WO2014141881A1

WO2014141881A1 - Method for associating between characteristic point sets, association device, and association program

Info

Publication number: WO2014141881A1
Application number: PCT/JP2014/054642
Authority: WO
Inventors: 井戸伸彦
Original assignee: Ido Nobuhiko
Priority date: 2013-03-11
Filing date: 2014-02-26
Publication date: 2014-09-18
Also published as: CN105190647A; HK1214388A1; CN105190647B; JP2014174780A; JP6154627B2

Abstract

In the present invention, when associating a characteristic point of one object and a characteristic point of another object by comparing the two objects, which have characteristic points as elements, such as handwritten input characters having strokes as elements, variations due to object distortion and variations that are the subject of normalization are absorbed, and accurate and high-speed associating is performed. A plurality of coordinate axes are stipulated that are not parallel with each other in an N-dimensional space in which the coordinates of characteristic points are defined, an integer representing the order of appearance in a row of characteristic points sorted by the coordinate values in the coordinate axes is considered to be the sequential coordinate value of the characteristic points in the coordinate axes, a calculation value that monotonically increases with respect to the difference in sequential coordinate values is included in the cost between characteristic points, and an association is performed that lessens the cost. The variations are absorbed by means of the sequential coordinate values not changing even if there are small variations in the N-dimensional space of the characteristic points, and by means of the sequential coordinate values not changing in other coordinate axes even if the sequential coordinate values do change in some coordinate axes.

Description

Association method between feature point sets, association device, and association program

The present invention compares two objects, which are sets of feature points having coordinates composed of N coordinate values in N (N is a natural number of 2 or more) dimensional space, and compares the feature points in one set with the other. The present invention relates to an associating method, an associating device, and an associating program for determining correspondence with feature points in the set.

N-dimensional (特徴 N is a natural number of 2 or more) 技術 The technology for associating each feature point in two objects in space has applications in various fields. A feature point here has a coordinate which consists of N coordinate values corresponding to N-dimensional space, and points out what is an element of an object. A feature point may have an attribute value other than coordinates.

For example, a handwritten input character expressed by a stroke on a two-dimensional plane and a template character referred to for comparison with the object may be an object, and the start point and the end point of each stroke may be considered as feature points. I can do it. The feature point association in this case is applied in the field of on-line character recognition, as can be seen in Patent Document 1.

Alternatively, an audio signal expressed on a two-dimensional plane of a spectrum with two axes of frequency and intensity can be used as an object, and a point on the path can be considered as a feature point. In this case, the correspondence between the feature points of the two audio signals is applied in the field of speech recognition, as seen in Patent Document 2.

Any of the above application fields are general technical fields described in textbooks (Non-Patent Document 1).

The association of two feature points in these applications means that the two feature points have the same mutual positional relationship on the two-dimensional plane with other feature points in each object. . That is, since the association is performed for other feature points, the correspondence can be established as a whole only after the mutual positional relationship in each object is the same.

In addition, the objects in these applications have fluctuations called “distortion” and “displacement” for each individual, making the association difficult. DP (dynamic programming) matching, which is one of methods called elastic mapping, is known as a method for performing association by absorbing fluctuations called distortion and deviation (Patent Document 1, Non-Patent Document 2).

In DP matching, costs are determined for all combinations of feature points in one object and feature points in the other object, and the total cost of the pair by one-to-one correspondence is calculated. It is a solution to the optimization problem to be minimized. As described in Non-Patent Document 2 (Section 2.1) as “local distance”, this cost is often referred to as distance. Actually, the cost value often includes the Euclidean distance between the two feature points in total. However, as described in Non-Patent Document 4 or Patent Document 1 as the prior art, a difference in shape may be expressed as a distance, and the term distance is not necessarily a definition of geometric distance. It does not mean something that fits. For example, the value based on the “writing point movement direction” in Non-Patent Document 3 (Section 5.1) is included in the distance.

