JPS6321233B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a device for verifying the identity of patterns composed of striped patterns such as fingerprints based on the characteristics of the patterns.

Conventionally, as features for matching striped patterns such as fingerprints, as shown in FIG. M _f (f=1, 2,...,
F) on the fingerprint screen 0 (X _f ,
Y _f ), direction D _f , relation (M _fr , R _fr ; r=1
~4), and the use of the crowding amount C _f was proposed.
A device for extracting these features is described in Japanese Patent Application No. 54-39648 (Japanese Unexamined Patent Publication No. 55-138174), and a matching device is described in Japanese Patent Application No. 54-98966 (Japanese Unexamined Patent Publication No. 56-1989).
- Publication No. 24675). Here, relation means that for each feature point M _f , each of the four quadrants (r = 1 to
For each nearest feature point M _fr in 4 ₎ , the number of ridge intersections on the straight line connected to the origin M _f is defined as R _fr , and the amount of density C _f is defined as This is a quantification of the density of neighboring feature points existing within a certain circular area. However, the said patent application
The matching device disclosed in the specification of No. 54-98966 does not strictly evaluate the correspondence of individual minutiae in the two fingerprints to be matched, so if the number of search files becomes large, the identification ability may not be sufficient. The big problem was that it wasn't.

An object of the present invention is to provide an apparatus that enables higher-performance fingerprint verification by strictly evaluating the corresponding minutiae pairs of two fingerprints to be verified.

Next, the present invention will be explained in detail with reference to the drawings. Before explaining the device, we will define the data format of the fingerprint to be verified. When comparing two fingerprints, one will be referred to as a search fingerprint and the other as a file fingerprint for ease of understanding. Normal fingerprint verification is performed as a process of searching for a file fingerprint that matches a search fingerprint from among a large number of file fingerprints.

In the following explanation, as a basic process of fingerprint matching, a comparison of one search fingerprint in question with one file fingerprint, which is sequentially selected from a file, will be explained. The feature of the search fingerprint is the area pattern
It consists of R _N and a minutiae list (N _s ; s=1 to S), and the characteristics of the file fingerprint consist of an area pattern P _M and a minutiae list (M _f ; f=1 to F). Since the data formats of both fingerprint features are the same, only the file fingerprint features are shown in FIGS. 2a and 2b.

The contents of the feature point list b in FIG. 2 have already been explained in correspondence with FIG. On the other hand, the area pattern P _M shown in FIG. 1A is a binary pattern indicating the pattern stamping area shown in FIG.
Then set the area pattern elements outside the pattern stamping area to “0”.
It is expressed as The area pattern extraction method is described in, for example, Japanese Patent Application No. 51-152377 (Japanese Unexamined Patent Publication No. 53-75824).
This can be achieved with the device shown in . Here, it should be noted that the feature point M _f is extracted only in the region indicated by "1" in the region pattern P _M , and not in the region indicated by "0".

Next, the configuration of an embodiment of the fingerprint matching device according to the present invention will be explained with reference to the block diagram shown in FIG. The fingerprint matching device according to the present invention has an area pattern storage 11N for storing characteristics of a search fingerprint for the control unit 10.
and minutiae list storage 12N, area pattern storage 11M for storing characteristics of file fingerprints, minutiae list storage 12M, paired inspection circuit 15, and counter-inspection circuit 1.
6, consisting of a deviation plane 17, a pair candidate list memory 18, and pair list memories 19N and 19M that hold paired minutiae of search fingerprints and file fingerprints, and an address signal line 110 for each of the above-mentioned memories and deviation planes.
N, 110M, 120N, 120M, 170, 1
80, 190N, 190M, write signal line 111
N, 111M, 121N, 121M, 171, 1
81, 191N, 191M, read signal line 112
N, 112M, 122N, 122M, 132N,
132M, 172, 182, 192N, 192M
And, for each of the above test circuits, the input signal line 1
51, 161, and output signal lines 152, 162, which are coupled to the control unit 10. Also provided are an input signal line 101 for inputting characteristic data of search fingerprints and file fingerprints as interface signals from the entire verification system, and an output signal line 102 for outputting verification results. Regarding the file fingerprint minutiae list storage 12M, the second
Referring to figure b, for the feature point M _f , X _f , Y _f , D _f ,
It retains the information of C _f , M _fr , R _fr (r=1 to 4), but
The coordinate conversion circuit 13M converts the coordinate-related data X _f , Y _f , D _f into the output signal line 131
It is provided in M. The signal line for setting the conversion amount of this coordinate conversion circuit 13M is 130M, and the output signal line is 132M. The minutiae list storage 12N for the search fingerprint has exactly the same structure.

