JPH0757084A - Fingerprint collating device - Google Patents

Abstract

PURPOSE:To increase collating speed by reducing a thinning processing when a collation is performed. CONSTITUTION:When a collation is performed, at first, a feature point which is the highest criticality is defined as the first feature point, a first window WD 1 is opened by defining the coordinate location in the pattern of a collated fingerprint where the feature point may exist as a center. A first candidate point P1 is extracted by performing a thinning within the WD 1. Next, the coordinate location of the P1 is regarded as the coordinate location where the first feature point should be and a WD 2 is opened by defining the coordinated location where a second feature point may exist as a center. A second candidate point P2 is extracted by performing a thinning within the WD 2. Subsequently, by performing the same processing as the above, a WD 3 and the succeeding ones are opened after the coordinate location of the P1 is defined as the coordinate location where the first feature point should be, the thinning is performed within the area and a candidate point is extracted.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a fingerprint collation device for recognizing an individual by using a fingerprint, and in particular, a ridge line or a valley line in a fingerprint pattern (in this specification, the ridge line and the valley line are generically named. The present invention relates to a fingerprint collation device that collates a collated fingerprint with a registered fingerprint based on the position, direction, type, etc. of these characteristic points, using the end points and branch points of the pattern line) as characteristic points.

[0002]

2. Description of the Related Art In recent years, in the fields that require personal recognition such as entry / exit control to / from computer rooms and important machine rooms, access control to computer terminals and bank financial terminals, etc. Therefore, fingerprint collation devices are being adopted. This fingerprint collation device includes an operation unit and a control unit, and the operation unit is provided with a ten-key pad, a display, and a fingerprint sensor. In this fingerprint collation device, the user's fingerprint is registered in association with the ID number. That is, when the user inputs the ID number using the numeric keypad and causes the fingerprint sensor to collect his / her fingerprint before the operation, the pattern of the collected fingerprint (registered fingerprint) has 256 gradations. It is given to the control unit as an image (image data). The control unit extracts a processing area from the image data, thins the entire area, defines an origin (for example, a center point of the processing area) in the processing area, and sets an orthogonal coordinate (X- Y) is determined, and data of all feature points (excluding pseudo feature points) in the processing area is extracted. That is,
For each feature point, the position, direction, and type (end point, branch point) of that feature point are extracted. Then, the data of each extracted feature point is stored in the memory as registration data. As a result, during operation, when the user inputs the ID number using the numeric keypad and causes the fingerprint sensor to collect his / her fingerprint, the pattern of the collected fingerprint (verified fingerprint) is sent to the control unit as image data. Given, the processing area is extracted in the same way as in the case of the registered fingerprint, the whole is thinned, and the data (verification data) of all the feature points (excluding pseudo feature points) in the processing area is extracted. Then, by comparing the extracted collation data with the stored registration data, the collation fingerprint and the registration fingerprint are collated. At the time of matching, the longest ridge width distribution waveform in the XY direction of the matching fingerprint is obtained, and this is compared with the longest ridge width distribution waveform in the XY direction of the registered fingerprint to perform position correction on the coordinates of the matching fingerprint. There is.

[0003]

However, according to such a conventional fingerprint collation device, when performing fingerprint collation,
Since the thinning process is performed on the entire processing area to extract the feature points, there is a problem that the thinning process takes time and the matching time is prolonged. Further, although the position of the collated fingerprint on the coordinates is corrected based on the longest ridge width distribution in the XY directions, the correction is made only in the XY directions, and the accuracy cannot be said to be high.

The present invention has been made in order to solve such a problem, and an object thereof is to reduce the thinning process at the time of matching and to speed up the matching speed. To provide. Another object of the present invention is to provide a fingerprint collation device capable of accelerating the collation speed and performing the position correction of the collation fingerprint on the coordinate with high precision to improve the collation precision.

