JPS63276681A - Device for collating fingerprint - Google Patents

Abstract

PURPOSE:To highly accurately collate a fingerprint by updating the registered data of feature point groups based on the feature point extracted from a fingerprint picture inputted at the time of collating. CONSTITUTION:The feature point groups to be extracted from the inputted fingerprint picture when it is decided that the collation coincides at the time of inputting the fingerprint are accumulated and stored in a second storing means 15. Accordingly, based on the feature point groups stored in the means 15, the history of the change of the input feature point groups can be decided. The feature point groups selected correspondingly to this history are used as the registered feature point groups, stored and set to a first storing means 14 as the registered data when the coincidence of the collation of the inputted feature point groups is decided. Then, in the means 14, the registered data in which the input feature point at the time of the latent collation is defined to be a reference is constantly updated and set. Based on this registered data, a collating operation is executed and the collation of the fingerprint of high accuracy is executed.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention is a method for comparing an input fingerprint image with a registered fingerprint, which can be effectively applied to, for example, a keyless entry system, an entry control system, etc. The present invention relates to a fingerprint verification device.

[Prior art] For example, without using a mechanically configured key,
When configuring a keyless entry system or a room entry control system, a fingerprint verification device has been considered that registers fingerprints and compares the registered fingerprints with, for example, the fingerprint of a person attempting to enter a room to confirm a match. In such a fingerprint verification device, it is desired that a sufficiently high recognition rate can be obtained even when, for example, a low-quality input fingerprint cannot be obtained.

As a means of obtaining fingerprint images, for example, the method of copying using ink has been known for a long time, but when used in management systems etc., it is necessary to read fingerprint images without using ink etc. For example, as such a fingerprint image input means, there is
As shown in Japanese Patent No. 85600, a total reflection method using a prism is known. In principle, this method utilizes the difference in refractive index between the convex and concave portions of a fingerprint, and as a result, the fingerprint image that is read varies greatly depending on the surface condition of the finger.

Furthermore, in the case of a fingerprint verification device, the performance of the verification device is determined by how the fingerprint data registered in the device is selected and how the verification is performed. becomes important.

In conventional fingerprint verification devices, if the quality of the fingerprint image obtained during registration is poor, the registered data includes pseudo minutiae that are not originally recognized as minutiae. This makes subsequent fingerprint verification extremely difficult.

As a countermeasure for improving the recognition rate for such low-quality images, for example, as shown in Japanese Patent Application Laid-Open No. 144982/1982, there is a means for optimally determining determination thresholds, feature points, etc. It is considered. In other words, this means selects a plurality of feature points from among the large number of feature points stored at the time of registration in descending order of matching performance from the history of matching results, and uses these selected feature points during matching. It is something.

However, in such a matching method, if the input image quality at the time of initial registration is poor, the matching result will inevitably be poor. However, it may be difficult to select the required number of feature points for verification. Further, even if selected, the quality of the selected feature points may not be sufficient, and subsequent matching may be difficult.

When performing fingerprint verification, the condition of the finger, especially the surface, on which the fingerprint is present is relatively unstable. If the surface condition of the finger changes and the image quality of the input fingerprint changes, the fingerprint will look significantly different from the input image when it is registered. In other words, the feature points extracted based on the fingerprint image input during verification often differ from the registered feature points, and these different feature points are deleted from the feature points used during verification. Become. Taking these points into consideration, it has been considered to additionally register the feature point with the next best performance as a feature point to be used during verification. However, these newly added feature points that are registered are feature points that originally do not have good performance, and therefore are not an essential measure against bad input images.

−6= [Problems to be solved by the invention] This invention was made in view of the above points, and each time a matching operation is performed, the features extracted from the fingerprint image input during this matching are The registered data of the feature point group is updated based on the point, and in particular, this update is not affected by the image quality at the time of registration, and is able to respond to changes in the image quality of the input image. Therefore, it is an object of the present invention to provide a fingerprint verification device having a registered data update function that can constantly increase the recognition rate.

