JP2898970B1 - 3D seal registration / collation device - Google Patents

Abstract

[Summary] [Object] To provide an automatic seal registration / collation device capable of accurately collating a registered seal with a seal to be inspected. [Structure] A seal is registered in three dimensions based on a distance from a certain surface, and a seal to be inspected is a two-dimensional binarized image, and the center position and the direction of rotation are superimposed to correspond to each red pixel of the seal to be inspected. Calculate the sum of the distances to the points on the registered seal, change the center position in the vicinity, adjust the direction of rotation, repeat the distance calculation, and find the minimum distance. Judge as a false seal.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic seal verification method for storing data of a registered seal in a three-dimensional manner and determining whether the seal imprinted on paper is the same as the registered seal. In countries such as Japan and China, mainly in the kanji area, seals are widely used as a means of personal identification. In government offices, etc., in the handling of important documents, visual seal verification is performed. At a bank post office or the like, the seal is visually checked to confirm the identity of the depositor. With the increase in the number of registered seals and the number of office work,
Visual collation is becoming a burden. In order to reduce the burden, there is a strong demand for realizing automatic seal verification by a machine.

What is desired for automatic seal verification by a machine is that the verification time is not different from the visual verification time. ● Never judge a false seal as a correct seal. And so on.

[0003] Furthermore, the registration of imprints is performed only once, and the quality does not deteriorate over time. ● Imprints to be inspected are stamped on paper under various imprinting conditions. ● Imprints of poor quality that cannot be verified visually are required to be imprinted again.

[0004] In order to satisfy these conditions, fluctuations of various imprint patterns that change according to the imprinting conditions can be absorbed, and simple matching using a small number of features can be performed to shorten the collation time, and false stamps can be reliably performed. It is desired that a robust collation that can be determined as a false mark be performed.

According to the present invention, a registered seal is obtained as distance data by a seal measuring device, and a seal image to be inspected is obtained as binary data by an image reading device, and a difference between the binary data and the distance data is obtained. The collation is performed.

[0006] In order to avoid confusion of terminology, a definition is given. A seal stamped on a seal registration application office or a seal stamped on a bankbook is called a registered seal stamp. At the time of registration, a reference stamped on paper is referred to as a registered seal imprint. A seal to be collated after registration is called an inspected seal. The seal imprinted on the paper at the time of collation is referred to as a seal to be inspected. Verification refers to checking whether the inspected seal is the same as the registered seal.

There are four concepts. There are a total of four concepts: two seals at the time of registration and collation, and two seals at the time of registration and collation. At the time of registration, only a registered seal and a registered seal are present. At the time of verification, there is no registered seal, and there is only the seal to be inspected, the seal to be inspected, and the registered seal. In the visual judgment, the seal to be inspected and the seal to be registered are visually compared to thereby determine whether or not the seal to be inspected and the seal to be registered are the same.

[0008]

2. Description of the Related Art The methods proposed so far for automatically verifying a seal are roughly classified into the following two methods. (A) A method using a registered seal as plane data. (B) A method using a registered seal as distance data. The current state of these two types of prior art will be described.

[(A) Method of Using Registered Seal as Plane Data] The method of using the registered seal as plane data is similar to the currently performed visual seal verification.
The seal imprinted on the paper as the registered seal is carefully compared with the seal to be inspected two-dimensionally. The registered seal impressed on the registered seal is used as reference data. Use the seal of the seal stamped by the seal. In many cases, binarization is used for comparison. The registered imprint is also binarized, and the inspected imprint is also binarized. The binarized registered imprint and the binarized seal imprint are collated based on the degree of coincidence. In this case, both the registered seal and the seal to be inspected are merely two-dimensional binary information pressed on a plane. It may not be binarized (ternary). Even in that case, the same two-dimensional thing is compared and contrasted. The visual judgment is simply applied to the automatic judgment by a machine. It is a very natural approach that traces human behavior. It can be called the imprint plane comparison method.

For example, Hiroshi Mieno, "Experiment and Consideration of Seal Imprint by Overlay", Information Processing Vol. 16, NO.
3, this method is proposed in PP205-211 (1975). The seal impressed on the seal is divided into meshes.
For example, it is divided into 50 × 50 meshes. The amount of light in each mesh is ternarized. D ₀ = 0 to 30%, D ₁ = 3
_0% to 70%, the amount of light as that D 2 = 70 to 100% is divided into three stages. The light quantity in each mesh is measured, and D _{0 to} D ₂ are stored in association with the mesh. This is registration. And the mesh division also subject _imprint, to correspond to D 0 to D _2. Same D on same mesh
j, weight 1 if different, such as D ₀ and D ₂ , weight p if different between D ₂ and D ₁ , weight p (0 <p <
The evaluation function E is calculated by adding 1).

The evaluation function is calculated many times by changing the correspondence, and the one that minimizes the evaluation function is obtained. Ingenuity can be seen in three stages instead of two. However, since the number of meshes is small, it is difficult to see the subtle differences in imprints. In addition, since the collation is performed by the evaluation function, the amount of calculation becomes enormous. The relative position must be derived in a direction that minimizes the evaluation function, but this is not easy. Increasing the number of meshes increases the amount of calculation again. The general method of the evaluation function has the advantage that alignment and rotation angle adjustment are not required, but it is not an advantage because the amount of calculation increases accordingly.

Various superposition methods of two-dimensional imprints have been proposed. In addition to the time-consuming method like the evaluation function, after adjusting the center position and the rotation direction,
There is also a method of checking the degree of coincidence between two imprints. Rather, it seems to be more mainstream. However, this method has significant disadvantages.

Various quality patterns are produced in the seal to be inspected depending on the conditions of the seal. Various seal imprints are generated from one seal. Because of such various patterns, it may not be possible to judge whether they are the same as the registered imprint. Or make a wrong decision. Various collation methods based on the superimposition of binary data have been devised, but the collation accuracy is extremely poor for imprints that are blurred or imprinted depending on the stamping conditions. It is not always the case that the seal is pressed carefully, and it is impossible to ask the customer to press the seal carefully.

For this reason, it is not possible to perform highly accurate seal verification by a simple method of performing verification based on two-dimensional matching between the seal imprint and the registered seal. A method that goes beyond this in view of the fact that static two-dimensional comparison cannot be used has been proposed. Unfortunately, none of them have been successful. A method has been proposed in which a seal imprint to be inspected is not compared as it is, but a pseudo imprint is estimated from an imprinted seal to be inspected, restored and collated with a registered seal imprint.

For example, IEICE vol. J77-DI
I, No. 10, p2027-2035 (1994). The registered imprint is modified variously, a number of modified imprints are generated, and the modified imprint and the inspected imprint are two-dimensionally compared.
However, even if it is corrected in a pseudo manner, the direction of the correction is not necessarily correct. Even if the direction of deterioration is not clear, it cannot be corrected accurately. Sometimes they are corrected by mistake. If incorrectly corrected, the false seal may be mistaken for the correct seal. This method is effective only for a limited number of imprints satisfying specific conditions. For many other imprints there is nothing less than helplessness.

All of these methods attempt to match the seal as the seal itself, that is, two-dimensionally. These two-dimensional methods cannot cope with an infinite pattern of the seal to be inspected, and cannot obtain sufficient matching accuracy. There is no help for the problem that the present invention is trying to solve.

