JPH08249426A - Data collating device - Google Patents

Abstract

PURPOSE: To provide a data collating device capable of accurately judging the similarity of a reference pattern with an input pattern even when the reference pattern is included in the valley of potential corresponding to the input pattern. CONSTITUTION: This device is equipped with a reference pattern storage part 13 which stores the reference pattern as an elastic body formed in such a way that a lattice is assumed, and a positive charge is arranged in a specific lattice point corresponding to a feature point representing the shape of the reference pattern, and also, a negative charge is arranged in a remaining lattice point. and adjacent point charges are coupled with a first spring, and also, the lattice point is coupled with the point charge arranged on the lattice point with a second spring, a potential generating part 12 which generates an electrostatic field provided with the valley corresponding to the shape of the input pattern, a dynamical equlibrium point evaluating part 16 which places the elastic body in the electrostatic field virtually and finds a position where balance between a Coulomb force and an elastic force is obtained. and an energy evaluating part 17 which calculates the whole energy of the system.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data collating device used for processing such as character recognition, graphic recognition and voice recognition.

[0002]

2. Description of the Related Art In order to recognize a character or a figure, unknown data and known data are collated (matched) and the similarity is determined by some standard.

The template matching method, which uses known data as a template (an example of a standard pattern) and measures the degree of matching with unknown data (input pattern), is the most basic matching method and is used for cutting out images. .
However, in the template matching method, since the standard pattern and the input pattern basically have to match exactly, it is not possible to deal with the positional displacement, rotation, deformation, etc., which are required for pattern recognition and the like.

In recent years, a mechanical recognition model has been studied as a method capable of coping with deformation and displacement. This considers the standard pattern as a stretchable elastic film such as rubber, and drops it into the potential field formed from the input pattern to determine the similarity between the standard pattern and the input pattern.
This is a method of determining by the total energy of the system when the elastic force of the elastic body and the force received from the potential field are balanced.

In an actual data collating apparatus, since numerical calculation is performed by a computer, a mass point model that uses a mass point and a spring, which is convenient for numerical calculation, is used instead of the virtual stretchable elastic film. In the mass model, as shown in FIG.
The feature points 201 representing the standard pattern 301 are regarded as mass points, and the mass points are connected by the spring 202. Also, each mass point 2
01 are also connected to the corresponding initial positions 200 by springs 203. Then, the energy of the system is calculated when the mass point 201 at the initial position 200 moves by the potential force and balances with the restoring force of the springs 202 and 203. Note that the potential field is not shown in FIG.

As a conventional data collating device employing such a mass point model, one shown in FIG. 14 is known. The standard pattern storage unit 103 represents and stores the standard pattern as an elastic body using a mass model. On the other hand, the potential generation unit 102 generates, from the input pattern 101, a virtual potential field having the valleys at the characteristic points of the pattern. The mechanical equilibrium point evaluation unit 108
Using the generated potential field data and the elastic body stored in the standard pattern storage unit 103, the mechanical force balance position is obtained.

Now, assuming image data, the potential is Φ (X, Y), the mass of a mass point is m, and the number of all mass points forming a standard pattern is n. The initial position of the i-th mass point in the standard pattern is represented by a position vector o _i , and the position at time t is represented by a position vector ξ _i (t). When the spring constant of the spring 203 connecting the initial position and the mass point is k _o , the restoring force vector n _i to the initial position is expressed as n _i = k _o (ο _i −ξ _i ) ... (1).

The spring constant of the spring 202 connecting the mass points is k _N. When the j-th mass point is connected to the i-th mass point, the restoring force vector Ν _i for displacement between both mass points is N _i = Σ _j k _N (| ξ _j −ξ _i | − | ο _j −ο _i |) (Ξ _j −ξ _i ) / | ξ _j −ξ _i | ... (2)

Further, the vector f _i of the force received by the potential Φ is expressed as f _i = −m∇Φ ... (3)

Letting the acceleration vector of each mass point be α _i (t), the equation of motion by the resultant force of formulas (1), (2), and (3) n _i + Ν _i + f _i = m α _i (t) ... (4) Consider exercise according to. Then, for all i, when | n _i + N _i + f _i | <ε (5), it is determined that the system is in a balanced state (dynamic equilibrium point). However, ε represents a sufficiently small value.

