JPS6334508B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION Technical Field The present invention relates to a fingerprint verification method. In the present invention, the term "fingerprint" is defined as a representative of so-called pattern patterns such as fingerprints, foot prints, and other striped patterns. Furthermore, "verification" means determining the identity of one fingerprint to be searched for and a previously registered fingerprint.

BACKGROUND TECHNOLOGY A fingerprint that is normally compared is called a search fingerprint, and a fingerprint that is registered in advance to be compared is called a file fingerprint. Each fingerprint has a special point in the ridge pattern, that is, a point M (end point) where the ridge K breaks off in Fig. 1.
and a branching or merging point m (branching point), and these points are generally called feature points.

Conventionally, in fingerprint matching, information about the type of minutiae, the position X, Y, and the direction D of the minutiae with respect to a certain reference coordinate system is coded and stored for each of the search and file fingerprints. This is done using fingerprint characteristics.

For example, the method of selecting a coordinate system that describes the position and direction of a search fingerprint minutiae left at a crime scene or the like is arbitrary and generally does not match the coordinate system representing the minutiae points of a file fingerprint to be verified. Therefore, during verification, it is necessary to perform a coordinate transformation of rotation and translation by the most probable amount for each partner file fingerprint to be verified for the coordinate system of the search fingerprint, and then perform a comparison test.

In general, when performing verification, there are a huge number of fingerprints of the other file to be verified, so high speed and certainty (reliability) of verification are important.

The conventional fingerprint verification method will be described below.

In general, the number of minutiae of the search fingerprint and the minutiae of the file fingerprint to be compared are different, and their correspondence is unknown. In order to correlate these, a quantity representing a local feature that is not affected by coordinate transformation is introduced for each feature point.

For example, the type of each feature point (end point and branch point)
is one of them, and the number of other feature points that exist within a certain radius from a certain feature point is also a quantity that represents local features. Furthermore, we create a local coordinate system with the original feature point as the origin and the direction D as the Y axis, and divide these quantities into the number of other feature points distributed in each quadrant of this coordinate system. In other words, it becomes a quantity representing a local feature with an even larger amount of information.

When these local features are introduced, by comparing the feature points of each file fingerprint with each feature point of the search fingerprint using these local features, the degree of approximation of each can be determined regardless of how the coordinate system is selected. strength can be determined. Therefore, pair candidates between each feature point are determined locally in the order of the strength of approximation as described above,
The amount of coordinate matching that best matches the local approximations of the feature points serving as pair candidates can be determined, and the final match value can be determined.

However, conventional verification methods and devices have the drawback that the speed and reliability of the verification described above are not necessarily sufficient.

OBJECTS OF THE INVENTION An object of the present invention is to provide a fingerprint matching method that has high reliability and enables high-speed matching between a search fingerprint and a file fingerprint.

Structure of the Invention A conventional fingerprint verification device detects the positions and angles of each feature point in a predetermined reference coordinate system for a search fingerprint and a file fingerprint to be matched, and feature points indicating relation data independent of the coordinate system. a minutiae list storage means for storing a list of minutiae points; a fingerprint area storage means for storing a fingerprint area indicating whether each minutiae is a bright area or an unknown area for each of the two fingerprints; On the other hand, it is comprised of means for selecting a pair of minutiae from the minutiae points of the two fingerprints by checking the position, angle, and relation data of the minutiae.

The fingerprint matching device of the present invention includes a pair candidate list storage means for storing a pair candidate list indicating the pair minutiae described above as pair candidates, and a pair candidate list storage means for storing a pair candidate list indicating the pair candidate minutiae as pair candidates; optimal coordinate matching amount generation means for generating an optimal coordinate matching amount from the deviation and angular deviation; coordinate conversion means for converting either of the reference coordinate systems of the two fingerprints into a coordinate system matched by the optimal coordinate matching amount; a pair list generating means for generating a pair list indicating the true minutiae and pair values determined by checking the consistency of the position, angle, and relation data of each of the coordinate-matched minutiae; It is characterized by comprising a matching value determining means for determining a matching value from the feature points and pair values.

Embodiments of the Invention The present invention will be described in detail below with reference to the drawings.

Referring to FIG. 2A, the system to which the fingerprint verification device of the present invention is applied includes a verification data input processing device 1, a verification data management processing device 2, a data storage device 3, a verification control device 4, and four verification devices 5. ，
It consists of numbers 6, 7 and 8. However, the number of matching devices is not limited to four. Referring to FIG. 2B, the verification devices 5, 6, 7 and 8
Each of
0 and the processor 50, and data lines 10a, 20a, 50a and control lines 10b, 2 for data transmission between the processor 50 and the unit 10.
It is composed of 0b and 50b.

Referring to FIG. 3, the control unit 10 shown in FIG. 2B includes a buffer memory 300 for storing data from the verification control device 4 and data from the secondary verification processor 50, and a buffer memory 300 for storing data from the verification control device 4 and for exchanging data with the verification control device 4. a processor interface circuit 302 and a control circuit 303 that exchange data with the processors 20 and 50;
and a transfer destination determining circuit 304.

Next, the operation of the system during fingerprint verification will be explained below with reference to FIG. 2A. Verification data management processing device 2
From there, a verification instruction and location information in the data storage device 3 of search fingerprints and file fingerprints to be verified are given to the verification control device 4. Upon receiving the verification instruction, the illuminator control device 4 first reads the search index 1 finger from the data storage device 3 and sends it all to the verification devices 5, 6, 7, and 8 in common, and then sequentially reads the file fingerprints from the data storage device 3. Verification device 5, 6, 7, 8
to start matching. That is, the verification device 5,
Reference numerals 6, 7, and 8 calculate a matching value by performing parallel matching between one search fingerprint and each different file fingerprint.
When the comparison between one search fingerprint and a large number of file fingerprints is completed, the comparison control device 4 controls each of the comparison devices 5, 6,
The verification results are read from 7 and 8, rearranged according to the verification values, and reported to the verification data management processing device 2. The verification data management processing device 2 edits the verification results and outputs them via the signal line 12.

Referring to FIG. 4, each of the primary matching processors 20, 30 and 40 shown in FIG.
Sequence control circuit 210, primary feature point list memory 211, coordinate conversion circuit 212, nearest special point restoration circuit 213, "pair" inspection circuit 214, "pair"
Candidate list memory 215, work area 216, coordinate matching amount determination circuit 217, and control memory 218
The primary minutiae list memory 211 is composed of a search fingerprint minutiae list memory 211a and a file fingerprint minutiae list memory 211b.

Referring to FIG. 5, each of the secondary processors 50 and 60 in FIG.
1, coordinate conversion circuit 512, pair candidate list memory 5
13, a work area memory 514, a pair list memory 515, an area pattern list memory 516, a candidate list memory 517, a control memory 518, and an arithmetic circuit 519.

Next, the operation of one embodiment of the present invention will be described in detail in the following order according to the flow shown in FIG.

(1) Reading operation of fingerprint minutiae data (2) Coordinate conversion operation (3) Nearest neighbor minutiae restoration operation (4) Example of test operation (5) Variation example of test operation (6) Generation of coordinate matching amount and matching operation (7) Precise matching operation by the secondary matching processor (8) Transfer destination processor determination operation (1) Reading operation of fingerprint minutiae data Figure 7 shows the structure of the minutiae data used in the fingerprint matching of the present invention. shows. The feature point data consists of identification information, a region pattern list, and a feature point list. The identification information includes information such as a fingerprint identification number assigned to each fingerprint, date of birth, gender, finger type, and pattern classification. The area pattern list is information indicating whether the fingerprint area at the time of feature point extraction in the input data processing device 1 is a bright area or an unknown area. The minutiae list stores information indicating the characteristics of the minutiae points M ₀ , M ₁ , ..., M _f ...M _i , ...M _z within the fingerprint area, and the data shown in Figure 8 is It has a structure.

Referring to Figure 8A, the th minutiae generally has information such as a type code indicating whether the minutia point is a branch point or an end point, Q information indicating the position, D indicating the ridge direction, and information about the surroundings of the minutia point. C indicating confluence
and R ₀ , R ₁ , R ₂ , and R ₃ indicating the number of ridges between the feature point M and its nearest neighbor feature point. The positions and directions on the minutiae list are all expressed using an orthogonal coordinate system having an origin at an appropriately determined point in the fingerprint area as a reference coordinate system. Generally, these reference coordinate systems differ depending on the fingerprint.
Regarding the feature point extraction method mentioned above, please refer to the patent application filed in 1983.
It is described in detail in 39648.

Figure 8B shows the feature points closest to the feature point M (referred to as nearest neighbor feature points) for each of the four quadrants in the local coordinate system whose y-axis is the direction D of the feature point M.
The number of ridges R ₀ , R 1 , M ₀ , M 1 , M ₂ , M ₃ and the number _of ridges between the minutiae M and each _of the nearest minutiae points,
The relationship between R ₂ and R ₃ is shown.

The above-mentioned input feature point data is input to the data input processing device 1 and the verification data management processing device 2 shown in FIG. 2A.
The data is stored in the data storage device 3 via the . Next, the verification data management processing device 2 instructs the verification control device 4 to read out the search fingerprint from the data storage device 3. In response to this instruction, this device 4 searches for an input feature point list from the data storage device 3 via the collation control device 4, the control unit 10 in FIG. 2, and the sequence control circuit 210 in FIG. 4.The feature point list memory 211a Enter. When the feature point list of the search index is read from the data storage device 3, the feature point list of the file fingerprint is read from the position specified by the read address. This read file fingerprint minutiae list is sent to the file minutiae list memory 21 in FIG. 4 via the same route as described above.
1b.

The feature point list memory 211 shown in FIG. 4 stores a search and file fingerprint input feature point list in the format shown in FIG. 8A. Information indicating the type code Q and location at the address indicated by each feature point M
Information D indicating X and Y, direction, information C indicating the amount of density, and the number of ridges R ₁ -R ₄ are stored.

The data read from the feature point list memory 211 is subjected to coordinate transformation in a coordinate transformation circuit 212 shown in FIG.

(2) Coordinate Conversion Operation Next, the coordinate conversion operation used in the fingerprint matching device of the present invention will be described in detail.

Since the file fingerprint data and the search fingerprint data are different in collection time, collection conditions, and collection environment, some correction work is required prior to comparing the two fingerprint data. In other words, from the standpoint of pattern recognition, the position and direction of the break in the fingerprint pattern are important points for minutiae, but in many cases file fingerprint data and search fingerprint data are different from each other for the reasons mentioned above. Since the coordinate systems expressing the positions of the break points are different, it is necessary to perform coordinate transformation from the coordinate system of the search fingerprint data to the coordinate system of the file fingerprint data to match the coordinates.

Furthermore, in addition to the minutiae list, the fingerprint data includes an area pattern list representing unknown areas of the fingerprint when the fingerprint is taken, but this area pattern list is not a target of conversion. It is necessary to refer to the area pattern list during conversion to the search fingerprint data and minutiae comparison processing with the file fingerprint data, but since the area pattern list of the search fingerprint data is the value before conversion, at this time must perform an inverse transformation from the coordinate system of the file fingerprint data to the coordinate system of the search fingerprint data.

Referring to FIG. 9, the orthogonal coordinate system (X _p −Q _p −
The coordinate position (x _p , y _p ) of point _P on the new orthogonal coordinate system (X _q −O _q −Y _q ) is the coordinate position (x _q , y _q ) of point P on
The converted state is shown. The origin O _p of both orthogonal coordinate systems is displaced by the interval (x, y) between O _q ,
And both orthogonal coordinate systems are rotated by an angle θ. It can be easily seen by a geometrical method that the following relationship holds between the coordinate position (x _p , y _p ) before transformation and the coordinate position (x _q , y _q ) after transformation.

x _q = (x _q - x)・cos θ+ (y _p − y)・sin θ y _q = (y _p − y)・cos θ− (x _p − x)・sin θ... Referring to Figure 10, the equation Therefore, the coordinate position (x _q , y _p ) after transformation from the coordinate position (x _q , y p )
The process for obtaining y _q ) is shown. In other words, the displacement (x, y) and rotation angle θ of both orthogonal coordinate systems
and the coordinate position (x _p , y _p ) to be transformed is given, the process to obtain cos θ and sin θ (phase), the process to obtain (x _p −x) and (y _p − y) (phase), Processing to obtain an intermediate result by multiplying the processing result by the processing result of the phase (Phase), and processing to obtain the final result by adding or subtracting the intermediate results (Phase)
The coordinate position (x _q , y _q ) transformed by

In addition, in the opposite direction to the above conversion direction, convert the coordinate position (x _q , y _q ) to the coordinate position (x _p , y _p ) (inverse transformation)
The formula for doing this is shown below.

x _p = x _q・cosθ−y _q・sinθ+x y _p =y _q・cosθ+x _q・sinθ+y... Conventionally, formulas and transformations expressed by formulas are all performed by software methods, so high-speed processing is possible. The disadvantage is that it is difficult.
In particular, it is difficult to speed up processing with phase unless a computer with special functions such as high-speed multiplication and division functions is used.

Referring to FIG. 11, the coordinate conversion circuit 21
2, three registers 1110, 1111 and 1112, six selection circuits 1120, 1121,
1140, 1141, 1170 and 1171,
5 adders/subtractors 1130, 1131, 1132,
1160 and 1161 and four multipliers 115
0, 1151, 1152 and 1153 and a read-only memory (ROM) 1180.

The ROM1180 has two orthogonal coordinate systems (X _p −O _p
A circular function value for the rotation angle θ between -Y _P ) and (X _q -O _q -Yq) is stored in advance.