[Means for solving the problem] That is, in the fingerprint verification device according to the present invention,
As schematically shown in FIG. 1, a fingerprint image of an individual to be verified is inputted from an input device 11, and feature points of the fingerprint are extracted from the input image by an extraction means 12. The extracted input feature point group is then compared by the matching means 13 with registered data consisting of the registered feature point group stored and set in the first storage means 14, and the extracted input feature point group is The point cloud is supplied to second storage means 15.

Then, when the matching means 13 determines that there is a match, the input feature point group supplied to the second storage means 15 is
They are added or updated to be stored in the second storage means 15 as a historical feature point group. Then, a registered feature point group obtained based on a plurality of historical feature point groups corresponding to each of a plurality of past input screens stored in the second storage means 15 is stored in accordance with a command from the updating means 16. This allows the first storage means 14 to store the updated information.

[Function] In the fingerprint verification device as described above, when a fingerprint is input and a match is determined to be a match, the fingerprint is extracted from the input fingerprint image at that time. A group of feature points is stored cumulatively. Therefore, the history of changes in the input feature point group can be determined based on the feature point group stored in this second storage means 15, and the selected feature points corresponding to this history can be determined. The group comes to be used as a registered feature point group, and is stored and set as registered data in the first storage means 14 upon determination of matching of the input feature point group. therefore,
The first storage means 14 is always updated with registered data such that the input minutiae at the time of the latest verification is used as a reference, and the verification operation is executed based on this registered data. This will enable highly accurate fingerprint matching to be performed.

[Embodiments of the invention] Hereinafter, one embodiment of the present invention will be described with reference to the drawings.

The fingerprint image to be compared in FIG.
The image data is converted into a digital signal and taken into the image bus 23. The input processing of this fingerprint image data is performed via the control bus 25 in response to a command from the microcomputer 24, and is stored in the image memory 26 as appropriate. A circuit 27 creates a fingerprint skeleton image from the input fingerprint image, and feature points are extracted based on this skeleton image.

At the time of fingerprint registration, the feature point group of the input image processed and extracted as described above is used as registration data, and the extracted feature point group is processed by the disk driver 27, for example. The data is written and stored in an external storage device 28 consisting of a floppy disk. Then, the input feature point group of the fingerprint image input during the matching operation is compared with the registered data to determine whether the input individual and the registered individual match.

FIG. 3 shows the configuration of the fingerprint verification device in more detail. In the image input device 21, although not shown in detail in this figure, the fingertip that comes into pressure contact with the verification surface to be identified. The fingerprint image is read and input using, for example, a CCD camera. The analog image signal read by this input device is converted into digital data by an A/D converter that constitutes the image input circuit 22, and is supplied to the image memory 26.

In this case, the fingerprint image input by the binarization circuit 30 is binarized, and the ridges of the fingerprint are, for example, "0".
The image is converted into a binary image as shown in FIG. 4 (A). This binary image contains slight noise in the background and the four parts of the fingerprint, and the shoreline holes present in the finger appear as white areas in the ridges. , so that these noise elements are removed.

Based on the image from which noise has been removed in this manner, the image data is skeletonized by the skeleton linearization circuit 31, and the inverted image of the binarized image from which noise has been removed is further processed by the inversion circuit 32. This inverted image is also skeletonized by the skeleton line forming circuit 31. In other words, as shown in (B) in Fig. 4, there is an output image in which the peaks corresponding to the ridges of the fingerprint image in Fig. 4 (A) are skeletonized, and (B) as shown in Fig. 4 (C) is the same.
A valley image is obtained by skeletonizing the valley portion of the fingerprint image obtained by inverting the image shown in the figure, and these image data (B) and (C) are stored in the image memory 2B.

Then, based on such a skeleton image, the feature points of the fingerprint skeleton are extracted in the feature point extraction circuit 33.