As in Non-Patent Document 4, a case is considered in which the Euclidean distance obtained from the coordinates of the two feature points in the N-dimensional space is included in the cost in DP matching. If the Euclidean distance is large, the cost increases, and it becomes negative to associate two feature points. That is, it means that the difference between the two feature points is large in the mutual positional relationship with other feature points in each object. Conversely, if the Euclidean distance is small, it means that there is little difference in the mutual positional relationship between two feature points. As described above, in the correspondence such as DP matching, the distance between the feature points obtained by the coordinates in the N-dimensional space is used as a measure of the difference in the mutual positional relationship between the feature points in the object. Can be interpreted. In the following description, only the Euclidean distance used as a measure of the magnitude of the difference in mutual positional relationship in the N-dimensional space of feature points in the object will be referred to as a distance. Of course, not only the Euclidean distance but also those that do not satisfy the definition of Manhattan distance or geometric distance are called distances. What evaluates the difference in the shape described in Non-Patent Document 3 as a distance is not called a distance.

From the viewpoint of correspondence, the above-mentioned “distortion” and “displacement” change the coordinates of the N-dimensional space even though the mutual positional relationship of the feature points within the object does not change, and as a result, Can be regarded as an event in which fluctuation occurs.

On the other hand, it is common to perform some pre-processing on the target object prior to association, regardless of what association method is applied. For example, Non-Patent Document 3 (Section 5.1) describes that “normalization” is performed on character data. Non-Patent Document 1 (Section 2.4.2) also describes that parallel movement and enlargement / reduction processing are used as processing for matching. For example, in the case of a character input by handwriting, the position and size of the drawn character are naturally not uniform, so correction is necessary for the size and position. That is, the object to be associated includes a variation that is a target of processing generally called normalization. As described in Non-Patent Document 3, the Euclidean distance between the feature points is usually calculated after normalization.

JP-A-9-179940

JP-A-6-51793

In the conventional association including the elastic mapping described in Non-Patent Document 3, the distance calculated on the N-dimensional space of the coordinate system that takes the coordinate value including the variation due to the distortion is used as the cost. Yes. For this reason, a non-simple method such as DP matching is required for the association.

When calculating the distance between feature points, if a coordinate system that takes the coordinate value that represents the essence of the mutual positional relationship is used instead of a coordinate system that takes a coordinate value that includes variation due to distortion, the association is easier. An effect that can be achieved by a method, or that a more accurate and high-speed processing can be achieved even if the same method is used is produced. Furthermore, fluctuations to be normalized are also absorbed if a coordinate system that takes coordinate values representing the essence of the mutual positional relationship is used.

The present invention converts a coordinate system that includes coordinate values including variations due to distortion and variations to be normalized into a new coordinate system that represents the essence of the mutual positional relationship, and converts the distance on the new coordinate system. It is an object to provide a method, an apparatus, and a program for associating feature points used.

The first configuration of the feature point associating method according to the present invention is an object having two or more feature points having coordinates composed of N coordinate values in an N (N is a natural number of 2 or more) dimensional space. When comparing the feature points of the first object and the feature points of the second object, the cost between the feature points to be matched is determined and In the feature point association method for determining the association so as to reduce the total value of the costs between the attached feature points, two or more non-parallel coordinate axes on the N-dimensional space are defined, and the plurality of coordinate axes For each coordinate axis, a coordinate value on the coordinate axis of the feature point that is an element of the first object is obtained, and the feature points are sorted in ascending or descending order according to the obtained coordinate value. A first ordered coordinate calculating step of obtaining an ordered sequence of feature points, and calculating an integer value indicating an appearance order of the feature points in the ordered sequence as an ordered coordinate value of the feature points on the coordinate axis; and The second order coordinate calculation step of calculating the order coordinate value of the feature point that is an element of the object, two feature points of the feature point of the first object and the feature point of the second object A cost calculating step for determining the cost between feature points by including a calculated value that monotonously increases with respect to a difference between the order coordinate values of the two feature points on each coordinate axis of the plurality of coordinate axes; and And determining the association using the cost determined in the cost calculating step.