When the mechanical equilibrium point evaluation unit 108 finds a state that satisfies the equation (5), the energy evaluation unit 107 calculates the total energy E of the system at that time. That is, as shown in the following equation (6), the potential energy mΦ and the elastic energy k _N (| ξ _j −ξ _i | − | ο _j −ο
_i |) ^2/2, and the initial position the spring of the elastic energy k _o | o _i
-Ξ _i | ^2/2 to be added for all of the mass point and the spring.

[0012] _{E = Σ i (mΦ + Σ} j k N (| ξ j -ξ i | - | ο j -ο i |) 2/2 + k o | ο i -ξ i | 2/2) ... (6) This The similarity between the standard pattern and the input pattern can be determined by the magnitude of the total energy E.

[0013]

By the way, FIG. 15 and FIG. 16 both show a standard pattern 30 having seven feature points 201.
2, 303 three-dimensionally shows a state in which each equilibrium point is reached on the potential surface 300 (z coordinate corresponds to the potential value). Note that these are shown in plan view in FIGS. 18 and 19 (in the drawings, the initial position of the feature point 201 is indicated by the “x” mark). FIG. 15 shows the standard pattern 3
When the characteristic point 201 of 02 is largely deviated from the valley 300a of the potential, FIG. 16 corresponds to the case where the characteristic point 201 of the standard pattern 303 is substantially coincident with the valley 300a of the potential.

When the minimum value of the potential is Φ _min = 0, the maximum value of the potential is Φ _max = 5 (the unit is arbitrary), and the mass of each mass point is m = 1, it is calculated according to the above equation (6). The energy value E becomes E = 26.49 in the case of FIG. 15 and E = 1.71 in the case of FIG.
That is, the higher the similarity between the input pattern (potential) and the standard pattern (FIG. 16), the smaller the energy E of the system and the expected result.

Next, as shown in FIG. 17, the characteristic points 201 of the standard pattern 304 exist only in a part of the potential valley 300a, and the similarity between the standard pattern and the input pattern is not so high. Imagine a case. The standard pattern 304 is represented by three feature points 201. It is to be noted that FIG. 20 shows this in plan view. When the energy E of the system in this case is obtained in the same manner as in FIGS. 15 and 16, E = 1.09, which is a very small value. As described above, according to the mass point model, when the number of feature points 201 forming the standard pattern is small and the standard pattern is included in the potential valley 300a,
Although the pattern similarity is not very high, the energy value E of the system is small. Therefore, there is a problem that the similarity cannot be correctly determined.

In the mass point model, starting from the initial position, the force acting on each mass point is calculated, then the motion of the mass point is calculated according to the equation of motion, and the position where the forces balance is obtained by the mechanical relaxation process. In this mechanical relaxation process, since there are many systems, it is necessary to sequentially calculate the movement amount for each minute time for each mass point. Therefore, there is a problem that the calculation takes time in the case of a practical scale.

Therefore, an object of the present invention is to accurately determine the similarity between the standard pattern and the input pattern even when the standard pattern is included in the potential valleys corresponding to the input pattern. The object is to provide a data collation device that can do so. It is a further object of the present invention to provide a data collating device capable of obtaining a position where forces are balanced at a higher speed without sequentially tracking the amount of movement based on the equation of motion.

[0018]

In order to achieve the above object, a data collating apparatus according to claim 1 is a data collating apparatus for determining the similarity between a standard pattern and an input pattern, and A positive charge is placed at a specific lattice point corresponding to the feature point representing the shape, a negative charge is placed at the remaining lattice points, and the row and column directions are placed between adjacent point charges placed on the lattice points. And a standard pattern storage unit that stores the standard pattern as an elastic body that is coupled by a first spring and the lattice points and point charges arranged on the lattice points are coupled by a second spring. , Generates an electrostatic field in the two-dimensional region in which the elastic body is to be placed, in which the potential value of the specific region is set lower than the potential values of the remaining regions, corresponding to the shape of the input pattern. Po When the initial generation unit and the elastic body stored in the standard pattern storage unit are placed in the electrostatic field generated by the potential generation unit, the Coulomb force received by the point charge from the electrostatic field and the 1
And a mechanical equilibrium point evaluation unit that obtains a position where the elastic force received from the second spring and the elastic force received from the second spring, and the electrostatic potential energy of the point charge and the elasticity of the first and second springs at the balanced position. An energy evaluation unit for calculating the sum of energy and energy is provided.