[0005]

In order to achieve such an object, in the first invention (the invention according to claim 1), the importance is determined for each feature point in the pattern of the registered fingerprint, and Registration data storage means for storing data of each feature point as registration data including information on the determined importance,
In order of increasing importance stored in the registration data storage means, a search area is defined centering on the coordinate position in the pattern of the collation fingerprint where the feature point is likely to exist, and the search area is thinned to extract the feature points. And a feature point extracting means for The second invention (the invention of claim 2)
In the first invention, the first search area is defined around the coordinate position in the pattern of the collation fingerprint in which the feature point of the first rank having the highest importance stored in the registration data storage means may be located, The first search area is thinned to extract the first feature point, and the coordinate position of the extracted first feature point is set to the first position.
Considering that the coordinate points should have the characteristic points of the rank, the second search area is defined around the coordinate position in the pattern of the collation fingerprint where the characteristic points of the second rank are supposed to be, and the second search area is defined. The second feature point is extracted by thinning and the extracted second
The feature point and the first feature point are used as a reference pair, and the position correction on the coordinate of the collation fingerprint is performed based on the reference pair.

[0006]

Therefore, according to the first aspect of the invention, at the time of collation, the coordinate positions in the pattern of the collation fingerprint which are likely to have the characteristic points are sorted in descending order of importance stored in the registered data storage means. A search area is defined as the center, and only the search area is thinned to extract feature points. Further, in the second invention, at the time of collation, the first search is centered on the coordinate position in the pattern of the collation fingerprint which is likely to have the feature point of the first rank having the highest importance stored in the registration data storage means. An area is defined, and this first search area is thinned to
Feature points are extracted, the coordinate position of the extracted first feature point is regarded as a coordinate position where the first-order feature point should exist, and the fingerprint pattern of the collation fingerprint that is likely to have the second-order feature point. A second search area is defined around the coordinate position in the center, the second search area is thinned to extract second feature points, and the extracted second feature points and first feature points are extracted. A point and a reference pair are used, and the position of the collation fingerprint on the coordinates is corrected based on the reference pair.

[0007]

EXAMPLES The present invention will now be described in detail based on examples. FIG. 2 is a block diagram of a fingerprint collation device showing an embodiment of the present invention. In the figure, reference numeral 1 is an operation unit, 2 is a control unit, and the operation unit 1 includes ten keys 1-1,
Fingerprint sensor 1-3 together with display (LCD) 1-2
Is provided. The fingerprint sensor 1-3 is a light source 1-31,
A prism 1-32 and a CCD camera 1-33 are provided. The control unit 2 is a control unit 2 including a CPU.
-1, ROM 2-2, RAM 2-3, hard disk (HD) 2-4, frame memory (FM) 2-5
And an external connection part (I / F) 2-6,
A registration program and a collation program are stored in M2-2.

[Registration of Fingerprint] The fingerprint of the user is registered in this fingerprint collation device as follows. That is,
Before operating, the user inputs the ID number assigned to him / her using the ten-key 1-1 and puts his / her finger on the prism 1-32 of the fingerprint sensor 1-3. Prism 1-
32 is irradiated with light from the light source 1-31, and the light from the light source 1-31 is totally reflected in the concave portion (valley line portion) of the fingerprint that does not contact the surface of the prism 1-32, and the CCD camera 1
To -33. On the contrary, in the convex portion (ridge portion) of the fingerprint that comes into contact with the surface of the prism 1-32, the condition of total reflection is broken, and the light source
The light from 31 is scattered. As a result, a fingerprint pattern with a high contrast is obtained in which the valley portion of the fingerprint is bright and the ridge portion is dark. The obtained fingerprint (registered fingerprint) pattern is converted into a grayscale image (image data) of 320 × 400 pixels and 256 gradations by A / D conversion, and the control unit 2
Given to.

When the ID number is given through the numeric keypad 1-1 (step 101 shown in FIG. 3), the control unit 2-1 of the control unit 2 follows the registration program stored in the ROM 2-2 and the CCD camera. The image data from 1-33 is taken in via the frame memory 2-5 (step 102), and the image quality is checked (step 1).
03). If the image quality is poor, the process returns to step 102 to re-acquire the image data, and if the image quality is good, the processing area is extracted from the acquired image data (step 104).

The extraction of the processing area in step 104 is 1
This is performed by setting 6 × 16 pixels as one block (local area), averaging the gray value of each block, and comparing with the predetermined level. That is, blocks of a predetermined level or lower are excluded as the background region, and blocks of a predetermined level or higher are extracted as the processing region.

