JP2002288676A - Device and method for image collation - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform image collation by a comparison between pixels at a practical speed and with high accuracy by performing a positioning with sufficient accuracy at a high speed. SOLUTION: A processing for searching the positions of a plurality of similar areas on an input image, corresponding to a plurality of small areas on a reference image, is performed in a search part 111. In this search processing, the degree of positional adoption on the positional combinations of a plurality of areas similar on the input image for a moved position, and a present position is evaluated by a positioning evaluation part 112, with the difference between the areas adjacent to each other being taken into consideration. Based on the evaluated results, the present position is moved controllably to the position to be moved. In an optimization part 113, optimization processing is performed for moving conditions, based on the evaluated results of the degree of positional adoption, so that a search range can be narrowed in the search process in the similar areas.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image collating apparatus and an image collating method, and more particularly to an image collating apparatus and an image collating method for collating two images based on a pixel value difference between the two images.

[0002]

2. Description of the Related Art An application for comparing two images, that is, a reference image and an input image, and judging coincidence / mismatch includes fingerprint collation. There are several methods of image matching in this case. One of the typical methods is a method of extracting features from each of two images and comparing the features. This method has a problem of feature extraction accuracy, and is not suitable for detecting a slightly different portion of a very similar image, but is generally fast.

In addition, there is a method of directly comparing corresponding pixel values of two images. This method is slow, because it is necessary to correct the displacement between two images before performing the pixel value comparison, and is slow, but is suitable for detecting a small difference portion of the images. For example, Japanese Patent Application Laid-Open No. H11-3431 discloses a technique for comparing images by such a direct pixel comparison method.

[0004]

However, in the conventional collation method in which pixels of two images are directly compared with each other, an enormous amount of time is required for positioning between the two images as described above. Is rarely used. Therefore, it has been considered to perform alignment by a relatively simple method of extracting a feature from the reference image, searching for an area having the same feature from the input image, and correcting a shift between the two images. However, with such a method, it is difficult to obtain sufficient alignment accuracy.

SUMMARY OF THE INVENTION The present invention has been made to solve the above-mentioned problem. The present invention makes it possible to execute positioning with sufficient accuracy at high speed, and to perform image matching by comparing pixels at a practical speed and at high speed. It is an object of the present invention to provide an image matching device and an image matching method that can be performed with high accuracy.

[0006]

According to the present invention, there is provided an image collating apparatus for collating a first image with a second image, comprising: means for dividing the first image into a plurality of regions; Moving the position of the corresponding similar area on the second image for each of the divided areas on the first image, while moving the position of the corresponding similar area on the second image on the second image. Searching means for searching for the positions of a plurality of similar regions, and a difference between a distance between adjacent regions on the first image and a distance between similar regions on the second image corresponding to the adjacent regions, respectively. A similar area to be moved by the search means based on pixel value differences between a plurality of areas on the first image and a plurality of similar areas on the second image corresponding to the plurality of areas. Moves when is at current position For each case of moving to a position, control is performed to evaluate the degree of position suitability for a combination of positions of a plurality of similar regions on the second image, and to control whether or not movement to the destination position is possible based on the evaluation result. Means.

In this image collating apparatus, when attention is paid to adjacent regions among a plurality of regions on the first image, similarities corresponding to the respective regions on the second image which is an input image. Considering that it is natural that the positions between the regions are adjacent to each other, the position matching degree regarding the combination of the positions of the plurality of similar regions on the second image is represented by:
Not only the difference in pixel values between the plurality of regions on the first image and each of the plurality of similar regions on the second image corresponding to those regions, but also the distance between adjacent regions on the first image and their neighbors The evaluation is also made using the difference between the distance between similar regions on the second image corresponding to the matching regions. As described above, by using the new evaluation scale regarding the positional relationship between the regions, the position matching degree between the current position and the destination position is evaluated at each movement operation to control whether or not the movement is possible, and thereby the difference between the pixel values is obtained. It is possible to perform the alignment more efficiently than in the case where the search control is performed by using only the search control.