Selection circuits 1120 and 1121 output the respective outputs of input terminals X and Y when the conversion mode instruction signal (mode signal) INV instructs forward conversion, and respectively output the outputs of input terminals X and Y when the mode signal INV instructs inverse conversion. It operates to select each output of adder/subtractor 1160 and 1161. Similarly,
Selection circuits 1140 and 1141 select adder/subtractors 1130 and 1131, respectively, during forward conversion, and select respective outputs of input terminals X and Y during inverse conversion.
During inverse transformation, adder/subtractor 1130 and 1 are used, respectively.
131, and adder/subtractor 1130,1
131, 1132, 1160, and 1161 operate to perform subtraction, subtraction, subtraction, addition, and subtraction, respectively, during forward conversion, and addition, addition, addition, subtraction, and addition, respectively, during inverse conversion.

First, input terminals X, Y, and D are given displacements x, y, and rotation angle θ, respectively, and a prefix signal is
When PAL is applied, displacement x, y and rotation angle θ
are registers 1110, 1111 and 1, respectively.
112. The rotation angle θ input to the register 1112 becomes the address for accessing the ROM 1180, and the sine value is obtained from the ROM 1180.
Read out sin θ and cosine value cos θ (phase in FIG. 10).

Next, the application of the prefix signal PAL is suspended, and the coordinate position to be transformed and the orientation of the feature point in the orthogonal coordinate system to be transformed are given to the input terminals X, Y, and D, respectively. At this time, in response to whether the mode signal INV is "0" or "1", it becomes a forward conversion instruction or an inverse conversion instruction, respectively, and each selection circuit and each adder/subtractor operate as described above.

Now, when the mode signal INV is "0", the adders/subtractors 1130 and 1131 are in the phase shown in FIG.
2 and 1153 are phases, and adders/subtractors 1160 and 1161 perform phase processing, and coordinate positions x _q and y _q after forward transformation are obtained from output terminals X _T and Y _T. The adder/subtractor 32 subtracts the rotation angle θ from the orientation of the feature point in the orthogonal coordinate system to be forward transformed, and outputs this value from the output terminal D _T as the orientation of the feature point in the orthogonal coordinate system after the forward transformation.

Next, when the mode signal INV is "1", the multipliers 1150, 1151, 1152, and 11
53 are x _q・cosθ, y _q・
sinθ, y _q・cosθ and x _q・sinθ multiplication, adder/subtractor 1160 and 1161 respectively x _q・cosθ−
Subtraction of y _q・sinθ and addition of y _q・cosθ+x _q・sinθ, adder/subtractors 1130 and 1131 respectively perform (x _q・sinθ
cosθ−y _q・sinθ)+x and (y _q・cosθ+x _q・sinθ)+
By performing each addition of y, the coordinate positions x _p and y _p after inverse transformation are obtained from the output terminals X _T and Y _T. Adder/subtractor 1
132 adds the rotation angle θ to the orientation of the feature point in the orthogonal coordinate system to be inversely transformed, and uses this value as the orientation of the feature point in the orthogonal coordinate system after the inverse transformation, and outputs it to the output terminal.
Output from D _T.

By employing the above configuration, coordinate positions for either forward or reverse transformation can be obtained in one motion, making it possible to perform bidirectional high-speed coordinate transformation. When the feature point list shown in FIG. 13 is stored in the feature point memory 211, the operation for restoring the nearest neighbor feature points described below is not necessary. However, when attempting to reduce the capacity of the data storage device, the feature point list shown in FIG. 8A must be stored in the feature point list memory 211. In this case, the following operation for restoring the nearest feature point is absolutely necessary.

(3) Restoration operation of the nearest feature point The coordinate conversion circuit 212 and the nearest feature point restoration circuit 213 shown in FIG.
As shown in the figure, Mf ₀ , Mf ₁ , Mf ₂ ,
Added as Mf ₃ . To explain this in detail, the degree of similarity between the search and file feature points is quantitatively expressed by the local features of each feature point, that is, the local features of each feature point that are not affected by the selection of the central coordinate system. Use approximation.

For example, Q representing the above-mentioned type is one of them. Furthermore, the number of other feature points within a predetermined distance from the target feature point can also be used as the local feature.

In particular, the following local features provide extremely powerful materials for creating a pair candidate list.

Now, select the arbitrary feature point of a certain pattern.
When M, a rectangular coordinate system with the position X and Y of this feature point M as its coordinate origin and the direction D as the direction of the X axis is defined as a local coordinate system determined by this feature point Mi.

In each quadrant of this local coordinate system (1st quadrant, 2nd quadrant, 4th quadrant, and 3rd quadrant), this coordinate origin (that is, the original feature point of interest)
If the numbers of the other minutiae points closest to M) are 0, 1, 2, and 3, then the original minutiae point M has other minutiae points in each quadrant of the local coordinate system determined by M). The nearest feature points M ₀ , M ₁ ,
This means that it has local characteristics of M, ₂ and M ₃ . Of course, this local feature is independent of the choice of central coordinate system.

So now, we need to find the arbitrary minutiae point of the search fingerprint.
0, 1, 2, and 3 as described above are obtained for M, and similarly for any j-th feature point Mj of the file fingerprint to be verified, each quadrant of the local coordinate system determined by Mj is Feature point number in
Assuming that the nearest feature points with k0, k1, k2, and k3 have been obtained, the local approximation between these two feature points M and Mj is the nearest point in the corresponding quadrant on each local coordinate system. The mutual relationship between feature points can be easily quantified by determining whether or not it is within the vicinity of a certain threshold. For example, find the difference in the X and Y coordinates of the nearest feature points in the respective local coordinate systems in the corresponding quadrant, and if this difference is within a certain threshold, the nearest feature in each quadrant It can be quantified by adding 1 to the weight of the degree of approximation for each point.

In this way, for each feature point, the process of finding the nearest feature point in each quadrant of the local coordinate system determined by this feature point and associating this with the original feature point (hereinafter referred to as relation generation processing) is as follows: As mentioned above, this is an important step in checking the identity of patterns.

In particular, in processing such as fingerprint matching, the number of file fingerprints that must be compared against one search fingerprint to determine identity is enormous, so it is important to perform such relation generation processing quickly and reliably. need to be executed.

Referring to FIG. 12, the nearest feature point restoration circuit 213 in FIG. 4 includes the feature point register 1201(Q),
1202(X), 1203(Y) and 1204
(D), address register 1205 (A), 120
5(B), 1206(E) and 1206(F),
Constant adder 1207, input selector 1208, square calculators 1209 (X ² ) and 1210 (Y ² ), adder 1211, register file 1212, comparator 1213, controller 1214, coincidence detector 121
5 and a prohibition gate 1216.

Now, the feature point list memory 211 used in this embodiment is configured as follows.

As shown in Figure 13, there are 256 row addresses specified by 8-bit binary numbers and 256 row addresses specified by 3-bit binary numbers.
It has a total of 2048 memory addresses consisting of 8 column addresses specified in base numbers. Each row address corresponds to each feature point of one pattern, and each information about a specific feature point is stored in each column address with the corresponding row address as described below. .

First, the 0th column address stores information (hereinafter referred to as Q) representing the type of the th minutiae (endpoint, branch point, singularity, etc.), and the 1st column address stores the information representing the type of the th minutiae (terminus, branch point, singularity, etc.) The X coordinate (hereinafter referred to as X) of
The direction D of the th feature point (hereinafter referred to as D) is stored in the third column address.

These X, Y, and D values are written in a specific XY coordinate system (hereinafter referred to as the center coordinate system when necessary), with the coordinate origin approximately at the center of this pattern. , the error during coordinate system extraction generally becomes quite large.

Furthermore, all the values of Q, X, Y, and D of a certain pattern have already been written to each memory address by the host device as described above prior to this processing. In this case, a specific end mark (E0D) is placed at the 0th column address (address that stores Q) at the next row address of the last feature point of this written pattern.
is stored and used to indicate the end of processing.

The four column addresses from the 9th to the 12th are memory addresses where the nearest minutiae generated by this embodiment are to be stored, and the nearest minutiae are as described below. .

Now, take a specific feature point and create a local coordinate system (hereinafter referred to as the XY local coordinate system) with the position X and Y of this feature point as the coordinate origin and the direction D as the X-axis direction, as shown in Figure 14. Suppose you create this
For each quadrant of the XY local coordinate system, take one other feature point that is closest to the origin of this local coordinate system (that is, the position of the th feature point), and assign the number of each feature point to the quadrant. M0, M1,
Assuming M3 and M2, these are the nearest feature points.

That is, in this embodiment, for any feature point,
In order to grasp the state of the vicinity of the feature point, it shows the operation of searching for the nearest feature point in each quadrant on the local coordinate system uniquely defined by this feature point, and associating them with the original feature point. . The nearest feature points obtained in this way have a characteristic that, unlike the X, Y, and D data, they are independent of how the central coordinate system is selected.

Now, these nearest feature points are stored in each column address shown below of the row address corresponding to the original feature point.

The number M0 of the nearest feature point in the first quadrant of this local coordinate system is stored in the ninth column address. The nearest feature point number M1 in the second quadrant is the 10th
The number M2 of the nearest minutiae in the 4th quadrant is set to the 11th column address, and the number M3 of the nearest minutiae located in the 3rd quadrant is set to the 11th column address.
Store each at the 12th column address. Note that the order of the quadrants and the column addresses to be stored are different from each other in order to simplify the hardware described later.

Next, the above-mentioned nearest neighbor feature point generation process will be explained in detail in relation to the operation of each circuit.

Inside the nearest neighbor feature point generation circuit, the control unit 1214
(FIG. 12) includes a control memory (not shown) that stores microprograms, and the processing proceeds by sequentially reading and executing the microprograms. The progress of this process is shown as a flowchart in FIG.

First, a start instruction is supplied from the sequence control circuit 210 via line 299 (STRT),
When the nearest neighbor feature point generation process starts (15th
), the control unit 1214 is the address register 120
5(A) and 1205(B) are set to initial values. The address register 1205 is a register that specifies the feature point from which the nearest neighbor feature point is to be found, and has a row address designation part 1205 (A) and a column address designation part 1205 (B). Both are cleared to 0 and set to point to the first memory address. (Figure 15a).

Next, the control unit 1214 controls the register 1205
Q _A , X _A , Y _A of the feature points from the row address of the feature point list memory 211 specified by the content A in (A)
and D _A , and X _A , Y _A and D _A as the 11th
These are loaded as transformation parameters into parameter registers 1110, 1111 and 1112 of the coordinate transformation circuit 212 shown in the figure. For this purpose, the control unit 1214 in FIG.
By supplying the contents of 205(A) and 1205(B) to the minutiae list memory 211 via line 12056(AD), the data can be read from the row address of the minutiae list memory 211 specified by the address register 1205(A). address register 1
4 bytes specified in 205(B), i.e.
Read Q _A , X _A , Y _A and D _A. The read values of X _A , Y _A and D _A are respectively applied to lines 1010 (X), 2020 (Y) and 10 in Figure 11.
30 (D) to each parameter register, and by providing a latch pulse on line 1000 (PL), these values are set in each register (FIG. 15C).

Next, the control unit 1214 in FIG. 12 determines whether or not the Q _A read above is an end mark, and if it is an end mark, it ends the process (FIG. 15 E), If not, proceed as follows (Fig. 15D).

Next, the control unit 1214 in FIG. 12 sets initial values for address registers 1206(E), 1206(F) and register file 1212 (FIG. 15(f)).

Address register 1206, like address register 1205, includes row address specification portion 12.
06(E) and column address specification part 1206
(F), which is a register for indicating the feature point number to be compared, and both are cleared to 0 by initial value setting.

Further, the register file 1212 consists of four registers 1212-0 to 1212-3, and each register has its own address field M.
0, M1, M2 and M3 and respective distance fields R0, R1, R2 and R3.

These address fields M0 to M3 contain the numbers of other feature points found at the current stage of processing that are closest to the feature point currently specified by the address register 1205(A). This is a field for storing each quadrant of the coordinate system separately.The point number of the first quadrant is set to M0, the point number of the second quadrant is set to M1, and the point number of the third quadrant is set to M0.
3, store the numbers of the points in the fourth quadrant in M2.

Further, the distance fields R0 to R3 are the origin of the local coordinate system (that is, the address register 1205
This field stores the square value of the distance from the feature point position specified in (A) to each of these feature points M0 to M3.

In the initial value setting described above, all bits of each of these registers in the register file 1212 are set to "1", address fields M0 to M3 represent that there is no corresponding feature point, and distance field R0 is set to "1". ~R3 is set to the farthest distance that can be represented.

Next, the control unit 1214 supplies the contents of the address registers 1206(E) and 1206(F) to the feature point list memory 211 via the line 12056(AD). The specified 4 bytes, that is, Q _E , X _E , Y _E and D _E are read out, and Q _E is stored in the feature point register 1201 (Q).

In addition, X _E , Y _E and D _E are converted into the coordinate conversion circuit 21
2 (Figure 11) lines 1010 (X), 102
0(Y) and 1030(D) respectively, the X _A Y _A local coordinate system (X _A ,
The coordinates are transformed into the respective values X _EA , Y _EA and D _EA expressed in the local coordinate system determined by Y _A and D _A , and the respective values obtained in this way are
20 (X'), 2130 (Y') and 1060
(D') through feature point register 1202 (X), 1
It is stored in 203 (Y) and 1204 (D) (Fig. 15 G).