Then, noise removal is performed on the extracted processing area (step 105). The noise removal here is performed by obtaining the average of the gray values of the surrounding pixels for every 3 × 3 pixels and setting this as the gray value of the pixel in the center. That is, in the 3 × 3 pixels shown in FIG.
By setting e = (a + b + c + d + f + g + h + i) / 8, noise removal is performed for every 3 × 3 pixels. And
The processing area from which noise has been removed is binarized for each block (local area) (step 106: see FIG. 5).

Next, in step 107, the image data of 320.times.400 pixels in which all the pixels in the background area of the binarized image in the processing area are set to "0" level, the XY direction (X: horizontal axis). Direction, Y: vertical axis direction), the longest ridge width distribution waveform is obtained.

That is, for the image data of 320.times.400 pixels shown in FIG. 6, y = n is scanned in the X-axis direction to find the number of consecutive pixels of "0" and "1" levels (shown in FIG. 7). Step 201). Then, the longest number of pixels is selected from the continuous "1" level, and the length Lx is selected.
Are stored corresponding to n (step 202). And
n is incremented (step 203), step 2
04, steps 201 to 204 are repeated until n ≧ 400 in step 204. As a result, the longest ridge width distribution waveform in the X direction as shown in FIG. 8A is obtained.

Next, x = m is scanned in the Y-axis direction to obtain the number of consecutive pixels of "0" and "1" levels (step 205). Then, the longest number of pixels is selected from the continuous "1" level, and the length Ly is stored corresponding to m (step 206). Then, m is incremented (step 207), the process proceeds to step 208, and step 2 is performed until m ≧ 320 in this step 208.
Repeat 05 to 208. As a result, the longest ridge width distribution waveform in the Y direction as shown in FIG. 8B is obtained.

Next, from the obtained longest ridge width distribution waveform in the XY directions, only the characteristic peaks (rims) are extracted according to a predetermined law, and the longest in the XY directions constituted by the extracted rims. The ridge width distribution waveform is obtained as the longest ridge width maximum value extraction waveform (step 108). This maximum ridge width maximum value extraction waveform is obtained as follows. That is, the main part of the longest ridge width distribution waveform in the X (Y) direction will be described with reference to FIG. 9, where is the minimum value candidate of the longest ridge width partial waveform. In this case, at the next maximum,
-| Is calculated. Since |-|> a (a = 5 pixels), is a candidate for the maximum value (not yet determined).
And Since is the maximum value candidate, is determined as the minimum value. Calculate |-| at the next minimum value. Here, since |-| <a, is removed. Next, at the maximum value, the magnitude relation between and is compared, and since it is larger than, it is removed and is set as the maximum value candidate. Then, |-| is calculated at the next minimum value, and since |-|> a, is determined as the maximum value, and is set as the minimum value candidate. ||| is calculated at the next maximum value. Here, since |-| <a, is removed. Then, it is compared with the next minimum value, and since is larger than, it is removed, and is continuously set as the minimum value candidate. The maximum ridge width maximum value extraction waveform in the X direction and the Y direction thus obtained (waveform in which only the maximum value among the determined minimum value and maximum value is extracted) is shown in FIG.
0 (a) and (b). Also, FIG. 20 and FIG.
FIG. 1 shows a flowchart for generating the longest ridge width maximum value extraction waveform. FIG. 20 shows the flow of determining the minimum value, and FIG. 21 shows the flow of determining the maximum value. Note that FIG.
In, since the longest ridge width distribution waveform always starts from the minimum value (0), the minimum value (0) is set as the first candidate of the minimum value A in step 501. When executing step 517, the determined maximum value B (B1) is set to "1", and the other portions not determined as the maximum value are set to "0".

Next, in step 109, the binarized image in the processing area is thinned. The thinning here is performed in order to determine an end point, a branch point, that is, a feature point, and follows Hilditch's sequential thinning processing. That is, since the ridges of the actual fingerprint data have a certain width, the feature points cannot be determined if they are used as they are. Therefore, a thinning process for narrowing the width of the ridge is performed so that the feature points can be determined. Figure 1
1 shows a part of the binarized image of the processing area after thinning.