In the control means, at the beginning of the search process, each of the plurality of similar regions is moved to an arbitrary destination position irrespective of the evaluation result of the degree of positional conformity, and thereafter, A search condition is determined by the search means based on the evaluation result of the positional suitability for each of the current position and the destination position so as to be limited to movement in a position direction in which the positional suitability increases. It is preferable to provide means for variably setting adaptively according to the progress of.

Thus, at first, each similar region moves from one position to another at any destination position, regardless of the degree of positional conformity temporarily reduced due to the movement. . Since the positioning is to optimize the relative positional relationship between similar regions, even if the movement of one similar region lowers the degree of positional conformity, the movement is later performed on another similar region. The movement may be optimal. By gradually tightening the movement conditions from such a state in which free movement is allowed, it becomes possible to limit the movement to only the direction in which the degree of position matching increases, and finally, the entire position matching Each similar area is converged to a position where the degree is highest. Therefore, even if the search range is not limited to a small range in advance, the time required for the search can be kept within a practical range, and positioning with sufficient accuracy can be performed at high speed.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows the configuration of an image matching system according to an embodiment of the present invention. This image matching system performs matching between a reference image and an input image, for example, to perform fingerprint matching or image tampering check. This image matching system is realized as software executed on a computer, and as its functional modules, as shown in the figure, an image input unit 11, an input image storage unit 12, a reference image storage unit 13, an image division unit 1
4, a collation processing unit 15 and a collation result output unit 16.

The image collation processing includes 1) input of an input image to be collated with the reference image, 2) alignment between the reference image and the input image, and 3) correspondence between the reference image and the input image. 4) Comparison of pixels to be performed is performed in the order of the comparison result output. In the present embodiment, particularly, by improving the alignment processing of 2), alignment with sufficient accuracy can be performed at high speed. Positioning is performed by dividing the reference image into a plurality of small areas, and determining the most suitable position on the input image for each of the plurality of small areas by the matching processing unit 15.

When performing image matching, first, an input image is input from the outside via the image input unit 11, and the input image is stored in the input image storage unit 12. A reference image is stored in the reference image storage unit 13 in advance, and the reference image is divided by the image division unit 14 into a plurality of mesh-shaped small regions. The matching processing unit 15 performs a process of searching, for each small area, a similar area having an image similar to the small area from the input images stored in the input image storage unit 12. The collation processing unit 15 includes a search unit 11 as shown in FIG.
1. It has an alignment evaluation unit 112 and an optimization unit 113.

The search unit 111 moves the position of the similar region on the input image for each small region, and moves the position of the similar region on the input image corresponding to each of the small regions on the reference image. Is searched for. In this search process, the destination positions of a plurality of similar regions on the input image are sequentially obtained, and the movement to the destination positions is performed. The next destination position (a candidate destination position) for each similar area is determined by generating a random number and adding the random number value to the coordinates of the current position. Each time the destination position of each similar area is determined, the position matching degree regarding the combination of the positions of the plurality of similar areas on the input image for each of the destination position and the current position is determined by the alignment evaluation unit 1.
The evaluation is performed by the control unit 12, and whether or not the current position is moved to the movement destination position is controlled based on the evaluation result.

The evaluation of the degree of positional conformity is performed based on the alignment evaluation index (D). The formula for calculating the alignment evaluation index (D) will be described later, but basically, the distance between adjacent small areas on the reference image and the distance between the input images corresponding to the adjacent small areas respectively. Calculated based on the difference between the distance between the similar regions and the pixel value difference between the plurality of small regions on the reference image and each of the plurality of similar regions on the input image corresponding to the small regions. Is done.