Next, the control unit 1214 executes the _process shown in FIG. , N), and when the end mark appears, the process starts on the side shown in FIG. 15 (h, Y in FIG. 15).

Upon entering the execution of FIG. 15(a), the following processing is performed.

That is, the respective contents X _EA and Y _EA stored in the feature point registers 1202 (X) and 1203 (Y) are calculated by the square calculator 1209 (X ² ).
and 1210 (Y ² ), are added together in adder 1211, and are sent to line 12110.
This produces an output of X ² _EA + Y ² _EA , which is output by comparator 1213.
is added to one input of On the other hand, the sign bits (represented by SX and SY, respectively) of the contents X _EA and Y _EA stored in feature point registers 1202 (X) and 1203 (Y) are 2-bit data (SY, SX). as line 12
023. The binary number represented by this 2-bit data (SY, SX) is X _EA ,
If the position of the feature point indicated by Y _EA is in the first quadrant, second quadrant, third quadrant, and fourth quadrant of the local coordinate system, respectively, (0, 0) = 0,
(0,1)=1, (1,1)=3 and (1,0)=
It will take a value of 2. The control unit 1214 sends a control signal to the input selector 1208 via a line 12140 to select the above-mentioned 2-bit data (SY,
SX) and supplies it via line 12080 to the register file as a file addressing signal at 1212.

As a result, in the register file 1212,
Contents R of the distance field of the register specified by (SY, SX) (that is, the register corresponding to the quadrant in which the minutiae stored in the minutiae register exists)
(SY, SX) is read out via line 12120 and provided to the other input of comparator 1213.

As a result, the comparator 1213 performs the above-mentioned R
Compare the value of (SY, SX) with the value of X ² _EA + Y ² _EA mentioned above, and calculate the value R (SY, SX) obtained so far.
If a feature point closer to the current coordinate origin appears in the same quadrant, that is, if R(SY, SX)>X ² _EA + Y ² _EA holds true, “1” is output to line 12130.

On the other hand, the content of the address register 1205 (A) and the content of the address register 1206 (E) are compared in the match detector 1215, and if they match, that is, the content of the address register 1206 (E) is compared with the original feature point. If the power feature points are the same, "1" is output to line 12150 and prohibition gate 1 is output.
Passage of 216 is prohibited.

As a result, the line 12 is set only when both conditions R(SY, SX)>X ² _EA + Y ² _EA A≠E (1) are satisfied.
160 becomes "1" (Fig. 15, Y side).

When "1" is output on line 12160, register file 1212 is controlled to the write state, and the contents of the register specified by the aforementioned 2-bit data (SY, SX) in register file 1212 are updated. That is, the content E of address register 1206 (E) specifying a new feature point is stored in address field M (SY, SX) via line 12060, and
The value of the output X ² _EA +Y ² _EA of 1 is stored in the distance field R (SY, SX) via line 12110 (FIG. 15C).

After this, the control unit 1214 controls the constant adder 1
Address register 1206 (E) using 207
is updated by incrementing the content by 1 to indicate the next feature point to be compared (FIG. 15, S), and returns to the process of reading out the next feature point (FIG. 15, G).

In addition, if the condition of equation (1) above does not hold, that is, the position of the feature point read this time to be compared is closer to the origin (original feature) than the already obtained position of the feature point in the same quadrant. point), or if the feature point to be compared is the same as the original feature point, the register file 12
Without updating the contents of 12 (Fig. 15,
N), the contents of the address register 1206(E) are updated by 1 to indicate the next feature point to be compared (FIG. 15), and the next feature point is read out again (FIG. 15). Return to g).

In this way, by repeating the processes of K, K, K, K, and S in FIG. 15, the control unit 1214 successively reads new feature points to be compared, and each For each quadrant,
Every time another feature point closer to the original feature point appears, the data of this feature point is used to create the file 12.
Update the contents of 12. This process is repeated until the end of the feature points stored in the feature point list memory 211 in FIG. 4 appears (until the end mark appears on the read _QE ).

When the above processing is completed, the register file 121
In the address fields M0 to M3 of 2, the number (row address) of the other minutiae closest to the original minutiae specified by the address register 1205(A) is stored for each quadrant. They would have been requested separately.

Therefore, next, the file 12 obtained in this way is
The contents of the 12 address fields M0 to M3 are
specified by address register 1205(A),
Nearest neighbor feature point storage position of feature point list memory 21
Store in M _A0 to M _A3 .

When proceeding to the nearest neighbor feature point storage process, the control unit 1214 sets the contents of the address register 1205 (B) indicating the column address of the original feature point as a value indicating the beginning of the nearest neighbor feature point storage position. The initial value is set to 4 (FIG. 15C).

At the same time, the output of line 12140 is switched so that input selector 1208 selects line 12060. Since the lower 2 bits of the address register 1205 (B) are supplied to this 12060, reading from each register in the register file 1212 is performed using the lower 2 bits of the address register 1205 (B) ( In other words, it is specified as the value obtained by subtracting 4 from B.

Now, the address field M (B-4) specified as described above in the register file 1212
The contents of are read out and line 12121
(WD) and is supplied as write data to the feature point list memory 211. On the other hand, the contents of address register 1205(A) and 1205
The contents of (B) are supplied as row address and column address designation data to the feature point list memory 211 in FIG. 4 via line 12056 (AD), respectively. Therefore, by supplying a write pulse to the feature point list memory 211 via the line 12141 (W), the control unit 1214 writes the data of M (B-4) of the above-mentioned file 1212 to the nearest neighbor of the feature point list memory 211. It can be transferred and stored at the feature point storage position M _AB (first
Figure 5).

When this transfer is completed, the control unit 1214 uses the constant adder 1207 to transfer the address register 1205
The contents of (B) are incremented by 1 and updated to indicate the next file address and column address (FIG. 15).

If the above process is repeated until the value of 1205(B) overflows and becomes 0 (see Figure 15,
Y), since the storage of all the nearest feature points to the original feature point specified in the address register 1205 (A) is completed, the control unit 1214 uses the constant adder 1207 to store the feature points in the address register 120.
The contents of 5(A) are updated by adding 1 to indicate the feature point from which the next relation data should be obtained (FIG. 15-S).

In this way, each data of this new source feature point is read out, and the coordinate conversion circuit 2 of FIG.
12 parameter registers 1110 to 1112, and the process returns to the process of configuring the local coordinate system for this new feature point (FIG. 15C).

In this way, the contents of the register 1205(A) are updated, and each time a new feature point is designated as the basis, a local coordinate system for that feature point is constructed. Through this loop, all other feature points are examined in order, and data indicating the nearest point in each quadrant of this local coordinate system is stored in the register file 1212.
generated inside. Once this is completed, these are transferred to the nearest feature point storage location of the feature point list memory 211 as the nearest feature points to the original feature point in the loop of S and C in FIG.

The process is repeated by the control unit 1214 in the manner described above, but once the register 1205 (A) has been specified for all feature points,
The generation of all the nearest feature points in the feature point list memory 211 is completed, and the next update of the register 1205(A) reads out the end mark (FIG. 15E, Y), and the process ends (FIG. 15E and Y). Figure 15 O).

As mentioned above, when the position and direction of each feature point of one pattern are given, the local coordinates determined for each feature point are used as information to understand the state of the vicinity of each feature point. By using the system to find other nearest feature points in each quadrant, the nearest feature points to the original feature point can be generated at high speed using a relatively simple algorithm.

Unlike the original data representing the positional direction of each feature point, the nearest feature points generated in this way are quantities that are unrelated to how the center coordinate system is selected, so two patterns whose center coordinate systems are not aligned To provide effective materials for creating a pair candidate list, which is required first when performing pattern matching.

(4) An example of test operation
An example of 14 includes a relation connection section 2141, a feature point storage section 2142, and a pair inspection section 2143.

Below, when we collectively refer to the feature point data for one feature point, the feature point number of the nearest feature point related to that feature point, and the basic relation data consisting of their relations, we will refer to this as total feature point data. When all the total minutiae data for the fingerprint or the total connected data described below are collectively referred to as fingerprint data, this is referred to as fingerprint data.

The relation connection unit 2141 converts the minutiae number of the nearest minutiae in the basic relation data of each minutiae of the fingerprint data stored in the minutiae list memory 211 to local coordinate position data and direction data regarding the minutiae. to generate relation connection data, which is stored in the feature point storage unit 2142.
are combined to send and store the data.

The total connection data stored in the feature point storage unit 2142 is obtained from the feature point data and relation connection data (relation rij, position data xij/yij, direction data dij) for each feature point, as shown in FIG. It's summery.

The “pair” detection unit 2143 is the feature point storage unit 2142
It has the role of extracting the total connected data of one minutiae of each of the search fingerprint and the file fingerprint from the search fingerprint and detecting the minutiae points that should form a "pair". The combination of total connected data regarding the feature points is sent to and stored in the pair candidate list memory 215 in FIG. 4.

Here, "pair" means that the total minutiae data or total connected data of all the minutiae points of the search fingerprint match the total minutiae data or total connected data of each minutiae point of the file fingerprint. A combination of feature points. At the time of the initial comparison and match determination, multiple minutiae points of the file fingerprint are compared and determined to be a match for one minutiae point of the search fingerprint, and the ratio is 1:n.
There may be a combination of (n≧2). This 1:
The combination of n is further processed to form a 1:1 “pair”
However, in the following explanation, the 1:n combination will also be referred to as a "pair".

The data stored in the feature point list memory 211 in FIG. 4 includes the feature point type Q1, density (Ci), position data (Xi, Yi),
It is a collection of one fingerprint of the total minutiae data including the direction data (Di) and all the basic relation data (rij, Mj) of the nearest minutiae points, and these are sent by the address signal from the relation connection unit 2141. 2146 and is sequentially outputted to the relation connection unit 2141 as a data signal 2144.

When the relation linking unit 2141 sequentially reads the basic relations of the nearest feature points from the feature point list memory 211a in response to the address signal 2146, the relation connecting unit 2141 stores the feature point number (Mj) included in this basic relation data. The data signal 21411 is converted into position data (xij, yij) and direction data (dij) regarding the feature point using local coordinates, and stored in the feature point storage unit 2142 along with the relation (rij).
Sequentially. The feature point storage unit 2142 stores the relation connection data sequentially sent by this data signal 21411, and configures the total connection data together with the feature point data in the form shown in FIG. 17.
It is stored in the search feature point memory 1802 in FIG.

When the above-described relation concatenation operation regarding one minutiae of the search fingerprint is completed, the relation concatenation unit 2141 reads out the total minutiae data of one minutiae of the file fingerprint from the minutiae list memory 211b and connects the aforementioned search fingerprint. In the same way as in the case of , the total connected data is configured and stored in the file feature point memory 18 in FIG.
Store in 03.

Search feature point memory 18 of feature point storage unit 2142
02 and the file feature point memory 1803, the relation linking unit 2141 outputs the command signal 21.
410, the “pair” detection unit 2143
, the “pair” detection unit 2143 sends the address signal 21431 and converts them into the data signal 2.
1421.

Meanwhile, the relation connection unit 2141 reads out the second total minutiae data of the file fingerprint during this time,
A relation connection operation is performed using the free space in the buffer memory of the feature point storage unit 2142, and the data in the file feature point memory 1803 is transferred to the “pair” detection unit 2.
143 and then stored.

The total connected data of the minutiae of the search fingerprint and the file fingerprint stored in the minutiae storage unit 2142 and read out to the “pair” detection unit 2143 is stored in the “pair” detection unit 2
At 143, the presence or absence of a "pairwise" relationship is checked, and when the total connected data of the minutiae of the search fingerprint and the total connected data of the minutiae of the file fingerprint match within a predetermined threshold, the total connected data of these The combination is sent to the pair candidate list memory 215 in FIG. 4 by the data signal 2147 and stored at the address specified by the address signal 2145.

The above-mentioned relation concatenation operation and "pair" detection operation are first performed for all the minutiae points of the file fingerprint with respect to the first minutiae point of the search fingerprint, and when this is completed, the second feature point of the search fingerprint is The same procedure is applied to all minutiae points of the file fingerprint.
This process is then repeated for all combinations of minutiae of the search fingerprint and file fingerprint.

When the above "pair" detection operation for all combinations of minutiae points of the search fingerprint and file fingerprint is completed and all "pairs" are stored in the pair candidate list memory 215, the sequence control circuit 210 performs "pair" detection. This is notified from the section 2143 by a command signal 2148.

Referring to FIG. 19, the relation connecting portion 2
141 is a shift register 401, an subtractors 403X, 403Y, and 403D that perform subtraction processing with position data or direction data input directly by data signal 2144;
a circular function generator 404 that receives the output of the D register and generates a circular function, and multipliers 405X, 405Y that receives the outputs of the subtracters 403X, 403Y and the outputs of the circular function generator 404 and multiplies them.
406X, 406Y and multipliers 405X and 40
6X, and an adder 407X and a subtracter 407Y that receive the outputs of multipliers 405Y and 406Y and perform addition or subtraction processing thereon.

When this relation connection unit 2141 receives the total feature point data of, for example, the feature point Ma sent from the feature point list memory 211 by the data signal 2144, the relation connection unit 2141 controls the shift register by the command signal 4001 from the control circuit 40101. 401 stores basic relation data ra, M~rak, Mk, and X register 402X, Y register 40
2Y and D register 402D contain position data X _a ,
Stores position data Y _a and direction data D _a .
At this time, the feature point data Qa, Ca, Xa, Ya,
Da is also stored in buffer memory 1801A of feature point storage section 2142 by data signal 21411 and address signal 21412.