(B) Method of Using Registered Seal as Distance Data This is the first inventor of the present invention. The three-dimensional shape of the registered seal is registered as it is. This is different from the conventional one because of the registration method. We do not stamp the registered seal on paper. Instead, the counter measures and registers the three-dimensional shape of the seal itself. The seal registration itself is completely different from the conventional method. The three-dimensional shape itself of the seal without the stamp is registered, not the stamp with the stamp. The right and left sides are reversed and the stamp itself with irregularities is measured and registered. This is a very novel method.

For example, IEICE, Vol. J80-DI
I, No. 1, p. 148-155 (1997). According to this method, an imprint condition is extracted based on a pattern of an imprint to be inspected, a registered imprint is virtually imprinted on a computer according to the condition, and a binary pattern of the virtually imprinted registered imprint is compared with the test pattern. Is what you do. Since the three-dimensional seal is registered, the inclination of the seal can be freely given, and the direction of the inclination can be expanded and the opposite direction can be narrowed. The two-dimensional imprints are automatically generated by giving the direction of the inclination and the force of the imprint as parameters. If the seal to be inspected is tilted and pushed to one side, the direction of the tilt can be known from the expansion, and a two-dimensional seal similar to that is generated from the three-dimensional seal. This is two-dimensionally compared with the seal image to be inspected. Since this method generates a virtual imprint corresponding to the seal imprint to be inspected, the matching accuracy is higher than the method using binary data as the registered seal imprint.

However, this method is not yet versatile. Even if you find the inflated side and make a seal imprinted in that direction from the three-dimensional seal, it may be that the crest on the inflated side originally expanded. The method of virtually generating a large number of two-dimensional seal imprints from one registered seal is a convenient way to make the seal closer to the seal to be inspected, but this blurs the range of the shape of the first registered seal. When a large number of virtual imprints are generated in this way from two different registered seals, there is a possibility that a common virtual imprint is generated. Then, there is a possibility that a different seal is mistaken for a correct seal. For this reason, the error matching rate is 6 to 7% even in this method. To make matters worse, a false seal may be determined to be a correct seal. Conversely, a positive seal may be a false seal. Furthermore, since various virtual imprints are generated and the virtual imprints continue to be produced unless collation is possible, there is no end. This is a very time-consuming method.

Even if a high-speed computer is used, it takes about one hour to extract the sealing conditions. It's not very practical because you can't make a customer wait one hour. It is meaningless that what can be done in a few seconds with the naked eye takes one hour using an oversized device. That it can be verified in a short time,
Therefore, the object of the present invention cannot be satisfied. It will be a dissertation but not worth it.

[0021]

There has been no seal stamp collation technology which has a high collation accuracy enough for practical use and can be processed in a short time. The binary registration imprint method is an application of visual seal stamp collation, however, it is not possible to accurately associate one registered seal imprint with an infinitely-examined seal imprint, and the matching accuracy is low. Although the registered seal distance data method is a subtle method, it requires a great deal of calculation time to extract the stamping conditions for virtual stamping on a computer, and is impractical. Further, since the matching itself is performed between the binary patterns, matching of a blurred portion or the like is insufficient as in the above-described method. Matching accuracy is not high.

Two conventional methods have been described for automatic verification of seals. In any case, accuracy, time, cost, etc. are insufficient. In each case, many trial calculations are repeated by trial and error.
Must compute the cost function over and over again.
Means that a large number of virtual candidates are generated from a registered seal, and the number of candidates and the seal image to be inspected are compared and evaluated. Even a high-speed computer takes too much time because it makes unreliable calculations according to unreliable policies.

Therefore, in the conventional technique, the seal verification cannot be automated yet. At present, automatic seal verification is not realized at all. As before, we have to rely on visual collation. An object of the present invention is to solve such a problem at once, and to provide an apparatus and a method capable of automatically collating any seal. The objects of the present invention are as follows.

(1) The time required for automatic collation is not different from the visual collation time. (2) Do not mistakenly determine a false seal as a correct seal.

According to a skilled person, a visual judgment is made in about several seconds. It doesn't matter if it takes an hour or even tens of minutes using a machine. Who pays a high price and installs the equipment if it reduces efficiency while installing equipment and taking up space? Automatic judgment must be completed in a few seconds. Since all of the conventional techniques of stamp verification required several tens to several hours, it cannot be taken to the order of a few seconds unless it is very innovative.

[0026] False and correct marks are not symmetric. Since a correct seal is a registered seal, there is no concept of a false seal for a registered seal. The seal to be inspected may be a correct seal or a false seal. The correct seal is the seal imprint of the same thing as the registered seal, and the false seal is the seal imprint of the different one from the registered seal. If you make a mistake, you can mistake the correct seal for a false seal,
There is a case where a false seal is mistaken for a correct seal. Both are incorrect, but mistakes for false marks for correct marks cause more serious problems. Unauthorized persons may acquire personal property or real estate. There must be no mistake in misrecognizing a false seal as a correct seal.

[0027]

SUMMARY OF THE INVENTION The present invention is based on the viewpoint that a seal surface (a convex portion of a seal engraving) corresponding to the vermilion portion of a seal to be inspected must exist. Therefore, the registered seal and the seal to be inspected are aligned, and the evaluation is made based on the sum of the distance between the vermilion portion of the seal to be inspected and the registered seal. If the seal is a positive seal, the instep that exists in the seal to be inspected must have its corresponding point on the mountain face of the registered seal. If all the intaglios of the seal to be inspected correspond to the points on the mountain surface of the registered seal, it means that the seal to be inspected is equal to the registered seal.

For this reason, a state where the registered seal is brought into contact with the seal to be inspected is imagined on a computer, and the red seal portion (set of insteps) of the seal to be inspected is compared between the seal to be inspected and the surface of the seal to be inspected. The distance is obtained and the sum is calculated. If the sum is smaller than a certain value, this is regarded as a positive mark. If the seal is correct and correctly aligned, the peak of the registered seal should completely overlap the vermilion attachment of the seal to be inspected.
Then, the sum of the distance between the two points at all points in the mountain portion should be zero. Actually, the sum of the distances is not 0 due to an alignment error or bleeding, but it should be a small value as long as it is a positive mark.

If it is a false seal, the valley surface and the cliff surface of the registered seal correspond to a part of the vermilion part of the seal to be inspected. Then the distance Kohei becomes longer. If there are several such points, the sum of the distances becomes long. A large sum of distances means that there is no seal surface corresponding to the vermilion portion of the seal to be inspected. If so, the seal cannot be a positive seal. If the sum of the distances is small, the mark is positive, and if the sum is large, the mark is false. Therefore, it can be said that the sum of the distances between the inspected seal and the registered seal surface in the vermilion portion of the inspected seal indicates the difference between the registered seal and the inspected seal.

There is only one seal to be inspected. Therefore, there is no room for generating a large number of virtual imprints as in the prior art. The prior art wasted time because it repeated vain trials of infinitely generating and comparing virtual candidates from registered seals. All of them are stupid to start from a registered seal. The present invention instead starts with a single seal imprint. Do not create virtual candidates. Comparison 1
The number is one to one. Therefore, the calculation time can be significantly reduced. This is one reversal idea.