Here, the minutiae of a fingerprint are, for example, the ends of the ridges indicated by a, b, and c in the fingerprint image of FIG. 4(A), and the branching point of the ridge indicated by d. In (B) of FIG. 4, which is a skeleton of the above-mentioned terminal points a to C, each of the terminal points a to C1 appears as terminal points a1 to C1.

Further, the branch point d appears as an end point d1 in the inverted image shown in FIG. 4(C). Therefore, the feature points of the fingerprint image can be obtained by extracting the end points of the skeleton image shown in FIGS. 4(B) and 4(C).

Some of the feature points extracted in this way include those that are not original feature points due to blurring or blurring of the image, and such pseudo feature points are removed by pseudo feature point removal. removed by circuit 34. As a means of removing these pseudo feature points, for example, as shown in Japanese Patent Application No. 61-171768 filed by the applicant, an interval skeleton line is traced from the detected end point, and in the process of this tracing, a new feature point is removed. When the existence of an end point or a branch point is confirmed, it is determined that this feature point is not an original feature point but a pseudo feature point.

For example, if there is a pseudo image such as a "beard" in the output image as shown in Figure 5 (A), if you trace the skeleton line from the end point x1 of this "beard", a branch point will appear in the middle. become. Therefore, when this branch point appears, the end point of this "whisker" is determined to be a pseudo feature point. In addition, as shown in (B) and (C) in Fig. 5, there is a break in the skeleton line, and the end points x2 and x3, as well as x4 to x7, are located close to each other. If detected, the endpoints X2 to x7 are determined to be pseudo feature points and removed.

Each feature point of the input feature point group from which the pseudo feature points have been removed from the feature points extracted in this way is used in the matching circuit 35 to determine each of the feature points of the registered data stored in the first storage device 36. When this matching circuit 36 determines that the matching result is a match, a command is given to the history feature fish school update circuit 37 to update the input feature point group from which the pseudo feature points have been removed. The second historical feature point group is
is stored in the storage device 38 of. This second storage device 38
, a plurality of feature point groups extracted from a plurality of past input images when a match was determined to be matched are cumulatively stored as a feature point group list, and an update command is given by the update circuit 37. When the input feature point group is inputted, the latest input feature point group is stored as update data in place of the oldest stored feature point group as necessary.

Furthermore, when the matching circuit 35 determines that there is a matching match,
Furthermore, a command is given to the registered data update circuit 39. The registered data update circuit 39 creates a registered feature point group based on the plurality of historical feature point groups stored in the second storage device 38 and the input feature point group, and converts this registered feature point group into registered data. The registered data is supplied to the first storage device 36 as the registered data.

In addition, when this fingerprint image is input at the time of registration in the initial state, the registration data creation circuit 4
0, and this registration data is stored in the first storage device 36. Here, the first and second storage devices 36 and 38 are the external storage device 28.
It is set to .

FIG. 6 shows the processing flow of the verification operation of the fingerprint verification device as described above. First, in step 100, a fingerprint image is inputted. The fingerprint image input in this manner is binarized in step 101, and further subjected to noise component removal processing in step 102. Step 1
In 03, skeleton lineization is performed based on the fingerprint image from which the noise components have been removed, as shown in (B) and (C) in Figure 4.
) to obtain a skeletal line image as shown.

Then, in step 104, a plurality of feature points are extracted from the skeleton line image corresponding to the end points or branch points, and in step 105, pseudo feature points are removed from the extracted feature point group. Become.

The input minutiae point group from which the pseudo minutiae points have been removed in this way is used as registration data based on the input minutiae point group in step 106 when registering a fingerprint, and is used as registration data based on the input minutiae point group in step 107. This is to perform storage registration processing in the storage device 36.

In addition, when performing a matching operation based on the extracted input feature point group, the registered feature point group in step 108 is compared with the extracted input feature point group, and in step 109, the matching process is performed. Determine the matching result. That is, the extracted input feature point group and the registered feature point group set as registered data are compared to calculate matching points, and when matching points are obtained that are equal to or greater than a set threshold, , it is determined in step 109 that there is a match.