On the other hand, the feature point type Qa is input to the control circuit 4010 by the data signal 4000, and the control circuit 4010
10 confirms that it is a minutiae point and gives a command to the shift register 401, and upon receiving this command, the shift register 401 sends the minutiae number M of the first nearest minutiae to the data signal 4011. Therefore, it is sent to control circuit 4010. Control circuit 4
010 thereby sends the address signal 2146 to the feature point list memory 211 and
Position data X, Y and direction data D are read out of the total feature point data of the primary nearest feature point M stored in 1.

The position data X, Y and direction data D read out from the memory 211 by the data signal 2144 are immediately sent to the subtracters 403X, 403Y.
and 403D, so the subtractor 403
At this time, X, 403Y and 403D are set to the X register 402 by a command from the control circuit 4010.
X, Y register 402Y and D register 402
Position data of feature point Ma input from D X _a , Y _a
and the direction data D _a , the difference ΔX _a ,
ΔY _a and ΔD _a are calculated, and subtracters 403X and 4
The output of 03Y is sent to multipliers 405X and 40, respectively.
6X and multipliers 405Y and 406X.

On the other hand, direction data D _a from D register 402D
is input to the circular function generator 404 in parallel, and the circular function value calculated by the circular function generator 404
cosD _a and sinD _a are sent by data signals 4041 and 4042 to multipliers 405X, 405Y and multipliers 406X, 406Y.

The four multipliers 405X, 405Y, 406X and 406Y are connected to the subtracters 403X, 406Y as described above.
03Y and the output from the circular function generator 404, ΔX _a cosD _a , ΔY _a cosD _a , ΔY _a
sinD _a and ΔX _a sinD _a are calculated and their outputs are sent to the adder 407X and the subtracter 407Y, so the adder 407X and the subtracter 40
7Y inputs these, calculates x _a = ΔX _a cosD _a + YX _a sinD _a y _a = ΔY _a cosD _a −ΔX _a sinD _a , and subtracts the results from the direction data d _a that is the output of 403D. At this time, the data signal 2141 uses the relation r _a output from the shift register 401 as relation concatenation data.
1 is sent to the buffer memory 1801A of the feature point storage unit 2142 and stored therein.

When the above-described relation connection operation and storage operation regarding the first nearest feature point M are completed, the control circuit 4010 sends a command signal 4002 to the shift register 401 to obtain the feature point number Mg of the second nearest feature point. , the above-described operations are repeated to execute the relation connection operation and its storage operation for all the nearest feature points.

When the above-described relation connection operation and storage operation for all the nearest minutiae points are completed, the relation connection unit 2141 continues to read out the total minutiae data of the first minutiae of the file fingerprint from the minutiae list memory 211. , performs the above-mentioned relation connection operation and stores it in buffer memory 1801B.

The minutiae storage unit 2147 stores the buffer memory 1 in response to a command signal 2149 from the control circuit 4010 during the storage operation of the relation connection data of the first minutiae of the file fingerprint described above.
The total connected data of the first minutiae of the search fingerprint stored in 801A is transferred to the search minutiae memory 1802.

When the relation connection operation regarding the first minutiae of the file fingerprint in the relation connection unit 2141 is completed and stored in the buffer memory 1801B, this total connection data is transferred to the control circuit 4010.
The data is transferred to the file feature point memory 1803 according to the command, and during this time a file connection operation is performed and the data is stored in the buffer memory 1801A.

When the total connected data of the search fingerprint and the file fingerprint are stored in the search minutiae memory 1802 and the file minutia memory 1803, respectively, a command signal 21410 is sent from the control circuit 4010 to the "pair" detection unit 2143, so that "pair" detection is performed. Section 2143
In response to this command signal 21410, search minutiae memory 1802 and file minutiae memory 140
The address signal 21431 is sent to 3 and the total concatenated data stored therein is read out and a "pair" detection operation is performed.

Search feature point memory 18502 and file feature point memory 1850 by “pair” detection unit 2143
When the reading from 3 is completed, the feature point storage unit 21
42 transfers the total connected data of the second feature point of the file fingerprint stored in the buffer memory 1801A to the file feature point memory 1803 according to a command from the control circuit 4010, and also transfers the third feature of the file fingerprint to the buffer memory 1801B. Stores the total connected data of points.

The above-mentioned relation connection operation, its storage operation, reading of the total connection data, and the resulting "pair" detection operation are performed in two buffer memories 1801.
Repeat using A and 1801B alternately,
When all the minutiae points of the file fingerprint have been examined for one minutiae point of the search fingerprint, a relation concatenation operation regarding the second minutiae of the search fingerprint is performed using the free buffer memory 1801A or 1801B. and stores it in the search feature point memory.

By performing the above-described operation on all minutiae points of the search fingerprint and the file fingerprint, the "pair" detection operation of one search fingerprint and one file fingerprint is completed.

Referring to FIG. 20, in FIG. 16, the "pair" detection unit 2143 inputs the total connected data of feature points that can be compared and judged from the search feature point memory 1802 and the file feature point memory 1803, and performs a dense amount or relay. Subtractors 701R, 701X, 70 that perform subtraction processing between the position data and the direction data
1Y, 701D and these subtractors 701R, 7
Comparators 702R and 70 input the outputs from 01X, 701Y, and 701D and compare their values.
2X, 702Y, 702D and these comparators 7
An AND circuit 706 that inputs the outputs of 02R, 702X, 702Y, and 702D, and this AND circuit 7
A counter 707 inputs the output of 06 and is reset by a command from the control circuit 700, and inputs the relations of the search fingerprint and file fingerprint, which are assigned when the nearest minutiae does not exist. A relation code detector 704 that determines whether or not it is a specific code; and a counter 70 that receives and counts the output of this relation code detector and is reset by the control circuit 700.
5, and a comparator 708 which inputs the output from the threshold generator 703 sent by inputting the output of the counter 705 and the output from the counter 707, and compares and determines the values thereof. It is composed of

“Pair” detection unit 2143 configured as described above
works as follows. That is, the control circuit 7
When 00 receives the command signal 21410 from the relation connection unit 2141, the address signal 2143
1 to the search minutiae memory 1802 and the ^file minutiae _memory ¹⁸⁵⁰³ , and sends the dense amount ( _C ^Sa _,
C ^F _a ), position data (X ^S _a , X ^F _a ; Y ^S _a , Y ^F _a ), and direction data (D ^S _a , D ^F _a ), respectively.
1X, 701Y, and 701D (the subscript S in M ^S _a , M ^Fa _, etc. indicates the search fingerprint, and the subscript F indicates the minutiae number or data regarding the file fingerprint), and subtractor 7
01R, 701X, 701Y, 701D are the absolute values of their differences |C ^S _a −C ^F _a |, |X ^S _a −X ^F _a |, |Y ^S _a −Y ^F _a
|, |D ^S _a −D ^F _a | are calculated and the values are sent to comparators 702R, 702X, 702Y, and 702D, respectively.

Comparators 702R, 702X, 702Y, 702
D is these subtractors 701R, 701X, 70
1Y, 701D and the command signal 7003 from the control circuit 700.
Threshold data sent from 03 (T _C , T _X ,
｜ ^{C S} _{a −C} ^F _a ｜T _C , ｜X ^S _a −X ^F _{a ｜} T _X ｜Y ^S _a −Y ^F _a ｜T _Y , ｜D ^S _a −D ^F _a | T _D is compared and outputted to the AND circuit 706, which outputs the result to the above comparator 702.
When the outputs from R, 702X, 702Y, and 702D are all on, the outputs are sent to the control circuit 700.

Upon receiving the output from the AND circuit 706, the control circuit 700 outputs command signals 7001 and 700.
2 to reset the counters 705 and 707, and at the same time send the address signal 21431 to retrieve the search feature point memory 1802 and file feature point memory 1803 from the first relation concatenated data (r ^S _af , x ^S _af , y ^S _af , d ^S _af and r ^F _af , x ^F _af , y ^S _af
, d ^S _af )
The subtracters 701R, 701X, 701Y, 701D and the ratio comparisons 702R, 702X, 702Y, 70 are read as in the case of the feature point data above.
By 2D, |r ^S _af −r ^F _af |Tr, |x ^S _af −x ^F _af |Tx, |y ^S _af −y ^F _af |Ty, |d ^S _af −d ^F _af |Td (Tr, Tx, Ty, and Td are relations, respectively.
(Threshold values for position data x, position data y, direction data d) are calculated and the results are sent to the AND circuit 706,
AND circuit 706 includes comparators 702R, 702X,
When the outputs from 702Y and 702D are all on, the outputs are sent to the control circuit 700. Control circuit 700 receives this output, sends command signal 702, and updates the contents of counter 707.

Prior to the above operation, the relations r ^S _af , r ^F _af
are input to the relation code detector 704, so that the relation code detector 704 determines the codes of the relations r ^S _af , r ^F _af and detects at least one of those codes. If the code is a specific code given when there is no minutiae feature point, the command signal 7040 is output to the control circuit 700 and the counter 705 is updated, and the control circuit 700 outputs the above-described AND circuit 7.
Command signal 700 regardless of the presence or absence of output from 06
2 is not output.

Relation connection data (r ^S _af ~ d ^S _af , r ^F _af ~ d ^F _af )
When the above operations related to
0 sends the address signal 21431 to the search minutiae memory 1802 and the file minutiae 1803 to read out the second relation connection data, performs the same comparison and judgment operation as described above for these data, and then continues to calculate the search fingerprint and file fingerprint. A comparison judgment operation is performed on all relation-connected data of the feature points.

When the above-described comparison and judgment operation regarding the relation connection data is completed, the control circuit 700 gives a command to the counter 705 to output the contents to the threshold value generator 703, and the threshold value generator 703 receives the information from the counter 705. Upon receiving the output, a threshold corresponding to the number of absent nearest feature points is output to the comparator 708.

A comparator 708 receives and compares the output from the threshold generator 703 and the output from the counter 707 output in response to a command from the control circuit 700, and determines that the value of the counter 707 is greater than or equal to the threshold. When the command signal 7080 is sent to the control circuit 700, the control circuit 700 receives the command signal 7080 and converts the internally held search minutiae number M ^S _a and file minutiae number M ^F _a into the pair shown in FIG. It is sent to the repair list memory 215 and stored at the address specified by the address signal 2145.

By performing the above operation for all combinations of all the minutiae points of the search fingerprint and all the minutiae points of the file fingerprint, the minutiae points that are "pairs" between the search fingerprint and the file fingerprint are searched, and a "pair candidate list" is created. The contents of memory 2151 are completed.

Note that in the above embodiment, the feature point storage unit 2142
is a device independent of the relation connection unit 2141 and the “pair” detection unit 2143, but this does not necessarily have to be independent, and the relation connection unit 2141 or the “pair” detection unit 21
It may be incorporated as a part of the components of 43.

(5) Modified example of test operation When referring to FIG. 21, the modified side 214' of the test circuit shown in FIG.
1' search feature point storage section 1802, file feature point storage section 1803, and composite pair detection section 2143'.

The composite relation connection unit 2141' converts the basic relation data of each minutiae of the fingerprint data stored in the minutiae list memory 211 shown in FIG. is converted into local coordinate position data and direction data regarding the feature point, and further, from the total feature point data regarding the first-order nearest neighbor feature point, the second-order relation data regarding the second-order nearest neighbor feature point is converted. Configure data and search for minutiae data, primary relation data, and secondary relation data (collectively referred to as total connected data) for each minutiae minutiae storage unit 1802 or file minutiae storage unit 1803 are combined to be sent to and stored.

The total connected data stored in the search minutiae storage unit 1802 or the file minutia storage unit 1803 is as follows:
As shown in Figure 22, for each feature point, feature point data and primary relation data (relation
rij, position data xij, yij, direction data dij) and secondary relation data (secondary relation rik, position data xik/yik, direction data dik).

The composite “pair” detection unit 2143′ includes the search feature point storage unit 1802 and the file feature point storage unit 1803.
It has the role of extracting one piece of total connected data from each and detecting feature points that should form a "pair", and the "pair" detected by this composite "pair" detection section
The combination of feature point numbers is stored in pair candidate list memory 2.
15 and stored.

The data stored in the feature point list memory 211 includes the feature point type (Qi), density (Ci), position data (Xi, Yi), direction data (Di), and first-order nearest neighbor for one feature point. It is a collection of one fingerprint sentence of the total minutiae data, which includes all the basic relation data (rij, Mj) of the minutiae points as one set.
These are sequentially output to the composite relation connection section 2141' by a data signal 2146.

The composite relation connection unit 2141' connects the feature point list memory 21 with the address signal 2144.
When basic relation data is read from 1, the feature point number Mj of the first-order nearest neighbor feature point included in this basic relation data is converted into position data xij, yij and direction data dij based on local coordinates regarding that feature point. constitutes the primary relation data together with the relation rij, and furthermore, the address signal 21
44, the total feature point data of the primary nearest neighbor feature point Mj is read out from the feature point list memory 211.
Position data xik, yik and directional data regarding the local coordinates of the first-order nearest neighbor feature point of the next-nearest neighbor feature point
dik and secondary relation rik are generated to form secondary relation data, and these are combined with feature point data as total connected data as data signal 21411.
The search feature storage unit 1802 sends this data signal 2141 to the search feature storage unit 1802.
1, the address signal 2142 causes the 22nd
Store it at the address specified in the format shown in the figure.

