JP2680882B2 - Fingerprint collation device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] A fingerprint collation device that performs feature extraction during registration to create a dictionary and performs partial pattern matching during collation, and an object thereof is to speed up the collation operation while maintaining the reliability of the collation result. A fingerprint image input means for inputting an input fingerprint image that requires matching, and a dictionary in which an image of a fingerprint portion including at least fingerprint feature points for each registered fingerprint is stored in advance as a registered image, Collating means for collating the registered image with a collating portion corresponding to the fingerprint portion of the input fingerprint image, a starting point setting means for setting a scanning start point of the collating portion in the collating means, a starting point setting means and a collating means. And a control means for controlling the matching portion, and the collating means performs interlaced scanning of the pixels from the scanning start point set by the start point setting means to check the coincidence with the registered image. The collation result indicating the degree is output, and when the collation result does not reach the predetermined threshold value, the control unit changes the scanning start point in the start point setting unit and the collation unit performs scanning after changing the collation portion. It has a configuration in which pixels are interlaced to scan from a starting point.

[Industrial applications]

The present invention relates to a fingerprint collation device that performs feature extraction during registration to create a dictionary and collates fingerprints by partial pattern matching during collation.

In recent years, how to secure safety has become an important issue as computers spread throughout society. For example, many questions have been raised from the viewpoint of ensuring the safety of ID cards and personal identification numbers that have been conventionally used as a means of confirming the identity when entering a computer room or using a terminal. On the other hand, fingerprints are the most powerful means of personal identification because they have the great features of "universality" and "lifetime immutability", and personal verification systems using fingerprints are being studied.

[Conventional technology]

 FIG. 8 shows the configuration of a conventional fingerprint matching device.

In the figure, a fingerprint sensor 10 is, for example, a holographic fingerprint sensor described in a research group material PRU88-38, pp.61-68 by Igaki et al., Which is pressed against a light guide plate such as a glass plate. A virtual image of a fingerprint is obtained by irradiating a fingerprint of a finger with a laser beam from an oblique direction and extracting light, which is scattered by ridges of the fingerprint and guided along the light guide plate, through a hologram formed in a part of the light guide plate. This virtual image is picked up by the imaging optical system and the image pickup device, and the obtained output image signal is temporarily stored in the grayscale image storage device 12 in the form of a grayscale image.

Fingerprints generally include endpoints that are the ends of ridges or valleys and bifurcations where the ridges or valleys diverge. These end points and branch points have different arrangements for each person, and are therefore called feature points. Therefore, the binarized pattern in the vicinity of the feature point in the fingerprint of each individual and the position coordinates thereof are registered in the dictionary 14 together with the identification number or name of each individual. In the fingerprint, a fingerprint portion including such an end point or a branch point and registered in the dictionary is called a window. Generally, multiple windows are provided for one fingerprint.

The grayscale image stored in the grayscale image storage device 12 is a line 15
Is supplied to the preprocessing device 16 in response to the request signal via.
The preprocessing device 16 binarizes the grayscale image signal, and the binarized image signal is stored in the image storage device 18. The image signal stored in the image storage device 18 is collated in the collation device 20 with the binarized feature point pattern of the window stored in the dictionary 14.

FIG. 9 is a flowchart showing the matching processing operation performed by the matching device 20. In step 1, the pattern of the first window in the dictionary is compared with the corresponding binarized image. The comparison is performed by dividing the window and the small area image corresponding to the window into pixels, and determining whether the image values are the same or not for each pixel. The proportion of pixels in the window that do not match is defined as the degree of mismatch. Generally, the fingerprint image output by the fingerprint sensor is accurately located in the position or pattern of the feature points registered in the dictionary due to the difference in the printing condition that occurs each time the fingerprint is printed on the light guide plate 1, for example, the difference in the finger angle or the printing force. It does not always correspond. In step 2, the degree of disagreement is compared with a predetermined threshold, and if the degree of disagreement is smaller than the threshold, it is determined that the binarized image may correctly correspond to the feature point of the window. Points on the screen that satisfy such conditions are called candidate points. In general, the first comparison (step 1) rarely encounters a candidate point for the reasons explained above, and the first window continues scanning within a predetermined range through steps 3 and 4 until a candidate point is found.

When the candidate point is found in step 2 after several trials, the movement vector between the scan start point and the candidate point in the first window is calculated. Although a plurality of windows are generally registered for each person in the dictionary, all the windows are translated by this movement vector based on the position coordinates of the candidate points.