A further feature of the present invention resides in that the distance between the corresponding points of the seal to be inspected and the registered seal is measured and collation is performed based on the sum of the distances, instead of measuring the overlay error.

On the contrary, the inspected seal is not viewed from the registered seal. This still has one problem. The distance from the registered seal is measured only for the point where the seal to be inspected exists (point A), and is added to the distance sum. The point S that exists in the registered seal but does not exist in the seal to be inspected is not calculated. Therefore, the presence of the point S in the registered seal that is not in the seal to be inspected does not increase the distance sum. Its presence does not increase the error and cannot be detected by this method. However, the presence of a point S in the registered seal that is not in the seal to be inspected should increase the possibility that it is a false seal. Because it is not counted, the method of the present invention must be described as one-sided in some respects. If the mountain face of the registered seal completely includes the seal to be inspected, the sum of the distances becomes zero. However, when the mountain face of the registered seal includes all the seals to be inspected, it is not the same as the seal to be inspected simply because the registered seal is thicker.

There may be such a case. Therefore, the area C _i of the vermilion part of the seal to be inspected and the area C of the mountain face of the registered seal
_r is compared, and if C _i / C _r is equal to or less than a certain value, the seal is rejected. Simply _hoarse, would _C i / C _r is also less than one thing. Otherwise, C _i / C _r may be smaller than 1 because the seal to be inspected is a fine character and is included from the registered seal. Rejecting the determination that this is less than or equal to a certain value smaller than 1 avoids erroneously recognizing a false mark as a correct mark.

Here, each part of the seal is defined. The convex part of the sculpture is called the mountain surface. The bottom surface (the concave portion of the sculpture) is called a valley surface. The slope between the mountain and the valley is called the cliff. The outline of the collation method of the present invention will be described. The registered imprint is a distance image (three-dimensional image) by the distance image measuring device.
Register as The xy coordinates are taken on the mountain surface, and the surface height z at the (x, y) point is obtained. An expression for the seal surface such as z (x, y) is obtained. If the measured seal surface has an inclination, the inclination is corrected. Set the peak to 1 and the valley to 0
Normalize as follows. In addition, the engraved character portion (mountain side) of the seal is thickened in consideration of the bleeding of vermilion due to the seal.
An average imprint pattern is extracted by binarization, and the area of the imprint portion is determined.

The seal to be inspected is optically read as binary data. From this, the area of the mountain surface is determined. The area of the mountain surface is
In the case where the seal is a certain percentage compared to the area of the mountain surface of the registered seal, a rejection process is performed as a seal pattern that is not sufficient for determination. Verification processing is performed on the seal to be inspected whose vermilion has a certain area or more.

The collation is performed while changing the position and the rotation angle.
This is performed by obtaining the degree of difference between the registration mark and the inspection mark, and obtaining the lowest degree of difference. The difference is a value obtained by normalizing the distance value from the vermilion image of the inspected seal to the uneven surface of the registered seal by the total number of vermilion pixels, and generally takes a value of 0 to 1.
Ideally, the degree of difference is 0 if the vermilion portion and the seal surface (mountain surface) all match. If they are all different, it becomes 1. If the degree of difference is lower than a certain determination value, it is determined to be a positive mark. If it is higher than a certain judgment value, it is judged as a false mark.

[0037]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention enables automatic verification of a seal. FIG. 1 is a schematic configuration diagram of a seal registration device and a seal verification device according to the present invention. By using the present invention, once the shape of the seal is registered, once the seal imprinted on the paper or the like is input by an image reading device such as an image scanner, it is determined whether the seal is correct or false. Automatic determination is possible. Further, since both the registration mark and the test mark are converted into digital data, it is possible to perform verification at a remote place by performing data communication.

The gist of the present invention lies in searching for a seal surface corresponding to the inlaid portion of the seal to be inspected.
Therefore, in the present invention, the collation result does not affect the quality of the seal imprint of the seal to be inspected. Even if the search fails, the correct seal is erroneously determined to be a false seal, and the false seal is not erroneously determined to be a correct seal. This is because it is impossible that a portion corresponding to the imitation of the false seal exists only on the surface of the seal.
This also means that the present invention is secure in terms of security.

The novel aspect of the present invention has been described above. Although it is desirable to use a distance image as the registered seal, a certain degree of collation can be performed by a similar collation method with a conventional two-dimensional image. If the seal surface has an inclination at the measurement stage, this is corrected. If it is necessary to consider the bleeding of the vermilion, a process of increasing the stroke of the registration mark is performed. If the inspected seal image is too faint or too thin to be judged, a rejection process is performed. The seal verification device of the present invention has the following configuration. First, list the items.

A. Registered image measurement device B. Registered image storage device C. Inclination correction mechanism D. Corrected image storage device E. F. Distance normalization mechanism G. Normalized image storage device Shape expansion mechanism Dilated image storage device I. Standard imprint feature extraction mechanism Standard imprint feature storage device Inspection image measurement device L. Test image storage device Rejection determination mechanism Initial value setting mechanism Optimal position / rotation angle determination mechanism Judgment value storage device Q. Judgment value correction mechanism Matching result judgment mechanism

Further, the operation of each mechanism will be briefly described.
You. A. Registered image measurement device
Coordinates x and y in the plane direction in three-dimensional coordinates defined in the device
Distance z_mFind (x, y) and measure as distance image
Yes B. Registered image storage device Distance data of each coordinate point (x, y)
Ta_m(X, y) is stored. C. Tilt correction mechanism Distance data in xyz space
Vote in the polar coordinate space, and from the peak value the plane of the mountain of the registered seal
Equation z_c(X, y) is obtained and the plane equation z_c(X,
Distance data z including inclination from y)_mZ minus
_rD. Obtain (x, y) and set the inclination of the registered seal to 0 D. Corrected image storage device Distance data z after tilt correction
_r(X, y) is stored. E. FIG. Distance normalization mechanism The top of the valley is set to 0 on the top of the registered seal.
Distance z with the deep point as 1 _rIs normalized. F. Normalized image storage device Distance data z after normalization
_RG. Store (x, y) Shape expansion mechanism One-pixel or two-line contour of mountain surface
The distance data between the fattened A and the fattened A is z
_R ^*(X, y). H. Dilation image storage device Distance data z after shape expansion_R
^*(X, y) is stored. I. Standard imprint feature extraction mechanism
Pixel value C_rIs counted. J. Standard imprint feature storage device Number of pixels used as standard imprint features
C_rIs stored. K. Inspection image measurement device
Measure as a value image. L. Test image storage device Binary data of each coordinate point of the test seal
Data w (x, y) is stored. M. Rejection judgment mechanism Number of pixels C_iTo
Count and use this as the number of pixels C on the mountain surface of the registered seal_rDivided by C
_i/ C_rIs calculated and this value is smaller than a certain constant from 0 to 1.
If not, reject the seal. N. Initial value setting mechanism
Match the rotation angles of the seal to be inspected and the seal to be registered
Let me know. O. Optimal position / rotation angle determination mechanism
Determine the appropriate rotation angle, and register with the seal
Obtain the distance of the registered seal and sum the sum S to the total number of pixels Q
Is calculated. P. Determination value storage device Stores the determination value D. Q. Judgment value correction mechanism Average value M of the distance of the mountain surface of the registered seal
, And M is subtracted from the determination value D to obtain a correction determination value.
You. R. Matching result judgment mechanism Correction judgment value is smaller than a certain threshold
If it is not, it is judged to be correct, and if it is larger than the threshold, it is judged to be false.
Set.