As described above, whether or not to move from the current position to the next destination position when searching for a position in a similar area is determined by
Since the control is performed based on the evaluation result of the position suitability for each of the current position and the next destination position of the similar area,
The position of each of the plurality of similar regions is searched so that the degree of position matching is the best. In this case, the optimizing unit 113 performs optimization processing on the moving condition based on the evaluation result of the degree of positional suitability so that the search range is narrowed down in the search process of the similar area. By this optimization process, the similar region initially moves within a relatively large search range, but the movement of the similar region is gradually restricted only in the direction in which the degree of positional adaptability increases from within the large search range. And finally converge to a position where the degree of positional conformity is the highest. In this way, the positioning between the reference image and the input image is performed in small area units.

Thereafter, the matching result output unit 16 compares corresponding pixels between the reference image and the input image, and outputs the comparison result as a matching result.

Hereinafter, specific processing contents will be described.

FIG. 2 shows how to check the correspondence between the reference image and the input image. In the present embodiment, the reference image (in FIG. _2, S 1 ~S ₄₎ fine meshed rectangular area _S 1 to S _n are divided into. And each area S
_{For i} (i = 1 to n), a part S ^′ _i having a similar pattern is selected from the input image. This means that alignment between the reference image and the input image is performed for each region.

After that, the sum D _P (S _i , S S of pixel value differences for each pair of the corresponding mesh areas S _i and S ^′ _i
^' _i ) is checked and the similarity (match / mismatch, degree of tampering, etc.) between the reference image and the input image is determined by adding the values to all regions.

[0020] Now, the case that on the reference image _{S i} and _{S j} are adjacent to each _other, to be written _(S i, S _j) and ∈L. If S _i and S _j are adjacent to each other,
_i , S ^′ _i on the input image corresponding to S _j
It is more natural to consider that and S ^′ _j are also located as close as possible. That is, if the difference between the center x coordinate of S _{i and} the center x coordinate of S _j is D _x (S _i , S _j ), D _x
(S _i , S _j ) must have substantially the same value as D _x (S ^′ _i , S ^′ _j ). Similarly, if the difference between the center y coordinate of S _{i and} the center y coordinate of S _j is D _y (S _i , S _j ), D _y (S _i , S _j ) becomes D _y (S ^′ _i , S ^′ _j ) must be almost the same value.

Therefore, in the present embodiment, the following value of D is used as the above-mentioned alignment evaluation index. D is a parameter indicating the naturalness of the alignment.

[0022]

(Equation 1)

Here, α is an empirically determined constant. The first term on the right-hand side of the above equation, adjacent small areas _S i, _S each set to their _S i of _{j, S} the distance between similar regions S between the _{j ^'i,} S
^' indicates the sum of differences for all sets when the difference from the distance between _j is calculated. The second term on the right-hand side indicates the sum of pixel value differences between the plurality of regions S _i (i = 1 to n) and the corresponding plurality of similar regions S ^′ _i (i = 1 to n). ing.

Adjacent small areas S_i, S_jMany combinations
There are several possible. For example, in FIG.₁And S₂The mutual
To S_i, S_jRelationship and S₃And S₄To each other S_i,
S_jMay be defined as₁And S₃The mutual
To S_i, S_jRelationship and S₂And S₄To each other S_i,
S_jMay be defined as Furthermore, S₁S _i
S when_jS₂, S₂S_iS when_j
S₄, S₄S_iS when_jS₃, S₃S
_iS when_jS₁, May be defined as follows.
In any case, the relationship between adjacent areas is
The same definition is used for the input image.

As can be seen from equation (1), in order to obtain the most natural alignment result, it is sufficient to determine a combination of S ^′ _j (i = 1 to n) that minimizes the above D. Become. However, the combinations of S ^′ _j (i = 1 to n) are infinite.

Therefore, in this embodiment, the formula (1)
Using a method called the mulated Aneeling method, an optimal combination that minimizes D in Expression (1) is obtained.

More specifically, first, in the initial state, S ′ _i (i = 1 to n) is determined almost randomly as shown in FIG. That is, S ′ ₁ to S ′ ₁ to S _{1 to} S ₄ corresponding to S ₁ to S ₄ , respectively.
The position of S ′ ₄ is determined at random. Of course, S ′ _{1 to} S ′ ₄ may be initially set at exactly the same positions as the respective coordinate relationships of S _{1 to} S ₄ .