FIG. 4 is a flowchart illustrating an overview of the operation of the matching device shown in FIG. 3. First, when the area pattern and feature point list, which are the characteristics of the search fingerprint and file fingerprint to be matched, are input into this matching device, all the feature points registered in each feature point list are brute force searched for their positions, The direction and relation are checked, and potentially corresponding feature point pairs are stored in a candidate pair list. Next, the amount of coordinate matching where the corresponding feature points most closely match is determined according to this candidate pair, and the coordinates of the feature point list of the file fingerprint are matched according to this amount of matching. Next, the feature point list after coordinate matching is inspected in terms of its arrangement, and candidate pairs are carefully selected. Furthermore, based on the carefully selected candidate pairs, correspondence by relation is checked, and the candidate pair value given to each candidate pair is corrected. After generating a pair list for determining the final pair minutiae from the candidate pair list obtained in this way, the pair values are corrected again by relation, and the area pattern and opposing minutiae are inspected.
A convergence check is performed to determine the final value of the pairwise values. The pair values are totaled and the matching value normalized by the matching feature points is calculated and output as the final output of the matching device. In addition, if an appropriate matching amount is not found during coordinate matching, matching is performed by omitting the matching process below the coordinate matching process and outputting a matching value that represents a convenient mismatch that has been set in advance as a matching value. Perform speed improvements. From now on, in the order of the above operation summary,
The configuration and operation of each part will be explained in detail.

Referring to FIG. 3, the area pattern P _N of the search fingerprint to be matched and the minutiae list (N _s ; s=1 to S)
is input via the input signal 101 (signals and signal lines will be used together hereafter as long as they correspond),
The control unit 10 generates an address signal 110N in the area pattern memory 11N, and writes and holds the area pattern P _N via the write signal 111N. Also, the feature point list is written and held via the write signal 121N by generating an address signal 120N in the feature point list storage 12N.

Next, for the area pattern P _M of the file fingerprint and the feature point list (M _f ; f=1 to F), address signals 110M and 120M are generated in the area pattern memory 11M and feature point list memory 12M, and a write signal 111M is generated. , 121M. As shown in FIG. 2a, the area pattern memories 11M and 11N are ordinary memories that hold two-dimensional grid-like binary data, and their address signals 11
0M and 110N are for selectively addressing grid elements, that is, area pattern elements. The feature point list memories 12M and 12N hold the feature point data group shown in FIG. 2b, and the address signal 1
20M and 120N are normal memories that specify feature points M _f and N _s as units. These memories are already known and do not require any particular explanation. When the search fingerprint and file fingerprint to be compared are held in this way, the actual matching operation is started.

The first step is the generation of a candidate pair list. That is, for each minutiae point N _s (s=1 to S) of the search fingerprint, the control unit 10 converts each minutiae point M _f (f=1 to F) of the file fingerprint into the address signals 120N, 120 in a brute force manner.
0M and read signals 132N, 122
Read via N, 132M, 122M. Among the feature point data, the position (X _s , Y _s ) and direction D _s , the position (X _f , Y _f ) and direction D _f are determined by the coordinate conversion circuit 13N,
13M from the signals 132N and 132M, and the density amount C _s , C _f and the relation N _sr ,
R _sr (r=1 to 4), and M _fr and R _fr (r=1 to 4) are read from the signals 122N and 122M. In this case, the amount of coordinate transformation is set in advance as the non-transformed amount (ξ=0, η=0, θ=0) via the signals 130N and 130M, and the coordinate transformation is performed in the coordinate transformation circuits 13N and 13M. Make sure there are no.

FIG. 5 is a block diagram showing one embodiment of the coordinate conversion circuit 13N. Note that 13N and 13M have the same structure, and N or M should be added as a subscript to each number, but for simplicity, for example, 131Nx
is simply written as 131x. In FIG. 5, the coordinate conversion circuit is a register 134 with selection input.
x, y, d, subtractor 135x, y, d, sine cosine constant ROM 133, multiplier 136c, s, 137
It is comprised of c, s, an adder 138x, and a subtracter 138y. The coordinate transformation amounts (ξ, η, θ) are data set via the signal 130 (i.e. 130N) from the control unit 10, or feature point data _Xs ,
Whether Y _s and D _s are set via the signals 131x, y, and d (each partial signal of 131N) is selected by a control signal from the control section (not shown because it is complicated), and the registers 134x, It is held in y, d. Also, during conversion operation, feature point data X _s , Y _s ,
D _s is the signal 131x from the feature point list memory 12N,
y, d, the subtractor 135x,
Coordinate transformation amounts (ξ, η, θ) preset in the registers 134x, y, d by y, d
(X _s −ξ), (Y _s −η), and (D _s −θ) are output, respectively. _s of signal 132d = (D _s −
Sine cosine ROM 133 addressed by θ)
outputs the pre-stored cosine value cos (D _s -θ) and sine value sin (D _s -θ) as signals 1330c,s. and multipliers 136c, s, 137c,
s and the adder/subtractor 138x, y, _s = (X _s - ξ) cos (D _s - θ) + (Y _s - η) sin
(D _s −θ) _s = (Y _s −η) cos (D _s −θ) − (X _s −ξ) sin
( _Ds - θ) is calculated, and the coordinate transformation results _s , _s , _s are output as output signals 132x, y, d. The above outputs 132x, y, and d correspond to 132N in FIG. 3. This concludes the explanation of the coordinate conversion circuit.