Here, the measurement of the registered image is stable,
If there is no need for tilt correction, tilt correction processes C to D and Q
Can be omitted. Steps G to H can be omitted if it is not necessary to consider bleeding or protrusion of vermilion. If rejection processing is not required, processing steps I to J related to rejection
And M can be omitted.

Each of the devices and mechanisms A to R shown in FIG.
I will explain each one one by one. A. Registered image measurement device This is a 3D measurement device (for example,
This is a device for measuring as a distance image according to (Inder).
FIG. 2 shows an example of the measured registered seal. Registered seal
With the character face facing the receiver of the rangefinder.
From the receiver (x, y, 0) to the seal surface
Distance is data z_mIt is measured as (x, y). Coordinate system
Is fixed to the range finder. Range finder
Is taken as xy coordinates, so z is the distance. Rangef
The vertical distance at each point (x, y) of the finder is z
_m(X, y). The mountain face of the seal is parallel to the xy plane
The distance between the rangefinder and the surface of the seal.
The sum of the separation h and the depth of the seal is the distance z _m(X, y). Is
To z_m(X, y) -h is the depth at (x, y)
It can be said that it manifests itself.

B. Registered image storage device The registered image storage device stores distance data z for each coordinate point.
_m (x, y) is stored.

C. Inclination correction mechanism It is desirable that the surface of the seal is measured without inclination by the registered image measuring device. However, depending on the environment,
It is possible that the surface of the seal is measured at an angle. If there is a tilt, the mechanism corrects the tilt. Therefore, this mechanism is not necessary when measurement is performed without inclination.

For example, the inclination of the three-dimensional data is corrected by applying the Hough transform to the three-dimensional image. Specifically, the measured three-dimensional data in the xyz plane is voted in the θ-φ-ρ space, and its peak value (θ ₀ , φ ₀ ,
ρ ₀ ). ρ is distance, φ is angle from xy plane,
θ is the angle from the x axis around the z axis. That is, θ, φ,
In the three-dimensional space of ρ, each point is distributed, and the point of the highest density of the distribution is set as a peak value. (X, y,
Even if voting is performed with the three-dimensional rectangular coordinates of z), there is no peak value, only dispersion.

Only after voting on the three-dimensional polar coordinates (θ, φ, ρ) does the set of points on the plane converge on the points. Since the target is a seal, there are two points with the highest density according to the mountain surface and the valley surface. The peak value closer to the range finder is adopted. The vector looking at that point is (cos
φ ₀ cos θ ₀ , sin φ ₀ cos θ ₀ , sin φ ₀ )
It is. Since the inner product of this and the three-dimensional vector (x, y, z) gives the distance ρ ₀ ,

[0048] _{ρ 0 = xcosφ 0 cosθ 0 +} ycosφ 0 sinθ 0 + zsinφ 0 (1)

The following equation holds. This is the plane equation of the mountain surface of the registered seal. Using this parameter, a plane equation of the inclination of the seal surface can be obtained.

Z _c (x, y) = ax + by + c (2)

[0051] _{However, a = -cosφ 0 cosθ 0 /} sinφ 0 (3) b = -cosφ 0 sinθ 0 / sinφ 0 (4) c = ρ 0 / sinφ 0 (5)

Is obtained. This plane is the formula for the seal face.
Difference between this plane and the measured 3D data

Z _r (x, y) = z _c (x, y) −z _m (x, y) (6)

By obtaining the above, an image whose inclination has been corrected can be obtained. Since the equation of the mountain surface is z _c (x, y),
Now _z m (x, y) by subtracting the _z r (x, y) is 0 in Yamamen, the valley surface should be equal to the depth of the valley. The inclination is corrected not by rotating but by calculating the height from the inclined plane.

D. Corrected image storage device The corrected image storage device stores distance data _zr (x, y) of each coordinate point after correction.

E. Distance normalization mechanism In this mechanism, the distance data on the seal surface is 0,
The distance increases toward the bottom, and the bottom approaches 1
The distance image is normalized to a small value (FIG. 3). This is by seal
This is to make the depth of the unevenness of the engraved portion unchanged.
The normalization is specifically performed as follows. First z_r(X,
the maximum value maxz of y) _rAnd the minimum value minz_rAsk for
You. Using this value, distance data z_r(X, y) is

[0057]

(Equation 1)

Is normalized to [0, 1]. That is, the value of z is 0 or 1, or a value between 0 and 1. There are no negative values or values exceeding 1.

F. The normalized image storage normalized image storage device, the distance of each coordinate point of the normalized data z _{R (x,} y) is stored. The point on the valley is roughly 1
In addition, the point on the mountain surface should be approximately 0. Cliffs 0 and 1
Between the values.

G. Shape expansion mechanism The distance image measured by the registered image measuring device faithfully represents the shape of the registered seal. The collation used in the present invention searches for the surface of the seal corresponding to the vermilion of the seal imprint to be inspected. There is a possibility that the surface of the seal does not exist, and the seal is determined to be a false seal. In particular, when pressed in a tilted direction, the tilted side tends to be slightly thicker than the mountain surface of the seal.

Therefore, the present mechanism performs an expansion process corresponding to the previously measured convex portion of the seal surface. The expansion process enables it to cope with vermilion bleeding. Therefore, the seal of the registered seal is slightly wider than the seal to be inspected. Therefore, this mechanism is not necessary in an environment in which phenomena such as bleeding and vermilion protrusion do not occur.

Normalized image z_R(X, y) at each point
Attention will be paid to the vicinity of the eight. Add 9 neighborhoods to current coordinates 9
The minimum value of the normalized value of one pixel is z_Rmin(X,
y) to z _R ^*(X, y). Focus on the inside of the mountain surface
If there is a coordinate, the value of all 9 pixels of 3x3 including it is
0. Therefore, the value of the coordinates is
That's weird. If there is a pixel of interest inside the valley surface, 9 images
All prime values are 1. Since the minimum value is 1, this process
The value of that coordinate is unchanged. One away from the mountain surface
If there is a pixel of interest on the cliff surface, the z coordinate of itself
Is between 0 and 1, but there is a peak pixel near 8
The small value is z = 0. Therefore, the z coordinate of the cliff pixel is written to 0.
Be replaced. All pixels on the cliff side at the boundary between the mountain and the cliff are mountains
It can be said that it is raised on the surface. This allows
The convex portion on the surface can be thickened by one pixel. However,
At the boundary, the maximum number of pixels having a value is set. This around
25 pixels (5 × 5: pixel of interest and its surrounding 24 pixels
Focusing on (primary), it is possible to increase the thickness by two pixels. This
It can be adjusted according to the application. With the above processing, the shape
Can be inflated.

H. The dilated image storage expansion image storage device, the distance of each coordinate point of the shape after inflation data z _{R *} ^(x, y) is stored.