According to this configuration, fluctuation due to distortion is absorbed by the following two effects.
The first effect is that the order coordinate values are maintained at the same value even if the coordinate values slightly change due to fluctuations due to distortion.

The second effect is that the order coordinate values on the remaining coordinate axes are maintained even if the order coordinate values may fluctuate on some coordinate axes of the plurality of coordinate axes. It is that the fluctuation | variation of is suppressed small.

Furthermore, size and position normalization prior to matching is not necessary. In addition to absorbing fluctuations due to strain, “using ordered coordinate values has an effect equivalent to elastic mapping such as DP matching” or “strain that elastic mapping handles has been removed. It can be interpreted that the ordinal coordinate values represent the essential mutual positional relationship.

According to a second configuration of the feature point association method of the present invention, in the first configuration, the first ordered coordinate calculation step and the second ordered coordinate calculation step include the ordered sequence of the feature points. Instead of an integer value indicating the order of appearance in, a value obtained by converting the integer value by a monotone function is used as the order coordinate value.

According to this configuration, even if the number of feature points that are the elements of the first object and the number of feature points that are the elements of the second object are greatly different, a reasonable association Can be obtained. As a monotonic function used for conversion, for example, if a function corresponding to the values of points arranged at regular intervals on a closed section on a number line is selected, the order coordinate value is normalized. When a complex shape is associated with a simplified shape, it is convenient to use the normalized order coordinate values.

According to a third configuration of the method of associating feature points according to the present invention, in the first or second configuration, the object is a character expressed as one or more strokes on a two-dimensional plane. Is the stroke, and the coordinates of the feature point are composed of coordinate values of the coordinates of one or more points on the two-dimensional plane representing the position of the stroke.

According to this configuration, since individual differences of various characters accompanying writing are absorbed, it is possible to effectively perform stroke correspondence that does not depend on the stroke order.

The first configuration of the feature point associating device according to the present invention is an object having two or more feature points having coordinates composed of N coordinate values in N (N is a natural number of 2 or more) dimensional space. When comparing the feature points of the first object and the feature points of the second object, the cost between the feature points to be matched is determined and In the feature point associating apparatus for determining the correspondence so as to reduce the total value of the costs between the attached feature points, two or more non-parallel coordinate axes on the N-dimensional space are defined, and the multiple coordinate axes For each coordinate axis, a coordinate value on the coordinate axis of the feature point that is an element of the first object is obtained, and the feature points are sorted in ascending or descending order according to the obtained coordinate value. First order coordinate calculation means for obtaining an order sequence of feature points, and calculating an integer value indicating an appearance order of the feature points in the order sequence as an order coordinate value on the coordinate axis of the feature points; The second order coordinate calculation means for calculating the order coordinate value of the feature point that is an element of the object, two feature points of the feature point of the first object and the feature point of the second object The cost calculation means for determining the cost between feature points by including a calculated value that monotonically increases with respect to a difference between the order coordinate values of the two feature points on each coordinate axis of the plurality of coordinate axes; and And means for determining correspondence using the cost determined by the cost calculation means.

According to a second configuration of the feature point associating device of the present invention, in the first configuration, the first ordered coordinate calculating means and the second ordered coordinate calculating means may be configured such that the ordered sequence of the feature points. Instead of an integer value indicating the order of appearance in, a value obtained by converting the integer value by a monotone function is used as the order coordinate value.

According to a third configuration of the feature point association apparatus according to the present invention, in the first or second configuration, the object is a character expressed as one or more strokes on a two-dimensional plane, and the feature point Is the stroke, and the coordinates of the feature point are composed of coordinate values of coordinates of one or more points on a two-dimensional plane representing the position of the stroke.

The configuration of the program according to the present invention is characterized by causing a computer to execute the feature point association method according to the first to third configurations.

As described above, according to the present invention, fluctuations due to distortion and fluctuations to be normalized are absorbed at the time of conversion to order coordinate values, so the distance based on the order coordinate values is used as a cost for association. Thus, there is an effect that accurate association can be obtained even if the procedure for associating feature points is simplified. Even when a procedure that is not simplified is used, there is an effect of improving the accuracy of correspondence and speeding up.