Thereafter, S '₁~ S '₄In the order of
Done. Here, as shown in FIG.₁The random
Move to new destination area S '_tIs assumed. And
S ' ₁Is S '_tD to Dt when moving to
D corresponding to the position is D0.

At this time, a random number p taking a value between 0 and 1 is generated and compared with exp ^{(D0-Dt) / T.} Where T
Is a value called a temperature parameter. As a result, Tara smaller than p is ^{exp (D0-Dt) / T} , S '1
_Is moved to _S't, and _S't is set as a new current position.
Otherwise, S _'1 is S' does not move to _t.

[0030] In this way, the search control process of determining whether to move to the target position by the relationship between p and ^{exp (D0-Dt) / T} decide the destination position, e.g. S _'1 to S' _{The process} is repeated many times, once in the order of ₄ . At this time, if T is a sufficiently large value, (D
0-Dt) / T, regardless of the magnitude relationship between the _{D 0} and _{D t} 0
Therefore, the value of exp ^{(D0−Dt) / T} is 1, and the random number p between 0 and 1 is always exp
^It becomes smaller than ^{(D0-Dt) / T.} Therefore, S ′ ₁ to S ′ ₁ to
S _'4 each go free to rapidly move to an arbitrary position.

On the other hand, when T is a small value, (D0−
Dt) becomes more dependent on (D0−Dt).
Is positive, that is, only when Dt is smaller than D0, the random number p becomes smaller than exp ^{(D0-Dt) / T.} Therefore, when T is a small value, S ′ _{1 to} S ′ ₄ must be D
Move in the direction in which the value of becomes smaller. (Strictly speaking, D0-
(Even if Dt is negative, it moves when the value of the random number p at that time is small and Dt is not much larger than D0, but the probability is sufficiently lower than when D0-Dt is positive.) .

Therefore, in this embodiment, in the initial stage of the search process,
At S ′ regardless of the value of (D0−Dt)₁~
S '₄Each is moved to an arbitrary destination position and
After that, only when Dt is smaller than D0
S '₁~ S '₄First, go through T so that the movement of
Set a large value determined experimentally, and gradually increase T
I will lower it. That is, for example, under a certain T condition
S '₁~ S '₄Execute each search control process once
Then, with T lowered by a predetermined value, S '₁~ S '
₄Each search control process is performed once, and T
With the predetermined value lowered, S '₁~ S '₄Each search system
The control process is performed once. Do
And finally T is a sufficiently small lower limit predetermined
At the point when (Tt) is reached, S ′ at which D becomes the minimum₁~ S '
₄Is obtained.

FIG. 5 shows how D changes when search control for each of S ′ _{1 to} S ′ ₄ is repeatedly performed while gradually changing the moving conditions by changing T.

[0034] As illustrated, initially not Ika reduced easily while variation value of D because S _'1 ~S' ₄ will move randomly extensive freely. However, if the value of T is gradually reduced, the degree of dependence on the value of (D0-Dt) starts to increase, so that random free movement and movement in the direction in which D decreases begin to coexist. Among S ′ _{1 to} S ′ ₄ , even if the movement of a certain similar area is random, the movement of the similar area leads to a movement that relatively reduces the value of D in consideration of the movement of another similar area. Can be obtained. And more T
As S decreases, S ′ _{1 to} S ′ ₄ move to the optimum positions, so that the value of D starts to decrease toward a minimum value at a certain T, and eventually D decreases to a minimum value that does not decrease any more. Converge. By defining the final lower limit temperature (Tt) of T to a temperature at which the value of D converges, S ′ ₁ to
Optimum combination of positions of S _'4 is obtained.

Next, referring to the flowchart of FIG.
A series of processing procedures will be described.