I. Standard Imprint Feature Extraction Mechanism If the inspected imprint is blurred or the vermilion is poor and thin, the rejection process may be performed assuming that the judgment cannot be made. As a comparison pattern for this purpose, an ideal standard imprint when vermilion is completely attached is created by binarizing the registered seal, and the number of pixels corresponding to the vermilion portion is calculated as a feature amount. Therefore, the present mechanism is unnecessary in an environment where rejection processing is not required.

The binarized image is obtained by separating the surface of the seal into a surface (mountain surface) and a bottom surface (valley surface), and giving a value 1 to the mountain surface and a value 0 to the valley surface. That is, z _{R *} ^(x, y) the pixel (x, t) values on the basis of the problem of clustering either of two classes. There are various known methods for clustering, which may be used. For example, a clustering method that reduces intra-class variance and increases inter-class variance

In IEICE (D), vol. J63-D, N
o. 4, p349-356, April (1980). The number of pixels corresponding to the vermilion portion (black) of the binarized image is counted and set to _Cr .

J. The standard imprint feature storage device standard imprint feature storage unit, the number of pixels C _r is stored as a standard imprint features. The operation at the time of registration of the registered seal has been completed. Thereafter, the operation for the seal to be inspected is performed as needed.

K. Inspection image measuring device The owner of the seal brings the seal and stamps it on paper. This is called a test seal. This is a two-dimensional thing because it was stamped on paper. The inspection image measuring device K reads the inspection seal image by an optical reading device (for example, a scanner).
This is a device for measuring as a binary image w (x, y). The measurement is performed at the same resolution as the registered image measuring device.

L. Test image storage device The test image storage device stores binary data w (x,
y) is stored.

M. Rejection determination mechanism Here, rejection processing is performed on data that is difficult to determine because the inspected seal image is blurred or faint. The rejected seal can be subjected to processing such as having the seal imprinted again.
It plays a role in preventing erroneous determination due to a small amount of information. In an environment where rejection processing is not performed, this mechanism is not provided.

First, the number C _r of red pixels (black) is obtained from the registered seal.
Ask for. This is the standard vermilion pixel count. From a test image, and counts the number C _i of ink pad pixels (black). Then by standard ink pad number of pixels C _r obtained from the registered seal, obtaining the value C _{i /} C _r obtained by dividing the ink pad pixel number C _i of the test image. If this ratio value is smaller than a certain value, it is determined to be rejected. For example, if you want to reject a seal that is less than 60% of the standard,

C _i / C _r <0.6 (8)

In this case, rejection may be determined. The above procedure is performed before the registration of the seal to be inspected and the registered seal.
This is an operation that can be performed without alignment because the number of all black pixels is compared.

N. Initial value setting mechanism It will be the alignment from now on. The registered seal stamp registered in three dimensions is aligned with the seal image to be inspected pressed on the paper.
In principle, the alignment can be performed by matching two points and matching the azimuth with one point. Even if two points match, two points must be determined in advance. In the case of seals, many are circular and it is difficult to determine two points. Because it is a seal, the only obvious feature point is the central point. Therefore, the center point is adjusted, and the azimuth around the center point is adjusted. The parameters are therefore center point and azimuth. It is good if these values are decided at once, but it does not go so. I will decide asymptotically. First, approximate values of the center point and the direction are given. This is called an initial value.

The initial value setting mechanism sets an initial value for searching for parameters (position and rotation angle) necessary for collation according to the present invention. By this initial value setting, parameter values appropriate for collation are narrowed down globally.

[Position Initial Value Setting]
This is considered to be equal to the difference between the center of the registered seal and the center of the seal to be inspected. Therefore, the center coordinates w (x ₀ , y ₀ ) of the seal image to be inspected
And this is used as the position parameter. Therefore, "position" means a center point position, and is not a general relative position. The center of the seal is different from the center of gravity of the vermilion. It is the center of a circle, ellipse, or rectangle determined by the frame of the seal. The seal frame has many circles but also a rectangle. There are also regular polygons.
In any case, there are twofold symmetry and mirror symmetry.

The initial value of the position parameter is determined by circumscribing a rectangle, such as a rectangle and a parallelogram, to the seal to be inspected as a diagonal intersection of the circumscribed rectangle. The diagonal intersection is set as the center of the imprint frame, that is, as an initial value of the position parameter.
This is shown in FIG. A stamp with Yamamoto is shown vertically inside the circular frame. The square circumscribes the seal. The intersection of the square diagonals is centered.

The circumscribed rectangle does not mean that any shape may be used. Must be rectangular or parallelogram. Since the direction is unknown, trapezoid-like asymmetric ones cannot be used. In the case of a circular frame, it can be circumscribed by a square. In the case of an elliptical frame, a square makes contact only at two points. The ellipse can be bounded by a rectangle. Even in the case of a rectangular frame (rectangle), the center point can be determined by circumscribing the rectangle. If the frame is more complex, for example a regular polygon, the center can be determined by a rectangle.

However, when a part of the seal is missing or noise is generated, the square may not circumscribe the regular outer frame of the seal. This is shown in FIG. The left frame is missing. The vertical rectangle is circumscribed, but the diagonal intersection is shifted to the right from the center point. In consideration of such a case, the center coordinates are also obtained as diagonal intersections for circumscribed rectangles having angles different by 45 degrees. As shown in FIG. If the two calculated center coordinates are separated by ε or more, two initial values are set, and the subsequent parameter search is performed in parallel. ε is an appropriate constant. FIG. 4 (b), FIG.
If the diagonal intersection of (c) has only a difference equal to or smaller than ε, one of them is set as the initial value of the position. Since the optimum value is determined later, the initial value may be arbitrary.

The above is the initial value setting of the center position of the seal to be inspected. The registration seal only needs to determine the center coordinate once. The center coordinates are known.

[Setting of Initial Value of Rotation Angle] Next, an initial value of a rotation angle parameter is set. This is for adjusting the angle (aligning the azimuth) between the registered seal and the seal to be inspected. Since the registered seal and the seal to be inspected are very similar and are compared, it is extremely important to match the orientation. Research on the rotational alignment of the test seal and the registered seal has been energetically performed, and effective results have been obtained. The result can be used. For example, IEICE (D), vol. J67-
D, No. 1, pp 133-140, Jan. 1984
The case where is used will be described.

First, the surface portion of z ^* _R (x, y) is binarized. The binarization method may be performed as a clustering problem as in (I. Standard imprint feature extraction mechanism). Next, the peripheral density of the binarized standard imprint is calculated. The peripheral density is calculated by dividing the imprint into N equal parts around the origin and counting the number of pixels of the vermilion portion existing in each fan-shaped area. This is shown in FIG. N = from X axis
.. Are divided into N fan-shaped areas such as 0, 1,. Only the pixels in the vermilion portion are counted and defined as the peripheral density. n = 0
In the figure, the portion of the upper right corner of the elliptical frame of Yamamoto and the bent portion of the lower right corner of the mountain shape are red pixels. The number of dots is the number of pixels. The white background (background: valley surface) is not counted. The calculated peripheral density of the standard registration imprint is {fr (n)} _{n = 0} ^N−1 . r means a registered seal. n is a number assigned to a fan shape divided into N equal parts (n = 0,..., N−1). fr
(N) is the number of pixels on the mountain surface present in the n-th sector of the registered seal. It is simply the number of pixels, and does not consider the distance from the center. There is one point near or far from the center.
The center coordinates of the registered seal are already determined, and if N is a constant value, fr (n) is determined. N is a number that engraves the entire circumference. The following is circular and has high symmetry, so that the correlation function does not show a difference unless it is finely cut. Therefore, N = 180 or 360, or a very large number. This increases the amount of computation but is unavoidable.