It is a notional block diagram of the feature point matching apparatus according to the embodiment of the present invention. A plurality of coordinate axis trains and feature points as numerical examples are illustrated on a secondary plane. 1 is a conceptual configuration diagram of a writing test apparatus according to an embodiment of the present invention. It is an operation example of a writing test apparatus.

Hereinafter, the best mode for carrying out the present invention will be described with reference to the drawings.

In the description, an example on a two-dimensional plane is used instead of a general N-dimensional space. In the following description, the variable N is used as the number of feature points of the first object unless otherwise specified.

FIG. 1 (b) is a conceptual block diagram of the feature point matching apparatus (100) according to the embodiment of the present invention, and exchanged data is also shown in the figure. The feature point associating device (100) is composed of an object input means (110), a cost calculation means (120), and an association determination means (130). Further, the cost calculating means (120) is composed of a distance cost calculating means (121) and another cost calculating means (125), and the distance cost calculating means (121) is an order coordinate value calculating means (122) and a distance calculating means. (123) and coordinate axis definition holding means (124).

Note that other cost calculation means (125) will be referred to in the description of FIG.

Referring to FIG. 1, the object input means (110) inputs a first object and a second object whose elements are feature points to be matched. For example, a handwritten input character represented by a stroke and a character referred to as a correct answer correspond to this.

From the object input means (110) to the order coordinate value calculation means (122) in the cost calculation means (120), the two-dimensional plane coordinates (191) of the feature points that are the elements of the object are passed as data. This data is defined as in equation (1). Data such as Expression (1) is input for each of the first object and the second object.

In the following explanation, specific numerical examples are shown in parallel with the definition in order to improve intelligibility. As a numerical example of equation (1), equation (2) is given.

The order coordinate value calculation means (122) inputs a plurality of coordinate axis definitions (195) from the coordinate axis definition holding means (124), and calculates the order coordinates (192) of the feature points. A general multiple coordinate axis definition (195) is shown in equation (3). In Equation (3), K coordinate axes are _defined as a coordinate axis array F = (f _k ). The coordinate axis f _k is represented by a function that gives the coordinate value of the point (x, y) on the coordinate axis. That is, the coordinate axis f given by the vector (a, b) is represented by f = ax + by.

Formula (4) shows a numerical example of formula (3). Represented by the formula (4), the _{f 0} is the x-axis, coordinate axes of which was rotated counterclockwise by [pi / _4, and has a _{_{f 1, f 2, f 3}} . Which coordinate axis should be defined depends on the application field.

It should be noted that an extension that uses a non-linear function for the coordinate axis definition function can be recalled, but it is not an essential extension because the same effect can be obtained by performing coordinate transformation in advance.

Next in order coordinate value calculating means (122), for each coordinate axis f _k, to calculate the coordinate values of the feature point p _n, to implement the sorting (ascending) the set of feature points by its coordinates. The sorted feature point sequence O _fk (P) is shown in Equation (5).

Formula (6) shows a numerical example of Formula (5). This value will be described with reference to FIG. FIG. 2 is a diagram illustrating a plurality of coordinate axis trains of Equation (4) and feature points of Equation (2) on a secondary plane as numerical examples. For example, when viewed in the direction of the vector f _{1 in} FIG. 2, the feature points are arranged in the order of p ₁ , p ₂ , p ₀ , p ₃ . This is the meaning of O _f1 (P) in formula (6).

When the order of _pn counted from 0 in the feature point O _fk (P) is represented as O _fk (p _n ), it can be expressed as in Expression (7). However, “[..]” in the expression (7) takes a value of “1” when “..” is true and “0” when it is false. Equation (5) does not clearly indicate how to determine the order when the coordinate values are the same, but in equation (7), the second term on the right side follows the subscript of the variable of the feature point. As such a tie-breaking method, there are other methods such as (a) following the order when the coordinate axes are slightly rotated, and (b) treating them in the same order. Unless there are many arrangements in which three or more points are arranged on the same straight line, there is no problem with the method of equation (7).

