JP2882327B2 - Line figure matching 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 line graphic matching apparatus, which selects a feature point corresponding to a standard pattern from a line figure input as online data, evaluates the structure of the corresponding line figure, and matches the pattern. The present invention relates to a line graphic matching apparatus for performing the following.

[0002]

2. Description of the Related Art Conventionally, this type of line / graphic matching device matches a line / graphic input from an input device such as a tablet with a standard pattern as online data, as typified by an online character recognition device. It is used to evaluate the degree of the matching. For example, in 1982, Kazuhiko Yamamoto, an example of a line figure matching device, recognizes handwritten educational kanji characters by the relaxation matching method (IEICE Transactions Vol.
l. J65-D No. 91116-1174). In this example, it is not line graphic matching for online data, but the apparatus can also be applied to line graphic matching of online data. The operation of the line graphic matching apparatus described in this paper will be described with reference to FIG. In this conventional example, an input data storage unit 31 that stores input data, a line segment feature extraction unit 32 that extracts a line segment feature for approximating a line segment of the input data, and a dictionary in which dictionary data is stored A data storage unit 34, a corresponding line segment candidate extraction unit 33 for selecting a line segment candidate corresponding to a line segment in the dictionary data and a line segment in the input data, and a line segment in the input data and a line segment in the dictionary data Processing unit 35 for finding the optimal correspondence
And a matching score calculation unit 36 for calculating a matching score.

An input data storage section 31 stores binarized image data and sends it to a line feature extraction section 32.

[0004] The line segment feature extraction unit 32 tracks the boundaries of the binary image data and uses the Davis method (L. Davi).
s. Shape Mathing Using Re
laxation Technologies, IEEE
Trans, Vol PAMI-1 No. 1 (197
9)), a contour point sequence is extracted, and polygon approximation is performed using a boundary line segment. The characteristics of each line segment include the coordinates of the start point of the line segment, the coordinates of the end point, the inclination of the line segment, and the length of the line segment, are stored in a line segment feature list, and sent to the corresponding line candidate extraction unit 33.

The corresponding line segment candidate extraction unit 33 reads out the dictionary data from the dictionary data storage unit 34, and obtains the difference d of the line segment inclination in all combinations of each line segment in the dictionary data and each line segment in the input data. _1, the difference d ₂ in length, the difference d ₃ of the displacement of the starting point, the difference d ₄ of positional deviation of the end point are compared, when satisfying the following equation, it is stored as the corresponding candidate line segments. Here, c ₁ , c ₂ , c ₃ , and c ₄ are given by constant values.

In addition to such a one-to-one correspondence, a plurality of input line segments and one dictionary line segment, and a plurality of dictionary line segments and one dictionary line
The comparison is also made in correspondence to the book input line segments. FIG.
By comparing the dictionary line segment i with the input line segment k as shown in (a), the difference d _{5 in} the length of the line segment (the difference between the length of the dictionary line segment i and the length of the input line segment k) is found. Under the condition that d ₅ > c ₅ , from the end point vicinity line list of the input line segment k, If there is a line segment k 'that satisfies the expression, two input line segments k and k' are stored as corresponding candidate line segments corresponding to the dictionary line segment i. Here, d ₆ is the gradient difference between dictionary line segment i and input line segment k, d ₇ is the gradient difference between dictionary line segment i and input line segment k ′, d
₈ the difference between the sum of the length and the input line segment lengths of k and k 'dictionary segment i, d ₉ is the difference between the positional deviation of the starting point of the input line segment k and dictionary segment i, d ₁₀ is the dictionary a line segment i input segment k 'difference of the positional deviation of the end point of the, d ₁₁ are input segment k and the end point of the input line segment k' indicates the difference of the positional deviation of the starting point of, c _6, c _7, c ₈ ,
c ₉ , c ₁₀ , and c ₁₁ indicate constant values.

[0007] Similar to FIG. 17 (b) also FIG. 17 (a), the comparison between the input line segment k and dictionary segment i, the difference between the line segment length d ₁₂
(The difference between the length of the input segment i and the length of the dictionary segment k) is d ₁₂ >
If c ₁₂ become conditions, from the end near line list of dictionary segments i ', If there is a line segment i 'that satisfies the expression, the dictionary line segments i and i'
The input line segment k is stored as a corresponding candidate line segment corresponding to.