A data collating apparatus according to a second aspect is the data collating apparatus according to the first aspect, wherein individuals that generate a group of individuals having a set of relative coordinates with respect to a lattice point of each point charge as a gene are generated. A genetic operation including at least one of selection, selection, mutation and mating is performed on the population generation part and the population of the individual to generate a new set of relative coordinates with respect to the lattice point of each point charge. And a genetic manipulation unit for generating a new population of individuals.

[0020]

The data collating apparatus according to the first aspect operates as follows.

The standard pattern storage unit stores standard patterns as elastic bodies in advance. That is, a positive charge is arranged at a specific lattice point corresponding to the characteristic point representing the shape of the standard pattern, and a negative charge is arranged at the remaining lattice points, and between the adjacent point charges arranged on the lattice point. The above-mentioned standard pattern is stored as an elastic body in which is connected by the first spring in the row direction and the column direction, and is connected by the second spring between the lattice points and the point charges arranged thereon. ing.

The potential generation unit compares the potential value of a specific area with the potential values of the remaining areas in the two-dimensional area where the elastic body is to be placed, corresponding to the shape of the input pattern input from the outside. To generate a low electrostatic field.

Next, when the elastic body stored in the standard pattern storage unit is placed in the electrostatic field generated by the potential generation unit, the mechanical equilibrium point evaluation unit determines that each point charge is A position (dynamic equilibrium point) where the Coulomb force received from the electrostatic field and the elastic force received from the first and second springs are balanced is obtained.

Next, the energy evaluation section obtains the sum of the electrostatic potential energy of the point charge and the elastic energy of the first and second springs at the mechanical equilibrium point.

Here, when the similarity between the standard pattern and the input pattern is high, the value of the energy sum becomes small, while when the similarity between the standard pattern and the input pattern is low, the energy sum becomes high. The value increases. Therefore, the similarity between the standard pattern and the input pattern can be determined by the magnitude of this energy sum.

Moreover, in this data collating apparatus, even if the standard pattern is included in the valley of the potential corresponding to the input pattern, the similarity between the standard pattern and the input pattern can be correctly determined. You can
That is, when the standard pattern is included in the potential valley corresponding to the input pattern, negative charges as well as positive charges exist near the bottom of the potential valley in the initial state. Due to the elastic force of the first and second springs, this negative charge does not escape from the potential valley even in the equilibrium state, and remains near the potential valley bottom even after moving to the equilibrium position. Therefore, as a result, it has a large potential energy. Therefore, when the characteristic points of the standard pattern exist only in a part of the potential valley, the contribution of the potential energy of this negative charge becomes large, and the energy of the entire system becomes large. As a result, the similarity between the standard pattern and the input pattern is correctly determined.

In the data collating apparatus according to the second aspect, the individual group generation unit generates a group of individuals having as a gene a set of relative coordinates with respect to the lattice point of each point charge. The mechanical equilibrium point evaluation unit places a point charge of the elastic body displaced in accordance with a set of relative coordinates of each individual belonging to the group in a two-dimensional area assumed by the potential generation unit. Thus, it is determined whether or not the point charges are in a position (dynamic equilibrium point) where the Coulomb force received from the electrostatic field and the elastic force received from the first and second springs are balanced. When an individual satisfying the balance condition is not found in the population, the genetic operation unit is
A genetic operation including at least one of selection, selection, mutation and mating is performed to generate a new population of individuals having a new set of relative coordinates with respect to the lattice point of each point charge as a gene. The dynamic equilibrium point evaluation unit is a two-dimensional region assumed by the potential generation unit again assuming that each point charge of the elastic body is displaced according to the set of relative coordinates of each individual of this new generation group. Put on. Thus, it is determined whether or not the point charges are in a position (dynamic equilibrium point) where the Coulomb force received from the electrostatic field and the elastic force received from the first and second springs are balanced. In this way, new generation groups are generated one after another by the genetic operation unit, and the equilibrium condition is determined by the mechanical equilibrium point evaluation unit. When an individual satisfying the equilibrium condition is found in the group, the energy evaluation unit determines the electrostatic potential energy of the point charge and the first electric charge at the equilibrium position.
And the sum of the elastic energy of the second spring and the elastic energy of the second spring.

As described above, this data collating apparatus can generate the position coordinates of each point charge by the method of the genetic algorithm. In this case, the dynamic equilibrium point evaluation unit does not need to analyze the equation of motion, and only needs to evaluate the force balance condition based on the set of relative coordinates of the individual. Moreover, since the genetic operation unit generates a plurality of individuals at a time, the mechanical equilibrium point evaluation unit can execute the evaluations in parallel based on the set of relative coordinates of the plurality of individuals. Therefore, the force balance position can be obtained at a higher speed than in the case of sequentially tracking the movement process over time.