Next, feature points are extracted from the binarized image of the thinned processing area (step 110). The extraction of this feature point will be explained using, for example, the binarized image of the processing region shown in FIG. 11, from (-n + 1, -n + 1) to (n
-1, n-1) is calculated by calculating the number of connections in 3 × 3 pixels. That is, 3 × 3 shown in FIG.
In the pixel, the coordinates of E are set to (-n + 1, -n + 1), and the coordinates of E are sequentially set to (n-1, n-1) by 3
By moving x3 pixels, the number of connections is calculated only for 3x3 pixels in which the coordinate of E is at the "1" level, and if the number of connections is 1, it is determined as an end point (feature point), and 3 is calculated. If there is, it is determined as a branch point (feature point). If the number of connections is 0, it is determined as an isolated point, if it is 2, it is a continuous point, and if it is 4, it is determined as an intersection.

The number of connections is obtained by subtracting the levels of adjacent pixels with respect to the surrounding pixels (A, B, C, D, F, G, H, I) centered on E and subtracting the levels. It is calculated by adding the values and dividing by 2. For example, in FIG.
In the 3 × 3 pixel shown in 2 (b), the level difference between G and H is 1, the level difference between H and I is 1, and the addition result of their absolute values is 2. 1 is obtained as a result, which is determined as the end point. In addition, FIG.
In the 3 × 3 pixel shown in (c), the addition result of the values obtained by subtracting the levels of the adjacent pixels is 6, so that 3 is obtained as the number of connections, and it is determined that this is a branch point. It

Next, for each extracted feature point, position / direction information of that feature point is extracted (step 111).
The extraction of the position / direction information of this feature point is performed as follows. First, the intersection of the center line of the upper and lower limits in the X direction and the center line of the upper and lower limits in the Y direction of the processing region is set as the center point of the processing region, and this center point is set as the origin (0, 0) on the coordinates of each feature point Find the position of. Then, as for the end point, as shown in FIG. 13A, the ridge line of the end point P _T is tracked by 10 pixels, and the last relative coordinate thereof is used as the direction information of the end point P _T. As for the branch point, as shown in FIG. 13B, a point obtained by tracing each ridge for 10 pixels from the branch point P _B is set as the end, and a triangle is formed by this end, and the shortest among the three sides. The relative coordinates of the directions on the diagonal of the side of are the direction information of the branch point P _B.

Next, the importance of each extracted feature point is determined (step 112). The importance of this feature point is determined as follows. [Rank: Standard without other characteristic points around (stability)] First, as shown in FIG.
A surrounding area S4 of × 20 pixels is defined, and it is checked whether or not another feature point exists in this surrounding area S4. If another feature point exists in the surrounding area S4, the feature point P
Are excluded from the ranking as pseudo feature points.
If another feature point does not exist in the surrounding area S4, 24 × 2
A surrounding area S3 of 4 pixels is defined, and it is checked whether or not another feature point exists in this surrounding area S3. If another feature point exists in the surrounding area S3, the parameter (first parameter: rank) corresponding to the stage of the surrounding area S is set to level "4".
And If there is no other feature point in the surrounding area S3,
A peripheral area S2 of 30 × 30 pixels is defined, and the peripheral area S
If another feature point does not exist in 2, the peripheral area S1 of 40 × 40 pixels is defined. If another feature point exists in the surrounding area S2, the rank is set to level "3", and if another feature point exists in the surrounding area S1, the rank is set to level "2".
If there is no other feature point in the surrounding area S1, the rank is set to level "1". FIG. 14A shows the relationship between the size of the surrounding area S and the rank. The surrounding area S is not limited to a quadrangle, but may be a circle or the like. [Distance: A measure of the proximity from the center point (reproducibility)]
Next, for the feature point P, the distance from the center point (0, 0) is obtained. If the distance from the center point is within 0 to 9 pixels, the parameter (second parameter: distance) corresponding to the distance to the center point is set to level "0". 10 to
If it is within 19 pixels, the distance is set to level "1". FIG. 14B shows the relationship between the distance from the center point and the distance. [Importance: (Rank + Distance)] A value obtained by adding the rank and the distance is determined as the importance of the feature point P. Then, in descending order of importance (in descending order of level),
Give priority to each feature point.

Next, in step 113, a registration file is created. In this registration file conversion, step 10
The maximum ridge width maximum value extraction waveform in the XY directions extracted by the position correction information extraction processing in 7 and 108, and the data (registered data) of each feature point extracted in the feature point information extraction processing in steps 109 to 112 , And stored in the hard disk 2-4 in association with the input ID number. That is, X
Along with the maximum ridge width maximum value extraction waveform in the Y direction, the position information, direction information, type, and priority of each feature point are registered in association with the input ID number. At this time, data is not registered for the feature points excluded from the ranking as pseudo feature points in step 112.