Now, since the amount of coordinate transformation is (ξ=η=θ=0), for the minutiae point N _{s s} = X _s , _s = Y _s , _s = D _s For the minutiae point M _f of the file fingerprint, _f = X _f , _f = Y _f , _f = D _f , meaning no conversion. The feature points N _s , M _f read out in this way, two sets of X _s , Y _s , D _s , C _s , and
X _f , Y _f , D _f , and C _f are immediately sent to the paired inspection circuit 12 via the input signal 151 .

FIG. 6 is a block diagram showing one embodiment of the pair test circuit 15. The test circuit includes a threshold ROM 153 that holds several threshold groups, and a difference absolute value unit 154.
x, y, d, z, comparator 155x, y, d, z,
and AND156. Its operation is based on two feature point data X _s , _Ys , _Ds ,
The absolute value of each difference is calculated for C _s , X _f , Y _f , D _f , and C _f using absolute difference calculators 154x, y, d, and z. The result is sent to the input signal 15 from the control unit 10.
The outputs 153x, y, d, z of the threshold ROM 153 selected by 1t, that is, the thresholds T _x , T _y , T _d , T _c
and are compared by comparators 155x, y, d, and z. This is |X _f −X _s |≦T _x , |Y _f −Y _s |≦T _y , |D _f −D _s |≦T _d , |
C _f −C _s |≦T _c . The output of this comparison result is AND156
and returns a test output 152 to the control unit 10 indicating whether or not all of the above conditions are satisfied. This concludes the explanation of the test circuit.

Minutiae data of search fingerprint and file fingerprint X _s ,
When Y _s , D _s , C _s and X _f , Y _f , D _f , C _f are tested by the pair test circuit described above, only the threshold T _c for the amount of concentration is strict, and T _x , Since T _y and T _d are before coordinate matching, a loose tolerance condition should be set to the extent that the fingerprints to be compared are not upside down. Pairs of minutiae that pass this kind of inspection
When N _s ; M _f is discovered, the control unit 10 immediately
X _s , Y _s , D _s and X _f , Y _f , D _f currently read from the feature point list are stored in the transformation amount registers (134x, y, d in FIG. 5) of each coordinate transformation circuit 13N, M. At the same time, relations N _sr , R _sr (r=
1 to 4), and M _fr and R _fr (r=1 to 4) are held inside the control unit. Then, for each quadrant r, address signals 120N, 1 are applied to the feature point list 12N12M.
Output feature point numbers N _sr and M _fr via 20M,
The feature point data X _sr , Y _sr , D _sr , and X _fr , Y _fr ,
D _fr is read out via signals 132N and 132M.
In this case, since the read feature point data has passed through the coordinate conversion circuits 13N and 13M, it has been transformed into a local coordinate system having the feature points N _s and M _f as origins, respectively. This converted feature point data and the relations R _sr and R _fr held in advance in the control unit are sent to the paired inspection circuit 15, and preset threshold values T _x ′, T _y ′，T _d ′，
At T _r ′, the difference is examined. That is, 〈M _f 〉〈N _s 〉 〈M _f 〉〈N _s 〉
〈M _f 〉〈N _s 〉 ｜X _fr −X _sr ｜≦T _x ′, ｜Y _fr −Y _sr ｜≦T _y ′, ｜D _fr −D
_sr ｜≦T _d ′ ｜R _fr −R _sr ｜≦T _r ′ (where 〈N _s 〉 is the feature point N _s
This means that the local coordinate transformation of the feature point Mf (where <M _f > means the local coordinate transformation of the feature point Mf) has been examined. The control unit 10 sequentially performs this operation for each quadrant r = 1 to 4, sets the number of successes as the candidate pair value W _sf , and if the candidate pair value W _sf is a predetermined number or more, the candidate pair feature for the feature point N _s is determined. The point number M _f is M _si and the candidate pair value W _sf is
The output is held in the candidate pair list storage 18 as W _si .
To aid understanding, for N _s ; M _f as shown in Figure 7, r = 3, that is, only N _s3 ; M _f3 is |Y _fr 〈M _f 〉−
This is denied by the test Y _sr 〈N _s 〉|≦T _y ′, in which case the candidate pair value becomes W _sf =3. That is, the meaning of the candidate pair value is that the larger the value, the higher the possibility of attaching a feature point.

The candidate pair list storage 18 may be a normal memory that holds contents as shown in FIG. 8, and the control unit 10 shown in FIG.
In the storage unit determined by N _s and i, the signal 1
A pair of M _si and W _si can be written and read through 81 and 182 . Here, the number of i is 1
This is the number of file fingerprint minutiae points that may be paired with one search fingerprint minutiae point _Ns , and usually approximately i=1 to 16 is appropriate. When storing the candidate pair minutiae M _{sf and} W _sf , the control unit 10 reads and writes the N _s rows so that W _s1 ≧W _s2 ≧...W _si ≧
The memory contents of the rows are sorted and stored in the order of the size of W _si so that W _si+1 ≧ .