Similarly, the imprint to be inspected is divided into N sectors around a certain center, and the number of vermilion pixels fb (n) included in the sector is counted. b is a suffix representing the seal image to be inspected. The sector of n = 0 may be anywhere. However, since the seal to be inspected and the registered seal are mirror-symmetrical, n is taken in the opposite direction. If the number of fan-shaped pixels fr (n) of the registered seal is clockwise, the number of the number of fan-shaped pixels fb (n) of the inspected seal is counterclockwise. Number of fan-shaped pixels fb of the seal to be inspected
(N) can be counted only when the center coordinates are determined. That is, it is a function of the center position. Since the center of the seal to be inspected (position parameter) changes variously, fb (n) also changes by the same number of times. That is, even if N is determined, fb
(N) is not determined. This is also different from the case of a registered seal. Let the peripheral density of the seal to be inspected be {fb (n)} _{n = 0} ^N−1
And The sum total of the product of the peripheral density fr (n) of the registered seal and the peripheral density fb (n + k) of the test seal imprinted by the number k is referred to as a cross-correlation function R (k). The cross-correlation function is now calculated.

[0084]

(Equation 2)

The sum of the products of the peripheral densities shifted by k-th is calculated. When n + k exceeds N−1, n + of fb (n + k) is calculated.
k can be read as n + k−N. That is the meaning of modN. A cross-correlation function R (k) is calculated for all k (= 0,..., N−1). K that maximizes R (k)
= K ₀ should equal the deviation of the rotational orientation of the registration seal and the test print image. So determining the initial value of the rotation angle k = k _0. Although the may turn either to match, registered seal is centered and fixed one from the beginning, because they determined the orientation both fixed with respect to the registration seal,
Turn the seal to be inspected. The rotation angle is 2πk ₀ / N. When the seal to be inspected is rotated this much, the orientations of the registered seal and the seal to be inspected match. Since the centers match and the orientations match, a one-to-one comparison can now be made.

R (k) Why does the maximum k give a deviation between the direction of the seal to be inspected and the direction of the registered seal? It states that. If the (n + k) th inspection seal imprint sector area is the same as the nth registered seal sector area, the number of vermilion pixels (the number of mountain face pixels) should be the same. That is, fr (n)
= Fb (n + k). This is because n is 0 to (N−
This holds true for all of 1). When there are several numbers and the sum of their products is calculated, the maximum is obtained when the same numbers are multiplied. Explained simply with two variables, an inequality such as a ² + b ² ≧ 2ab holds.
This is the same even if there are N variables. Σa _k ² ≧
Inequality, such as the _{Σa k a j (k ≠ j} ) is always true.
An equal sign occurs when all values are equal. Therefore, k, which maximizes the cross-correlation function R (k), represents a shift in the rotational direction between the registered seal and the seal to be inspected, and if the seal to be inspected is rotated by k fan-shaped regions, it should have the same orientation as the registered seal. .

Although it is correct, a seal has a high central symmetry in the first place. The number of vermilion pixels in the fan-shaped area (k = 0 to N-1) is almost similar and does not change much.
Then, there is a possibility that the azimuth is determined erroneously because there is not much difference between the cross-correlation functions. If N is small, no difference appears. This means that the number of steps must be large, such as N = 360. This increases the calculation time. The definition of the cross-correlation function in the direction of rotation is simple, and is a natural idea for determining the rotation angle.

A seal image such as "Yamashita" can be calculated with a rotation angle different by 180 degrees since the peripheral density is almost point symmetric. Therefore, two parameters are taken in order from the maximum value of R (k) in descending order, and the subsequent parameter search is performed in parallel with the rotation angles k ₀ and k ₁ as initial values.

O. Optimal Position / Rotation Angle Determining Mechanism This mechanism calculates a determination value used for collation while determining parameters of an optimal position and an optimal rotation angle. Here is the heart of the invention. It is hard to understand. Center position (position parameter) and rotation angle (k ₀ ) of the seal to be inspected
Was determined by the processing up to the previous stage. However, this is a tentative one, and repeated calculations are repeated to asymptotically approach the optimum position and rotation angle.

Since the center and the direction are temporarily determined, the seal to be inspected can be superimposed on the registered seal. In other words, the stamped paper is brought close to the seal and brought into contact at one point. This is shown in FIG. Even though the contact is made, both the registered seal and the seal to be inspected are data, so the contact is not actually made. Contact by calculation. And the red pixel P of the seal image to be inspected
Then, the distance from the registered seal surface having the same value (x, y) is obtained. Ideally, the vertex α of the registered seal is located where the vermilion pixel P of the seal to be inspected exists. Ideally, if contact is made without inclination, the distance is 0 because Pα overlaps. The distance between points of the corresponding registered seal is obtained for the vermilion pixels P of all the seals to be inspected, and the sum of the distances is calculated. Let S be the sum of the distances. In the case of a positive seal, if there is no inclination and no bleeding, the vermilion pixel P of the seal to be inspected always touches the mountain face of the registered seal stamp, so the distance is 0 everywhere and the sum is also 0.

If the valley surface of the registered seal corresponds to the vermilion pixel P of the seal image to be inspected, the sum S increases by 1 at this point. When the pixel of the seal image to be inspected corresponds to the valley or cliff surface of the registered seal, S increases accordingly. Further, if the registered seal has a slope, S increases by the gap due to the slope even if the inspected seal imprint pixel P corresponds to the mountain surface α of the registered seal. If the white background of the seal to be inspected corresponds to the mountain surface of the registered seal, S does not increase. This is because the distance is not measured and added to the distance in the white background portion of the seal image to be inspected. This is also because the registered seal is viewed from the seal to be inspected. The evaluation scale D is a value obtained by dividing the distance S between all vermilion pixels of the inspected seal and the registered seal by the number Q of vermilion pixels.

Center pixel (position parameter) of the seal to be inspected
When the rotation angle is determined, the sum S of distances and the evaluation scale D can be calculated immediately. However, the center and the rotation angle of the seal to be inspected are not always optimal. Since the initial value has been determined earlier, the center (position parameter) is shifted to the vicinity of 8 of the initial value of the position parameter (center), and N points of the cross-correlation function are calculated with that point as the center and k ₀ giving the maximum value Then, the rotation angle is determined, and the distance S between each pixel of the seal image to be inspected and the registered seal is obtained. Then, an evaluation scale D is obtained. In this way, the center is shifted from the initial value to the vicinity of 8, and the same is performed. The rotation angle is determined at each point, and S and D are obtained.