[0029]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The data collating apparatus of the present invention will be described in detail below with reference to embodiments.

FIG. 1 shows a block configuration of a data collating apparatus according to an embodiment. This data collating apparatus includes a potential generation unit 12, a standard pattern storage unit 13, a mechanical parallel point evaluation unit 16, and an energy evaluation unit 17. The individual population generation unit 14 and the genetic manipulation unit 15 will be described later for convenience.

This data collating apparatus employs a point charge model in which a virtual elastic film in a mechanical recognition model is represented by a point charge placed on a lattice and a spring. That is, in this point charge model, a lattice is assumed as illustrated in FIGS. Then, among the lattice points arranged in a matrix, positive charges (indicated by “●” in the figure) 41 are arranged at specific lattice points corresponding to the characteristic points representing the standard pattern.
All the remaining lattice points are negatively charged (indicated by "○" in the figure)
43 is arranged. The adjacent point charges arranged on the grid point are coupled by the first spring 42 in the row direction and the column direction along the grid. Further, each point charge is also coupled to a corresponding lattice point (initial position indicated by "x" mark in the figure) by a second spring (not shown).

In this embodiment, it is assumed that a binary image is handled, and the positive point charge is +1 and the negative point charge is-.
A binary value of 1 is assumed. When handling a multi-gradation image, an appropriate threshold value may be set depending on the gradation to set positive charge and negative charge and set the charge amount.

The potential generator 12 generates a potential field corresponding to the shape of the input pattern 11.

For example, in the binary image representing the input pattern, the value of the black pixel (feature point) is set to the minimum potential value Φ.
_min = 0, the white pixel value is the maximum potential value Φ _max = 3
And Then, as illustrated in FIGS. 6 and 7, a white surface pixel adjacent to the black pixel (feature point) is subjected to a linear functional conversion process to generate the potential surface 30. In the example of FIG. 6, the minimum potential value Φ _min = 0 for two black pixels adjacent in the row direction near the center, the potential value 1 for the white pixels surrounding the two black pixels, and the outside of the white pixel. A potential value of 2 is given to the surrounding white pixels. Further, the value of the white pixel on the outside is the maximum potential value Φ _max = 3. In the example of FIG. 7, the minimum potential value Φ _min for two diagonally adjacent black pixels near the center and another black pixel separated by two pixels in the row direction.
= 0, the potential value 1 is given to the white pixels adjacent to these black pixels and surrounding each of them. The white pixel having the potential value 1 forms a connected closed region. The potential value 2 is given to the white pixels surrounding the outside of this closed region, and the value of the white pixels outside that is the maximum potential value Φ _max = 3. In this way, the potential generation unit 12 forms the potential surface 30 corresponding to the input pattern 11 and has the potential valley linearly inclined toward the black pixel (valley bottom) and the flat portion connected to the outside thereof. To generate.

The potential generation unit 12 uses the data representing the generated potential surface 30 as the mechanical equilibrium point evaluation unit 16.
Transfer to

On the other hand, the standard pattern storage unit 13 stores point charge patterns corresponding to various standard patterns, and sequentially sends data representing the point charge patterns to the mechanical equilibrium point evaluation unit 16.

The mechanical equilibrium point evaluation unit 16 applies the potential surface 30 provided by the potential generation unit 12 to
The point charge pattern sent from the standard pattern storage unit 13 is dropped, and the force balance is evaluated as follows.

That is, as in the case of the mass point model, the potential is Φ (X, Y) and the number of all point charges forming the standard pattern is n. The initial position of the i-th point charge in the standard pattern is represented by a position vector oi, and the position after movement is represented by a position vector ξi.

Assuming that the spring constant of the spring connecting the initial position and the point charge is k _o , the restoring force vector n _i to the initial position is expressed as n _i = k _o (ο _i −ξ _i ) ... (1) To be done.

The spring constant of the spring 42 connecting the point charges is k _N. When the j-th point charge is connected to the i-th point charge, the restoring force vector Ν _i for the displacement between the point charges is N _i = Σ _j k _N (| ξ _j −ξ _i | − | ο _j −ο _i |) (ξ _j −ξ _i ) / | ξ _j −ξ _i | ... (2)

The vector f ^Q _{i of the} force received by the potential Φ is expressed as f ^Q _i = -q ▽ Φ (3A) (where q = + 1 or q = -1).