In a state in which it is determined that there is a match, the registration data is updated in step 110 using the extracted feature point data.

In this step 110, the feature point list accumulated and stored in the second storage device 38, that is, a plurality of history feature point groups corresponding to each of the plurality of fingerprint images that were matched in the past, and the currently extracted input Among the input feature point groups extracted this time, the feature points with a high probability of existing in the historical feature point group are compared with the actual registered features among the input feature point groups extracted this time. Feature points recognized in the point cloud are extracted. If the number of feature points extracted in this way is more than a certain number, the registered data is updated using the extracted feature points as a registered feature point group, and the oldest feature point group in the feature point list is updated. is replaced with the input feature point group extracted this time.

Figure 7 shows the flow of the registration data update process. In this process, the registered data is compared with the feature points extracted from the input fingerprint image, and if there is a match, it is determined that the person is the real person. In this case, this registration data update process is executed.

That is, in this update process, first step 200
Then, a feature point list consisting of a plurality of historical feature point groups stored in the second storage device 38 is read out. This second storage device 38 stores a plurality of input feature point groups when the person was judged to be the real person in the past, and stores the input feature point groups for multiple sides, and the input feature point groups extracted this time in step 201 are stored. Performs alignment with the existing history feature point group. The state of this alignment is determined in step 202, and as a result of this alignment, the feature points of the input feature point group currently extracted are:
If there is a large positional deviation from the feature point position in the accumulated feature point list, the feature point extracted from the current input image is determined to be inappropriate as registered data to be used for the next verification. However, no update operation is performed. if,
This is because, if registration data created based on images with a large positional shift is used as new registration data to be updated, the recognition rate will be extremely lowered thereafter.

If it is determined in step 202 that the amount of positional deviation is below a certain value, the process proceeds to the next step 203, where the coordinates and directions of each feature point are compared. That is, in this step 203, it is compared whether or not the currently matched feature point data exists in the feature point data in the accumulated feature point list, and this matching operation is performed for all feature points. The process is repeated until it is confirmed in step 204 that the comparison with all feature point data existing in the feature point list has been completed.

In this way, the feature point data that has been verified above is aligned with the feature point data of all the accumulated screens, and the coordinates and directions are compared. The number of times a corresponding feature point exists is counted in the history feature point data of each of a plurality of past screens accumulated in the step 205, and in step 205, feature points whose value is equal to or greater than the specified number are extracted.

In step 20B, the verification result obtained during the verification in step 108 is read out, and further in step 2
In step 07, feature points determined to match the actual registered data are extracted from the group of 19 input feature points from the current matching results. In step 208, the above step 20
Step 20 is applied to the feature points with high existence probability extracted in step 5.
The matched feature points extracted in step 7 are added, and it is determined whether the number of extracted feature points exceeds a certain number.

If it is determined in step 208 that the number of extracted feature points is less than a certain number, updating of the registered data is stopped.

Further, when it is determined that a certain number or more of feature points have been extracted, the feature points obtained from the currently matched image are added to the feature point list stored in the second storage device 38 in step 209. Using the input feature point group added as a history feature point group and extracted, update registration data to be used for the next verification is created, and the registration data stored in the first storage device 36 is updated in step 210. Let the process execute.

As mentioned above, the coordinate positions and directions of all the feature points used in the matching operation have been determined, and the feature point data is constituted by a combination of these coordinates and direction data.

That is, as shown in FIG. 8, the coordinates of Xm and ym can be obtained for each feature point M, and the direction dm can also be determined. This direction dm is approximated using the least squares method and is determined by the following equation.

dm=(nZxy-ΣxZy)/(nΣx2-(Σx)2)...(1) After removing the pseudo minutiae from the minutiae data extracted in this way, the input fingerprint image for matching is created. The feature point data of
It is created as shown in the figure. In this figure 9, M□,
Ml, M2... are numbers assigned to each feature point, and the X and Y coordinates corresponding to each feature point,
Further, the directions are set in combination, and each feature point data is configured.