First, S 'to be moved_nNext move about
The destination area (destination frame St) is represented by a random number and S^' _iCurrent location
(Step S101). next,
S ' _nWhen D is moved to the next destination area (St)
Is calculated by equation (1) (step S102). this
Then, a random number p that takes a value between 0 and 1 is generated, and p and exp
^{(D0-Dt) / T}Is compared with the destination area (S
It is determined whether the movement to t) is possible (steps S103, S
104). p is exp^{(D0-Dt) / T}Less than
In this case, movement to the movement destination area (St) is performed (step
Step S105), otherwise, no movement is performed. that's all
The search control process of S ′₁~ S '₄Done for all of
It is.

Then, S '₁~ S '₄Search for everything
When the rope control process makes one round, the value of the temperature parameter T
The value K is subtracted (step S106). And T
Until the value reaches the lower limit value Tt, steps S101 to S101
The search control processing at 104 is S ' ₁~ S '₄About everything
It is executed repeatedly.

When the value of T reaches the lower limit value Tt, the alignment processing ends, and the positions of S ′ _{1 to} S ′ ₄ at that time are set to S.
It is determined as the optimum combination corresponding to ₁ to S _4. After dividing into small regions and performing accurate positioning in this manner, by comparing pixels between S _{1 to} S ₄ and S ′ _{1 to} S ′ ₄ , for example, fingerprint matching or It is also possible to accurately perform a collation process that requires detection of a slight difference between the reference image and the input image, such as a tampering check of a handwritten signature.

As described above, according to the present embodiment, 1) S ′ _{1 to} S ′ ₄ by performing search control while evaluating the degree of conformity using the difference in distance between adjacent regions.
Can be efficiently set to the optimum position. 2)
By the movement condition variably set using a Simulated Aneeling method, without limiting narrowly the search range, S ^_'j
It is possible to quickly find the optimum combination from the infinite combinations (i = 1 to n).

Since all the functions of the image collating system of the present embodiment can be realized by software, a program for causing a computer to execute the above-described processing procedures is prepared, and stored in a computer-readable storage medium. The same effect as in the present embodiment can be easily obtained only by introducing the program into the computer through the storage medium and executing the program.

Further, the present invention is not limited to the above-described embodiment, and can be variously modified in an implementation stage without departing from the gist thereof. For example, in the above-described embodiment, when a certain similar area is moved, all the small areas S _i (i = 1 to n) and all the corresponding similar areas S ^′ _i (i = 1 to n) are moved. Although the evaluation value was obtained in consideration of the sum of the differences between the pixel values between the similar regions, the difference between the pixel values may be only the difference between the pixel values between the similar region and the corresponding small region. The difference between the distances is also the only difference between the distance between the similar region and another similar region determined to be adjacent to the similar region and the distance between the adjacent regions on the corresponding reference region. May be.

In the above description, the shape of the area to be divided first is rectangular, but the image may be divided into hexagonal areas instead of the rectangle. FIG. 7 shows the relationship between the reference image and the input image in this case. In FIG. 7, the image is searched in units of hexagonal small areas, and a more natural alignment result can be obtained because the degree of freedom of combination between the areas is greater than in the case of a rectangle. Become.

Further, the above embodiments include inventions at various stages, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent features. For example, even if some components are deleted from all the components shown in the embodiment, the problem described in the column of the problem to be solved by the invention can be solved, and the effects described in the column of the effect of the invention can be solved. Is obtained, a configuration from which this configuration requirement is deleted can be extracted as an invention.

[0044]

As described in detail above, according to the present invention, it is possible to perform a positioning with sufficient accuracy at a high speed, and to perform an image comparison by comparing pixels with a practical speed with a high accuracy. Becomes possible.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a functional configuration of an image matching system according to an embodiment of the present invention.

FIG. 2 is an exemplary view showing a correspondence relationship between a reference image and an input image in the embodiment.

FIG. 3 is an exemplary view for explaining an initial state position of each similar region in the embodiment;

FIG. 4 is an exemplary view for explaining a current position and a destination position of a similar area in the embodiment.