With the acceleration vector of each point charge being α _i (t), the equation of motion n _i + Ν _i + f ^Q _i = mα _i (m α _i (m) for each point charge based on the resultant force of equations (1), (2) and (3A) t) Consider an exercise according to (4A). Then, for all i, when | n _i + Ν _i + f ^Q _i | <ε (5A), it is determined that the system is in a balanced state (dynamic equilibrium point). However, ε represents a sufficiently small value.

When the mechanical equilibrium point evaluation unit 16 obtains the state satisfying the expression (5A), the energy evaluation unit 17 calculates the total energy E of the system at that time. That is, as shown in the following equation (6A), the potential energy qΦ and the elastic energy k _N (| ξ _j −ξ _i | − | ο _j −ο _i
|) ^2/2, and the initial position the spring of the elastic energy k _o | o _i
-Ξ _i | ^2/2 every point summation over charge and spring.

[0044] _{E = Σ i (qΦ + Σ} j k N (| ξ j -ξ i | - | ο j -ο i |) 2/2 + k o | ο i -ξ i | 2/2) ... (6A) this The similarity between the standard pattern and the input pattern can be determined by the magnitude of the total energy E.

FIGS. 2 and 3 show three-dimensionally the state where the standard patterns 32 and 33 in the point charge model reach the equilibrium points on the potential surface (z coordinate corresponds to the potential value) 30, respectively. It is to be noted that these are shown in plan view in FIGS. 4 and 5 (in the drawings, initial positions of the point charges 41 and 43 are indicated by “x” marks). In FIG. 2, the feature points (4) 41 having 32 standard patterns are the potential valleys 30.
It is present only in part of a, and shows the case where the similarity between the standard pattern and the input pattern is not very high. On the other hand, FIG. 3 shows a case where the standard pattern and the input pattern substantially match.

The minimum value of the potential is Φ _min = 0, the maximum value of the potential is Φ _max = 5 (the unit is arbitrary), and the energy value E calculated according to the above equation (6A) is E in the case of FIG. = 16.92, E = 1 in the case of FIG.
It becomes 0.40. That is, the higher the similarity between the input pattern (potential) and the standard pattern (FIG. 3), the smaller the energy E of the system and the expected result.

The reason for this is as follows. In this point-charge model, positive charges (feature points) receive a force from the potential field to the potential valley, and negative charges receive a force from the potential field to the potential peak from the potential field. However, due to the elastic force of the spring, the negative charge 43, which was near the valley of the potential in the initial state, cannot escape from the valley 30a of the potential even in the equilibrium state, and remains near the valley of the potential even after moving to the equilibrium position. . Therefore, as a result, it has a large potential energy. Therefore, when the characteristic point 41 of the standard pattern exists only in a part of the potential valley 30a as shown in FIG. 2, the contribution of the potential energy of this negative charge becomes large and the energy E of the entire system becomes large. Of.

As described above, according to the point charge model, the similarity between the standard pattern and the input pattern is correctly determined even if the standard pattern is included in the potential valley corresponding to the input pattern. be able to.

In this data collating apparatus, the mechanical equilibrium point can be obtained at high speed by operating the individual population generating section 14 and the genetic operating section 15 shown in FIG.

Specifically, first, the individual population generating section 14
M objects (M is a natural number) as targets for genetic manipulation
To generate an individual. Each individual has, on its chromosome, a shift from the initial position (corresponding lattice point) of each point charge, that is, a set of relative coordinates of each point charge as a gene.

For example, as shown in FIG. 9, among the lattice points arranged in a matrix in the initial state for the standard pattern,
Negative charges m ₁ and m ₂ are arranged at lattice points whose (X, Y) coordinates are (1,1) and (1,2), and (X, Y) coordinates are (2,2).
Positive charges (feature points) at the lattice points 1) and (2, 2)
It is assumed that m ₃ and m ₄ are arranged. These four point charges m ₁ ,
When m ₂ , m ₃ , and m ₄ move to the positions shown in parent 1 in FIG. 11, the chromosome of this individual (parent 1) has four genes as shown in FIG. 10 (a). A set of relative coordinates {(0.1, -0.4), (0.1, 0.7), (0.
2, -0.5), (0.1, -0.4)}. For example, the first (0.1,
-0.4), the relative coordinates for the grid points of the point charge m ₁ (1,1), i.e., in the X-direction point charge m ₁ from the grid point (1,1) + 0.1, in the Y direction -0 It means that it is shifted by .4. The second (0.1, 0.7)
Shows that the point charge m ₂ is +0.
It shows that it is deviated by 0.7 in the 1 and Y directions.
Thus, the gene has four point charges m ₁ , m ₂ , m ₃ ,
It is represented by the deviation from the initial position (lattice point) of m ₄ . Note that when these four point charges m ₁ , m ₂ , m ₃ , and m ₄ move to the positions shown in parent 2 in FIG.
As shown in, the chromosome of this individual (parent 2) is a set of four relative coordinates {(0.1, -0.3), as a gene.
(0.7, 0.0), (-0.1, 0.4), (-0.
2, −0.6)}.