Here, a local coordinate system in which the direction Do of each feature point MQ coincides with the X axis can be defined as shown in FIG. 10. In this local coordinate system, there are multiple feature points around the feature point M□, and the nearest feature points M□, M2, M3, and M4 in the four quadrants of this local coordinate system are extracted. Each feature point is converted into coordinate and direction data in this local coordinate system using the following equation.

xi = (Xi −X□) cos D□+ (Y
i −Y□ ) sin D□ ---(2) yi =
-(Xi -Xo) sin D□ten(Yi -Y
□ ) cos D(,・-(3)di =Di-D□
・・・・・・・・・・・・・・・・・・・・・・・・(
4) Here, (Xo% YO% D□) indicates the coordinates and direction of the central feature point in the absolute coordinate system, and (Xi, Yi, Di) indicate the coordinates and direction of the nearest feature point in the four quadrants. It represents coordinates and directions in an absolute coordinate system.

For all feature points, the coordinates and directions of the nearest feature points in the four quadrants are determined in the same way as above, and by using such a data group of feature points, registered data as shown in Figure 11 is created. Let them create it. The format of this registered data is, for each feature point, the feature point number, the coordinate value and direction in absolute coordinates, and the number of the nearest feature point within the four quadrants in the local coordinate system determined by the direction of this feature point. Furthermore, it is composed of the coordinate position and direction in this local coordinate system.

When registering such registration data, this registration data is stored in the first storage device 36, and input feature point group data as shown in FIG. 9 is also set in the second storage device 38. Register it in the feature point list.

When comparing input fingerprints, the registered data as described above is compared with minutiae points extracted from the input fingerprint image. Points are added when the number of nearest feature points exists, and points are subtracted when the number of nearest feature points is less than two, so that the input image matches the registered fingerprint based on the total score of each feature point extracted above. Determine whether or not.

In such a matching operation, the registered data sets a local coordinate system corresponding to each feature point, selects the nearest feature point corresponding to each of the four quadrants of this local coordinate system, and Since the coordinates and direction data of a point are set in combination, even if there is a positional shift or a rotational angle shift in the input image, there is no factor that would reduce the recognition rate. Further, there are cases where distortion occurs in the input fingerprint image, and the amount of distortion increases in proportion to the distance. Therefore, the determination may be made by widening the allowable range of deviation according to the distance from the central feature point.

The specific determination method is as shown in FIG. 12, in the local coordinate system determined by the central feature point M□ at the time of feature point registration, the coordinates of the first neighboring feature point M1 in the first quadrant are (
xi, 3'l) and its direction is dl,
During matching, the coordinates of neighboring minutiae points in the local coordinate system determined by the direction of central minutiae MO' are (X1', X1'
), and the direction is d1', the following equation determines whether the feature point M1 matches the feature point M'.

(xi'-xi)2+(yl'-yl)2<(r□+v
'77p 薯V SHOULDER') sinθ)・・・・・・・・・
...... (5) di'-di<Δd...
・・・・・・・・・・・・・・・・・・(6) Here,
θ is determined by the magnification determined by the image input unit used.

Then, as shown in FIG. 13, the matching results include, for example, the central feature point Mi at the time of registration, the feature point number mk that was determined to match this feature point at the time of matching, and the vicinity of the four quadrants of this feature point. The feature point numbers mk'l to mk4 are output.

Next, updating of registered data will be explained. Conventionally, registered data is not updated unless it is re-registered. Therefore, if the fingerprint image input at the time of registration is bad, all of the pseudo minutiae cannot be removed, so some of the pseudo minutiae will be included in the registration data. Therefore, the recognition rate inevitably decreases. Furthermore, the surface condition of the finger easily changes depending on the season, occupation, injury, etc., and it is difficult to deal with changes in the surface condition of the finger using fixed registered data.