When the above operation is repeated and the storage of the total connected data of all minutiae regarding one search fingerprint is completed, the composite relation connection unit 2141' sequentially reads out the fingerprint data of the file fingerprint and relays it as in the case of the search fingerprint. After performing the connection connection operation, the total connection data is outputted to the file feature point storage unit 1803 as a data signal 21411. The file minutiae storage unit 1803 stores the total connected data at a designated address according to the designation of the address signal 21413, and repeats this operation to store the total connected data of all the minutiae points of the file fingerprint.

The fingerprint data stored in the search minutiae storage unit 1802 and the file minutia storage unit 1803 is checked by a composite “pair” detection unit 2143′, which will be described in detail later, to determine whether or not there is a “pair” relationship for all combinations of those minutiae points. is examined, and when the total connected data of the minutiae of the search fingerprint and the total connected data of the minutiae of the file fingerprint match within a predetermined threshold, the combination of those minutiae numbers is transmitted as the data signal 214.
5 to the pair candidate list memory 2151 and stored at the address specified by the address signal 2148.

FIG. 21A shows the relationship between the first nearest feature points Mf, Mg, Mh, Mk with respect to the reference feature point Ma, and the second nearest feature point with respect to the first nearest feature point. ing. For example, Ma
The nearest feature point Mf in the first quadrant in the coordinate system whose origin is Mf is the nearest feature point Ml in the first quadrant in the local coordinate system whose origin is Mf, and this feature point Ml is This is one of the second-order nearest neighbor feature points.

In addition to verification using general primary relation data, verification using secondary relation data is an extremely effective method for performing highly reliable verification. This is because it is unavoidable that some errors will occur in the center and direction of the reference coordinates or local coordinates during the setting process. In this case, for example, as shown in FIG. 21A, the feature points
Local coordinates Xa, Ya are local coordinates Xa′,
When set to Ya′, the first nearest neighbor feature point Mf,
Among Mg, Mh, and Mk, the feature point Mk is the coordinate axis Xa′,
For Ya', since it is in the same quadrant as the first-order neighboring feature point Mf, it does not become the first-order nearest neighbor feature point, but instead the feature point Mu becomes the first-order nearest feature point. Harm occurs. In this regard, it is possible to eliminate the above-mentioned damage by adding the second-order nearest neighbor feature point and comparing the first-order relation data and between the first-order relation data and the second-order relation data. Become.

The structure of the composite relay circuit section 2141' is the same as that of the relay connection section 21 shown in FIG.
The configuration is the same as that of No. 41.

This operation is performed as follows.

Referring to FIG. 19, when the composite relation connection unit 2141' receives the total feature point data of, for example, the feature point Ma sent from the feature point list memory 211 by the data signal 2144, the composite relation connection unit 2141' receives a command from the control circuit 4010. Basic relation data raf,
Stores Mf~rak, Mk and X register 40
2X, Y register 402Y and D register 40
Position data X _a , position data Y _a and direction data D _a are stored in 2D. At this time, the feature point data Qa, Ca, Xa, Ya, Da are data signals 2141
1 and address signal 21412, search feature point storage unit 802 or file feature point storage unit 1
803. (See Figure 22).

On the other hand, the feature point type Qa is input to the control circuit 4010 by the data signal 4000, and
010 confirms that it is a minutiae point and gives a command to the shift register 401, and upon receiving this command, the shift register 401 sends the minutiae address Mf of the first-order nearest neighbor minutiae as a data signal. 40
11 to the control circuit 4010. The control circuit 4010 thereby receives the address signal 214.
6 to the feature point list memory 211 to read out the total feature point data of the first-order nearest neighbor feature point Mf stored in the feature point list memory 211, and among these, the basic relation data is transferred to the shift register 40.
Store in 1.

On the other hand, among the total feature point data read out by the data signal 2144 from the feature point list memory 211, position data Xf, Yf and direction data Df
immediately subtracters 403X, 403Y and 4
Since it is input to 03D, subtracters 403X, 40
3Y and 403D are the control circuit 4010 at this time.
The differences ΔXaf, ΔYaf, ΔDaf are calculated based on the position data X _a , Y _a and the direction data Da of the feature point Ma input from the X register 402X, Y register 402Y and D register 402D according to the command from
The outputs of subtractors 403X and 403Y are sent to multipliers 405X, 406Y and multipliers 405Y, 406Y, respectively, and subtracter 403D
The output of
Sent to 03.

On the other hand, the direction data from the D register 402D
D _a is input to the circular function generator 404 in parallel,
Circular function value calculated by circular function generator 404
cosDa and sinDa are sent by data signals 4041 and 4042 to multipliers 405X, 405Y and multipliers 406X, 406Y.

The four multipliers 405X, 405Y, 406X and 406Y are connected to the subtracters 403X, 406Y as described above.
03Y and the output from the circular function generator 404, ΔXafcosDa, ΔYafcosDa,
ΔYafsinDa and ΔXafsinDa are calculated and their outputs are sent to a multiplier 407X and a subtracter 40.
7Y, so adder 407X and subtractor 407
Y inputs these, calculates xaf = ΔXafcosDa + ΔYafsinDa yaf = ΔYafcosDa − ΔXafsinDa, and sends the results to the data signal 214.
11 to the search feature point storage unit 1802 or file feature point storage unit 1803, and the search feature point storage unit 1802 or file feature point storage unit 1
803 uses address 21412 or 21419 to combine these data, direction data daf output from subtractor 403D, and relation raf sent from shift register 401 as primary relation data regarding nearest feature point Mf. Store at the specified address. (See FIG. 22) When the above-described relation connection operation regarding the first primary nearest feature point Mf is completed, the control circuit 40
10 is a command signal 4002 to the shift register 401
The feature point number Mg of the second nearest neighbor feature point is sent
, the above-described operation is repeated and executed for all first-order nearest neighbor feature points.

When the above-described relation connection operation corresponding to the first-order nearest neighbor minutiae is completed, the composite relation connection unit 2141' stores the minutiae number of the first second-order nearest neighbor minutiae already stored in the shift register 401.
The position data Xl, Yl and direction data Dl are read out by Ml, and the position data constitutes the secondary relation data using exactly the same procedure as described above.
In addition to calculating and transmitting xal, yal and direction data dal, relation information is sent from the shift register 401.
rfl is sent and input to a secondary relation calculator (not shown), and the primary relation raf is separately stored in a primary relation register (not shown).
(The calculation method is a combination of ordinary addition and subtraction, so detailed explanation will be omitted.) The result is output and the feature point Ml
Search feature point storage unit 1802 or file feature point storage unit 1803 as a secondary relation regarding
Store in. (See FIG. 22) This operation is then performed for all secondary nearest neighbor feature points to complete the relation connection operation regarding the feature point Ma.

When the above operations are completed for all minutiae points of one fingerprint, the control circuit 4010 outputs the command signal 2.
145 is output to the composite "pair" detection section 2143' to notify that all operations have been completed.

Referring to FIG. 23, the composite pair detection unit 214
3' has a pair detection section 214 shown in FIG.
The configuration is almost the same as No. 3.

The difference is that the address signal 21 from the control circuit 700
In addition to 43, an address signal 74 with a changed address designation is outputted, and in addition to the address signal 2148, a data signal 2145 is outputted.
The operation of this circuit will be explained in detail below.

The control circuit 700 is connected to the composite relation connection section 21
When command signal 21410 is received from 41', address signals 21431 and 21432 are sent to search minutiae storage unit 1802 and file minutiae storage unit 1.
803, the density amount ^{(C S a , C F a} _{) ,} ^position _{data (} ^{X S} _a ^, _{_} ^_ _{_} ^_ _{_} ^_ _{_} ^_ _{_}
Input to 701D and subtracter 701R, 701X, 7
01Y and 701D are the absolute values of their differences |C ^S _a −C ^F _a |, |X ^S _a −X ^F _a |, |Y ^S _a −Y ^F _a
|, |D ^S _a −D ^F _a | are calculated and the values are sent to comparators 702R and 702R, respectively.
Send to 02X, 702Y, 702D.

Comparators 702R, 702X, 702Y, 702
D is these subtractors 701R, 701X, 70
1Y, 701D and the command signal 7003 from the control circuit 700.
Threshold data sent from 03 (T _C , T _X ,
T _Y , T _D ), respectively |C ^S _a −C ^F _a |T _C , |X ^S _a −X ^F _a |T _X , |Y ^S _a −Y ^F _a |T _Y , |D ^S _a −D ^F _a |T _D and outputs the result to the AND circuit 706, which in turn compares and determines the comparator 702
When the outputs from R, 702X, 702Y, and 702D are all on, the outputs are sent to the control circuit 700.

Upon receiving the output from the AND circuit 706, the control circuit 700 outputs command signals 7001 and 700.
2 to reset the counters 705 and 707, and change the address signals 21431 and 21432 to store the search feature point storage unit 1802.
and the first primary relation data (r ^S _af , x ^S _af , y S _af , d ^S _af , and r ^F af , x ^F _af , y ^F _{af ,} d ^F _af ) from the file ^feature point storage unit 1803 , and subtracters 701R and 701 as in the case of the feature point data above.
X, 701Y, 701D and comparator 702R,
By 702X, 702Y, 702D |r ^S _af −r ^F _af |Tr, |x ^S _af −x ^F _af |Tx, |y ^S _af −y ^F _af |Ty, |d ^S _af −d ^F _af | Td (Tr, Tx, Ty, Td are relations,
(Threshold values related to position data x, position data y, and direction data d) are calculated and the results are sent to the AND circuit 706,
AND circuit 706 includes comparators 702R, 702X,
When the outputs from 702Y and 702D are all on, the outputs are sent to the control circuit 700. Control circuit 700 receives this output, sends command signal 702, and updates the contents of counter 707.

Prior to the above operation, the relations r ^S _af , r ^F _af
are input to the relation code detector 704, so that the relation code detector 704 determines the codes of the relations r ^S _af , r ^F _af and detects at least one of those codes. If the code is a specific code given when the nearest feature point does not exist, a command signal 7040 is output to the control circuit 700 and the counter 705 is updated, and the control circuit 700 outputs the above-described AND circuit 7.
Command signal 700 regardless of the presence or absence of output from 06
2 is not output.

First primary relation data r( ^S _af ~d ^S _af , r ^F _af
~d ^F _af ) is completed, the control circuit 700 sends the address signal 21431 and the address signal 21432 with changed address specification to the search feature point storage unit 1802 and the file feature point storage unit 1803, and Part 1802
Then, the first primary relation data (r ^S _af ,
In addition to reading x ^S _af , y ^S _af , d ^S _af , the second primary relation data (r ^F _ag , x ^F _ag , y ^F _ag , d ^F _ag ) is read from the file feature point storage unit 1803. The same comparison and judgment operation as described above is performed on these, and the combination of the primary and secondary relation data of the minutiae of the search fingerprint and the primary and secondary relation data of the minutiae of the file fingerprint is then continued. The same comparison and determination operation as described above is performed.

1 Regarding minutiae of both search and file fingerprints
The operation of comparing and determining the combinations of the next and second-order relation data may be performed for all combinations of the first-order and second-order relation data, but in order to increase the efficiency of matching, The process may also be performed for combinations of primary relation data and secondary relation data.

When the above comparison and judgment operation is completed for all predetermined combinations between the primary relation data and the secondary relation data, the control circuit 700 gives a command to the counter 705 and outputs the contents to the threshold generator 703. threshold generator 70
3 is a comparator 7 which receives the output from the counter 705 and sets a threshold value corresponding to the number of absent nearest feature points.
Output on 08.

A comparator 708 receives and compares the output from the threshold generator 703 and the output from the counter 707 output in response to a command from the control circuit 700, and if the value of the counter 707 is equal to or greater than the threshold, A command signal 7080 is sent to the control circuit 700, and upon receiving this command signal 7080, the control circuit 700 determines the combination of minutiae numbers (M ^S _a , M ^F _a ) of the search fingerprint and file fingerprint held internally as ""pair" is sent to the "pair" candidate list memory 215 and stored by data signal 2145 and address signal 2148.

By performing the above operation for all combinations of all minutiae points of the search fingerprint and all minutiae points of the file fingerprint, the minutiae points that will be a "pair" of the search fingerprint and the file fingerprint are searched and stored in the pair candidate memory 2.
15 contents are completed.

Pair Candidate List Memory An example of the pair candidate list memory created as a result of the pair testing operation described above will be described below.

Referring to FIG. 24, there are a total of 64 different row addresses and 16 different column addresses. Each entry indicated by an arbitrary row address i and column address j has a pair candidate instruction field Mij and a similarity weight storage field.
Wij is divided into two fields, and each of these fields stores data in the following format.

First, each row address corresponds to a search fingerprint minutiae having the same number. That is, the minutiae point of the file fingerprint that has the highest local approximation to the i-th search fingerprint minutiae is selected as the first pair candidate for the i-th search fingerprint minutiae by the number of this pair candidate (that is, the candidate The file fingerprint minutiae point number) is stored in field Mi0. At the same time, the strength of the degree of approximation of this pair of candidates is stored as a weight in field Wi0. Next, the number of the file fingerprint minutiae with the next strongest local approximation to the same i-th search fingerprint minutiae is set as the second pair candidate, and the field Mij
, and the strength of its approximation is also stored as a weight in Wij. In this way, candidate pairs of file fingerprint minutiae for the i-th search fingerprint minutiae are stored in successive column addresses in order of the strength of local approximation. When the strength of the local approximation becomes lower than a certain threshold value, the pair candidate list for this row address i is truncated there, and a specific end mark is stored in the weight field to indicate the cessation of processing. Ru.