In step 5, an image comparison is further performed with the corresponding binarized image for each of the windows thus corrected in position, and the degree of disagreement described above is obtained. This operation will be described in more detail later. In Step 6, Step 5
The degree of disagreement obtained in step 1 is compared with a predetermined threshold (this may be the same as the threshold used in step 2), and if the degree of disagreement is a value equal to or less than the threshold, the feature point pattern of the window and the binarized image Correspondence is accepted and conversely is rejected if large. If it is a pass, the number of windows passed in step 7, that is, the total number of passes is incremented by 1 and then the total number of passes is compared with another threshold value, and if the total number of passes is equal to or more than the other threshold value, the personal verification is established. It is said to have been done. On the other hand, if it is determined in step 7 that the total number of passes does not reach the threshold value, it is determined in step 8 whether or not the comparison and comparison of the feature point pattern and the binarized image have been completed for all windows. For example, in step 9, the next window is selected and the operations in step 5 and subsequent steps are repeated. If it is judged as failed in step 6, step 10
Then, it is determined whether or not the failures have continuously occurred in the previous window and the current window. If so, it is highly likely that the candidate point selection made in step 2 is incorrect, and the next candidate point search is made by repeating step 3, step 4, step 1 and step 2. On the other hand, if such a series of rejects is not detected in step 10, step 8 is skipped, in other words, step 8 is executed without increasing the total number of passes in step 7. In step 8, it is determined whether or not the comparison of the feature point pattern and the binarized image has been completed for all windows, and if it is determined to be no, step 9 is executed again.

On the other hand, if it is determined in step 8 that the comparison has been completed for all windows, there is a strong possibility that the selection of candidate points is erroneous, and the position movement consisting of step 3, step 4, step 1 and step 2 Done. If it is determined in step 4 that the position of the candidate point is out of the predetermined movement range, it is concluded that the feature point collation cannot be established, and, for example, the person's entry into the computer room or access to the terminal is denied.

As described above, the fingerprint image obtained by the fingerprint sensor slightly changes for each imprinting due to the difference in fingerprint imprinting conditions.
Therefore, even if the position coordinate of each window is modified in step 5 based on the position coordinate of the candidate point in step 2, it does not always completely match the position coordinate of the corresponding binarized image. For this reason, conventionally, as shown in FIG. 10, the position coordinates of the window after the parallel translation of each window are used as the starting point, and the window is scanned while shifting the pixel with respect to the binarized image by one pixel at a time and the characteristic point pattern Collation of the pattern of the binarized image is performed. In the illustrated example, the position of the window is spirally moved pixel by pixel, and the degree of disagreement corresponding to each position is obtained.

[Problems to be solved by the invention]

As can be seen from FIG. 9, the scanning as shown in FIG. 10 is performed for each window, in other words, in a loop from step 8 to step 9 and back to step 5, and outside of this loop, from step 8 A loop for returning to step 1 through steps 3 and 4 is provided. For this reason, there is a problem in that the collating operation of the entire apparatus becomes very slow if scanning without jumping as shown in FIG. 10 is performed in step 5. In the scanning shown in FIG. 10, fingerprints with large line widths and thin fingerprints are uniformly scanned while performing matching, so that it takes a lot of time especially for fingerprints with large line widths.

The present invention has been made in view of the above points, and an object of the present invention is to provide a fingerprint matching device that can speed up the matching operation while maintaining the reliability of matching results.

[Means for solving the problem]

 1 (A) and 1 (B) show the principle of the device of the present invention.

In the figure, a fingerprint image input means 1 inputs a fingerprint image of a person who needs collation.

The dictionary 2 stores in advance, as a registered image, an image of a fingerprint portion including at least a feature point of the fingerprint for each registered fingerprint.

The matching means 3 matches the registered image with the matching portion corresponding to the fingerprint portion of the input fingerprint image.

The starting point position setting means 5 sets the scanning start point of the matching portion in the matching means 3.

The control means 4 controls the starting point setting means and the matching means.

Further, the collating means 3 uses the collating portion as the starting point setting means 5
Pixels are interlaced from the scanning start point set by, and a collation result indicating the degree of coincidence with the registered image is output.

The control means changes the scanning start point in the starting point setting means when the collation result does not reach the predetermined threshold value, and the collation means starts the collation portion from the changed scanning start point and scans before the change. Then, the interlaced scanning is performed in a phase different from the combination of the pixels for which the collation is performed.

[Action]

In the device of the present invention, since the collating means 3 collates the fingerprint image with the fingerprint partial image in the dictionary while interlacing the fingerprint partial image with the interlaced scanning, the operation is fast.
In addition, the collation of the combination of the fingerprint image and the fingerprint partial image skipped by such an interlacing scan can be performed while shifting the scanning start point by the start point setting means 5 controlled by the control means 4, which is overlooked. Never.