FIG. 5 is a view showing a state of a change in the degree of positional conformity when the search control for a similar area is repeatedly executed while the moving conditions are gradually tightened in the embodiment.

FIG. 6 is an exemplary flowchart illustrating the procedure of a positioning process in the embodiment.

FIG. 7 is a view showing another example of the correspondence between the reference image and the input image in the embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 11 ... Image input part 12 ... Input image storage part 13 ... Reference image storage part 14 ... Image division part 15 ... Matching processing part 16 ... Matching result output part 111 ... Search part 112 ... Alignment evaluation part 113 ... Optimization part

Claims (8)

[Claims] 1. An image collating apparatus for collating a first image with a second image, comprising: means for dividing the first image into a plurality of regions; Searching means for searching for the position of a plurality of similar regions on the second image corresponding to each of the plurality of regions on the first image while moving the position of the corresponding similar region on the second image; A difference between a distance between adjacent regions on one image and a distance between similar regions on the second image respectively corresponding to the adjacent regions, and a plurality of regions on the first image and corresponding to the regions; Based on the pixel value difference between each of the plurality of similar regions on the second image and the similar region to be moved by the search means, when the similar region to be moved is at the current position and when the similar region is moved to the destination position. about,
Control means for evaluating a position suitability for a combination of positions of the plurality of similar regions on the second image, and controlling whether or not movement to the destination position is possible based on the evaluation result. Image collating device. 2. The control unit according to claim 1, wherein in the initial stage of the search process, each of the plurality of similar regions is moved to an arbitrary destination position irrespective of the evaluation result of the degree of position suitability, and thereafter, the position is determined. A movement condition based on the evaluation result of the position suitability for each of the current position and the destination position is set according to the progress of the search process by the search means so that the movement is limited to the movement in the position direction in which the suitability increases. 2. The image collating apparatus according to claim 1, further comprising means for adaptively variably setting. 3. The image matching apparatus according to claim 1, wherein the search unit sequentially determines a destination position of each of the plurality of similar regions based on the current position and a random value. . 4. The method according to claim 1, further comprising calculating a difference between pixel values between the plurality of regions on the first image and the corresponding plurality of similar regions searched on the second image. The apparatus according to claim 1, further comprising a unit configured to output a result of comparison between the first image and the second image. 5. An image collating method for collating a first image with a second image, comprising: dividing the first image into a plurality of regions; A search step of searching for a position of a plurality of similar regions on the second image corresponding to each of the plurality of regions on the first image while moving a position of a corresponding similar region on the second image; A difference between a distance between adjacent regions on one image and a distance between similar regions on the second image respectively corresponding to the adjacent regions, and a plurality of regions on the first image and corresponding to the regions; Based on a pixel value difference between each of the plurality of similar regions on the second image, and a case where the similar region to be moved is at the current position and a case where the similar region is moved to the destination position by the search step. about Evaluating the degree of position suitability for a combination of the positions of the plurality of similar regions on the second image, and controlling whether or not the movement to the destination position is possible based on the evaluation result. Image matching method. 6. The control step includes: in the initial stage of the search process, each of the plurality of similar regions is moved to an arbitrary destination position irrespective of the evaluation result of the degree of position suitability, and thereafter, the position A movement condition based on the evaluation result of the position suitability for each of the current position and the destination position is set in accordance with the progress of the search process by the search step so that the movement is limited to the movement in the position direction in which the suitability increases. 6. The image matching method according to claim 5, further comprising a step of adaptively variably setting. 7. The image matching method according to claim 5, wherein in the searching step, a destination position of each of the plurality of similar regions is sequentially determined based on the current position and a random number. . 8. The method according to claim 8, further comprising calculating a difference between pixel values between the plurality of regions on the first image and the corresponding plurality of similar regions searched on the second image. The image matching method according to claim 5, further comprising a step of outputting a matching result between the first image and the second image.