Here, d ₁₃ is a difference between the inclination of the dictionary segment i and the input line k, d ₁₄ is a difference between the dictionary segment i ′ and the input segment k, and d ₁₅ is the dictionary segment i, i. ', The difference between the sum of the lengths and the length of the input line segment k, d ₁₆ is the difference in the displacement of the starting point between the dictionary line segment i and the input line segment k, d ₁₇ is the dictionary line segment i' and the input line segment The difference d ₁₈ between the end point of the dictionary line segment i and the start point of the dictionary line segment i ′ is denoted by d ₁₈ , and c ₁₃ , c ₁₄ ,
c ₁₅ , c ₁₆ , c ₁₇ , and c ₁₈ indicate constant values.

For all dictionary line segments, correspondence candidate line segments are obtained based on the above conditions, and the degree of correspondence is calculated as follows.

Dictionary line segment i and input line segment k correspond to degree P _ik
Is given by ( _Equation 4) P _ik = max ｛1−W ₁ × max (| d ₁₉ | −c ₁₉ ,
0) −W ₂ × max (| d ₂₀ | −c ₂₀ , 0) −W ₃ × m
ax (| d ₂₁ | −c ₂₁ , 0) −W ₄ × max (| d ₂₂ |
−c ₂₂ , 0), 0｝, which is obtained by all the corresponding candidate line segments, and sent to the relaxation processing unit 35.

Here, d ₁₉ is the inclination difference between dictionary segment i and input segment k, d ₂₀ is the difference in length between dictionary segment i and input segment k, and d ₂₁ is the difference between dictionary segment i and input segment k. positional displacement difference of the start point of the input line segment k, d ₂₂ represents the positional deviation difference of the end point of the input line segment k and dictionary segment _{i, W 1, W 2,} W 3, W 4, c 19, c ₂₀ , c
₂₁ and c ₂₂ indicate constant values.

In the relaxation processing unit 35, an input line segment corresponding to the dictionary line segment i is set to k, and a matching coefficient r when the input line segment corresponding to the neighboring dictionary line segment j to the start point of the dictionary line segment i is set to l.
_{Let sij} (k, l) be an adaptation coefficient _reij '(k, l') where l 'is an input line segment corresponding to a nearby dictionary line segment j' with respect to the end point of dictionary line segment i. , ( _Equation 5) r _sij (k, l) = max ｛min (1−W ₅ × | d ₂₃
| + C ₂₃ , 1), 0｝ + max ｛min (1-W ₆ × |
d ₂₄ | + c ₂₄ , 1), 0｝ _reij (k, l) = max ｛min (1−W ₇ × | d ₂₅
| + C ₂₅ , 1), 0｝ + max ｛min (1-W ₈ × |
d ₂₆ | + c ₂₆ , 1), 0}, and is calculated and obtained for all the corresponding candidate line segments.

Here, d ₂₃ is the difference between the difference in the x direction at the starting point between dictionary line segment i and input line segment k and the difference in the x direction at the end point between dictionary line segment j and input line segment l, d ₂₄ Is the difference between the difference between the dictionary line segment i and the input line segment k in the y direction at the start point and the difference between the dictionary line segment j and the input line segment l in the y direction, and d ₂₅ is the dictionary line segment i and the input line. the difference between the difference in the x-direction of the end point of the partial k and the end point of the x-direction difference and dictionary segment j 'to the input line segment l', d ₂₆ is the end point of the input line segment k and dictionary segment i Difference in y direction, dictionary line segment j 'and input line segment l'
, C ₂₃ , c ₂₄ , c ₂₅ ,
c ₂₆ denotes a constant value.

Next, the degree of the above-mentioned correspondence is updated by the above-mentioned adaptation coefficient according to the following equation. (Number _{6) P ik = (q ik} × P ik) / Σ (q ik '× P ik') q ik = (q sik + q eik) / ΣL j + ΣL j 'q sik = Σ {L j × max ( r _sij (k, l) ×
P _jl )｝ q _eik = Σ ｛L _j ′ × max (r _sij ′ (k, l ′) ×
P _j ′ _l ′)｝ where L _j is the length of dictionary line segment j, L _j ′ is the length of dictionary line segment j ′, and l is the correspondence of all neighboring line segments to the starting point of dictionary line segment i. The input line segment l 'indicates the corresponding input line segment for all the neighboring line segments with respect to the end point of the dictionary line segment i.