The order coordinates (192) in FIG. 1 are defined as in Expression (8) using _Ofk (p _n ) in Expression (7). The order coordinate as shown in Expression (8) is calculated by the order coordinate value calculation means (122) for each of the first object and the second object.

Equation (9) shows a numerical example of equation (8). For example, the value ( _{0, 2, 3, 3} ) of the value of O _F (p ₀ ) is “0” when the order of p ₀ in the expression (6) is O _f0 (P), and “0” when O _f1 (P) is “ 2 ”, O _f2 (P) is“ 3 ”, and O _f3 (P) is“ 3 ”.

When considering the sequential coordinates as in equation (9), the coordinate axes based on the reverse direction vector are only reversed in order, so the multiple coordinate axis definition (195) given by equation (4) is based on the origin. It can be understood that only the coordinate axes corresponding to the non-parallel vectors having the end points in the first quadrant and the second quadrant should be considered.

The distance calculation means (123) that receives the order coordinates (192) of the feature points of the first and second objects calculates the cost (193) between the feature points using this. As the calculation method, various calculation methods based on the difference between the order coordinate values may be used. In the present embodiment, the cost d _O is obtained by taking “the sum of the differences in the order coordinate values” shown in Expression (10). In addition to Expression (10), a method using a calculation method such as “geometric mean of difference in order coordinate value” or “square of sum of squares of difference in order coordinate value” is also conceivable. As long as the difference in order coordinates increases, the cost increases. That is, any calculation value that monotonously increases with respect to the difference in order coordinates may be used. In Expression (10), P corresponds to the first object, and Q corresponds to the second object.

For example, when a second object having a numerical example such as Expression (11) is considered, the cost d ₀ (p based on the distance between the feature point p ₁ in the first object and the point q _{2 in} the second object is considered. ₁ , q ₂ ) is as shown in Equation (12). Furthermore, the costs {d _O (p _n , q _m )} for all combinations of feature points in the first object and points in the second object are as shown in Table 1.

In the numerical example shown in Table 1, the number of feature points is the same between the first and second objects. However, when these are different, by normalizing as shown in Equation (13), for example, A complicated shape can be associated with a simplified shape. The reason why the denominator of Expression (13) is set to N−1 instead of N is to make the lowest order “1”. That is, the normalized order coordinate value is not an integer value but a rational number in the closed interval [0, 1]. Even when normalized, the definition of the ordinal distance follows the equation (10).

In Expression (13), the normalized order coordinate value is obtained by a linear function with respect to the rank O _fk (p _n ). However, the function to be used is not necessarily linear, and may be a monotone function. For example, the function of Expression (14) is monotonous and may be used. When Expression (14) is used, the difference in the ranking is sensitively reflected in the normalized order coordinate value near the center, and the degree of reflection is reduced as the position approaches the highest or lowest position. It is reflected in the same way. In addition, equations (13) and (14) are both monotonically increasing functions. There is no problem with using a monotonically decreasing function, but it is meaningless because the order is reversed.

In FIG. 1, the association determining means (130) inputs the cost including the cost (193) {d _O (p _n , q _m )} between the feature points according to the equation (10) as in the numerical example of Table 1. Then, the association (194) is determined. The feature of the present invention is that the order coordinate value is used for calculating the cost, and the procedure used in the association determining means (130) does not have to be specific. In the present embodiment, a simple association procedure is adopted in which association is performed in order from the lowest cost. That is, the smallest d _O (p _{n ′} , q _{m ′} ) is selected from the costs {d _O (p _n , q _m )} in which both _pn and q _m are not associated, and this p _The procedure of repeating associating _{n ′} and q _{m ′} until all feature points are associated is adopted. This procedure of associating in order from the lowest cost does not necessarily obtain the optimum association that minimizes the total cost. However, since the cost based on the order coordinate values works effectively, accurate association can be obtained even using such a simple association procedure.

When the association determining means (130) performs association according to the cost of the numerical example of Table 1 using the procedure of associating in ascending order from the above-mentioned lowest cost, the association (indicated by a circle in Table 1) 194) is obtained.