For such points, the matching device shown in the above embodiment extracts only feature points that are expected to appear in the future from the matched images, and creates subsequent registration data. I am trying to update the registered data. Therefore, the quality of the input fingerprint image at the time of registration is not affected, and the registration data is updated to a high quality one, and it is possible to sufficiently cope with changes in the surface of the finger.

Next, the actual updating operation of registered data will be explained. At the time of initial registration, feature points are extracted from the input image for registration to create registration data based on the input feature point group data as shown in Figure 11. The storage settings are made to be stored in the storage device 8θ of No. 1. Further, the input feature point group data used as this registered data is written as one of the history feature point groups at the Oth point in the feature point list shown in FIG. Here, for example, two pieces of information nt and nx are written at the beginning of the time point list. This information nt indicates the number of screens of feature point group data written and stored in the feature point list, and nx indicates the number of the feature point list in which the feature point group data will be stored next. Then, each time the additional storage of one feature point group data is completed, the above nt and nx are both “+
1'', and nt is increased to the maximum number N of data stored in the feature point list. Then, nx is returned to "0" after the historical feature point group data is stored up to the Nth feature point list, and from then on, it is incremented by "+1" every time this historical feature point group data is updated. The position where the oldest feature point data among the feature point data stored in the feature point list is stored is expressed. Then, at the time of the next update, the feature point data stored in the position specified by this nx will be rewritten with the new feature point data that was collated at this time.

The registration data stored in the first storage device 36 is updated when it is determined that the input data matches the registration data as described above. First, the matched feature point group is aligned with the feature point groups of all the screens stored in the feature point list at that time, and the feature points in the matched feature point group are matched. A comparison is made to check whether a point exists as a feature point in a group of feature points in the feature point list. The amount of positional deviation after the above alignment is X. % 3'
If O%do, it is determined that a feature point corresponding to the following case exists.

l xa −xi −xo 1<Δx =-= −=
(7) l ya −yi −yo l <Δy −=
=--(8)l da -di -d(+l <Δd
・・・・・・・・・(9) In the above formula, Xa % 3'a
% da indicates data of each matched feature point, and xi % yi % di is feature point data stored in the feature point list.

Here, the positional deviation amount xO or yO is its allowable value S
If the value exceeds xmax or Symax, it is determined that the matched image has an extreme positional shift, and the update is stopped.

The same comparison operation as above is performed for all the feature point groups stored in the feature point list, and feature points with a high probability of existing among the feature points of each feature point group stored in the feature point list are selected. . In addition, the feature points that match during actual matching and appear in the matching results have a very high possibility of existing in the next input image, so the feature points that appear in the matching results are unconditionally selected. . If the number of minutiae points selected in this way is less than the required number, the image quality of the fingerprint image that was verified this time will be significantly lower than the image quality of the image stored in the minutiae list. It is predicted that something bad will happen, and it is judged that the data is not suitable for future registration, and the data update is interrupted.

If the number of feature points that appear in the matching result is selected as more than the number deemed necessary, registered data is created using the selected feature points, and updates are performed using this registered data. .

Also, the input feature point group at the time of matching is added to/updated to the feature point list, and the information nt and nx are rewritten.

In other words, by implementing the registration data update function in this way, even if the image quality of the fingerprint image input at the time of registration is poor, inappropriate minutiae will be deleted from the registration data as the matching operation is repeated. Become so. Then, registration data is created only from feature points with a high probability of appearance, making it possible to match fingerprints with a high recognition rate. In addition, each time the matching operation is repeated, new registered data is created based on the matched feature point data, which is effective against changes in image quality due to changes in the surface condition of the finger, for example. It is now possible to do so, and there is no need to re-register fingerprints.