(6) Generation of coordinate matching amount and matching operation Referring to FIG. 25, the coordinate matching circuit 21
One circuit example of No. 7 is the differential plane display memory 601.
(DIF), register files 602 (RF1) and 603 (RF2), adders 604, 605 and 606, 1 adders 607 and 608, registers 609 (DX), 610 (DY), 611 (RK0),
612 (RK1), 613 (RK2), 614 (MD
1F), 615 (XM) and 616 (YM), input selectors 617, 618, 619, 620 and 621, comparator 622 and AND gate 623
Contains.

The operation of this embodiment uses the contents of the feature point list memory 211 and the pair candidate list memory 215, rotates the coordinate system of the search fingerprint by γΔθ, and changes the X coordinate to ΔX _T or the Y coordinate. The purpose is to find values of γΔθ, ΔX _T , and ΔY _T that will best match the minutiae of the search fingerprint and the minutiae of the file fingerprint when they are translated by ΔY _T respectively. However, γ is a positive or negative integer, and Δθ is preset to a specific value depending on the accuracy to be obtained. (For example, Δθ=5.6°) In order to perform such a desired operation, a microprogram (microcode) is stored in a control memory memory (not shown), and the control memory 21
The control section 8 (not shown) reads this microprogram one after another from its specific starting address and executes it to proceed with the processing described below.

First, the outline of this process will be explained.

First, the control unit controls the feature point list memory 21
1 and the above information stored in the pair candidate list memory 215 and process it, a difference plane weight map is stored in the difference plane display memory 601 (DIF) included in the coordinate matching circuit 217. generate.

This difference plane weight map is as follows.

The coordinate system of the search fingerprint minutiae is rotated by a certain angle γΔθ, and the values of the X and Y coordinates of the specific i-th search fingerprint minutiae after rotation are Xsi and Ysi, respectively. Next, file fingerprint minutiae is a specific j-th pair candidate for this search fingerprint minutiae (the number designating this file fingerprint minutiae is registered in the pair candidate designation field Mij of the pair candidate list memory 215). Let the values of the X and Y coordinates of The weight registered in the weight storage field Wij of the pair candidate list memory 215 is added to the memory address of DIF). All valid i,j in the memory 215
The sum of the values is the difference plane weight map.

The difference plane weight map generated in this way is assigned to the memory address of the memory 601 (DIF) corresponding to the position on the difference plane specified by arbitrary quantized ΔX and ΔY, and the above-mentioned A weight pattern is formed by storing the weights accumulated in this way. Note that the weight pattern of this map differs depending on the above-mentioned rotation angle γΔθ of the search fingerprint coordinate system.

Next, on the weight map generated in this way, find the ΔX coordinate value X _M and the ΔY coordinate value Y _M for which the accumulated weight is maximum by searching the entire surface of this map. . However, when performing this search, as will be explained in detail later, the weights at the positions X _M and Y _M to be searched, and the weights at the adjacent positions before and after them, are each multiplied by a specific weight coefficient, and then they are A load search is performed in such a way that the total weight of the loads is maximized. Once the values X _M and Y _M of the ΔX and ΔY coordinates that give the maximum load weight are found, record the values of X _M and Y _M and the value of the maximum load weight MDIF found at that time, respectively. I'll keep it.

Next, after changing the value of γ of the rotation angle γΔθ of the search fingerprint coordinate system a little more than before, do the same as before,
A difference plane weight map is generated, and from this difference plane weight map, the values X _M and Y _M of the ΔX and ΔY coordinates at which the load weight is maximum and the maximum load weight MDIF at that time are determined as described above, and the values calculated this time are The maximum load weight MDIF of the octopus is compared with the maximum load weight MDIF remaining from the previous time to determine whether it is larger than the previous one. If it is larger than before, then X _M , Y _M
and MDIF values are updated with the respective values obtained this time.

By repeating these operations, the coordinate system of the search fingerprint can be changed using a predetermined fineness Δθ as a unit.
By swinging within a predetermined angle range, the value of the angle γΔθ at which the above-mentioned maximum load weight MDIF becomes maximum and the X _M and Y _M at that time are determined, and this is the coordinate matching amount γΔθ, ΔX _T and ΔY _T
It becomes.

Next, this will be explained in detail in relation to the operation of each circuit.

First, the weight map generation process for generating the above-mentioned difference plane weight map will be described.

The control unit of the control memory 218 in FIG. 4 converts the X, Y, and D data of a specific i-th search fingerprint minutiae in the minutiae list memory 211 to the coordinate conversion circuit 212 under the control of a microprogram stored in the memory. This is read out via the control memory 218
are stored in registers (not shown) included in the control of X _si , Y _si and D _si , respectively.

As shown in FIG. 11, this coordinate conversion circuit 212 has respective parameter registers 1110 for storing conversion parameters ΔX _A , ΔY _A , and Δθ _A.
(ΔX _A ), 1111 (ΔY _A ) and 1112
(Δθ _A ), and these registers are stored in the control memory 21 before this weight map generation process is started.
The values of ΔX _A =0, ΔY _A =0, and Δθ _A =γΔθ have already been stored by the control unit No. 8, respectively.

The function of this coordinate conversion circuit 212 is that the memory 21
The values of X, Y, and D read from 1 are displayed in a new coordinate system created when the coordinate origin is moved to the position of ΔX _A ′, ΔY _A , and each coordinate axis is rotated counterclockwise by Δθ _A. This is a circuit that converts into values X', Y' and D'. _{In other} words _{, from} X _, Y _{, and} D _, 2) It is expressed as D' = D - Δθ _A. In the present example, this circuit 212 merely rotates the coordinate system by Δθ _A =γΔθ.

Next, the control unit of the control memory 218 obtains the number of the j-th pair candidate from the pair candidate instruction field Mij having a specific row address i and a specific column address j of the pair candidate list memory 215,
The X, Y, and D values of the minutiae of the file fingerprint having this number are read directly from the minutiae list memorandum 211 without going through the coordinate conversion circuit 212. Let these values be expressed as Xmij, Ymij, and Dmij, respectively.

Next, the control section of the control memory 218 uses the arithmetic circuit of the sequence control circuit 210 to calculate the above-mentioned X _si , Y _si
and D _si and Xmij, Ymij and
From Dmij, ΔX, ΔY, and ΔD are calculated as follows: ΔX=Xmij−X _si ΔY=Ymij−X _si ΔD=Dmij−D _si . Next, the absolute values of ΔX, ΔY and ΔD thus obtained are compared with specific predetermined threshold values Tx, Ty and Td (relatively large values, for example Td=45°), Only when all of these are smaller than their respective threshold values, the weight storage field of the row address i and column address j of the pair candidate list memory 215 is
Wij is read out and added to the current contents of the differential plane display memory 601 (DIF) of the circuit 217 at the memory address specified by ΔX and ΔY.

To do this, the control section of control memory 218 sets the value of ΔX to line 6170 (ΔX) of coordinate matching circuit 217 shown in FIG.
By supplying the value of ΔY to 180 (ΔY) and setting line 6181 (CT) to “1”, input selectors 617 and 618 receive the value of ΔX supplied via lines 6170 and 6180, respectively.
and ΔY are selected and supplied as the row address and column address of the differential plane display memory 601 (DIF), respectively. As a result, the current contents of the memory addresses corresponding to the difference plane coordinate positions ΔX and ΔY of the memory 601 (DIF) are read out via line 6010.
This is supplied as one input of the adder 604,
As the other input, the value read from the weight storage field Wij is provided via line 6040 to form the sum of both, and this sum is returned to the original memory address specified by ΔX and ΔY. Store. As a result, the memory 601
It is possible to easily perform weight accumulation processing in which the contents of Wij are accumulated on the difference plane displayed by (DIF).

Now, in order to generate the above-mentioned difference plane weight map using this, first, the difference plane display memory 60
The contents of 1 (DIF) are all initialized to 0, and the values of i and j are each initialized to 0 first. However, the values of i and j are stored in a parameter register (not shown) provided in the control section in the control memory. In this way, the control unit starts the above-mentioned weight accumulation process from i=j=0, successively increments the value of j by 1, and continues until the above-mentioned end mark appears in the weight storage field, and when the end mark appears, i The value of j is increased by 1, the value of j is returned to 0, and the process continues from there.Finally, when the end mark appears in the Q field of the search fingerprint minutiae, the process ends there.

The weight map generation process described above is shown in FIG. 26 as a flowchart.

FIG. 27 shows an example of the generated difference plane weight map.

As shown in FIG. 27, the difference plane weight map of this embodiment has 16 levels from -8 to 7 as the values of ΔX and ΔY, and is relatively coarsely quantized. Therefore, for ΔX and ΔY, the upper 4 bits of the value obtained by the above calculation are used.

Next, a weight concentration position search process will be described in which the position of maximum concentration of weights is searched on the difference plane weight map thus generated and the values of the coordinates X _M and Y _M of that position are found.

This process involves finding the position of maximum concentration of weights on a difference plane weight map as shown in FIG. In order to avoid this, in this embodiment, instead of directly comparing and searching the weights stored at the individual coordinate positions of the quantized difference plane, the following weight search is performed. In other words, the 28th
A load search is performed by multiplying by a specific load coefficient as shown in the figure and searching while comparing the total.

In order to perform such a load search at high speed, this embodiment has a dedicated coordinate matching circuit 217 shown in FIG. 25.

Now, when the difference plane weight map as shown in FIG. 27 is completed in the memory 601 (DIF) and the search operation is started, the processing proceeds as follows.

First, the coordinate matching circuit 217 for search processing is initialized, and all memories and registers of the circuit 217 except the memory 601 (DIF) are set to their respective initial values.
That is, register files 602 (RF1) and 603 (RF2), registers 611 (RK0), 6
12 (RK1), 613 (RK2) and 614
(MDIF) are all cleared to 0. Further, -8 is set in registers 609 (DX) and 610 (DY) as search start positions, respectively. In this embodiment, the difference plane display memory 60
1 (DIF) can be specified as the quantized ΔX value from -8 to 7 and the ΔY value from -8 to 7, as shown in Figure 27, but the search start position is determined by this difference. Plane ΔX=-8, ΔY=-8
choose. Note that the contents of the register 609 (DX) are used to designate the address corresponding to ΔX in the difference plane display memory 601 (DIF), and the contents of the register 610 (DY) are used to designate the address corresponding to ΔY. Also, in the search process, line 61
81 (CT) is set to "0", so selectors 617 and 618 are set to register 60, respectively.
Control is performed to select the inputs on the 9 (DX) and 610 (DY) sides to specify the address of the memory 601 (DIF). Similarly, in the search process, the memory 601 (DIF) is always set to the read state under the control of the line 6181 (CT).

Now, the coordinate matching circuit 217 has control memory 218
It has two timings T0 and T1 which are controlled by control information provided via line 6110 (T0) from the control circuit of.

The timing when the line 6110 (T0) is "1" is defined as T0, and the following operation is performed at this timing.

First, selectors 619, 620, and 621 are connected to the left input, that is, memory 601, respectively.
(DIF), is controlled to select the output of file 602 (RF1) and file 602 (RF1) and file 603 (RF2), as well as registers 611 (RK0), 612 (RK1) and
The input side of 13 (RK2) is enabled.

As a result, as is clear from FIG. 25, the following operation is performed.

RK0 DIF (DX, DY) + 2RF1 (DY) + RF2
(DY) RK1 RK0 RK2 RK1 RF1 (DY) DIF (DX, DY) RF2 (DY) RF1 (DY) However, in all the operation displays above, the right side of the arrow represents each content before update, and the arrow The left side represents each updated content. (In other words, even if the symbol is the same, the contents are different on the right side and the left side.) It is assumed that these updates are performed at the same time.

For example, according to , register 612
The contents of (RK1) before the update are changed to register 6 due to the update.
Transferred to register 613 (RK2)
This is the updated content of register 611.
The contents of (RK0) before the update are changed to register 6 due to the update.
Transferred to register 612 (RK1)
This indicates that the content will be updated.

Also, DIF (DX, DY) is the register 609
(DX) and the contents of the differential plane display memory 601 (DIF) with the memory address specified by the contents of the register 610 (DY), and similarly RF1
(DY) and RF2 (DY) are register 610
(DY) represents the contents of file 602 (RF1) and file 603 (RF2), respectively, having the memory address specified by the contents. For example, the operation display shows that the updated contents of file 603 (RF2) whose address is specified by the contents of register 610 (DY) are file 602 whose address is specified by the contents of register 610 (DY).
Indicates that it will be rewritten with the content before the update of (RF1).

Furthermore, as shown in the operation display, the adder 605 is an adder that doubles the output from the selector 620 side and adds it to the output from the selector 621 side.

Next, at timing T1, line 6
110 (T0) becomes “0” and line 6230 (T1)
becomes “1”. As a result, at timing T1, the following operation is performed.

First, registers 611 (RK0), 612
The input sides of (RK1) and 613 (RK2) are disabled. As a result, the contents of these registers remain unchanged during this timing T1 period. Next, selectors 619, 620 and 6
21 are the right inputs, namely registers 611 (RK0), 612 (RK1) and 613, respectively.
Controlled to select the output from (RK2). Also, a "1" is supplied to line 6230 (T1), and AND gate 623 is enabled.

As a result, as is clear from FIG. 25, the following operation is performed. In other words, the value RK0+2RK1+RK2 appears at the output of the adder 606, and this and the value of the register 612
(MDIF) is compared with the contents before update by comparator 422, and if MDIF≦RK0+2RK1+RK2 is established, line 6231 becomes "1", and as a result, registers 614 (PDIF), 615 (XM) and 616 ( YM) input side is enabled and MDIF RK0+2RK1+RK2 XM DX YM DY processing is performed. In other words, only when the following conditions are satisfied, registers 614 (MDIF) and 615
(XM) and 616 (YM) are listed above.
Updated as indicated by the behavior display and otherwise unaffected.