Next, the genetic operating unit 15 receives the received M
Genetic operations such as selection, selection, mutation, and mating are performed on each individual to generate M individuals of the next generation (details will be described later). However, the first time, M received
The genes of the M individuals are sent to the mechanical equilibrium point evaluation unit 16 as they are without applying a genetic operation to the individual individuals.

The mechanical equilibrium point evaluation unit 16 evaluates the force balance based on the set of relative coordinates received from the genetic operation unit 15 as individual genes. In this example, the left side of equation (5A) | n _i + Ν _i + f ^Q _i | for all point charges
Calculate the value of and sum these to obtain the evaluation value (total force) u
And

For example, in the example of FIG. 9, as shown in the figure,
Corresponding to the input pattern, the potential valley (0.5 <Y <
2.5) and a flat surface (Y <0.5 or Y> 2.5) connected to the outside of the potential surface 30 is generated. Then, the mechanical equilibrium point evaluation unit 16
Is assumed to have received the set of relative coordinates of the two individuals (parent 1 and parent 2) shown in FIG. At this time, the evaluation value u of the parent 1 is 20.114 and the evaluation value u of the parent 2 is 19.790.

The mechanical equilibrium point evaluation unit 16 compares the calculated evaluation value u with a predetermined reference value s (s = 1 in this example), and when the evaluation value u is larger than the reference value s. Judges that the balance condition is not satisfied, while it judges that the balance condition is satisfied when the evaluation value u is smaller than the reference value s.

For example, in the case of parent 1 and parent 2, it is determined that the balance conditions are not satisfied.

If none of the M individuals satisfy the balance condition, the genetic manipulating section 15 performs the genetic manipulation on the M individuals to generate M individuals of the next generation. To generate.

FIG. 8 shows "selection", "selection", "mutation", and "mating" as genetic operations for M = 8 individuals.
The following shows an example of sequentially executing. In this series of genetic operations, the reciprocal 1 / u of the evaluation value u is used as the fitness w of the individual.

More specifically, in "selection", the individual having the lowest fitness w is eliminated from the group of individuals. In this example, the individual 7 is extinguished.

In "selection", about 75 out of M = 8 seats
Individuals are selected and applied to the six seats corresponding to% with a probability proportional to the fitness w from the group. At the same time, individuals are randomly selected and applied to the remaining two seats corresponding to about 25% of the M = 8 seats, regardless of the fitness w, from the group.

In the "mutation", the value of the relative coordinates of the individuals in the population is randomly changed by random number processing.
The change of the coordinate value by the random number process is performed in the unit of a number representing the relative coordinate of each point charge. Therefore, some individuals do not mutate in the population. Further, even if an individual has a mutation, not all values of the set of relative coordinates are changed. In this example, individuals 3 and 6 are mutated.

In “mating”, a pair of individuals is selected from a group of M = 8, the values of the relative coordinates of specific parts are exchanged,
Create a new pair of individuals.

For example, assume that the pair of parent 1 and parent 2 shown in FIG. 10A is selected. First, as shown in FIG. 10B, each value of the relative coordinates of each of the parent 1 and the parent 2 is represented by a 4-digit binary number. The most significant digit of the four digits represents the sign of the relative coordinates. That is, 1 corresponds to + and 0 corresponds to-. Of the four digits, the remaining three digits represent a binary number that is obtained by multiplying the relative coordinate value by ten. That is, the least significant digit of these three digits corresponds to the first place after the decimal point of the relative coordinates.
In this way, each set of relative coordinates of the parent 1 and the parent 2 is represented by 32 bits. Then, by exchanging a specific bit, in this example, the 15th bit to the 26th bit, 2
One individual (child 1, child 2) is generated. As shown in FIG. 10 (c), the chromosome of this child 1 has a set of four relative coordinates {(0.1, −0.4), (0.1, 0.
4), (-0.1, 0.4), (-0.1, -0.
4)}. On the other hand, the chromosome of child 2 is a set of four relative coordinates {(0.1, -0.3), (0.
7, -0.3), (0.2, -0.5), (0.2,-
0.6)}. The pair of relative coordinates of the child 1 and the child 2 corresponds to the positions of the point charges m ₁ , m ₂ , m ₃ , and m ₄ shown in FIG. 12 on the potential surface.