(Equation 6) is repeated a certain number of times t, and the input line segment k having the largest P _ik is selected as the line segment corresponding to the dictionary line segment i.

The matching score calculation unit 36 calculates and calculates a matching value between the input data and the dictionary data based on the degree of the correspondence, and outputs it.

[0017]

In this conventional line figure matching apparatus, when matching input data of a line figure approximated to a line segment expressed by extracting a feature point or a representative point with dictionary data, When the line figure having the same structure as the line figure of the dictionary data is deformed, the number of feature points of the input data and the dictionary data may not match, and a plurality of feature points may be associated with one point, Since a feature point to be associated cannot be found, there is a drawback that an erroneous correspondence is likely to occur in the correspondence between the feature point of the input data and the feature point of the dictionary data. In order to solve this problem, it is necessary to perform processing such as integration or division of feature points on a line figure whose number of feature points has changed due to deformation. In this case, it is sufficient to use only a combination of about one to two.
A larger number of combinations must be considered for a line figure having a larger number of line segments such as a pair 3, a 1: 4, and a 2: 3, and the amount of calculation becomes enormous as a whole process.

Another problem is that if there is no correct combination among the predetermined combinations of line segment correspondences, erroneous correspondences may occur. In order to solve this problem, as a method of matching the number of feature points, there is a method of equally dividing a stroke. In this case, if the number of strokes does not match, the number of feature points does not match. There is a problem that causes an erroneous correspondence because a feature point to be associated with one point or a feature point to be associated cannot be found.

[0019]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, a line graphic matching apparatus according to the present invention has an input data storage unit for storing line graphic data input as online data, and a matching of the input data. A dictionary data storage unit for storing dictionary data information prepared in advance to calculate the degree of the feature point, and extracting the feature points with the fineness that can save the shape of the original figure as much as possible from the input data output from the input data storage unit. A feature point extraction unit that extracts and adds feature information and stores the feature points, and compares and supports features of all combinations of each feature point in the input data output from the feature point extraction unit and a feature point in the dictionary data A feature point matching calculation unit for calculating and obtaining a similarity indicating the degree of the degree, and generating a similarity table; and an input corresponding to a feature point in the dictionary data based on the similarity table. A corresponding feature point selecting unit for selecting a feature point in the input data, and a line of the input data based on the similarity between the feature point in the dictionary data selected by the corresponding feature selecting unit and the feature point in the input data. A matching score calculation unit that calculates a matching degree between the figure and the line figure of the dictionary data.

[0020]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, embodiments of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing one embodiment of the present invention. Referring to FIG. 1, a first embodiment of the present invention is shown.
Is an input data storage unit 11 for storing input data (coordinate point sequence), and a feature for extracting feature points such as end points, intersections, and inflection points with a fineness that can preserve the shape of the original data as much as possible. A point extraction unit 12, a dictionary data storage unit 14 in which dictionary data is stored, and a similarity between a feature point in input data and a feature point in dictionary data are calculated for all pairs. A feature point matching calculation unit 13 to be obtained, a corresponding feature point selection unit 15 for selecting feature points corresponding to each feature point of the dictionary data based on the similarity, and a line figure after selection of the corresponding feature point. A matching score calculation unit 16 that evaluates the structure and calculates a matching score is provided.

Next, the operation of this embodiment will be described with reference to FIG.

[0022] The input data storage unit 11, 1 to N-number of stroke _{S = (S 1, S 2} , ..., S N) and the i-th stroke, M _i pieces of coordinate point sequence {(x _1i , y _1i ),
Input data composed of (x _2i , y _2i ),..., (X _Mi , y _Mi )} is given and stored as an input, and is sent to the feature point extracting unit 12.

The feature point extracting section 12 reads out the strokes from the first stroke to the Nth stroke, and sets the step threshold d to 5 for each stroke so that the original shape of the input data can be preserved as much as possible. Using the threshold value d, the length D ₁ of the line segment from the start point of the first coordinate point to the next second coordinate point is determined. The length D ₂ of the line segment from the second coordinate point to the next third coordinate point is determined, and D = D ₁ + D ₂
Is smaller than the threshold value d, and if it is smaller, D (= D
₁ + D ₂ +... + D _i ) ≧ d, the above operation is repeated until the (i + 1) th coordinate point is extracted, and the (i + 1) th coordinate point is extracted as a feature point.
Similarly, sampling for sequentially extracting the next feature point is performed. The feature point sequence obtained by the sampling is used as the feature point sequence of the input data in addition to the start point and the end point of the stroke.

