, and P (0) and P (9) are the t-points of E ₀ and E ₁ , respectively.

Step 124 : Connection of Primitive Sequences In FIG. 12, two primitive sequences E ₀ and E ₁ are
Share the A ₂ which is the first element of the series. In this way, two primitives that are adjacent to each other may be connected to each other by sharing the first element of the series or connected to each other by sharing the last element of the series. -Connected, called t-connected.

For example, E ₀ and E ₁ are h-connected. this,

[Equation 14] (In case of t-connection, h is replaced with t).

Step 125 ; Structure of singular point The singular point is detected by the singular point detection unit 11 as a set of points as described with reference to FIG. Here, the structure of the singularity is
As shown in FIG. 13, there are four types. In the figure, the dotted line represents the flow of strokes. 13 (1), (2),
The types of singular points shown in (3) and (4) are respectively T-
Type, X-type, K-type, and L-type.

The structure of the singularity v is given by the binary relation between two primitive sequences e _i and e _j containing v

[Equation 15] It is described as. Where χ ∈ {X, K, T, L}
Indicates the adjoining structure of e _i and e _j on v, and σ, τ∈ {h
p, tp, h, t, φ} is the primitive sequence e _{i of} v
The position on (e _j ) is shown. σ is

[Equation 16] And when e _i consists of one primitive, (m
= 0),

[Equation 17] Becomes τ is the same as σ.

For example, in FIG. 6, the singular point (P (1),
P (10)) is of type T and is on the primitives A ₀ and A ₆ , which are on the primitive series E ₀ and E ₂ , respectively, as described above, so this singularity is ,

[Equation 18] It is described as.

The processing results of steps 124 and 125 describe the structure of the online input character shape (character shape structure description).

From the above processing, the structure of the character shown in FIG. 3 is described as follows.

[Formula 19]

Matching Unit 13 and Character Shape Model Unit 14 The character shape model unit 14 is composed of a structural model of the character shape described as shown in Expression 19. The matching unit 13 identifies a character by matching the structure description output from the curve structure analysis unit 12 and a structure description model included in the character shape model unit 14 in advance. As can be seen from the structure description shown in Expression 19, since the matching depends only on the primitive series that form the stroke and their connections, and the structure of the singular point,
It does not depend on the order in which the strokes are written or the direction of stroke flow, such as when a stroke is written from left to right or from top to bottom.

[0034]

According to the present invention, the shape of each stroke of an online input character is described by a combination of primitive series consisting of a chain of simple curve components (primitives) and a connection relationship between the primitive series, and , Sloak intersections and contact points are singular points, and the singular points are described by stroke intersections and contact situations, so that when recognizing an online input character, information of singular points can be obtained from stroke data input online. It can be adapted to various cases and easily extracted. As a result, the description of the positional relationship such as the branching of the stroke of the online input character, the contact, and the like becomes precise, so that the similar character can be easily identified and the accurate recognition becomes possible.

[Brief description of drawings]

FIG. 1 is a functional block diagram of an embodiment of the present invention.

FIG. 2 is a diagram illustrating a process in a singular point detection unit.

FIG. 3 is a diagram showing an example of online input characters.

FIG. 4 is a processing flow chart of a curve structure analysis unit.

FIG. 5 is a diagram illustrating types of primitives.

FIG. 6 is a diagram showing a result of decomposing the example of FIG. 3 into primitives.

FIG. 7 is a diagram illustrating a combination of primitives.

FIG. 8 is a diagram illustrating a combination rule of primitives.

FIG. 9 is a diagram illustrating a combination rule of primitives.

FIG. 10 is a diagram illustrating a combination rule of primitives.

FIG. 11 is a diagram illustrating a combination rule of primitives.

FIG. 12 is a diagram showing a configuration example of a primitive series.

FIG. 13 is a diagram illustrating a structure of a singular point.

FIG. 14 is a diagram showing another example of singular points.

[Explanation of symbols]

 10 Online character input device 11 Singularity detection unit 12 Curve structure analysis unit 13 Matching unit 14 Character shape model unit

Claims (1)

[Claims] 1. The shape of each stroke of an online input character is described by a combination of primitive series consisting of a chain of simple curve components (primitives) and a connection relationship between the primitive series, and at the intersection of sloak and contact. By using a point as a singular point and describing the singular point according to the situation of crossing or contact of strokes, in the process of recognizing an online input character, when the coordinates of two points of two strokes match, these two points In the case of a set of points, if one point of the other stroke exists in contact with or close to the inside of a line segment connecting two consecutive points of one stroke, the point on the other stroke and the point on the one stroke If a line segment that connects two consecutive points of two strokes intersects a pair of points that are closer to each other, then the point with the smallest distance between the strokes A singularity detection method for on-line character recognition, which is characterized by detecting a pair, as a singularity.