As described above, the candidate pair list is completed by N _S ×M _F operations, which is the product of the number of feature points of the search fingerprint and the file fingerprint.

The second process of fingerprint matching is a coordinate system matching process using a deviation plane.

Regarding the positional deviations ΔX, ΔY, the deviation plane 17 receives an address signal 17 as shown in FIG.
It is a normal memory in which writing and reading are performed using two-dimensional addresses ΔX and ΔY quantized at 0. Regarding the direction deviation ΔD, a part of it is used at the quantized one-dimensional address ΔD as shown in Figure 9. The configuration is such that it can also be used.

The control unit 10 sends N _s (s=1 to S) and i (i=1,
2...) are sequentially generated and their contents M _si and W _si are read out, N _s is stored in the feature point list storage 12N, M _si is stored in the feature point list storage 12M as address signals 120N,
120M, coordinate conversion circuit 13N,
13M without conversion, read the directions D _s and D _si of the feature points, and use the four arithmetic operators in the control unit to calculate ΔD=D _si −
D _s is calculated, ΔD is supplied to the address signal 170 and W _si is added to the corresponding deviation plane element via the write signal 171 . (When performing this operation,
It is necessary to initialize all the relevant elements of the deviation plane 17 to "0". The initialization operation is a well-known memory clearing function and will not be specifically explained. ) By repeating such operations, the deviation plane becomes as shown in FIG. 9 for all N _s ; M _si candidate pairs stored in the candidate pair list. In this figure, the black dots schematically indicate the size of W _si , and in reality, W _si is accumulated in the quantization unit of ΔD. If the search fingerprint and the file fingerprint are the same fingerprint, the candidate pair N _s is registered in the candidate pair list;
M _si has a high probability that it is a true pair according to the relation test, and the minutiae direction of such a candidate pair
It is expected that D _s and D _si are concentrated at a certain directional deviation ΔD ^* as shown in FIG. On the other hand, there may be false pairs in the candidate pair list, but these deviations ΔD are statistically scattered on the ΔD deviation plane,
I don't concentrate. This method is well known as "characteristic extraction using Hough transform."

The control unit 10 reads and compares each element of the deviation plane 17 to determine the maximum concentration area as the deviation ΔD ^*
Then, ξ=0, η=0, θ=ΔD ^* are immediately set in the coordinate conversion circuit 13M via the signal 130M. As a result, the amount of matching between the search fingerprint and the file fingerprint is determined only with respect to coordinate rotation.

The control unit 10 calculates ΔX=X _si -X _s and ΔY=Y _si -Y _s by the same process as the deviation ΔD, and places the concentration points of the deviations ΔX and ΔY as shown in FIG. 10 on the deviation plane 17. By determining the concentration points ΔX ^* and ΔY ^* , the amount of coordinate matching ξ = ΔX ^* , η
=ΔY ^* is set in the coordinate conversion circuit 13M. The determination of the coordinate matching amount is completed in the above manner,
When determining the deviation ΔD ^* and the deviations ΔX ^* , ΔY ^* , if the cumulative sum of the concentrated portion of the deviation plane is smaller than a preset threshold, the control unit 10 determines that the search fingerprint and the file fingerprint are the same. It is determined that it is not a fingerprint, and a convenient matching value representing non-match matching is outputted as a signal 102.
, and the subsequent matching process is aborted. Also, when a sufficient cumulative sum is obtained, the above coordinate matching amount (ξ=
ΔX ^* , η = ΔY ^* , θ = ΔD ^* ), feature point data X _f , Y _f , D _f (f = 1 to
F) All coordinates are transformed. This operation is performed by the control unit 1
M _f (f=1~
F), and after reading via the read signals 132M and 122M, it is only necessary to write again via the write signal 121M.

This concludes the explanation of the coordinate matching process.

The third process of pair-fingerprint matching is a selection process of candidate pairs. After resetting the coordinate conversion circuit 13M to the non-conversion state (ξ=η=θ=0), the control unit 10 re-sets all the candidate pairs N _s ;M _si held in the candidate pair list storage 18. Read and supply N _s , M _si as addresses to the address signals 120N, 120M of the feature point list storage 12N, 12M, and read the corresponding feature point data X _s , Y _s , D _s , and X _si , Y _si , D _si
is a stricter threshold T _x ″ when generating the candidate pair list,
T _y ″, T _d ″ are tested using the paired test circuit 15 (in this case, Tz ″ is a don't care test, so
The value is set equal to the maximum output of the difference absolute value unit 154z. ) This is because the search fingerprint and the file fingerprint have completed coordinate matching, so the arrangement of minutiae should match, except for local distortions that occur when the fingerprint is imprinted. Candidate pair N _s that is determined not to be a pair in this pairwise test; M _si is the candidate pair value
It is deleted from the candidate pair list storage along with W _si .
This is the process of selecting candidate pairs.