The mechanical equilibrium point evaluation unit 16 again evaluates the force balance by the above-mentioned method based on the set of relative coordinates of the new generation individuals from the genetic operation unit 15. Then, the calculated evaluation value u is compared with a predetermined reference value s (s = 1 in this example), and when the evaluation value u is larger than the reference value s, it is determined that the balance condition is not satisfied. On the other hand, when the evaluation value u is smaller than the reference value s, it is determined that the balance condition is satisfied.

For example, in the generation including the above child 1 and child 2,
As shown in FIG. 10C, the evaluation value u of the child 1 is 0.69.
6, the evaluation value u of the child 2 is 25.050. in this case,
Child 1 shows a mechanical equilibrium point.

As described above, if there is an individual satisfying the balance condition among the M individuals, the dynamic equilibrium point evaluation unit 1
6 is an energy evaluation unit 1 that indicates a set of relative coordinates of the individual.
Send to 7.

The energy evaluation unit 17 evaluates the total energy E of the system at the above-mentioned balanced position based on the set of relative coordinates of the individual. Finally, of all the standard patterns stored in the standard pattern storage unit 13, the one giving the smallest total energy E is specified.

As described above, this data collating apparatus can operate the individual population generating section 14 and the genetic operating section 15 to generate the relative coordinates of each point charge by the method of the genetic algorithm. In this case, the mechanical equilibrium point evaluation unit 16 does not need to analyze the equation of motion,
Only the force balance condition needs to be evaluated based on the set of relative coordinates of the individual. Moreover, since the genetic operation unit 15 generates a plurality of individuals at once, the dynamic equilibrium point evaluation unit 16
Can perform evaluations in parallel based on the set of relative coordinates of multiple individuals. Therefore, the force balance position can be obtained at a higher speed than in the case of sequentially tracking the movement process over time.

In the actual processing of the genetic operation, the number of elements of the set of relative coordinates held by the individual as a gene matches the number of point charges arranged at the lattice points.

Of course, the positions may be changed so that the potential field is formed from the standard pattern and the elastic body is formed from the input pattern.

[0071]

As is apparent from the above, in the data collating apparatus according to the first aspect, the dynamic equilibrium point evaluation unit generates the elastic body stored in the standard pattern storage unit by the potential generation unit. When placed in the electrostatic field, the position (dynamic equilibrium point) at which the Coulomb force received by each of the point charges from the electrostatic field and the elastic force received by the first and second springs are balanced is obtained. The energy evaluation unit obtains the sum of the electrostatic potential energy of the point charges and the elastic energy of the first and second springs at the mechanical equilibrium point. When the similarity between the standard pattern and the input pattern is high, the value of the energy sum becomes small, while when the similarity between the standard pattern and the input pattern is low, the value of the energy sum becomes large. Therefore, the similarity between the standard pattern and the input pattern can be determined by the magnitude of this energy sum.

Further, in this data collating apparatus, when the standard pattern is included in the potential valley corresponding to the input pattern, the negative charge as well as the positive charge stays near the bottom of the potential valley even after moving to the equilibrium position.
Therefore, the contribution of the potential energy of this negative charge becomes large, and the energy of the entire system becomes large. Therefore, the similarity between the standard pattern and the input pattern can be correctly determined.

According to another aspect of the data collating apparatus of the present invention, the individual group generation unit generates a group of individuals having a set of relative coordinates with respect to the lattice point of each point charge as a gene. And
By the genetic operation unit, new populations of individuals having new sets of relative coordinates for the lattice points of the above point charges as genes are generated one after another, and the equilibrium point evaluation unit can determine the equilibrium condition. . When an individual satisfying the balance condition is found in the group, the energy evaluation unit
The sum of the electrostatic potential energy of the point charges and the elastic energy of the first and second springs at the balanced position is obtained.