In the embodiment described above, the procedure of associating in ascending order of cost is used on the premise that one-to-one correspondence is performed. However, the present invention is not necessarily limited to only one-to-one correspondence. Absent. For example, when a complex shape is associated with a simplified shape by performing normalization as in Expression (13), the correspondence is one-to-many.

Next, an embodiment will be described in which a kanji writing test is performed on an apparatus having a touch panel by associating character images by using the feature point association apparatus (100) of the first embodiment. .

FIG. 3 is a conceptual configuration diagram of the writing test apparatus (300) according to the embodiment of the present invention, and exchanged data is also shown in the drawing. In FIG. 3, the correlating device (100) shown in FIG. 1 is used by the writing test device (300) from the outside, but for the sake of consistency with FIG. Of course, the form corresponding to the function of the associating device (100) may be incorporated in the writing test device (300).

The writing test apparatus (300) includes a position input means (320) and a display means (330) in the touch panel (310), a handwriting input control means (340), a writing problem management means (350), and a scoring means (360). Has been. Further, the scoring means (360) is composed of a midpoint calculating means (361), an angle calculating means (362), and a correctness determination means (363).

In FIG. 3, the writing problem management means (350) outputs a writing question (391) to the display means (330), and the user (399) visually recognizes it (392), and position input means (320). Input (393) by drawing is performed. In response to this, the position input means (320) outputs the coordinate information (394) of the drawing input to the handwriting input control means (340). The handwriting input control means (340) displays the character image (395) by the display means (330), and outputs the stroke information (396) of the handwritten input characters to the scoring means (360) when the drawing input ends and scoring starts. To do.

Subsequently, in the scoring means (360), the midpoint calculating means (361) is applied to the stroke information (396) of the handwritten input characters and the stroke information (397) of the correct answer characters input from the writing problem management means (350). And angle calculation means (362). The midpoint calculation means (361) obtains the midpoint of each stroke in the stroke information. When the xy coordinates of the midpoint are input to the associating device (100), the coordinate value of the feature point, that is, the value of Expression (1) is obtained.

The angle calculation means (362) obtains the angle with respect to the x-axis at the start point and end point of each stroke in the stroke information. This is used to calculate the cost in the other cost calculation means (125) in FIG. 1 when input to the associating device (100). In the present invention, in addition to the cost due to the distance based on the difference between the order coordinate values, which is a feature of the present invention, other costs are also added as necessary to be input to the association means (130). The angle with respect to the x-axis at the start point and end point is an example. In this embodiment, since only the midpoint of the stroke obtained by the midpoint calculating means (361) cannot be distinguished from the first and second strokes of the “10” character, for example, this is another cost. Costs due to different angles are required. Formula (15) gives the definition of cost d _S by angle.

In the associating device (100) in FIG. 3, of the stroke information that has been processed by the midpoint calculating means (361) and the angle calculating means (362) in the scoring means (360), the stroke information of the correct character. As a first object, stroke information of handwritten input characters is input as a second object. The association is performed by the processing described with reference to FIG. 1, and the value d of Expression (16) obtained by adding Expression (10) and Expression (15) is used as the cost at this time.

As a result of the processing by the associating device (100), the associating (194) is output to the correctness determination means (363) in the scoring means (360). The correctness / incorrectness determination means compares the associated strokes or the relationship between the strokes to determine whether the correct answer characters and the handwritten input characters match at a certain level or more, and determines correct / incorrect answers. The result (398) is output to the display means (330).

In the embodiment described with reference to FIG. 3, a character composed of a stroke is used as an object, and the order coordinate value is calculated using the coordinates on the plane of the midpoint of the stroke that is the feature point. There are other methods than using the coordinates of a single point on the plane. For example, the order coordinate value may be calculated in the four-dimensional space (xs, ys, xe, ye) using the xy coordinates of the start point (xs, ys) and the end point (xe, ye) of the stroke. In this case, if F = (xs, xs + ys, ys, −xs + ys, xe, xe + ye, ye, −xe + ye) is taken as the coordinate axis sequence, the processing corresponds to the ordering of the start point and the end point separately.