〔Example〕

 FIG. 2 shows a block diagram of an embodiment of the apparatus of the present invention.

In the figure, a fingerprint sensor 10 is a fingerprint sensor similar to the fingerprint sensor of FIG. 8, and outputs analog image information of the fingerprint imprinted on the light guide plate, for example. The output analog image information is supplied to a grayscale image storage device 12 similar to the grayscale image storage device of FIG. 8 and stored in the form of digital grayscale image information, for example. The binarization device 16 corresponds to the preprocessing device shown in FIG. 8 and sends an area designation signal designating a fingerprint area to be fingerprint-matched to the grayscale image storage device 12 to request grayscale image information corresponding to the fingerprint area. . The binarization device 16 further performs binarization by comparing the supplied grayscale image information with a predetermined threshold value. The image information binarized in this manner is stored in the binarized image storage device 18. On the other hand, a dictionary similar to the dictionary of FIG.
The image of the characteristic points of each individual, that is, the end points and bifurcation points of the fingerprint and their vicinity, is binarized and stored together with the position coordinates.

The binarized image stored in the binarized image storage device 18 and the feature point image stored in the dictionary 14 are supplied to the matching device 40, which will be described in detail later. Further, a start point setting device 42 for setting a start point of a matching operation described later cooperates with the matching device.

Further, the fingerprint collation device of this embodiment is supplied with the collation result from the collation device 40, and the binarized image storage device 18, the dictionary 14, the collation device 40, the starting point setting device 42 and an external device, such as a door of a computer room, are opened. It includes a control device 44 that controls a door opening / closing device 48 that opens and closes.

 The operation of the apparatus shown in FIG. 2 will be described below.

As described above, when a person who receives fingerprint collation imprints a fingerprint on the fingerprint sensor 10, the binarized image information is stored in the binarized image storage device 18. On the other hand, the controller 44 controls the dictionary 14
The fingerprint feature point pattern and its position coordinates registered in (1) are called to the collation device 40, and the binarized image storage device 18 is controlled to output the binarized image to the collation device 40. Further, the control device 44 outputs a start point movement signal to the start point setting device 32,
The start point setting device 42 sets the scanning start point of the window performed by the matching device 40.

Hereinafter, the third operation of the collating device 40 will be described in which the image information is supplied from the binarized image storage device 18 and the dictionary 14, and the scanning start point is supplied by the control device 44 and the start point setting device 42.
This will be described with reference to the flowchart in FIG.

After the operation of the fingerprint matching device is started in step 11, the fingerprint sensor 10 detects the fingerprint of the person who receives the fingerprint matching in step 12. Further, in step 13, the image of the characteristic points of the person registered in the dictionary 14 and the coordinates thereof are called. Steps
At 14, the feature point image is stored in the binarized image storage device 18 after being converted into a binarized image by the binarization device 16. Steps
In 15, the control device 44 controls the number of movements N of the starting point, which will be described later.
Is set to zero, and the starting point setting device 42 sets the position of the starting point.

Step 16 is the same step as the position matching step 1 in FIG. 9, and for example, for the first window registered in the dictionary, the feature point image of this window and the binarized image of the position corresponding to this window A comparison is made to determine the degree of disagreement. In step 17, it is determined whether or not the degree of disagreement obtained in step 16 is less than or equal to a predetermined threshold value A, and candidate points are obtained. If the obtained dissimilarity exceeds the threshold value A, step 18 corresponding to step 3 and step 4 in FIG.
And step 19 is executed, and the next candidate point is searched within the predetermined range.

When a candidate point is found in step 17, pattern matching corresponding to step 5 in FIG. 9 is executed in step 20 by the matching device 40 while moving the position of the window. FIG. 4 is a somewhat detailed flowchart of step 20.
Referring to FIG. 4, at the beginning of step 20, step 101
Specifies an arbitrary window other than the candidate points.

Further, in steps 102 and 103, the start point is input and the window scanning start point is set. The starting point is set based on the following relational expression.

The coordinates of the first window in the dictionary are _i (i = 1 to N; N is the number of windows).

Regarding the _first window: ₁ : start point before movement, ′ ₁ : start point after movement, _N : start point, where n is the movement vector at movement, ₁ = ₁ (1) ₁ = ₁ + _N = ₁ + _It is expressed as _N (2).