Now, continuing from the above, the control section of the control memory 218 sends timing information to the line 6070 (T1') and registers 610 (DY) and 609 (DX).
Update. Updating of these registers is performed as follows.

In general, the timing information on line 6070 (T1') is used to update the contents of register 610 (DY) so that 1 is added to the contents before update.

However, if the content of the register 610 (DY) before updating is the upper limit of 7, the content of the register 610 (DY) is updated to -8 based on this timing information, and at the same time, the carry from the circuit 607 is transferred to the circuit 608. added to. As a result, the register 609 (DX) is further updated so that 1 is added to the contents before the update. However, circuit 607
In the general case where no carry occurs, the contents of register 609 (DX) are kept unchanged by updating.

Also, if the carry from circuit 607 is applied to circuit 608 when the content of register 609 (DX) before updating is the upper limit of 7, circuit 608 generates a carry, which is controlled via line 6080. The information is supplied to the control unit of the storage 218 and used as information for terminating the search process.

Now, registers 610 (DY) and 6 according to the timing information on line 6070 (T1') mentioned above.
When the update of 09 (DX) is completed, the T1 timing ends, and the control circuit of the control memory 218 returns to the above-mentioned T0 timing control again, and uses the updated contents of each register and memory to perform the above-mentioned ~
Perform the following actions.

In this way, the control unit controls the coordinate matching circuit 217.
The search process is advanced by sequentially replacing the information on the T0 timing and the T1 timing. Then, upon receiving the search end information from the aforementioned line 6080, the alternation is stopped there, and the registers 615 (XM) and 616 (YM) at that time are
and reads the contents of 614 (MDIF). This completes the search process based on the load search. Note that FIG. 29 shows this weight concentration position search process in the form of a flow chart.

Through the above operations, the coordinate matching circuit 217
The difference plane weight map shown in Fig. 7 is searched for the weight using the load coefficient shown in Fig. 28, and the weight
The values XM and YM of the difference plane coordinates of the point where the MDIF is maximum and the value of this MDIF are determined, and the reason for this is as follows.

As is clear from the above description, the search scan starts from the lower left corner of the difference plane shown in FIG. 27 and scans vertically upward (in the direction in which ΔY increases),
When the upper limit is reached, ΔX is increased by 1 and the column is moved to the right column by one column, which is similarly scanned from the bottom to the top. Repeat this to cover the entire surface.

Now, such a scan, for example, ΔX=3,
Consider the state where the process has progressed to the point ΔY=5. At this time, registers 608 (DX) and 610
(DY), files 602 (RF1) and 603
The contents of (RF2) are as shown below.

DX=3, DY=5, RF1(5)=DIF(2,5),
RF2(5)=DIF(1,5).

In other words, the contents of file 602 (RF1) are the ΔX file one column before the ΔX=3 column currently being scanned.
It stores the contents of memory 601 (DIF) corresponding to column 2, and also stores the contents of file 603 (RF2).
The contents are the memory 601 (DIF) corresponding to the column with ΔX=1 two positions before the column currently being scanned.
It stores the contents of. As a result, the above-mentioned operation at timing T0 at this time is RK0
DIF(3,5)+2DIF(2,5)+DIF(1,5).

On the other hand, RK1 contains the value of RK0 at the previous scanning point (when DY = 4), and RK2 contains the value of RK0 at the two previous scanning points (when DY = 3) by updating. As is clear from the above formula, the updated RK1 and RK2 values are RK1=DIF(3,4)+2DIF(2,4)+DIF
(1,4) RK2=DIF(3,3)+2DIF(2,3)+DIF
(1, 3) becomes.

Therefore, the content of (RK0 + 2RK1 + RK2) of the output of adder 706 at timing T1 is DIF (3, 5) + 2 DIF (2, 5) + DIF (1, 5) +
2DIF (3, 4) 4DIF (2, 4) 2DIF (1, 4) +
DIF (3, 3) + 2 DIF (2, 3) + DIF (1, 3). From this, it is clear that in the above processing, the difference plane weight map is multiplied by the weight coefficient shown in FIG. 28 and integrated, and the result is set as an MDIF candidate, and the search is performed to find the maximum one.

However, the center position of the load is not the position specified by the current DX and DY, but is 1
(in the example above, DIF(3,5)
(instead of DIF (2, 4)), the maximum weight concentration position is XM and YM obtained by the above processing.
The value is obtained by subtracting 1 from each value.

Now, in this embodiment, the coordinate system of the search fingerprint minutiae is rotated using a specific minute angle Δθ as a unit,
For each γΔθ within a specific angular range, the above-described difference plane weight map generation process and the weight concentration position search process for this are executed, and the above-described weight MDIF
The value of γΔθ that makes the largest and the value of γΔθ at that time
The optimum amount of coordinate matching is determined by finding the values of XM-1 and YM-1.

The overall processing for this proceeds as follows. The control section of the control memory 218 includes a working memory area 216, and this area is provided with registers (not shown) for storing the following parameters.

That is, Δθ which is the unit of swing, multiple γ for specifying the swing angle, maximum value γmax of γ, rotation angle θ _M , read from the circuit 217.
MDIF (the value stored in this register is MDIF'), XM-1 (the value stored in this register is
X _M ), YM-1 (the value stored in this register is Y _M ).

Now, when the process of determining the amount of coordinate matching is started, the control unit of the control memory 218 sets the predetermined Δθ and γmax in each of the corresponding registers mentioned above, and initializes MDIF′ and γ to 0. do.

Next, a product γΔθ is created using the arithmetic circuit of the sequence control circuit 210 and stored in the parameter register 1112 (γθ _A ) of the coordinate conversion circuit 212.

After the above steps are completed, the control unit of the control memory 218 executes the above-mentioned difference plane weight map generation processing and weight concentration position search processing, and uses the read MDIF and the above-mentioned
Compare with MDIF′. If MDIF is larger than MDIF′, MDIF′ is the MDIF that read this
and replace X _M with XM−1 and Y _M with YM
−1 and also replace θ _M with γΔθ.

Next, by changing γ and repeating the above process, |γ
The above steps are performed for all γ values of |≦γmax, and the process ends.

This process is shown in FIG. 30 as a flowchart.

The values of θ _M , X _M and Y _M when the above-described processing is completed are the desired coordinate matching amounts γΔθ, ΔX _T and ΔY _T.

When the final result MDIF value is below a certain threshold level, it is determined that the coordinate matching is not successful. In other words, the processing is interrupted as there is no similarity between the two target fingerprints. When the coordinate matching amount is below a certain threshold level, processing is taken over to a secondary matching processor for more precise matching.

Information such as a feature point list, a region pattern list, identification information, and a pair candidate list is sent from the primary verification processor to the secondary verification processor via a common bus. In this case, the secondary feature point memory 511 stores a list of feature points whose coordinates have been transformed using the coordinate matching amount. Further, the area pattern list is stored in the area pattern list memory 516, and the pair candidate list is stored in the pair candidate list memory.

As described above, when this embodiment is used, the coordinates of each feature point of two pattern patterns to be matched,
Given a pair candidate list containing weights of pair candidates based on the local proximity of each feature point, and when the coordinate systems of both are not necessarily consistent, a special difference plane weight is calculated. By using a unique method of creating a map and performing a load search on this map for the location of concentration of weights, it is possible to align both coordinate systems by simply testing the rotation of the coordinate systems and without attempting translation. The amount of coordinate alignment required for this purpose can be determined quickly and with high reliability.

In this embodiment, the case of fingerprint matching has been described in detail, but the present invention is not limited to fingerprints, and can be applied to pattern pattern matching having a plurality of specific minutiae.

Furthermore, in this embodiment, specific configurations and capacities are used for the feature point memory, pair list memory, and difference plane display memory, but this is merely an example and is not intended to be limiting. do not have.

Further, although a specific configuration was used for the load coefficients in order to perform a load search for the location where the weights are concentrated, this is also an example and is not limited to this. If the configuration of this load coefficient changes, the configuration of the coordinate matching circuit will also change accordingly, but this can be easily realized by applying this embodiment.

(7) Precise verification operation by secondary verification processor Next, referring to FIG. 5, the control circuit 510
After resetting the coordinate transformation circuit 512 to the non-conversion state, all of the pair candidate list memory 513 are
N _s : Read M _si again and read its minutiae data X,
Y, D are set to a stricter threshold T _x than when generating the paired candidate list,
Inspect at T _y and Td. This is because coordinate matching between the search fingerprint and the file fingerprint has been completed, so the minutiae should already have the same configuration except for the distortion of the fingerprint imprint. The candidate pair N _s , M _si that is denied in this pairwise test is deleted from the pair candidate list together with its candidate pair value W _si . This is the process of selecting candidate pairs.

Furthermore, the following modification of candidate pair values is performed for all candidate pairs N _s :M _f of the carefully selected pair candidate list. That is, referring to FIG. 31, for the two candidate pairs N _s :M _f , N _s and M _f are respectively supplied to the address line 5110, and the second feature point memory 51
1 to each nearest feature point (Nsr, Mfr: r=0
~3) and store them in the pair candidate list memory 513.
Check whether Mfr is registered in the N _s line.
If M _f ′=M _fr is registered, the candidate pair value W _b is added to the candidate pair value W _a of N _s and M _f that is the basis. When the modification of the candidate pair value is completed, the pair candidate list memory 51 is stored according to the size of the new candidate pair value.
The contents of 3 are sorted every N _s rows. With the above steps, the final candidate list is completed. Pair lists 515a and 515b are generated in the pair list memory 515 based on this pair candidate list.

The two pair lists 515a and 515b in the pair list memory 515 have similar structures, and the third
As shown in Figure 2, N _s (s=1 to S) and M _f
(f=1 to F), addresses are designated via the address line 5150, and these are two sets of memories that can hold feature point numbers and paired values M _f , V _s and N _s , V _f .

The pair lists 515a and 515b are generated by initially initializing with a negative constant as a pair value, and then scanning the pair candidate list multiple times in the order of N _s (s=1 to S) to generate the highest pressure end i= 1, that is, the maximum candidate pair value N _s :M _s1 is selected from the pair list 515a, 515b in descending order of reliability based on the difference between the candidate pair value W _s1 itself and the second candidate pair value W _s2 . Transfer the candidate pair value to as a pair value. The candidate pairs registered in the pair list are deleted from the entire pair candidate list. The contents of all pair candidate list memories 513 are stored in the pair list memory 5.
15, the pair value of the pair N _s :M _f is υ _s =
υ _f = N _s : M _f ). Further, the negative initial value of the pair value of the feature point in the pair list that has not been transferred by the candidate pair value remains as it is.

Next, the pair value is determined by two processes: modifying the pair value and relaxing the non-pair value.

The pair lists 515a and 515b are sequentially checked and if the pair value υ is positive, the pair value is modified. That is, as shown in FIG. 33, for example, when the pair value υ _s of N _s in the pair list 515a is positive, the paired minutiae
Based on M _f , the secondary feature point list memory 511 is accessed in the same manner as the modification of candidate pair values, and the nearest neighboring feature points (N _sr , M _sr : r=0 to 3) are read out.
The pair of N _sr in the pair list 515a is M _sr and the pair value υ _sr
is positive. If positive, the pair value υ _sr is added to the pair value υ _s . Similar processing is performed on the pair list 515b side independently of the pair list 515a side. On the other hand, when a positive pair value is not stored in the pair list as shown in FIG. 34, for example, the pair list 515
When a certain M _f ′ in b has a negative pair value υ _f ′, supply this M _f to the address line 5110,
The feature point data of M _f ′ is read out via the line 5120, and its (X _f ′, Y _f ′) is added to the area pattern list 5.
16a on line 5160. As a result, if the area value read from the output 5161 is “0”, the above feature point M _f ′ is within the unknown area on the search fingerprint side, and the pair value υ _f ′ is don′t
It is rewritten to the value "0" which means care. On the other hand, if the area value is "1", the pair value υ _f ' remains unchanged. On the other hand, when performing the above processing using the pair list 515a, the coordinate transformation circuit 512 inputs the amount of inverse transformation (ε=△X ^* , η=
-ΔY ^* , θ=-ΔD ^* ), and check the contents of the area pattern list memory 516.

Feature points where the pairwise value υ _f ′ or υ _s ′ is kept as a negative value
Regarding M _f ′ or N _s ′, the following test is performed.
For example, for M _f ′, the nearest feature point (M _fr ′; r
= 0 to 3), the number of inner ridges R _f ' ₁ = 0, the arrangement of which is close to each other, and the direction of which is directly opposite is searched for by the facing inspection process in the arithmetic line 519. If there is, the feature point M _f ′ _r is added to the pair list 515
If the pair value υ _f ′ _r is a negative value when checked with b, both the pair value υ _f ′ _r and the above M _f ′ are set as don't care and changed to the pair value "0". This is as shown in Figure 35.
This test examines opposing adjacent minutiae points, and these two sets of minutiae points are unstable minutiae points that may or may not be drawn depending on the condition of fingerprint imprinting or mechanical automatic feature point drawing.

FIG. 36 is a block diagram showing one embodiment of the counter inspection circuit. That is, the difference absolute value unit 163x, y,
d, complement value 163c, comparator 164x, y, d, direction ROM 165, absolute difference value unit 166a, b, comparator 167a, b, AND gate 168 and threshold value
It consists of ROM169.