At the time of data registration, not the longest ridge width distribution waveform but the longest ridge width maximum value extraction waveform in which only its characteristic peaks are extracted is stored as position correction information, and pseudo feature points are excluded. By doing so, the memory consumption of the hard disk 2-4 can be reduced.

[Fingerprint Collation] The fingerprint collation device collates the user's fingerprint as follows. That is, during operation, the user inputs the ID number assigned to himself / herself using the ten-key pad 1-1, and then the fingerprint sensor 1
Put a finger on the prism 1-32 of No. -3. As a result, similarly to the case of registering the fingerprint, the pattern of the collected fingerprint (collation fingerprint) is given to the control unit 2 as a grayscale image (image data) of 320 × 400 pixels and 256 gradations.

The control unit 2-1 of the control unit 2 receives the ID number from the numeric keypad 1-1 (see FIG. 15).
(Step 301), the image data from the CCD camera 1-33 is fetched through the frame memory 2-5 according to the collation program stored in the ROM 2-2 (Step 302), and the image quality is checked (Step 303). ). If the image quality is poor, the process returns to step 302 to re-acquire image data, and if the image quality is good, a processing area is extracted from the acquired image data (step 304). Then, noise removal is performed on the extracted processing region (step 305) and binarization is performed for each block (local region) (step 30).
6). The pre-processing in steps 304 to 306 is the same as the pre-processing in steps 104 to 106 shown in FIG.

Next, the process proceeds to step 307, and the image data of 320 × 400 pixels in which all the pixels in the background region are set to the “0” level in the binarized image in the processing region is shown in FIG.
Similar to step 107 shown in (1), the longest ridge width distribution waveform in the XY directions is obtained. And this XY
Based on the longest ridge width distribution waveform in the direction and the longest ridge width maximum value extraction waveform in the XY direction registered corresponding to the input ID number, on the coordinates of the collation fingerprint represented by the binary image in the processing area. Rough position correction in step (step 30
8).

In this rough position correction, X
Each rim forming the longest ridge width maximum value extraction waveform in the X (Y) direction by comparing the longest ridge width distribution waveform in the (Y) direction with the longest ridge width maximum value extraction waveform in the X (Y) direction Is calculated with respect to the waveform values of ± 2 pixels before and after the longest ridge width distribution waveform in the X (Y) direction at the position corresponding to the rim. Then, the one having the smallest difference is obtained as a deviation amount for each rim, and the deviation amount is added and divided by the number of rims to obtain an average deviation amount per rim. Then, this shift amount is set as the position shift amount in the X (Y) direction, and the coordinate position of the collation fingerprint in the processing area is corrected so as to minimize the position shift amount in the X (Y) direction.

Incidentally, in the rough position correction described above,
The difference between the waveform values of ± 2 pixels before and after in the longest ridge width distribution waveform at the position corresponding to each rim of the longest ridge width maximum value extraction waveform is taken, and the one having the smallest difference is obtained as the deviation amount for each rim. As a result, it is possible to absorb the distortion of the pattern of the collation fingerprint, which becomes thin or thick depending on how the pressure is applied. Further, by comparing the longest ridge width distribution waveform in the X (Y) direction and the longest ridge width maximum value extraction waveform in the X (Y) direction, the positional deviation amount in the X (Y) direction is obtained. , X (Y) direction longest ridge width distribution waveforms are compared with each other to obtain a positional deviation amount in the X (Y) direction, the processing speed becomes faster.

After the rough position correction is performed in step 308, the process proceeds to step 309 to perform the collation determination process.
In this collation determination process, the registration data of each feature point registered corresponding to the input ID number is searched (step 401 shown in FIG. 16), and the feature point with the highest priority is set as the first feature point. , The first window (search area) is opened around the coordinate position (registered coordinate position) in the pattern in which the first feature point is supposed to be, with respect to the collation fingerprint subjected to the rough position correction. (Step 4
02: See FIG. 1 (a)). The sizes WD-X and WD-Y of the first window WD1 are determined from the amount of positional deviation in the X and Y directions (calculated value of rough position correction) obtained in the rough position correction in step 308.