The fourth process of pair fingerprint matching is a process of correcting candidate pair values. The following candidate pair value correction is performed for all candidate pairs N _s ;M _si in the selected candidate pair list. FIG. 11 shows the stored contents of the candidate pair list memory 18. Since the subscripts will become complicated in the following, the file fingerprint minutiae M _si that is a candidate for the search fingerprint minutiae N _s is simply expressed as M _k and the candidate The pairwise value W _si is simply written as W _k . The control unit 10 sets N _s to the address 180.
(s=1 to S) and i, one candidate pair N _s ; M _k or N _s ; M _si in the candidate pair list store 18
is selected, M _k is read out, and supplied to the address signals 120M and 120N along with N _s , and relation feature points N _sr and M _kr (r=1 to 4) are selected from the respective feature point list memories 12N and 13N. Read out. Next, the control unit 10 sequentially reads the N _sr rows of candidate feature points in the candidate pair list storage 18 and _checks for each r
(In this case, the matching of the feature points M _r is
It is simply that the minutiae numbers themselves match).
If a match with M _kr is found, the candidate value
Add W _Kr to the candidate pair value W _k of the underlying M _k .
The above operation is performed for all candidate pairs in the candidate pair list storage 18. The significance of the candidate pair value correction is that, as shown in the explanatory diagram in Fig. 11, when a certain feature point pair N _s ;M _k is a true pair, it is a relation feature point.
M _sr ; M _kr (r=1 to 4), that is, N _s1 ; M _k1 , N _s2 ;
There should be four feature point pairs M _k2 , N _s3 ; M _k3 , N _s4 , M _k4 , and therefore,
M _k1 should be registered in the N _s1 line, M _k2 should be registered in the N _s2 line, and so on. When the existence of the feature point pair described above is confirmed by searching the candidate pair list, the reliability of N _s ; M _k as a pair increases. The improvement in reliability of the feature point pair is corrected by increasing (+W _kr ) the candidate pair value W _k of N _s ;M _k . The candidate pair list is sorted in units of N _s rows according to the size of the new candidate pair value (the order is permuted in the order of W _k ). With the above steps, the final candidate pair list is completed.

The fifth process of pair fingerprint matching is a pair list generation process. The two paired list memories 19N and 19M are
As shown in Figure 12, it has the same tabular structure, is addressed by N _s (s = 1 to S) or M _f (f = 1 to F), and can hold feature point numbers and paired values. 2
It is a list memory of pairs. The list element of N _s or M _f specified by the address signal 190N or 190M from the control unit 10 is connected to the signal line 191N,
191M, 192N, and 192M, it has a structure that allows writing and reading.

In the process of generating the pair list, first, the control unit 10 generates the address signal 191N, and while addressing from N ₁ to N _S , the write signal 191N is set to "0" as the minutiae number and as a predetermined pair value as constant initial value data. All list elements of the pair list storage 19N are initialized while supplying the previously set negative pair value V _- . Similarly, the paired list memory 19M is also initialized. This work writes initial value data (0, V _- ) that means unpaired for all paired list memories,
This is to prepare for registration of feature point pairs from the next candidate pair list. That is, for the pair list element N _s or M _f that has not been registered from the candidate pair list,
The purpose is to clearly indicate that they are unpaired by leaving the initialization contents (0, V _- ) remaining. After the initialization process, the control unit 10 sends N _s (s
= 1 to S), and read out each element M _s1 and W _s1 (s = 1 to S) of i=1 as a signal 182.
Read via. Read W _s1 (s=1 to S)
are compared in the controller 10, and the maximum W _s1 , that is,
N _s and M _s1 are selected. Additionally, the selected
Read out W _s2 of N _s rows and check whether W _s1 − W _s2 is sufficiently large or whether two or more file fingerprint minutiae M _si that are paired with N _s are in conflict. (If W _s1 W _s2 , it is determined that there is a conflict).

N _s , M _s1 , and W _s1 when the test is satisfied are dually registered in pair list memories 19N and 19M. That is, for example, if M _s1 is a feature point M _f , the address N _s
M _f =M _s1 and V _f =W _s1 are written to the list element. At the same time, N _s and V _s =W _s1 are written to the list element of address M _f =M _s1 in the pair list storage 19M. When the pair list is registered and written as described above, the minutiae pair N _s ; M _s1 has been determined, so the line addressed by N _s in the candidate pair list memory 18 is erased (specifically, The feature point number "0" is written to M _s1 ), and all elements equal to M _s1 among the candidate pair feature points in all candidate pair list storage 18 are deleted and re-sorted row by row.

When one feature point pair N _s ;M _s1 is processed, the control unit 10 again repeats the search for the next feature point pair having the maximum candidate pair value and the registration in the pair list storage.

As described above, when all the elements of the candidate pair list storage are completed, the pair list generation process ends. Candidate pair list storage can generate min (S, F) feature point pairs at most, so registration is not written to the entire pair list, and while the pair list is being stored, there are list storage elements that remain in the initialized state. remain. This is an unpaired minutiae.