As in the example in the four-dimensional space, the coordinates of the feature points may be composed of coordinate values on the two-dimensional plane representing the stroke position. That is, it is not necessary to use only the coordinate value of a single point on the two-dimensional plane as in the case of using the coordinate value of the midpoint of the stroke. Even when the coordinate value of a single point on the two-dimensional plane is used, it does not have to be a middle point.

There is also a method of defining multiple costs based on order coordinate values. That is, other cost calculation means (125) in FIG. For example, when a character is an object, a method may be used in which one distance cost calculation unit is provided for the start point and another distance cost calculation unit is provided for the end point. This method is similar to the above example using the coordinate axis sequence F = (xs, xs + ys, ys, −xs + ys, xe, xe + ye, ye, −xe + ye), but the point corresponding to the coefficient in equation (16) can be set. Etc. are different.

FIG. 4 shows an operation example of the writing test apparatus described in FIG. In FIG. 4, a correct answer character image (401) and a handwritten input character image (402) are characters handwritten on a writing test apparatus equipped with an actual touch panel. Numbers are given in the order of strokes, and the numbers are associated with each other as an index.

In FIG. 4, the path (SVG) (411) of the stroke of the correct answer is the path information corresponding to 401, written in the notation of SVG (Scalable Vector Graph) adopted in HTML5. Similarly, 412 is path information 402. In these path information, the curve is expressed by a quadratic Bezier curve. In SVG, the y-axis is directed toward the bottom of the character and is opposite to a general xy coordinate plane. The coordinate values of the

path information

411 and 412 and the coordinate values and order coordinate values appearing in the following description follow the y-axis direction of the SVG.

In FIG. 4, the coordinates of the midpoint of the stroke and the ordinal coordinates (420) are defined as the coordinates of the midpoint of each stroke (image) extracted from the path information of 411 and 412, and these are defined by equation (8) and the like. It is the value converted to ordinal coordinates. Note that the order coordinates in FIG. 4 indicate values that are not normalized. An approximate value is used for the coordinate value of the midpoint for speeding up the processing.

The order distance (430) between strokes in FIG. 4 is a table showing the result of obtaining the order distance according to the equation (10) with respect to the value of 420. The stroke of the correct answer character corresponding to 401 is written in the vertical direction of the table, and the stroke of the handwritten input character corresponding to 402 is written in the horizontal direction. A combination of images that are associated with the value of 430 using the procedure for associating in order from the lowest cost is indicated by a circle in the table. In addition, when the writing test apparatus operates in a practical situation, the costs shown in Expression (15) are added together to perform association, but Expression (15) is omitted because it is not directly related to the present invention.

It can be confirmed that the strokes (midpoints), which are the characteristic points of the

characters

401 and 402, are correctly associated as indicated by the circles in the table 430. Although the image of the handwritten input character shown in 402 is considerably distorted, the association is correctly performed by the simple procedure of associating in order from the above-mentioned one with the lowest cost.

In the embodiment described above, the writing test apparatus using handwritten characters is dealt with. However, the associating apparatus according to the present invention can be used in a form similar to online character recognition. Of course, in addition to handwritten characters, various objects such as images, audio signals, and sentences can be applied as objects.

It can be used for correspondence between two objects for various objects such as images, audio signals, and texts, and can be applied to uses such as authentication, recognition, and search.