Here, the evaluation scale D is obtained as follows. First, all the pixels of the seal to be inspected are evaluated one by one. The sum S is initially 0. If the pixel with the seal to be inspected is a vermilion portion (vermilion pixel), the coordinates w of that pixel
Distance data z between the seal image to be inspected and the registered seal at (x, y)
_R (x, y) is added to the sum S of distances. This evaluation is performed for all the pixels of the seal imprint to be inspected, and a total sum S of the distance between the vermilion portion and the registered seal imprint is obtained. This is shown in FIG. The seal to be inspected contacts the registered seal somewhere, where the distance is zero.
If there is no inclination, the inspection seal imprint plane and the registered seal imprint plane are the same plane. However, it is not necessarily because of the inclination. The sum is obtained by adding the distance to S for all vermilion pixels. In order to obtain the average distance of each pixel, the total number of distances S is divided by Q, where

D = S / Q (10)

Is obtained. D is used as an evaluation scale. The value of D is lower for a seal imprinted with a correct seal, and the value of D is larger for a seal imprinted with a similar seal. Therefore, D can be said to be the degree of difference.

Further, it may be desired to perform matching by focusing on a pattern at a specific place such as similar characters such as "tree" and "book". In such a case, the difference S is obtained by appropriately weighting the distance S according to the coordinates.

[Procedure 1] Eight neighborhoods of the position parameter value (nine places including the position parameter) are evaluated. Nine rating scales are obtained. Let the minimum value be D _min . From the smallest evaluation value D _{m in} the range of up to D _min + T, truncate all positional parameters without the value of D. That is, D _min
Discard those that give D greater than + T. These are unsuitable as centers. Move to the next procedure for the remaining position parameters. Here, T is a threshold value.

[Procedure 2] Before and after the rotation parameter (3 in total)
Angle) is evaluated for all the position parameters detected in step 1, and from the minimum evaluation value _Dmin , _Dmin + T
The set of the rotation angle parameter not including the value of D and the position parameter at that time is discarded in the range up to. In step 1 center, D Decide rotation angle, because choosing what is in the range of D _min of T, is obvious from D _min in the range of T relative rotation angle. However, D is also obtained for the front-rear angle k ₀ ± 1 of the optimum rotation angle k ₀ , and the number of D is tripled. The rotation angle k ₀ ± 1 does not maximize the cross-correlation function, but may be more suitable as the alignment angle. Therefore, the range of the rotation angle is expanded by ± 1.

[0099] [Step 3] [Step 2] in the remaining of the parameters, compares the minimum evaluation value D _newMin, the minimum evaluation value D _Oldmin prior iteration time. If the minimum rating is improved,
Reduce T to 1/10 and return to [Procedure 1]. If it has not been improved, the parameter search is terminated assuming that the optimum position and rotation angle have been determined, and D at that time is set as a determination value.

The fact that the minimum judgment value has been improved means that a center point that is more suitable than the previous center point exists near eight. Therefore, there may be points near 8 that further reduce D. Therefore, it is necessary to repeat steps 1 and 2. In this way, the user follows the vicinity of 8 in the direction in which D decreases. Even if it traces the neighborhood of eight, every time the center point changes, the rotation angle must also be calculated, and the calculation amount is large. Eventually the center (position parameter) becomes 8
There may be a point that D does not decrease even if moved to the vicinity. Here, the center point and the rotation angle are determined. These are the optimal position and rotation parameters. It is evaluated by D at that time.

P. Judgment value storage device The judgment value storage device stores a judgment value D of the collation result.

Q. Judgment value correction mechanism The inclination of the seal surface is C.I. Although the tilt is corrected by the tilt correction mechanism, the correction may not be sufficient depending on the sampling error at that time. In that case, the entire degree of difference fluctuates depending on the type of seal. In addition, since a part of the mountain surface of the seal is brought into contact with the seal image to be inspected, even if the inclination is 0, if the mountain surface has irregularities, an offset is included in the distance value.
For this reason, it is difficult to determine whether the seal is a positive mark or a false mark, even if the comparison is performed using a fixed threshold. Therefore, the present mechanism corrects the determination value using a value depending on the distribution of the distance values of the registered imprint. This mechanism is unnecessary if the seal surface is measured or corrected without tilting accurately.

The distribution of registered seal imprint distance values is generally divided into two categories: the surface (mountain surface) and the bottom surface (valley surface) of the seal. Therefore, after clustering these into two categories, the points (mountains) belonging to the surface category
The average value M of the distances is determined. If this value is not 0, it means that a part of the registered seal is apart from the seal to be inspected. Even when the mountain surface of the registered seal has irregularities, by setting the average height of the mountain surface to 0, the degree of difference D can be increased only when the point corresponding to the seal imprint to be inspected hits the valley cliff. The difference D ^* corrected using this value is

D ^* = S / Q-M

Are represented as follows. The clustering method is
As in the case of binarization, a conventionally proposed method may be used.

R. Collation Result Judgment Mechanism In this mechanism, the collation result is calculated based on the judgment value. The collation result is a positive mark when the judgment value D ^* is smaller than the collation threshold,
If it is larger, it is determined to be a false mark. The threshold value is appropriately set to about 0.046 in experiments performed by the present inventor.

[0107]

An embodiment using the apparatus of the present invention will be described. SUN Spark Station as the CPU of the device
20 was used. Five types of registered seals, "Yamamoto", "Yoshinaga", "Nakai", "Masaki", and "Yoshikawa", were prepared as seal stamp data. These five seals were measured by a registered image measuring device as image data of 256 × 256 points per seal. That is, the measurement accuracy is such that the sampling interval in the X and Y directions is 0.05 mm, and the distance accuracy in the Z direction is 0.
005 mm.

The seal to be inspected is measured using an image scanner (Nikon AX-1200) which is a seal to be inspected measuring apparatus. The scanner measures 256 × 256 points of binary data. FIG. 7 shows an example of a seal image pattern to be inspected stamped with a registered seal. 20 for one 3D registration imprint
0 types of seal patterns to be inspected (100 types of positive marks, 1 similar marks)
00 type). 5 types of registered imprints for a total of 1
000 patterns are collated.

Table 1 summarizes the results of 1000 types of collation. In order to compare the performance with a known method, a method using a two-dimensional registration imprint and a three-dimensional registration imprint are used. As a collation method, IEICE (D-II), vol. J80-D-
II, No. 1, pp. 148-155, Jan. (19
An experiment was also performed using the same test seal imprint when the method of 97) was used. Experimental results show that the proposed method improves the matching accuracy. For example, the apparatus of the present invention was able to correctly collate an imprint having a severe defect as shown in FIG.

[0110]

[Table 1]

Eight erroneous collations were confirmed out of 1,000 collations. FIG. 9 shows all imprints in which the erroneous verification has occurred. In all of the eight erroneous verifications, the inspected seal imprinted with the correct seal was determined to be a similar seal (fake imprint). This is because sufficient information could not be obtained because the vermilion portion of the seal imprint was shaved. These samples were difficult to determine even visually, and were inevitable samples. In such a case, rejection processing is performed so as to obtain the seal again, rather than making a determination in either one. If rejection determination is made for less than 60% of the imprint patterns to be inspected, these can be rejected, and the erroneous verification can be reduced to 0%. In addition, it was confirmed that the seal imprinted with the similar seal was never erroneously recognized as a correct seal (real seal). The security from the aspect of security was confirmed.