The sixth step in the fingerprint matching process is a matching value determination process. The determination of the paired value is performed by two processes: correction of the paired value and correction (relaxation) of the unpaired value.

For the pair list storage 19N, the control unit 10 inputs an address N _s (s=1 to S) to the address signal 190N.
is generated, and the pair list element (M _f , V _f ) is read out via the read signal 192N, and when the pair value V _f is positive, the following correction of the pair value is performed. That is, as shown in FIG. 13, for example, pair list memory 1
When the value V _f of the address N _s in 9N is positive,
Using the paired minutiae N _s ; M _f as the address, store the minutiae list 1 in the same way as the correction of the candidate paired values described above.
Access 2N and 12M and their relations
N _sr and M _fr (r=1 to 4) are read out, and it is checked whether the pair feature point of N _sr stored in the pair list storage is M _fr and the pair value V _fr is a positive value. If it is a positive value, the address signal 190N of the paired list memory 19N is sent for each r.
Read the paired list storage element by changing the address to N _sr . The pair feature point of the read pair list storage element is M _fr , and it is further checked in the control unit 10 whether or not the pair value is a positive value. If the test is satisfied, the pair value V _fr is added to the pair value V _f .
Similar to the correction of candidate pair values described above, when the neighboring feature points indicated by the relation are a pair (N _sr ; M _fr ), this pair value correction is performed based on the original feature point pair (N _s ;
This process reflects the fact that M _f ) is a more reliable feature point pair.

Similar processing is performed on the paired list storage 19M side independently of the paired list storage 19N side.

Next, when the feature point M _f ' on the pair list storage 19M is an unpaired feature point that does not have a positive pair value as shown in _FIG .
The feature point data of M _f ' is read out via the signals 132M, 122M, and the position data (X _f ', Y _f ') is supplied to the signal 110 as a two-dimensional address of the area pattern storage 11N.
Supply to N. As a result, if the area pattern element read from the output 112N is "0", the above feature point M _f is outside the fingerprint imprinting area on the search fingerprint side, and the contrast value V _- means don't care. It is rewritten to the value “0”. On the other hand, if the area pattern element is "1", the pair value V _- remains unchanged. The above process is similarly performed in the paired list storage 19N.
However, in this case, when reading the feature point data from the feature point list storage 19N, the coordinate transformation circuit 13
N is the amount of inverse transformation (ξ=-ΔX ^* , η=-ΔY ^* , θ=
-ΔD ^* ) and examine the area pattern 11M. This means that (ξ=
ΔX ^* , η = ΔY ^* , θ = ΔD ^* ) was applied to match the overlapping position and direction of the pattern features of the search fingerprint and file fingerprint, but such matching was not possible for area pattern 11M. This is because I haven't been there yet. Therefore, when inspecting the search fingerprint minutiae N _s with the area pattern P _N of the file fingerprint, a near transformation (ξ=-ΔX ^* , η=
−ΔY ^* , θ=−ΔD ^* ) must be applied to each feature point data of the feature point _Ns .

Next, for the minutiae whose paired value is still maintained as a negative value V _- , an opposing minutiae test is performed. For example, in FIG. 14, the control _unit 10 sequentially generates addresses 190M, reads out the _pair values, and checks them.
Supply Mg to and read out the relation. Among the relations M _gr (r = 1 to 4), if R _gr = 0, check whether the direction D _gr (r = 1 to 4) is directly opposite to the direction D _g of the original feature point M _g . Search using the opposing inspection circuit 16. Once discovered, the feature point M _g ,
Store each position and direction of M _gr in a feature point list 12M
The input signal 161 of the opposing inspection circuit 16 is read from
supply to.

FIG. 16 is a block diagram showing one embodiment of the opposing inspection circuit 16. That is, the difference absolute value unit 163x,
y, d, complement value unit 163c, comparator 164x,
y, d, direction ROM 165, difference absolute value unit 166
a, b, comparators 167a, b, AND gate 16
8 and a threshold ROM 169, which are connected to an input signal 161, an output signal 162, and wiring as shown in the figure.

Two sets of minutiae M _g to be inspected “oppositely”;
The minutiae data of M _gr are expressed in simplified form (X,
Y, D) and (X', Y', D'). When each feature point data is supplied to partial signals 161x, y, d and 161x', y', d' of the input signal 161, only the direction D' is the direction complement, that is, π radian, in the complement value 163c. After being inverted, the absolute value of the difference is calculated by the difference absolute value unit 163x, y, d,
Threshold specification signal 1 which is a partial signal of input signal 161
61t, the output of the specified threshold ROM 169 and the comparators 161x, y, d: |X-X'|≦H' _x , |Y-Y'|≦H' _y , |D-D'
A comparison is made such that +π|≦H′ _d , and its output is input to AND gate 168. Here, in the direction component, π radian corresponds to the most significant bit (MSB) of the difference absolute value unit 163d, and correct calculation with respect to periodicity in the direction calculation is guaranteed. On the other hand, the difference absolute value output of the difference absolute value unit 163x, y and the difference sign (sign of the calculation result)
5 _{as an address, and output D XY from} the direction ROM 165 ^as follows: D On the other hand, the direction D-D' and the _above D
The absolute value of the difference is calculated at a and b, and the comparator 16
7a and 7b, the threshold value is compared with a predetermined threshold value from the threshold ROM 169, and the result is applied to the AND gate 1.
68. AND gate 168 returns a "counter" test affirmation signal to controller 10 via output signal 162 when all inputs are asserted.