FIG. 18A shows the relationship between the calculated value Av-m1 for rough position correction in the X direction and WD-X.
The calculated value Av-m2 of the rough position correction in the Y direction is shown in (b).
And WD-Y. That is, in the present embodiment, when the rough position correction is largely performed, when the first window WD1 is wide open and when the first window WD1 is small,
The first window WD1 is opened small.

Then, the opened first window WD1
The binarized image is subjected to thinning and feature points are extracted.
Then, with respect to the extracted feature points, position information and direction information are extracted and listed as a first candidate point corresponding to the first feature point (step 403). In this case, there may be a plurality of first candidate points. Figure 1
(A) shows the case where the number of the first candidate points is one.
If there is no first candidate point, in response to "NO" in step 404, the process immediately proceeds to step 416, and a display indicating that the collation fingerprint and the registered fingerprint do not match is displayed. If there is a first candidate point, the process returns to step 401, and the coordinate position of the first candidate point is set to the coordinate position where the first feature point should be (for the feature point having the second highest priority (second feature point)). (Registered coordinate position), the second window is opened with the coordinate position in the pattern of the collation fingerprint where the second feature point is likely to be the center (step 405: see FIG. 1B). That is, the coordinate position of the first candidate point P1 is regarded as the coordinate position where the first feature point should be, and the second window WD2 is opened with the coordinate position of the first candidate point P1 as the starting point. The size of this second window WD2 is constant.

Then, the opened second window WD2
The binarized image of is thinned and the characteristic points are extracted. Then, with respect to the extracted feature points, position information and direction information are extracted. In this case, there may be a plurality of second candidate points. FIG. 1B shows the case where there is one second candidate point. If the second candidate point P2 is found, the matching score is calculated for the candidate point P2. This matching score has each error value of the position, direction, and type of the second candidate point P2 with respect to the second feature point (for example, if the feature point which should be an end point is a branch point, the error value is 10). Is calculated by adding. Then, the first candidate point P1 and the second
When the candidate point P2 is used as a reference point pair and there are a plurality of first candidate points, a reference point pair is selected as described later (step 406), and "YE" in steps 407 and 408 is selected.
According to "S", the process proceeds to step 409, and the coordinate position of the first candidate point P1 is determined as the coordinate position where the first feature point should be (step 409). By doing this, step 3
The high-precision repositioning is performed with respect to the rough position correction in 08.

In the second window WD2, the second window WD2
If no candidate point is found, in response to “NO” in step 408, the process immediately proceeds to step 416, and a display indicating that the collated fingerprint and the registered fingerprint do not match is displayed. If there are a plurality of first candidate points in the first window WD1,
In response to “NO” in step 407, the process returns to step 405, the coordinate position of the next first candidate point is regarded as the coordinate position where the first feature point should be, and the second window is opened. The binarized image of the window is thinned to extract feature points, and the extracted feature points are used as second candidate points to calculate the matching score. In the same manner, the above-described processing is repeated until the counter in step 407 reaches the number of the first candidate points, and the second candidate points forming a pair and the matching score thereof are obtained for all the first candidate points. Then, while obtaining this matching score, the pair with the lowest matching score is selected as the reference point pair (step 406). If the counter in step 407 reaches the number of the first candidate points, then through step 408. Proceeding to step 409, the first candidate point of the reference point pair is determined as the first feature point, and its coordinate position is determined as the coordinate position where the first feature point should be.

After the re-position correction is performed in this way, the process returns to step 401, and the feature point having the next highest priority (third feature point) is centered on the coordinate position where the feature point may be located. , Third window WD3 is opened (step 410: see FIG. 1 (c)). Then, the binarized image of the opened third window WD3 is thinned to extract characteristic points, and position information and direction information are extracted for the extracted characteristic points (third candidate points) P3.
Then, the matching score of the third candidate point P3 is calculated (step 411). Similarly, when the feature points (candidate points) are not extracted even if the window is opened, the window is added and opened, and the number of feature points (candidate points) becomes 13 in step 412. The above processing is repeated until. FIG. 19 illustrates windows WD1 to WD13 opened in the pattern of the collation fingerprint.