Here, when an arbitrary candidate point detected by the first window is taken, the movement vector is: =-' ₁ =-( ₁ + _N ) (3) Then, the scanning of the i (i = 2 to N) th window Starting point _i
Is given as ₁ = _i + _N + = _i + ₁ +.

Here, _N is an amount that is determined when the start point movement count N is specified, the window coordinate vector _i is stored in the dictionary 14, and the candidate point movement vector or position vector is previously obtained in step 16. . In other words, in step 102, the window number i, the start point movement count N, and the coordinates of the candidate point movement vector or the candidate point position vector are supplied, and in step 103, the scanning start point is set by the equation (4).

In step 104 each pixel in the window is compared with the corresponding pixel in the corresponding area and in step 105 the degree of disagreement is determined. In step 106, the obtained disagreement degree is the predetermined threshold value A.
Is compared with the threshold value A, and if the threshold value is equal to or less than the threshold value A, it is determined to pass. on the other hand,
If it is determined not to pass in step 106, it is checked whether or not the position of the window is within a predetermined scanning range, and if the result is YES, the window is moved in the predetermined direction by a predetermined number of pixels in step 108. To be made.

FIG. 5 shows the interlaced scanning of windows, which is one of the features of the present invention, which is performed by step 108. In the figure, "the portion shown as a window indicates the starting position designated in step 103 in FIG. 4 when N = 0. In FIG. 5, the position of the window is represented by the position of the pixel at the upper left corner of the window. Therefore, the starting position is indicated by the symbol "1". After the first comparison, that is, steps 104-107 were made first, the window position
At 108, the pixel is moved to position "2", that is, to the right by 3 pixels. At this position steps 104-107 are executed again,
As a result, if the degree of disagreement is still larger than the threshold value A and the position of the window is within the predetermined scanning range, the window is further moved downward by 3 pixels in step 108. In this way,
If it is not judged as acceptable in step 106, and if there is a window within the scanning range, the position of the window is interlaced by 3 pixels in a spiral manner as shown in FIG. If it is determined in step 107 that the window is out of the scanning range, the window is rejected. Obviously, by performing the interlaced scanning as shown in FIG. 4, the speed of the pattern matching performed in step 20 of FIG. 3 is significantly improved. For example, if interlaced scanning is performed at intervals of n pixels, the time required for pattern matching is reduced to 1 / n ² .

When step 20 in FIG. 3 ends, step 6 in FIG. 9
21 is executed, and then steps 22 to 25 corresponding to steps 7 to 10 in FIG. 9 are executed.
Is executed. This process is the same as that described with reference to FIG.

If it is determined in step 22 that the total number of passes exceeds the threshold value, it is determined that the identity verification is established as in the case of FIG. On the other hand, if it is determined in step 22 that the total number of passes does not reach the threshold, the same process as described in FIG. 9, that is, steps 23, 24, 25 and steps 18 and 19 are executed. In the conventional example of FIG. 9, if it is determined to be no in step 4, it means that the collation is unsuccessful and the person's access is denied, but in this example, in step 20 of FIG. 3,
More specifically, since the interlaced scanning is performed in step 108 in FIG. 4, there is a possibility that the correct correspondence between the window and the small area may be overlooked in this process. Therefore, in the present invention, the value of N is incremented by 1 in step 26, in other words, the position of the starting point is shifted by a predetermined pixel, and steps 27 to 36 similar to steps 16 to 25 are executed. This process itself includes steps 16 to 25 and steps 1 to 9 in FIG.
The description is omitted because it is the same as 10. The process of step 31, like step 20, corresponds to the process of FIG.

If it is determined no in step 30 and if the value of N is smaller than the maximum value MAX of N in step 37, there still remain points skipped by the interlace scanning in FIG. At N, the value of N is further increased by 1 and the position of the starting point is further shifted by a predetermined pixel, and then the operations after step 27 are repeated.

FIG. 6 shows the movement of the starting point of the first window carried out in steps 26 and 37 in FIG. Referring to the drawing, the locus shown by the thick solid line represents the first time, that is, the interlaced scanning of the window at 3 pixel intervals corresponding to step 108 in FIG. 4 when N = 0, and the square shown by the thick line is N = 0. In the case of, the start position where the interlaced scanning is started is shown. Similarly, N =
The dotted line locus indicated by 1 represents the second time, that is, window interlace scanning with a 3-pixel interval corresponding to step 108 in FIG. 4 when N = 1, and the quadrangle indicated by the dotted line starts interlace scanning when N = 1. Show points. Furthermore, the trace of the thin solid line indicated by N = 2 is the third time, that is, the step of FIG. 4 in the case of N = 2.
A quadrangle indicated by a thin solid line for the interlaced scanning corresponding to 108 represents the interlaced scanning start point when N = 2.