[Effects of the Invention As described above, according to the fingerprint verification device according to the present invention,
When the input minutiae point group obtained based on the input fingerprint image is collated with the stored registered data and it is determined that there is a match, the occurrence probability = 30 - of the registered data based on high minutiae points. Since the update operation is executed, it is converted into registered data with a high recognition rate each time the verification is repeated, and especially when the surface condition of the fingertip changes, the Since the registered data is updated, the recognition rate can be effectively improved. Therefore, fingerprint verification operations in keyless entry systems, room entry control systems, etc. can be performed easily and reliably.

[Brief explanation of the drawing]

FIG. 1 is a block diagram explaining the present invention in detail, FIG. 2 is a block diagram explaining a fingerprint verification device according to an embodiment of the present invention, and FIG. 3 is a block diagram showing the above embodiment in further detail. The configuration diagram and FIGS. 4(A) to 4(C) are diagrams explaining the states of minutiae corresponding to the fingerprint image, respectively, and FIGS. 5(A) to (C)
are diagrams each explaining an example of pseudo feature points, and FIG. 6 is a flowchart explaining the overall operation flow of the above device.
FIG. 7 is a flowchart explaining the flow of the registration data update process, FIG. 8 is a diagram explaining the state of feature point direction calculation, and FIG. 9 is a diagram explaining the state of feature point data.
FIG. 10 is a diagram for explaining the local coordinate system for feature points, FIG. 11 is a diagram for explaining the state of registered data extracted based on the local coordinate system, and FIG. 12 is a diagram for explaining the state of the matching threshold. FIG. 13 is a diagram for explaining the matching results, and FIG. 14 is a diagram for explaining the contents of a feature point list that stores a group of historical feature points. 11.21... Image input device, 12... Extraction means (
input feature point group), 13... matching means, 14... first
storage means (registered feature point group), 15... second storage means (historical feature point group), 16... updating means, 22...
・Image input circuit, 24...Microcomputer, 2
5... Image memory, 26... Image processing circuit, 28.
...External storage device, 35...Verification circuit, 36...First storage device, 37...History characteristic fish school update circuit, 38
. . . second storage device, 39 . . . registered data update circuit, 40 . . . registered data creation circuit. Applicant's agent Patent attorney Takehiko Suzue (A) A1 (B) 01 (C) Figure 4

Claims (3)

[Claims] (1) An input device for inputting a fingerprint image of an individual to be verified; and a feature point extraction means for extracting a plurality of feature points that can be read from the skeleton line as an input feature point group based on the skeleton line of the input fingerprint image. and a first storage means for storing a group of registered feature points of a registered individual registered in advance as registered data, and a first storage means that stores a group of registered feature points of a registered individual registered in advance, and collates this first group of registered features with the input feature point group, and compares the group of registered feature points with the individual to be matched. a matching means for determining a match with the registered individual; and a second storage means for cumulatively storing the input feature point group as a historical feature point group of the registered individual in a state where the matching is determined by the matching means. , the probability of existing as a feature point of the input feature point group determined to be a match among the feature points of the plurality of historical feature point groups cumulatively stored in this second storage means. Registered data updating means for extracting high probability feature points and storing a feature point group including the extracted high probability feature points in the first storage device as a new registered feature point group; A fingerprint verification device characterized by: (2) The updating means selects the high probability feature point group including a plurality of matching feature points that are determined to match by matching the input feature point group with the registered feature point group as a new registered feature point group. 2. The fingerprint verification device according to claim 1, wherein the fingerprint verification device is stored in the first storage means. (3) Each feature point of the feature point group to be stored in the first and second storage means is expressed by a feature point number, a coordinate position in the absolute coordinate system, and information about the direction in which the skeleton extends, In addition, this minutiae data is created by setting a coordinate axis corresponding to the direction in which the skeleton extends from the origin, with each minutiae as the origin, and the coordinate position of the nearest minutiae that exists within each of the multiple quadrants separated by this coordinate axis. 2. The fingerprint verification device according to claim 1, wherein the fingerprint verification device is expressed together with a plurality of minutiae information including information on the extension direction of each minutiae point.