When there are a plurality of feature points in the opened window, the feature point with the smallest score is selected as the candidate point. If the calculation of the matching score of the 13th candidate point is completed,
In response to “YES” in step 412, step 413
Proceed to and sum the matching scores of the first to thirteenth candidate points.
Then, the process proceeds to step 414, the total of the matching scores is compared with a predetermined value that is set in advance, and if it is equal to or less than the predetermined value, it is determined that the matching fingerprint and the registered fingerprint match, and a message to that effect is displayed ( In step 415), the output for the electric lock (for the automatic door) is output via the I / F 2-6. If it is equal to or more than a predetermined value, it is determined that the collated fingerprint and the registered fingerprint do not match,
A message to that effect is displayed (step 416).

As described above, according to this embodiment,
At the time of matching, the window W is centered on the coordinate position in the pattern of the matching fingerprint that is likely to have its characteristic points in order of importance.
D is opened, and only the window WD is thinned to extract the feature points, and the thinning process at the time of matching is reduced, and the matching speed is increased. In addition, according to the present embodiment, at the time of verification, the first
The window WD1 is opened centering on the coordinate position in the pattern of the collation fingerprint where there may be characteristic points in order, the window WD1 is thinned to extract characteristic points P1, and the coordinates of the extracted characteristic points P1 are extracted. The position is regarded as a coordinate position where the first-order feature point should be, and the window WD2 is opened around the coordinate position in the pattern of the collation fingerprint where the second-order feature point is supposed to be, and the window WD2 is a thin line. And the feature points P2 are extracted, and the extracted feature points P2
2 and the feature point P1 are used as a reference pair, and the repositioning of the collation fingerprint on the coordinates is performed based on this reference pair, and the collation accuracy is improved.

Further, according to the present embodiment, the matching data of the corresponding feature points in the pattern of the collated fingerprint is sequentially extracted from the feature points of high importance in the pattern of the registered fingerprint, and in the pattern of the collated fingerprint. Since the collation is completed when the number of extracted feature points of (1) reaches a predetermined number, the collation speed is remarkably faster than in the conventional case. Further, according to the present embodiment, since the importance is determined as a value obtained by adding the rank and the distance, the stability and reproducibility at the time of matching are excellent, and the fingerprint at the time of registration and the fingerprint at the time of matching are excellent. Even in the case where the areas are largely deviated, the comparison can be performed with high accuracy by comparing a small number of feature points. That is, according to the present embodiment, it is possible to speed up the matching speed without lowering the matching accuracy, as compared with the conventional fingerprint matching device that matches all the feature points.

In this embodiment, the data is registered for all feature points except the feature points excluded from the ranking as pseudo feature points, but up to a predetermined order of high importance (for example, The data may be registered only for the 50th feature points). By doing this, the hard disk 2-4
It is possible to reduce the memory consumption in.

Further, in the present embodiment, if the feature points are not extracted even if the window is opened, it is assumed that the additional windows are opened and the matching score of 13 feature points (candidate points) is calculated. However, if the feature points are not extracted, the collation score may be increased and a window may be added so that the window is not opened.
In this case, the data may be registered only for the first to thirteenth feature points having high priority, and the memory consumption of the hard disk 2-4 can be further reduced. , Processing speed will be faster.

[0039]

As is apparent from the above description, according to the present invention, the first aspect of the invention has the characteristic points in the descending order of importance stored in the registered data storage means at the time of collation. Deaf collation A search area is defined centered on the coordinate position in the fingerprint pattern, and only that search area is thinned to extract feature points. Will be faster. Further, in the second invention, at the time of collation, the first search is centered on the coordinate position in the pattern of the collation fingerprint which is likely to have the feature point of the first rank having the highest importance stored in the registration data storage means. A region is defined, the first search region is thinned to extract a first feature point, and the coordinate position of the extracted first feature point is defined as the coordinate position where the feature point of the first rank should be. Assuming that there is a second-order feature point, a second search area is defined around the coordinate position in the pattern of the collation fingerprint, and the second search area is thinned to form the second feature point. Is extracted, the extracted second feature point and the first feature point are set as a reference pair, and the position correction on the coordinate of the collation fingerprint is performed based on this reference pair. After performing a rough position correction based on the longest ridge width distribution waveform, etc. If the position correction is performed, the position correction on the coordinate of the collation fingerprint can be performed with high accuracy, and in addition to the effect of the first invention, the collation accuracy can be improved. Play.

[Brief description of drawings]

FIG. 1 is a diagram illustrating a situation in which a window is sequentially opened from a feature point having a high priority to extract feature points in a matching determination process.