FIG. 2 shows an example of sampling using Hiragana "tsu" as an example.

For each feature point, a stroke number S to which the feature point belongs, a coordinate value (x, y), a start point, an intersection point, a bending point 1 (right-turn in the writing direction) as shown in FIG. Bend point 2 (turn left with respect to the writing direction) and end point are classified into five types of attributes, and coded attributes g as shown in FIG.
And the characteristic of the writing angle t toward the next characteristic point is obtained,
The expression of the th feature point is represented by A _i = (s _i , x _i , y _i , g
_i , t _i ).

FIG. 5 shows a conceptual diagram of a feature at one feature point.

The structure of the input data obtained by performing the sampling and the feature point expression for all strokes is expressed as a sequence of N feature points A = (A ₁ , A ₂ ,..., A _N ). It is sent to the point matching calculation unit 13.

The feature point matching calculation unit 13 reads the dictionary data stored in the dictionary data storage unit 14, and obtains M feature points B = (B ₁ , B ₂ ,..., B _M ) of the dictionary data.
And N feature points of input data A = (A ₁ , A ₂ ,..., A
_In all combinations with _N ), the features are compared and the similarity of feature points indicating the degree of matching of feature points is calculated and obtained. The i-th feature point expression of the input data is represented by A _i =
(S _ai , x _ai , y _ai , g _ai , t _ai ), and the j-th feature point expression of the dictionary data is represented by B _j = (s _bj , x _bj , y
_bj , g _bj , t _bj ), the similarity S between the i-th feature point of the input data and the j-th feature point of the dictionary data
_ij is given by the following equation: (Equation 7) S _ij = W ₁ × f ₁ (d ₁ ) + W ₂ × f
₂ (d ₂ ) + W ₃ × f ₃ (g _i , g _j ).

Where d ₁ = (x _ai −x _bj ) × (x _ai −
x _bj ) + (y _ai −y _bj ) × (y _ai −y _bj ), d ₂ = | t
_ai −t _bj |, and W ₁ , W ₂ , and W ₃ are each 1. In the present embodiment, assuming that sqrt represents a square root, f ₁ (x) and f ₂ (x) are f ₁ (x) = sqrt
(X), f ₂ (x) = sqrt (x), but any type may be used as long as a monotone increasing function as shown in FIG. 6 is used. Further, f ₃ (a, b) indicates a value of 1 when a = b, and indicates a value of 0 otherwise.

For all combinations of the feature points in the input data and the feature points in the dictionary data, the similarity is calculated and obtained, and a similarity table as shown in FIG. It is sent to the unit 15.

The corresponding feature point selection unit 15 selects feature points in the input data that should correspond to each feature point of the dictionary data based on the similarity table.

FIGS. 8, 9, 10, 11 and 12
The corresponding feature point selection unit 15 will be described with reference to FIG.

First, FIGS. 9, 10, 11 and 12
The symbols inside are explained. The values will be described together with the description of FIG. j (= 1, 2,..., M) indicates a feature point number in the dictionary data, and i (= 1, 2,..., N) indicates a feature point number in the input data. Dictionary feature point priority G (=
G ₁ , G ₂ ,..., G _M ) are composed of M as many as the number of feature points in the dictionary data, and G _j represents the priority of the j-th feature point in the dictionary data. Shall be given. The search flag F that limits the search space is
N search flags F as many as the number of feature points in the input data
(= F ₁ , F ₂ ,..., F _N ) and is given by two values of 0 and 1. The search flag F _i represents a search flag for the i-th feature point in the input data. When the value is 0, it indicates that the feature point is selectable. When the value is 1, it is not selectable. Represents the feature point of The corresponding feature point number sequence C is composed of M corresponding feature point number sequences C (= C ₁ , C ₂ ,..., C _M ) as many as the number of feature points in the dictionary data.
C _j indicates the feature point number in the input data corresponding to the j-th feature point in the dictionary data, and indicates the final result of the correspondence between the feature points in the dictionary data and the feature points in the input data. .

First, in initialization A2 in FIG. 8, k = 1,
_{F 1 = F 2 = ... =} F N = 0, C 1 = C 2 = ... = C M = -
Set to 1.