DESCRIPTION OF SYMBOLS 100 Feature point matching apparatus 110 Object input means 120 Cost calculation means 121 Distance cost calculation means 122 Order coordinate value calculation means 123 Distance calculation means 124 Coordinate axis definition holding means 125 Other cost calculation means 130 Association determination means 191 Feature point 2 Next plane coordinate value 192 Feature point order coordinate 193 Feature point cost 194 Corresponding 195 Multi-coordinate axis definition 300 Writing test device 310 Touch panel 320 Position input means 330 Display means 340 Handwriting input control means 350 Writing problem management means 360 Scoring means 361 Midpoint calculation means 362 Angle calculation means 363 Correct / error determination means 391 to 398 Arrow 399 indicating the flow of data User 401 Image of correct answer 402 Image of handwritten input character 411 Stroke of correct answer Roke Pass (SVG)
412 Handwritten input character stroke path (SVG)
420 Stroke midpoint coordinates and order coordinates 430 Order distance between strokes

Claims

Comparing two objects having two or more feature points as elements having coordinates composed of N coordinate values in N (N is a natural number of 2 or more) dimensional space,
When associating the feature point of the first object with the feature point of the second object,
In the feature point associating method for determining the cost between the feature points to be matched, and determining the matching so as to reduce the total value of the costs between the matched feature points.
Defining two or more non-parallel axes on the N-dimensional space;
For each coordinate axis of the plurality of coordinate axes, a coordinate value on the coordinate axis of the feature point that is an element of the first object is obtained, and the feature points are sorted in ascending or descending order according to the obtained coordinate value. A first ordered coordinate calculating step of obtaining an ordered sequence of the feature points, and calculating an integer value indicating an appearance order of the feature points in the ordered sequence as an ordered coordinate value of the feature points on the coordinate axes; A second order coordinate calculation step for calculating the order coordinate value of the feature point which is also an element of
The cost between two feature points of the feature point of the first object and the feature point of the second object is the order coordinates of the two feature points on each of the plurality of coordinate axes. A cost calculating step that includes a calculated value that monotonously increases with respect to a difference in values;
Determining an association using the cost determined in the cost calculating step;
A feature point associating method characterized by comprising:
In the first ordered coordinate calculating step and the second ordered coordinate calculating step, instead of an integer value indicating the appearance order of the feature points in the ordered sequence, a value obtained by converting the integer value by a monotone function is used. The method for associating feature points according to claim 1, wherein the method is used as an order coordinate value.
The object is a character expressed as one or more strokes on a two-dimensional plane, the feature points are the strokes, and the coordinates of the feature points are one or more on the two-dimensional plane representing the positions of the strokes. 3. The method for associating feature points according to claim 1 or 2, comprising coordinate values of the coordinates of the points.
Comparing two objects having two or more feature points having N coordinate values in N (N is a natural number of 2 or more) dimensional space as elements,
When associating the feature point of the first object with the feature point of the second object,
In the feature point associating device that determines the cost between the feature points to be matched, and determines the matching so as to reduce the total value of the costs between the matched feature points.
Defining two or more non-parallel axes on the N-dimensional space;
For each coordinate axis of the plurality of coordinate axes, a coordinate value on the coordinate axis of the feature point that is an element of the first object is obtained, and the feature points are sorted in ascending or descending order according to the obtained coordinate value. A first order coordinate calculation means for obtaining an order sequence of the feature points, and calculating an integer value indicating an appearance order of the feature points in the order sequence as an order coordinate value of the feature points on the coordinate axes; Second order coordinate calculation means for calculating the order coordinate values of the feature points that are also elements of the object,
The cost between two feature points of the feature point of the first object and the feature point of the second object is expressed as the order coordinates of the two feature points on each coordinate axis of the plurality of coordinate axes. A cost calculation means that includes a calculation value that monotonously increases with respect to a difference in value;
Means for determining correspondence using the cost determined by the cost calculation means;
An apparatus for associating feature points characterized by comprising:
The first ordered coordinate calculating means and the second ordered coordinate calculating means, instead of an integer value indicating the appearance order of the feature points in the ordered sequence, a value obtained by converting the integer value by a monotone function, The feature point associating device according to claim 4, wherein the feature point associating device is used as an order coordinate value.
The object is a character expressed as one or more strokes on a two-dimensional plane, the feature points are the strokes, and the coordinates of the feature points are one or more on the two-dimensional plane representing the position of the stroke. 6. The feature point associating device according to claim 4, wherein the feature point associating device comprises coordinate values of point coordinates.
A program that causes a computer to execute the feature point association method according to any one of claims 1 to 3.