FIG. 7 shows such a starting point shift. The numbers in the figure correspond to the value of N, and N is used when interlaced scanning is performed every 3 pixels.
By setting the starting point position to 9 positions as shown in the drawing corresponding to = 0 to 8, the same result as when the window is moved pixel by pixel can be obtained. In other words, the moving direction and amount of the starting point is the maximum number of times the starting point moves.
Is reached, the window and the binarized pixel are selected so as to be matched even if they are shifted by one pixel from each other. Of course, when scanning is performed at all start positions N = 0 to N = 8, the number of calculations is the same as when scanning one pixel at a time, but in the case of a person whose fingerprint is registered in the dictionary There is a high probability that the identity verification will succeed in step 33 before the scanning of the start position is performed, and the number of calculations and the matching time can be substantially reduced. In the flowchart of FIG. 3, the movement of the starting point is performed only before the position matching in step 27, and the starting point is not moved every time the pattern matching for each window in step 31 is performed. However, of course, it is also possible to perform it before the pattern matching for each window. In general,
In the case of a fingerprint with a thick ridge width, it is often possible to perform matching with N = 0. On the other hand, in the case of a thin fingerprint, matching is performed while increasing the values of N = 1, 2 and N.

Referring again to FIG. 3, in step 37, it is determined whether or not the value of N has reached the maximum value, that is, 8 in the example of FIG. 48 access denied.

〔The invention's effect〕

As described above, according to the fingerprint collation apparatus of the present invention, the scanning start point of the collation part is set by the start point setting means, and the control means scans the pixels from the scanning start point set by the starting point setting means. If the collation result does not reach a predetermined threshold value, the scanning start point in the start point setting means is changed, and the collation means scans the collation portion starting from the changed scanning start point and before the change. By performing the interlaced scanning at a position different from the combination of the pixels for which the collation is performed, the fingerprint collation can be performed at high speed without lowering the reliability of the collation result, which is extremely useful in practice.

[Brief description of the drawings]

1 (A) and 1 (B) are principle diagrams of the device of the present invention, FIG. 2 is a block diagram of the device of one embodiment of the present invention, FIG. 3 is a flow chart for explaining the operation of the device of FIG. 2, and FIG. Is a flow chart showing a part of the flow chart of FIG. 3 in detail, FIG. 5 is a diagram showing an example of interlaced scanning performed in the apparatus of FIG. 2, FIG. 6 is a diagram showing movement of a start point of interlaced scanning, and FIG. FIG. 8 is a diagram showing the moving direction and order in moving the starting point of FIG. 6, FIG. 8 is a block diagram showing the configuration of a conventional fingerprint collation device, and FIG. 9 is a part of the operation of the device of FIG. A flow chart, FIG. 10 is a diagram showing an example of scanning performed in the apparatus of FIG. In the figure, 1 is a fingerprint image input means, 2 is a dictionary, 3 is a matching means, 4 is a control means, 5 is a starting point setting means, 10 is a fingerprint sensor, 12 is a grayscale image storage device, 14 is a dictionary, and 15 is a line. , 16 is a pre-processing device, 18 is an image storage device, 20 is a collation device, 30 is a fingerprint sensor, 32 is a grayscale image storage device, 34 is a binarization device, 36 is a binarized image storage device, 38 is a dictionary, Reference numeral 40 is a collation device, 42 is a starting point setting device, 44 is a control device, and 48 is an external device.

Claims (1)

(57) [Claims] 1. A fingerprint image input means (1) for inputting an input fingerprint image that requires collation, and an image of a fingerprint portion including at least fingerprint feature points for each registered fingerprint, for each fingerprint. A fingerprint collation device including a dictionary (2) stored in advance as at least one registered image, and collating means (3) for collating the registered image with a collating portion corresponding to the fingerprint portion of the input fingerprint image. A start point setting means (5) for setting a scanning start point of the matching portion in the matching means (3), and a control means (4) for controlling the starting point setting means and the matching means. The collating unit interlaces the pixels from the scanning start point set by the start point setting unit to output a collation result indicating the degree of coincidence with the registered image, and the control unit outputs the collation result. Reaching a certain threshold If it is not, the scanning start point in the starting point setting means is changed, and the collating means starts from the scanning start point after the change of the collating portion, and is different from the combination of pixels scanned and collated before the change. A fingerprint matching device characterized by interlaced scanning in phase.