FIG. 9 shows that the number of feature points in the dictionary data is 5,
Using the similarity table when the number of feature points in the input data is 11, the states of the dictionary feature point priority order G, the search flag F, and the corresponding feature point number sequence C in the initialization A2 are shown.

At A3 in FIG. 8, the j-th feature point in the dictionary data where G _j = k and N search flags F =
The i-th data that has the highest similarity with all the feature points in the input data that has a value of 0 at (F ₁ , F ₂ ,..., F _N )
The _max-th feature point selected as the corresponding feature point, i _max is stored in the j-th corresponding feature point number C _j.

In the case of k = 1 in FIG. 10, in the second feature point in the dictionary data having the feature point priority order G _j = 1, the feature point in the input data corresponding to the search flag of 0 is set. each similarity between _{(S 21, S 22, ...} , S 211) the maximum similarity feature points in the third input data when a S ₂₃ in the is selected and the corresponding feature points, the corresponding feature points Number C ₂ to 3
Indicates a state where is stored.

[0038] In A4 in FIG 8, to change the value of the search flag F _imax that correspond to the feature points of the corresponding feature point number C _j selected in step A3 to 1.

In the example of FIG. 10, the third feature point in the input data corresponding to the second feature point in the dictionary data is selected as the corresponding feature point, and the third feature point corresponding to the selected corresponding feature point is selected. the search flag F ₃ shows a state in which change from 0 to 1.

At A5 in FIG. 8, 1 is added to k.

At A6 in FIG. 8, it is determined whether or not k is greater than or equal to a threshold value K that takes the same value as the number of feature points in the dictionary data. If so, go to step A7.

At A7 in FIG. 8, the search flag F (=
F _1, F _2, ..., examine the value of F _N), all go to step A5 time of 1, otherwise go to step A8.

At A8 in FIG. 8, a corresponding feature point corresponding to a feature point in the (j-1) -th or (j + 1) -th dictionary data in the same stroke as the j-th in the dictionary data of G _{j = k} It is determined whether or not exists. That is,
whether the corresponding feature point C _{j-1 of} s _j = s _{j- 1} is a positive value,
Alternatively, it is checked whether the value of the corresponding feature point C _{j + 1 such} that s _j = s _{j + 1} is positive, and when the value indicates a positive value, the corresponding feature of the adjacent feature point to the feature point of the j-th dictionary data It is determined that a point exists, and the process goes to step A9.
It is determined that there is no corresponding feature point of an adjacent feature point with respect to the feature point of the second dictionary data, and the procedure goes to step A3.

In the example of FIG. 11 where k = 2, it is assumed that the first feature point in the dictionary data whose feature point priority is G _j = 2 is in the same stroke (s ₁ = s ₂ ). the second corresponds to the neighboring feature points corresponding feature points indicates that the corresponding feature is present because C ₂ is 3 to.

At A9 in FIG. 8, the j-th feature point in the dictionary data of G _j = k and the search flag F (= F ₁ ,
F ₂ ,... F _N ), the local similarity is calculated and obtained for all the feature points in the input data indicating a value of 0. When the value of the corresponding feature point number C _j-1 is positive, the local similarity between the j-th feature point in the dictionary data and the i-th feature point in the input data is given by: _ij = S _ij + W ₃ + f ₃ (d ₃ ) + W ₄ × f ₄ (d ₄ )
-Pn (i, j).

Here, d ₃ is the difference between the angle from the C _j-1 th feature point in the input data to the i th feature point and the j th writing angle feature t _j in the dictionary data. Indicates that
d ₄ is the length of the line segment between the C _j-1 th feature point and the i th feature point in the input data and the length of the line segment formed by the j−1 th and j th features in the dictionary data The absolute value of the difference between
pn (i, j) is a constant value E when the C _j-1 th feature point and the i th feature point in the input data have the same stroke.
And a value of 0 for different strokes. Also,
W ₃ and W ₄ are both 1 and E is 50. In this embodiment, if sqrt represents a square root, f
₃ (x) and f ₄ (x) are f ₃ (x) = sqrt
(X), f ₄ (x) = sqrt (x), but any type may be used as long as a monotonic increasing function as shown in FIG. 6 is used.

[0047] In A10 in FIG. 8, a first i _max 'th feature point in the input data having the highest value among the local similarity obtained in step A9, the in G j _{= k} becomes dictionary data The j-th feature point is set as a corresponding feature point, and i _max ′ is stored in the j-th corresponding feature point C _j .