An example of the opposing inspection circuit has been described above.

When the opposing inspection circuit 16 detects that two sets of feature points M _g , M _gr having a negative pairwise value V ₋ are opposed, the control unit 10 controls the feature points M g , M _gr to be opposite to each other.
Pair value of pair list memory 19M with M _gr as address
Both V _- are changed to "0", the opposite value of don't care.
As shown in Fig. 15, this is a test of opposing adjacent feature points, and these two sets of feature points are determined based on the condition of the fingerprint impression and mechanical automatic feature extraction.
This is because they are unstable feature points that may or may not be extracted.

The above processing is also performed completely dually for the pair list storage 19N.

Pair list stores 19N and 19 with negative pair values
As a result of opposing feature point processing in M, if the above-mentioned "opposing" feature point is not found, the feature point list storage 1 is finally
By reading the density amount C _f (or C _s ) of 2M (or 12N) feature points M _f (or N _s ), the control unit 1
A threshold value is compared at 0, and if it is very large, the corresponding pair value in the pair list storage is also set to "0". The significance of this process is that, due to the pattern characteristics of a fingerprint, it is rare for a large number of feature points to be locally unevenly distributed (in other words, the amount of density is large). This is because noise extraction is often used. Summarizing the above, as schematically shown in Figure 17, the relaxation of the unpaired value V _- is
Unpaired minutiae existing outside the common imprint area of the area patterns P _N and P _M after coordinate matching between the search fingerprint and the file fingerprint, and the "opposing" minutiae and density indicated by dotted lines within the common imprint area This is an attempt to reduce the unpaired value V _- of feature points with a large value without care.

In the relaxation of the unpaired value V _- detailed above, we assumed that we write don't care "0" for the sake of simplicity, but more precisely, we do not limit it to "0" and write various test outputs. Correspondingly, a suitable negative pair value close to "0" can be introduced.

When the pair values of the pair list memories 19N and 19M are completed in the above manner, a comparison value is calculated as the final process of fingerprint comparison.

The control unit 10 sequentially reads out the paired list memories 19N and 19M as a collation value q. is calculated and output via the output signal 102. Here, S and F indicate the number of minutiae points of the search fingerprint and the file fingerprint whose paired values are not "0". These calculations are performed inside the control unit 10 using a general four arithmetic calculation circuit. Details are omitted.

The details of one embodiment of the fingerprint matching device according to the present invention have been explained in detail above, but the configuration of the control unit 10 can be easily accomplished by those skilled in the art based on the explanation of the operation so far, and therefore will not be explained. Not needed.
In order to simplify the explanation, correction of the candidate pair values in the candidate pair list memory 18 and pair list memory 19
In the correction of the pair values of N and 19M, regarding the test of N _sr ; M _fr (r = 1 to 4), which corresponds to the relation, we have only explained the test for equal r, but in general, the characteristics of the fingerprint Due to the distortion of the point configuration, it is not necessarily possible to expect correspondence between N _sr and M _fr with the same r, and M _fr ′ with a different r′ may correspond to N _sr , so N _sr ; M _fr ( r = 1 to 4) is not compared in quadrant r units (i.e., every equal r), but (N _sr ; r = 1 to 4) and (M _fr ; r' = 1 to 4) 16
It is accurate to check the correspondence by round robin. Correspondence tests in this case are local coordinate systems with origins at N _s and M _f respectively, and preset thresholds H _x ″, H _y ″,
By H _d ″〈N _s 〉〈M _f 〉 〈N _s 〉〈M _f 〉
〈N _s 〉〈M _f 〉 ｜X _sr −X _fr ′｜≦H _x ″, ｜Y _sr −Y _fr ′｜≦H _y ″, ｜D _s
A threshold value comparison such as _r −D _fr ′|≦H _d ″ should be performed just to be sure. These can be performed when generating the candidate pair list by using the coordinate conversion circuits 13N, 13M and the pair checking circuit 15. It is possible as well.

As is clear from the description of the operation, in the embodiment of the present invention, an off-the-shelf microcomputer is used to control the pair test circuit 15, the counter test circuit 16, and the control unit 10 with the processing device, and the area pattern with the memory device. Equivalent outputs can be obtained by allocating the memories 11N and 11M, the feature point list memories 12N and 12M, the deviation plane 17, the candidate pair list memories 18 and the pair list memories 19N and 19M, and the present invention The scope of the claims is not limited to such changes in hardware configuration.