FIG. 2 is a block configuration diagram of a fingerprint matching device showing an embodiment of the present invention.

FIG. 3 is a flowchart showing a flow of a registration program in this fingerprint collation device.

FIG. 4 is a diagram illustrating 3 × 3 pixels for explaining noise processing for a processing region.

FIG. 5 is a diagram showing one block in a binarized processing region after noise processing.

FIG. 6 is a diagram for explaining a method of extracting a longest ridge width distribution waveform in XY directions from image data.

FIG. 7 is a flowchart for explaining a method of extracting a longest ridge width distribution waveform in XY directions from image data.

FIG. 8 is a diagram showing a longest ridge width distribution waveform extracted from image data.

FIG. 9 is a diagram for explaining a method of extracting a longest ridge width maximum value extraction waveform from a longest ridge width distribution waveform.

FIG. 10 is a diagram showing a longest ridge width maximum value extraction waveform extracted from the longest ridge width distribution waveform.

FIG. 11 is a diagram showing a part of a binarized image of a processing region after thinning.

FIG. 12 is a diagram showing 3 × 3 pixels for explaining a method of extracting feature points from a binarized image of a thinned processing region.

FIG. 13 is a diagram illustrating a method of extracting direction information from feature points (end points and branch points).

FIG. 14 is a diagram showing a relationship between a size and a rank of a surrounding area S which is determined stepwise around a feature point, and a relationship between a distance from the center point and a distance.

FIG. 15 is a flowchart showing the flow of a collation program.

FIG. 16 is a flowchart showing a collation determination process in a collation program.

FIG. 17 is a diagram showing a peripheral region S determined stepwise with a feature point as a center.

FIG. 18 shows the relationship between the size WD-X (WD-Y) in the X (Y) direction of the first window WD1 and the calculated value Av-m1 (Av-m2) for rough position correction in the X (Y) direction. FIG.

FIG. 19 is a diagram illustrating windows WD1 to WD13 opened in the pattern of the collated fingerprint in the processing area.

FIG. 20 is a diagram showing a flowchart (minimum value determination flow) for generating the longest ridge width maximum value extraction waveform.

FIG. 21 is a diagram showing a flowchart (maximum value determination flow) for generating the longest ridge width maximum value extraction waveform.

[Explanation of symbols]

1 Operation part 1-1 Numeric keypad 1-2 Display 1-3 Fingerprint sensor 2 Control part 2-1 Control part 2-2 ROM 2-3 RAM 2-4 Hard disk 2-5 Frame memory 2-6 External connection part S (S1 -S4) Surrounding area P (P1, P2, P3, P _T , P _B ) Feature point WD (WD1 to WD13) window

Claims (2)

[Claims] 1. An origin is appropriately set in a fingerprint pattern taken as a registered fingerprint, orthogonal coordinates are defined around the origin, and data of a plurality of feature points in the pattern is extracted and extracted. The data of each feature point is stored as registration data, and the data of a plurality of feature points in the pattern is extracted as the verification data from the fingerprint collected as the verification fingerprint in the same manner as the registered fingerprint. In a fingerprint collation device that collates the collated fingerprint with the registered fingerprint based on the collation data and the registered data, an importance is determined for each feature point in the pattern of the registered fingerprint, and information regarding the determined importance The registration data storage means for storing the data of each feature point as registration data including, and the feature points are stored in the registration data storage means in descending order of importance. Cormorant wherein defining a search area around the coordinate positions in the A pattern collation fingerprint, the fingerprint collation apparatus characterized by comprising a feature point extracting means for extracting feature points by thinning the search region. 2. The method according to claim 1, wherein the first position is centered on the coordinate position in the pattern of the collation fingerprint, which is likely to have the first-ranked feature point with the highest importance stored in the registration data storage means.
Of the first search area, the first search area is thinned to extract the first feature point, and the coordinate position of the extracted first feature point is defined as the coordinate position where the feature point of the first rank should be. Considering that, the second search area is defined around the coordinate position in the pattern of the collation fingerprint where the feature point of the second rank may be located, and the second search area is thinned to extract the second feature point. The extracted second feature point and the first feature point are used as a reference pair, and position correction means for correcting the position of the collation fingerprint on the coordinates based on the reference pair is provided. Fingerprint matching device.