In the example of FIG. 11 where k = 2, the local similarity with the feature point in the input data at the first feature point in the dictionary data having the feature point priority order G _j = 2 is the largest. Is L _ll , and the l-th data in the corresponding input data is
This shows a state where the th feature point is selected and 1 is stored in the corresponding feature point _Cl .

At A11 in FIG. 8, a search flag F _r between the corresponding feature points C _j-1 and C _j (where r = C _j-1 )
The search flag indicating a value of 0 in the range from to F _imax ′ is set to 1, and the procedure goes to step A5. In the example where k = 2 in FIG. 11, the first feature point in the input data corresponding to the first feature point in the dictionary data whose feature point priority is G _j = 2 is selected, This shows a state where the values of the search flags F ₁ and F ₂ have changed from 0 to 1.

After performing the processing of each of the above steps,
12, the similarity table and the corresponding feature point sequence C (= C
_{1, C 2, ..., C} M) is sent to the recognition score calculating unit 16.

FIG. 12 shows the results of the correspondence between the feature points in the dictionary data and the feature points in the input data, the search flag F, and the corresponding feature point sequence C at k = 5.

FIG. 13 shows an example of the result of the corresponding feature point selection unit for Hiragana "i". The matching score calculator 26 calculates the similarity S between each feature point in the dictionary data and the corresponding feature point C (= C ₁ , C ₂ ,..., C _M ) in the input data.
Is corrected. The C _j-th similarity S _rj of feature points in the input and the j-th feature point in the dictionary data Data (however,
The correction of r = C _j ) is defined as ( _Equation 9) S _rj = S _rj + W ₅ × f ₅ (d ₅ ) + W ₆ × f ₆ (d ₆ ).

[0053] Here, r = C _j, d ₅ is the line segment connecting the line segment which connects the j-th feature point in the dictionary data and the C _j-th feature point in the corresponding input data , Represents the difference in the length of the corresponding line segment, and d ₆ represents the angle difference of the corresponding line segment.
Further, W ₅ and W ₆ are both set to 1. In this embodiment, sq
If rt represents a square root, f ₅ (x), f
₆ (x) is f ₅ (x) = sqrt (x), f ₆ (x)
= Sqrt (x), but any value may be used as long as a monotonically increasing function as shown in FIG. 7 is used.

The final matching degree R representing the degree of matching between the input data and the dictionary data is defined by the following _equation (10). R = _{ΣS rj} / M × h / H × (N−u) / N Calculate and output.

Here, r = C _j , H denotes the number of strokes in the dictionary data, and h denotes the number of surplus strokes (the number of strokes in which the search flags corresponding to each feature point in the strokes in the input data are all 0). ), M is the number of feature points in the dictionary data, N is the number of feature points in the input data, u is the number of uncorresponding feature points in the input data (a search flag F indicating a value of 0).
(= F ₁ , F ₂ ,..., F _N ).

Next, a second embodiment of the present invention will be described with reference to FIG.

FIG. 14 is a block diagram showing a second embodiment of the present invention. Referring to FIG. 14, the second embodiment of the present invention is different from the first embodiment shown in FIG. 1 in that a local feature correction unit 26 is provided. The operation of this embodiment will be described with reference to FIG. Input data storage unit 21, feature point extraction unit 22, feature point matching calculation unit 23, dictionary data storage unit 2
4. Corresponding feature point selection unit 25 and matching score calculation unit 27
Is the same as that of the first embodiment, and the description is omitted.

In this embodiment, after performing the corresponding feature point selecting section 25, the local feature correcting section 26 superimposes the positions between the corresponding feature points as shown in FIG. The similarity S is corrected based on the local error due to the displacement. The correction of the similarity S _ij of the i-th corresponding feature point in the input data corresponding to the j-th feature point in the dictionary data is represented by ( _Equation 11) S _ij = S _ij + W ₇ × f ₇ (d ₇ + D ₈ ).

Here, d ₇ indicates the distance between the j−1th feature point in the dictionary data and the corresponding feature point on the input side,
d ₈ is the distance between the feature points of the input side corresponding to the (j + 1) -th feature point in the dictionary data. W ₇ is set to 1. In this embodiment, assuming that sqrt is a square root, f ₇ (x) is f ₇ (x) = sqrt (x). However, if a monotone increasing function as shown in FIG. Such a thing may be used.