As described above, since the data collating apparatus can generate the relative coordinates of each point charge by the method of the genetic algorithm, the mechanical equilibrium point evaluation unit does not need to analyze the equation of motion. Only the force balance condition needs to be evaluated based on the set of relative coordinates of the individual. Moreover, since the genetic operation unit generates a plurality of individuals at a time, the mechanical equilibrium point evaluation unit can execute the evaluations in parallel based on the set of relative coordinates of the plurality of individuals.
Therefore, the force balance position can be obtained at a higher speed than in the case of sequentially tracking the movement process over time.

[Brief description of drawings]

FIG. 1 is a diagram showing a block configuration of a data collating apparatus according to an embodiment of the present invention.

FIG. 2 is a diagram schematically showing a state in which an elastic body that represents one standard pattern by a point charge model is in a balanced position in a potential field corresponding to an input pattern.

FIG. 3 is a diagram schematically showing a state in which an elastic body representing another standard pattern by a point charge model is in a balanced position in the potential field corresponding to the input pattern.

FIG. 4 is a plan view showing the state of FIG.

5 is a plan view showing the state of FIG. 3. FIG.

FIG. 6 is a diagram illustrating a method of generating a potential field from one input pattern.

FIG. 7 is a diagram illustrating a method of generating a potential field from another input pattern.

FIG. 8 is a diagram illustrating processing of a genetic operation by a genetic operation unit.

FIG. 9 is a diagram illustrating an initial position and a potential distribution of an elastic body representing a standard pattern.

FIG. 10 is a diagram illustrating a process of mating by a genetic operation unit.

FIG. 11 is a diagram showing a state in which a pair of relative coordinates possessed by an individual parent 1 and parent 2 corresponds to the position of the point charges m ₁ , m ₂ , m ₃ , and m ₄ with respect to the lattice point.

FIG. 12 is a diagram showing a state in which a pair of relative coordinates possessed by an individual child 1 and child 2 corresponds to the position of the point charges m ₁ , m ₂ , m ₃ , and m ₄ with respect to the lattice point.

FIG. 13 is a diagram showing an elastic body in which a standard pattern is represented by a mass point model.

FIG. 14 is a diagram showing a block configuration of a conventional data matching device.

FIG. 15 is a diagram schematically showing a state in which an elastic body that represents one standard pattern by a mass model is in a balanced position in the potential field corresponding to the input pattern.

FIG. 16 is a diagram schematically showing a state in which an elastic body representing another standard pattern by a mass model is in a balanced position in the potential field corresponding to the input pattern.

FIG. 17 is a diagram showing a state in which a standard pattern is included in a potential valley corresponding to an input pattern.

FIG. 18 is a plan view showing the state of FIG. 15.

FIG. 19 is a plan view showing the state of FIG.

20 is a plan view showing the state of FIG.

[Explanation of Codes] 11 Input Pattern 12 Potential Generation Unit 13 Standard Pattern Storage Unit 14 Individual Population Generation Unit 15 Genetic Manipulation Unit 16 Mechanical Equilibrium Point Evaluation Unit 17 Energy Evaluation Unit

Claims (2)

[Claims] 1. A data collation device for determining the similarity between a standard pattern and an input pattern, wherein positive charges are arranged at specific lattice points corresponding to feature points representing the shape of the standard pattern, and the remaining Negative charges are arranged at the lattice points, and adjacent point charges arranged on the lattice points are connected by a first spring in the row direction and the column direction, and the lattice points and the point charges arranged on the lattice points are connected to each other. As the elastic body formed by connecting the two parts with the second spring, the standard pattern storage section for storing the standard pattern and the two-dimensional area in which the elastic body is to be placed correspond to the shape of the input pattern. A potential generation unit that generates an electrostatic field in which the potential value of the specific region is set lower than the potential values of the remaining regions, and the elastic body stored in the standard pattern storage unit,
When placed in the electrostatic field generated by the potential generating section, a mechanical equilibrium point for obtaining a position where the Coulomb force received by the point electric field from the electrostatic field and the elastic force received by the first and second springs balance each point charge. An evaluation unit and an energy evaluation unit that calculates the sum of the electrostatic potential energy of the point charges and the elastic energy of the first and second springs at the balanced position are provided. Data collation device. 2. The data collating apparatus according to claim 1, wherein an individual group generation unit for generating a group of individuals having a set of relative coordinates with respect to a lattice point of each point charge as a gene; Then, a genetic operation including at least one of selection, selection, mutation and mating is performed to generate a new population of individuals having a new set of relative coordinates as a gene with respect to the lattice point of each point charge described above. A data collating device comprising a dynamic operation section.