[Brief explanation of the drawing]

Figure 1 is a diagram explaining fingerprint characteristics, Figure 2 is
FIG. 3 is a diagram illustrating an area pattern storage that defines a fingerprint feature, the area pattern that is its contents, a minutiae list storage, and a minutiae list that is its contents. 4 is a schematic flow diagram showing an outline of the operation of the present invention, FIG. 5 is a block diagram showing an example of a coordinate conversion circuit, FIG. 6 is a block diagram showing an example of a counter-inspection circuit, FIG. 7 is a diagram illustrating the content of candidate pair inspection, FIG. 8 is a diagram showing candidate pair list storage and the structure of the candidate pair list, which is the content thereof, and FIG. 9 is a diagram illustrating the direction in coordinate matching. FIG. 10 is a diagram showing the deviation plane of positional matching as above. FIG. 11 is a diagram explaining the candidate pair value emphasis process of the candidate pair list. FIG. FIG. 13 is a diagram showing the structure of the pair list that is stored and its contents; FIG. 13 is a diagram explaining the pair value modification of the pair list; FIG. 16 is a block diagram illustrating an example of a counter inspection circuit, and FIG. 17 is a diagram illustrating an example of a minutiae to which asymmetrical value relaxation is applied. It is. In the figure, a control unit 10, a search fingerprint area pattern storage 11N, a minutiae list storage 12N, a file fingerprint area pattern storage 11M, a minutiae list storage 12M, coordinate conversion circuits 13N, 13M, a paired inspection circuit 15, and a counter inspection circuit. 16, deviation plane 1
7. Candidate pair list memory 18 and pair list memory 1
9N and 19M are shown respectively.

Claims (1)

[Scope of Claims] 1. A fingerprint matching device that determines whether two collected fingerprints are the same based on pattern characteristics of the two fingerprints, comprising: (1) two-dimensional quantization pattern storage; (2) area pattern storage means comprising address, write and read signal lines for each element of the pattern storage and holding a pattern indicating a pattern effective area among the pattern features; (2) a plurality of feature points of the pattern features; Feature point list storage for holding data groups in a tabular format, addresses for each feature point data to single or multiple feature point data elements, write and read signal lines;
and a feature point list storage means having a built-in coordinate transformation circuit that performs a predetermined coordinate transformation on the coordinate-related data readout signal among the readout signals; (3) a coordinate deviation of the feature point data of the two pattern features in two dimensions (4) a characteristic point of one of the two pattern features; Candidate pair list storage means for holding candidate pairs in tabular form consisting of the other minutiae number and candidate pair value for each candidate pair, and candidate pair list storage means consisting of addresses, write and read signal lines for each candidate pair; , (5) For the feature point pairs selected from the candidate pairs, dually create a feature point pair consisting of the feature point number and pair value of the other for each feature point, corresponding to the two pattern features. Two list memories held in tabular format,
and pair list storage means comprising addresses, write and read signal lines for each of the feature point pairs in each of the pair list memories; (6) characteristics for the two pattern features held in the feature point list memory; (7) a pair inspection circuit that inputs point data pairs, compares the difference between each minutiae data element with a threshold value, and calculates a candidate value; (8) an opposing inspection circuit that inputs feature point data on the feature point list storage specified by the feature point and compares coordinate-related elements of the feature point data with a threshold value, thereby determining and outputting an opposing feature point; Connected to each address, write and read signal line of the area pattern storage means, each list storage means and deviation plane storage means, connected to input and output signal lines of each of the inspection circuits, and further connected to the two patterns from the outside. a control section which has a signal line for inputting the features and outputting a matching value of the two pattern features as a final output, and has built-in functions for order control, address generation, temporary storage, four arithmetic operations, and comparison; , each address, write and read signal line according to the order control of the control unit,
and input/output signal lines to (a) input the two pattern features and store them in the area pattern memory and feature point list memory, and (b) use the pair inspection circuit to input the two pattern features. (c) selecting candidate pairs from feature point data of features according to a preset threshold and storing them in the candidate pair list storage; (c) calculating deviations of feature point data of the selected candidate pairs onto the deviation plane; Coordinate transformation of the feature point data for coordinate matching is performed using the maximum cumulative deviation obtained from the accumulated results; (d) Again, the feature point data pair specified by the candidate pair is passed through the pair checking circuit. (e) For the re-selected candidate pairs, select candidates based on the neighborhood relationships existing in the candidate pairs. correcting the pair values; (f) using the updated candidate pair values as pair values and storing feature points selected by the pair values in the pair list memory; (g) re-reading the pair list. After correcting the pair value from the neighborhood relationship of the stored pair of minutiae points, (h) for the unpaired minutiae points that do not form a pair in the pair list storage, unpaired minutiae points on the same fingerprint are added to the minutiae list; read from the memory and correct the paired values by testing the opposing minutiae with a preset threshold value using the opposing testing circuit; From the area pattern storage output by
After correcting the paired values outside the common matching area, (v) a fingerprint matching device that calculates and outputs a matching value from the paired value group corresponding to the two pattern features and the number of each feature point.