The process of the local feature correcting unit 26 and the process of the corresponding feature point selecting unit 25 are repeated a certain number of times T (T is 3 in this embodiment), and the matching score calculating unit 27 outputs a matching value.

[0061]

As described above, the line / graphic matching apparatus according to the present invention includes a corresponding feature point selecting section for selecting a feature point in input data corresponding to a feature point in dictionary data in a one-to-one correspondence. Therefore, the feature points in the dictionary data and the corresponding feature points in the input data always correspond one-to-one, and the correspondence with a plurality of feature points such as one-to-two, one-to-three, two-to-three, etc. This has the effect that processing for thinking is not required.

When selecting a feature point in the input data corresponding to a feature point in the dictionary data, the stroke order information is used only for the writing direction in the stroke to select a corresponding feature point. Is provided, there is an effect that matching can be performed even on a line figure in which the stroke order and the number of strokes vary.

Further, since a feature point extracting unit for extracting input data expressing original data in more detail than dictionary data is provided, it is possible to select a feature point in the input data corresponding to a feature point in the dictionary data. This has the effect that attachment is possible.

[Brief description of the drawings]

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

FIG. 2 is a diagram showing sampling in a case where Hiragana “tsu” is taken as an example.

FIG. 3 is a diagram showing attributes of feature points.

FIG. 4 is a diagram in which attributes of feature points are coded.

FIG. 5 is a conceptual diagram showing features at an i-th feature point.

FIG. 6 is a diagram showing an example of a monotone increasing function.

FIG. 7 is a diagram showing a similarity table created from all combinations of feature points in dictionary data and feature points in input data.

FIG. 8 is a flowchart showing the operation of a corresponding feature point selecting unit.

FIG. 9 is an example diagram showing a state of a corresponding feature point sequence C and a search flag F in initialization using a similarity table.

FIG. 10 is an example diagram showing a feature point correspondence result, a corresponding feature point sequence C, and a search flag F at k = 1 using a similarity table.

FIG. 11 is an example diagram showing a feature point correspondence result, a corresponding feature point sequence C, and a search flag F at k = 2 using a similarity table.

FIG. 12 is an example diagram showing a feature point correspondence result, a corresponding feature point sequence C, and a search flag F at k = 5 using the similarity table.

FIG. 13 is a diagram when the result of the corresponding feature point selected by the corresponding feature point selecting unit is an example of Hiragana “I”.

FIG. 14 is a block diagram showing a functional configuration of a second embodiment of the present invention.

FIG. 15 is a conceptual diagram of local positioning of a local feature correction unit.

FIG. 16 is a block diagram showing a functional configuration showing an example of a conventional technique.

FIG. 17 is a diagram showing an example of correspondence between dictionary segments and input segments.

[Explanation of symbols]

 Reference Signs List 11 input data storage unit 12 feature point extraction unit 13 feature point matching calculation unit 14 dictionary data storage unit 15 corresponding feature point selection unit 16 matching score calculation unit 21 input data storage unit 22 feature point extraction unit 23 feature point matching calculation unit 24 dictionary Data storage unit 25 Corresponding feature point selection unit 26 Local feature correction unit 27 Matching score calculation unit 31 Input data storage unit 32 Line segment feature extraction unit 33 Corresponding line segment candidate extraction unit 34 Dictionary data storage unit 35 Relaxation processing unit 36 Matching score calculation Department

Claims (2)

(57) [Claims] An input data storage unit for storing line graphic data input as online data, and a dictionary data storage unit for storing dictionary data information prepared in advance for calculating the degree of matching with the input data. A feature point extraction unit that extracts feature points from input data output from the input data storage unit and adds and stores feature information; and a feature point and a dictionary in the input data output from the feature point extraction unit. A feature point matching calculation unit for comparing and calculating the similarity indicating the degree of correspondence by combining the features with the feature points in the data and creating a similarity table; and dictionary data based on the similarity table. A corresponding feature point selecting unit that selects a feature point in the input data corresponding to the feature point in, a feature point in the dictionary data selected by the corresponding feature selecting unit, and a feature point in the input data. Line figure aligning apparatus characterized by comprising a matching score calculator for calculating a degree of match line figure line drawing and the dictionary data of the input data based on the similarity. 2. A local feature correction unit for correcting a similarity between a feature point in dictionary data selected by the corresponding feature selection unit and a corresponding feature point in input data using a local feature. Line figure matching device.