In the present invention, the average calculation time per collation was about 20 seconds. For example, IEICE (D
-II), vol. J80-D-II, No. 1, pp
148-155, Jan. In (1997), since about one hour was used for one collation using a CPU device having the same performance, the excellent effect can be confirmed. Such a high-precision collation in a short period of time is a major breakthrough. Only the present invention can achieve such excellent results.

[0113]

The present invention is embodied in the form described above, and has the following effects. In the comparison, since only the vermilion portion is evaluated, there is no problem of similarity reduction due to a random sesame salt pattern of the vermilion portion as occurs in the normal overlay evaluation.

Since the structure of the registered imprint is stored as the distance image, and the structure is matched, the change in the imprint pattern can be absorbed and correct evaluation can be performed. In the case of a markedly stamped seal, it cannot be evaluated sufficiently.
Although there is a possibility that the positive seal may be erroneously collated with the false seal, the rejection can be determined based on the proportion of the vermilion portion, and the problem can be solved by requesting the user to seal again.

Further, in order for a false mark to be determined to be a correct mark, a vermillion area that matches the stroke of the correct mark and is equal to or greater than a certain value is required. Such seals are visually determined to be the same. Therefore, in the present apparatus, what is visually determined to be a false mark is not determined to be a correct mark.
Since evaluation is performed one pixel at a time, evaluation can be performed more strictly than by visual observation. Most of the processing is required to adjust the position and the rotation angle, but the processing time is about 20 seconds, which is a practical level. As described above, it is possible to perform automatic seal verification instead of visual inspection.

[Brief description of the drawings]

FIG. 1 is a schematic configuration diagram of components of a three-dimensional seal registration / collation device of the present invention.

FIG. 2 is a perspective view of a seal for indicating that a seal is registered in three dimensions.

FIG. 3 is a vertical cross-sectional view for explaining that three-dimensional seal registration is performed, and a mountain surface is normalized to 0 and a deepest portion of a valley surface is normalized to 1;

FIG. 4 is a view for explaining that a rectangle is circumscribed around a seal to be inspected, and the center of the seal to be inspected is obtained and set as an initial value of a position parameter. FIG. 4A shows that the square circumscribes the outline of the circle “Yamamoto” and the center is determined by the diagonal intersection.
FIG. 4B shows a case where a part of the outline of the circle is missing and the diagonal intersection of the circumscribed rectangle is not at the center. FIG. 4C shows that a rectangle inclined at 45 degrees is circumscribed by a circle, and the center is obtained by a diagonal intersection.

FIG. 5 is a diagram showing an inspection imprint by N half straight lines passing through the center;
FIG. 11 is a diagram of sector division for explaining that the image is divided into a plurality of sector regions, the number of vermilion pixels belonging to the sector regions is counted, and a cross-correlation function is calculated.

FIG. 6 explains how to match the center and rotation angles of the registered seal and the seal to be inspected so that they are partially in contact with each other, and calculate the total distance between all the pixels of the seal to be inspected and the corresponding registered seal stamp. FIG.

FIG. 7 shows an example in which five regular seals and five fake seals were prepared in an embodiment in which an experiment was performed using the seals of Yamamoto, Yoshinaga, Nakai, Masaki, and Yoshikawa, and two imprints of each of the test seals were pressed. FIG.

[FIG. 8] An imprint with many blurring while using a positive seal of Yamamoto, Yoshinaga and Masaki, and an imprint with many blurring imprinted by a false seal of Yamamoto, Yoshinaga and Masaki Nakai.

FIG. 9 is a diagram of a seal image to be inspected when a false seal is recognized as a correct seal.

Claims (5)

(57) [Claims] 1. A seal stamp registration device and a seal stamp collation device, wherein the seal stamp registration device three-dimensionally measures the irregularity shape of the seal stamp, and in a three-dimensional coordinate defined by the device, a distance z _m for each coordinate x, y in a plane direction. (X,
y), a registered image measuring device for measuring as a distance image, a registered image storage device for storing distance data z _m (x, y) of each coordinate point (x, y), and a mountain surface of the registered seal as 0. A distance normalizing mechanism for normalizing the distance z _r (x, y) with the deepest point of the valley surface as 1, and a normalized image storage device for storing the normalized distance data z _R (x, y) The stamp verification device includes: a test image measuring device that optically reads a seal image to be measured and measures it as a binary image; and binary data w (x, y) of each coordinate point of the seal image to be tested. An image storage device to be inspected, an initial value setting mechanism that determines the center coordinates of the seal image to be inspected, matches the center coordinates and adjusts the rotation angle of the seal image to be inspected and the registered seal image, and a registered seal image that matches the center position and the rotation angle And the seal to be inspected are partially touched, and the distance between the seal to be inspected and the registered seal is determined by the red pixel of the seal to be inspected. A determination value D is calculated by dividing the sum S and dividing the total sum S by the total number of pixels Q in the vermilion portion, and the center position is shifted for a pixel in the vicinity of the center position. And the seal to be inspected are partially contacted, and the distance between the seal to be inspected and the registered seal is determined by the vermilion pixel of the seal to be inspected, and the sum S is divided by the total number of pixels Q of the vermilion portion to calculate the determination value D. An optimal position / rotation angle determining mechanism for _obtaining a minimum determination value D _min in a neighboring pixel and a determination value storage device for storing the minimum determination value D _min at that time; A three-dimensional seal registration / collation device, comprising: a collation result judging mechanism for judging a fake seal when the value is larger than a threshold value. 2. Following the normalized image memory, the contour of the mountain surface fattened by one pixel or two pixels, the distance data between the fattened ^{_{A z R * (x, y}} ) to form expansion a mechanism, the distance data _z ^R * after the shape expansion (x, y)
2. An inflation image storage device for storing the inscription is registered, and the profile of the mountain surface of the registered seal is thickened and registered.
The three-dimensional seal registration and verification device described in 1. 3. Following the expansion image storage device seal registration device, and the standard seal image feature extraction mechanism for counting the number of pixels C _r of Yamamen binarizes the distance data of registered seal, the number of pixels to be standard imprint features provided a standard imprint feature storage unit for storing a C _r, the seal collation device, divided by the number of pixels C _r mountain face registration seal counts the number of pixels C _i of ink pad pixels of the test print image C _i
/ The C _r calculates a three-dimensional seal registration and matching according to claim 1 or 2, characterized in that this value is provided to dismiss determination mechanism for rejecting a test print image is smaller than the constant of 0-1 apparatus. 4. An average value M of the distances between the mountain surfaces of the registered seal stamps is obtained from the judgment values D and M following the judgment value storage device of the seal stamp collation device.
4. The three-dimensional seal registration / collation apparatus according to claim 1, further comprising: a judgment value correction mechanism for obtaining a correction judgment value by subtracting a reference value. 5. Following the registered image storage device, the distance data in xyz space is voted in the polar coordinate space of θφρ, and the plane equation z _c (x, x,
seeking y), and inclination correction mechanism plane equation _{z c (x, z r (} x minus the distance data z _m including slope from y), the inclination of the registration seal sought y) to 0, after the inclination correction 5. The three-dimensional apparatus according to claim 1, wherein a correction image storage device for storing the distance data _zr (x, y) is provided to correct the inclination of the registered seal. Seal registration / collation device.