Two sets of minutiae data (X, Y, D) and (X', Y', D') to be inspected "opposed" are supplied to input signals 161x, y, d and 161x', y', d, respectively. Then, in the complement value 163c, only the direction D' is inverted by the direction complement, that is, by π radians, and then the absolute value of the difference is calculated by the difference absolute value unit 163x, y, d, and the threshold value which is the output of the threshold value ROM 169 is calculated. and are compared by comparators 164x, y, and d. That is, |X−X′|≦T _x , |Y−Y′|≦T _y , |D−D′+
π|≦Td is tested and its output is input to AND gate 168. Here, in the direction component, π radian is connected so as to be opposite to the most significant bit (MSB) of the difference absolute value unit 163d, so that correct calculation with respect to periodicity in the direction calculation is guaranteed. On the other hand, the difference absolute value output and the difference sign (sign at the time of subtraction) of the difference absolute value units 163x, y are input as an address to the direction ROM 165, and the specified direction △D is outputted and supplied to the difference absolute value units 166a, b. do. In other words, △D=T ^-1 _ar (Y-Y')/(X-X') From the condition that △D is close, input all combinations of relatively small X-X' and Y-Y'. It has been made into a ROM. The absolute value of the difference between one direction D-D' and the above △D is calculated by the difference absolute value unit 166a, b, and the threshold value ROM1 is calculated by the comparator 167a, b.
compared with another predetermined value from 69;
The result is provided to AND gate 168.
AND gate 168 returns a "counter" test positive signal to sequence control circuit 510 via output signal 162 when all inputs are asserted.

In the above-mentioned difference absolute value calculators 166a and 166b, the difference of π radian between the directions D and △D and between D' and △D corresponds to 0 radian, so the most significant bit (MSB) of the subtraction is π/2. It is wired so that

In the above-mentioned difference absolute value calculators 166a and 166b, the difference of π radian between the directions D and △D and between D' and △D corresponds to 0 radian, so the most significant bit (MSB) of the subtraction is π/2. It is wired so that

An example of the opposing inspection circuit has been described above. Note that this opposing inspection circuit can be replaced by a general-purpose processor. When checking a pair list that has a negative pair value, if it is not the above-mentioned "opposite" minutiae, the density amount C of that minutiae point is finally checked, and if it is found to be very large by value comparison. is again reset to a value close to the relative value "0" or negative "0". Summarizing the above,
As schematically shown in FIG. 37, relaxation of unpaired values is achieved by unpaired minutiae existing in uncommon areas of area patterns 516a and 516b after coordinate matching between search fingerprint and file fingerprint, and in common areas. This is intended to reduce the unpaired values of "opposing" feature points and feature points with a large amount of density shown by dotted lines to don't care or negative paired values close to it.

When the pair values of the pair list are completed in the above manner, the control unit circuit 510 sequentially reads the pair list 515 and sets it as the matching value q. is calculated from the arithmetic circuit 519 and the work area 514. Here, S and F indicate the number of feature points in the common area of the area pattern 516 of the search fingerprint and file fingerprint. These operations can be performed using general arithmetic operation circuits, so the details will be omitted.

To simplify the explanation, we will explain the correspondence test of (N _sr , M _fr ; r=0 to 3) corresponding to the nearest feature points in the candidate pair value modification of the pair candidate list and the pair value modification of the pair list. However, in general, due to the distortion of the configuration of the fingerprint minutiae,
It is not possible to expect a correspondence between N _sr and M _fr , but N _sr
Since _there are cases where M _fr ′, which is a different r _{′ ,} corresponds to ) and (M _fr ′; r=0 to 3) 16 times in a round-robin manner. The correspondence check in this case is based on local coordinate systems based on N _s and M _f, respectively.

This is done by These can be performed by the coordinate conversion circuit 512 and the pair inspection process in the same way as the process when generating the pair candidate list.

When the verification value 8 finally calculated by the secondary verification processor is above a certain threshold level, when the file fingerprint identification number and the verification value are below a certain threshold level, it is determined that there is no similarity; The verification to be registered in the candidate list is completed.

(8) Transfer destination processor determination operation One embodiment of the present invention is based on the configuration shown in FIG. 2B, but by adopting the device configuration shown in FIG. 2C, the matching performance (one-to-one matching time )
It has the effect of improving.

The finger verification device shown in FIG. 2C includes a control unit 10, m primary verification processors (m
1), secondary verification processor n units (n1) and control unit 10, primary verification processor and 2
It consists of a common bus 70 for transmitting data between the next collation processors. In the embodiment, m=3, i.e., primary collation processors 20, 30, 40, and n=2, i.e., secondary collation files 50, 6.
The case of 0 is shown. In such an apparatus, since each processor performs one-to-one comparison between a common search fingerprint and each different file fingerprint, parallel processing is performed between the processors and high-speed comparison is possible.

FIG. 3B is a diagram showing a detailed configuration of the transfer destination determining circuit 304 in the control unit 10 in the matching device of FIG. 2C.

When the verification device 5 receives a verification instruction from the verification control device 4, the control unit 10 outputs a data transfer request 11 to the internal first decision circuit 100 as shown in FIG. 3B. The first decision circuit 100 inputs the busy signals 21, 31, and 41 indicating that the primary verification processors 20, 30, and 40 are performing fingerprint verification, performs the logical operation expressed by the following equation, and outputs the signals 22, 32. and 42 are output to perform the first determination of the primary verification processor to which the file fingerprint data is to be transferred.

Signal 22 = (Data transfer request 11) AND (Busy signal 21) Signal 32 = (Data transfer request 11) AND (Busy signal 21) AND (Busy signal 31) Signal 42 = (Data transfer request 11) AND (Busy signal 21 ) AND (busy signal 31) AND
(Busy signal 41) The second decision circuit 200
each data transfer request 25, 35 and 45 from 0, 30 and 40 to the secondary collation processor;
The next verification processors 50 and 60 input the busy signals 51 and 61 indicating that processing is in progress, perform the logical operation expressed by the following equation, and input the signals 26, 36, 46, 52, and 62 and the inhibition signal
1 to determine the secondary verification processor to which the file fingerprint data after the primary verification is to be transferred, and to suppress data transfer to the primary verification processor when data is transferred to the secondary verification processor. and signal generation.

Signal 26 = (Data transfer request 25) Signal 36 = (Data transfer request 25) AND (Data transfer request 35) Signal 46 = (Data transfer request 25 (AND (Data transfer request 35) AND (Data transfer request
45) Signal 52 = [(Data transfer request 25) OR (Data transfer request 35) OR (Data transfer request
45)] AND (Busy signal 51) Signal 62 = [(Data transfer request 25) OR (Data transfer request 35) OR (Data transfer request
45)] AND (Busy signal 61) Inhibition signal 71 = [(Data transfer request 25) OR (Data transfer request 35) OR (Data transfer request 45)] AND (Busy signal 51) AND
(Busy signal 61) OR [(Data transfer request 25) AND (Data transfer request 35)
AND (data transfer request 45)] The third decision circuit 300 outputs a signal 2 which is the AND of the signals 22, 32 and 42 and the inhibition signal 71.
3, 33, and 43, signals 26, 36, 46, 52, and 62 determined as described above, and busy signals 51 and 61, the logical operation expressed by the following equation is performed to transfer the file fingerprint data. 1 to permit the use of the common bus 70 for
The next verification processor or one of the secondary verification processors is determined.

Bus usage decision signal 24 (signal 23) OR (signal 26) to primary verification processor 20 Bus usage determination signal 34 = (signal 33) OR (signal 36) to primary verification processor 30 Bus use decision signal 44 = (signal 43) OR (signal 46) Bus use decision signals 53, 21, 31, 41, 51 and 61 to the secondary collation processor 50 are variables, and each processor uses different files. This suggests that parallel fingerprint matching can be performed on fingerprint data.

As is clear from the description of the operation of the embodiment of the present invention, a commercially available microcomputer is used to control the counter-inspection circuit, counter-inspection circuit, and control unit by the processing device, and to control the area pattern and the control unit by the memory. Equivalent outputs can be obtained by assigning a feature point list, a difference plane, a pair candidate list, and a pair list.

[Brief explanation of the drawing]

FIG. 1 is a diagram explaining minutiae on a fingerprint pattern, FIG. 2A is a diagram showing an example of a fingerprint verification system, and FIG. 2B is a fingerprint verification device which is an embodiment of the present invention shown in FIG. 2A. Figure 2C showing the configuration of
The figure shows the configuration of a further expanded fingerprint matching device, FIG. 3 shows the configuration of the control unit in the device shown in FIGS. 2B and 2C, and FIG. FIG. 5 is a diagram showing an embodiment of the next verification processor.
6 is a flowchart showing the fingerprint matching method of the present invention; FIG. 7 is a diagram showing minutiae data for one finger used in fingerprint matching; FIG. 8A 8B is a diagram showing an example of a minutiae list used in fingerprint verification.
9 is a diagram showing a coordinate system for explaining a general coordinate transformation operation. FIG. 10 is a flowchart for explaining a general coordinate transformation operation. FIG. 11 is a diagram showing the configuration of a coordinate conversion circuit according to an embodiment of the present invention used in the primary verification processor and the secondary verification processor;
FIG. 12 is a diagram showing the configuration of a nearest neighbor feature point restoration circuit showing an embodiment of the present invention, FIG. 13 is a diagram showing a feature point list after nearest neighbor feature point restoration, and FIG. 14 is a diagram showing the nearest neighbor feature point , FIG. 15 is a flowchart showing the nearest neighbor feature point restoration operation, FIG. 16 is a diagram showing a paired test circuit which is an embodiment of the present invention, and FIG. 17 is a flow chart showing the feature point used in the paired test circuit. A diagram showing an example of the list, FIG. 18 is a diagram showing a feature point memory section in the paired test circuit, FIG. 19 is a diagram showing a relation connection part in the paired test circuit, and FIG. 20 is a diagram showing the paired test circuit. FIG. 21A is a simplified detailed enlarged view of a fingerprint pattern for explaining secondary relation data used in the present invention, and FIG. 21B is an embodiment of the present invention. Figure 22 is a diagram showing a pair test circuit using the following relation data.
FIG. 23 is a diagram showing an example of the feature point list used in the test circuit shown in FIG. B; FIG.
FIG. 24 is a diagram showing an embodiment of the pair candidate list memory, FIG. 25 is a diagram showing an embodiment of the coordinate matching amount determining circuit of the present invention, FIG. 26, FIG. 29, and FIG.
The figure is a flowchart explaining the coordinate matching amount determining operation, FIG. 27 is a diagram showing an example of a weight map generated by the coordinate matching amount determining circuit, and FIG. 28 is a diagram showing the weight coefficients used in the coordinate matching amount determining operation. , FIG. 31 is a diagram for explaining the emphasis processing of paired candidates, and FIG.
33 is a diagram showing the relationship between the pair list and paired feature points, FIG. 34 is a diagram showing an example of the contents of the pair list, and FIG. 35 is a diagram showing an example of the pair list memory used in the present invention. 36 is a diagram showing an implementation of the opposing inspection circuit; FIG. 37 is a diagram showing the relationship between search and file fingerprints and area patterns; FIG. 38 is a block diagram showing an embodiment of a transfer destination determining circuit within the control unit.

Claims (1)

[Scope of Claims] 1. In a fingerprint matching device that determines the identity of a search fingerprint and a file fingerprint, the positions and angles of minutiae of the search and file fingerprints based on an orthogonal coordinate system with an appropriately determined point as the origin; and a minutiae list storage means for retaining relation data; and coordinate transformation into a local coordinate system having each minutiae as the origin, and storing the position, angle, and relation data between each minutiae of the two fingerprints. Pair candidate list storage means for holding a pair candidate list by checking matching; and optimal coordinate matching amount generation for generating an optimal coordinate matching amount for the two fingerprints by the feature point list storage means and the pair candidate list storage means. means, and checking the matching of matched position, angle, and relation data between each minutiae of the two fingerprints under a coordinate system matched by the optimal coordinate matching amount, thereby determining the pair candidate list. a pair list storage means for determining and retaining true paired minutiae and paired values; and candidate fingerprint storage means for determining and retaining matching values of the two fingerprints from the paired minutiae and paired values. A fingerprint verification device featuring: 2. In a fingerprint matching device that determines the identity of a search fingerprint and a file fingerprint, in a local coordinate system with the fingerprint minutiae as the origin, the position of the first neighboring minutiae closest to the origin;
a first storage means for storing first-order relation data including angles and relation data; and a second-order nearest neighbor closest to the origin in a local coordinate system with the first-order nearest feature point as the origin. and a second storage means for storing secondary relation data including positions, angles, and relation data of feature points.
Fingerprint verification device as described in section. 3. A fingerprint matching device that matches a search fingerprint with a file fingerprint, comprising: a first feature point list storage means for storing the position, angle, and relation data of each feature point in a reference coordinate system; the pair candidate list storage means; a second feature point list storage means for storing position, angle, and relation data of feature points determined by coordinate transformation of paired candidate feature points in another coordinate system by the coordinate transformation means; , optimal coordinate matching amount generation means for generating an optimal coordinate matching amount from the first feature point list storage means and pair candidate list storage means, the first feature point list storage means, pair candidate list storage means, and a coordinate conversion means connected to the optimum coordinate matching amount generation means; and a coordinate conversion means connected to the second minutiae list storage means, the paired candidate list storage means, and the fingerprint area storage means, and the coordinate conversion means is connected to the second minutiae list of the two fingerprints. and the pair list storage means for generating a pair list generated by checking the consistency of the position, angle, and relation data of each coordinate-aligned isomorphic point. The fingerprint verification device according to scope 1.