JPH06301768A - Fingerprint collation device - Google Patents

Abstract

PURPOSE:To improve the accuracy of positioning at the time of collating fingerprints and to improve fingerprint collation performance. CONSTITUTION:The fingerprint image of the inputted fingerprint is detected by a fingerprint input means. A pattern direction extraction means 11 extracts the directions of rising line patterns in the respective parts of the input fingerprint image, a specific point extraction means 12 extracts a specific point provided with the overall features of the rising line patterns of the input fingerprint image based on the directions of the rising line patterns in the respective parts of the input fingerprint image and a window image for positioning deciding means 13 decides a window image for positioning including the specific point. Then, by using the window image for the positioning including the specific point, the window image of the registered fingerprint and the inputted fingerprint image are positioned by a fingerprint collation means.

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, and more particularly to a fingerprint collation device used for various access controls.

Access control is required in various fields from opening and closing of a door to entry of a workstation. Locking and personal identification of computer systems are important tools for security management. However, they can easily be compromised by copying, theft or forgetting.

Therefore, a method using a personal identification technique using a part of the human body is an important alternative. Fingerprint matching technology is an example of this personal identification technology.

In the fingerprint collation device using this fingerprint collation technology, further improvement in collation performance is required.

[0005]

2. Description of the Related Art In a fingerprint collation device, a fingerprint is converted into image data by an input device and the fingerprint image data is registered. At the time of collating the fingerprint, the registered fingerprint image data is collated with the input fingerprint image data.

The fingerprint collation device collates the fingerprint by examining the distribution and shape matching of fingerprint feature points such as branch points and end points when collating the fingerprint. In the conventional fingerprint collation device, for example, the fingerprint is registered and collated by the moving window method shown in FIGS.

FIG. 11 is a flow chart showing a fingerprint registration procedure in the conventional moving window method, and FIG. 12 is a flow chart showing a fingerprint verification procedure in the conventional moving window method.

When registering a fingerprint, an input device is used to
The fingerprint image is detected (step 201), the detected fingerprint image is binarized (step 202), and the binarized fingerprint image is thinned (step 203).

From the fingerprint image subjected to the thinning process, feature points (branch points or end points) are extracted (step 204). A plurality of fingerprint images including the extracted feature points are cut out as window images with a window having a predetermined size (step 20).
5).

FIG. 13 is an explanatory view of feature point extraction and windows. As shown in FIG. 13A, the feature points in the fingerprint image 81 are extracted, and a plurality of rectangular windows including the feature points are set. 13B and 13C show an example of a window image including the feature points cut out by the window. 13B is an example of a window image including branch points, and FIG. 13C is an example of a window image including end points.

As described above, after cutting out a plurality of window images including feature points, the positional relationship of each window is kept,
The window image is registered in the fingerprint dictionary (step 206).

As shown in FIG. 12, when collating fingerprints, first, a fingerprint image is detected by the input device (step 21).
1) The detected fingerprint image is binarized (step 21)
2).

The binarized fingerprint image is collated using the registered window image. FIG. 14 is an explanatory diagram of pattern matching using the window image of the fingerprint dictionary and the alignment window.

As shown in FIG. 14A, in the fingerprint dictionary, a plurality of window images are registered corresponding to each ID number. One of the plurality of window images registered corresponding to each ID number is set as a positioning window, and the other plurality of window images are set as matching windows. The positioning window is used to match the registered window image with the two-dimensional position of the input fingerprint image. Each window of the fingerprint dictionary has a window image and a two-dimensional position (X
i, Yi) is registered.

In order to verify the fingerprint, first, the registration window is scanned on the input fingerprint image 82 to perform the registration. At this time, as shown in FIG. 14 (B), the scanning of the alignment window is started from the position (X1, Y1) at the time of registration, and a point that coincides with the window image is searched for in a spiral scanning locus. Then, the found point is used as a temporary alignment point. (Step 213).

In the search for the temporary registration point, it is judged whether or not the temporary registration point is found in the range of the searchable image (step 214). If no provisional alignment point is found in step 214, it is determined that the fingerprint is a human fingerprint different from the registered fingerprint of the person, and processing is performed when the person is not the person (step 215), and collation processing is performed. To finish.

If a temporary alignment point is found in step 214, the matching process is performed at the temporary alignment point using a plurality of matching windows.

In pattern matching using the matching window,
By grasping the amount of deviation between the position of the temporary alignment point and the position at the time of registration of the alignment window, the input fingerprint image is collated at the position where the collation window is displaced by this displacement amount.

When collating with this collating window,
In order to deal with the distortion of the fingerprint due to the softness of human skin, the matching window is two-dimensionally scanned by a slight amount, the matching position of the image is searched, and the pattern matching is performed (step 216).

It is judged whether or not the images match by the pattern matching of the matching window (step 217).

If the images do not match in step 217, it is determined whether matching has been completed for all matching windows (step 221). In step 221,
When the collation is completed for all the collation windows, it means that the window images equal to or larger than the predetermined threshold do not coincide with each other. Therefore, the temporary alignment point is moved, and the process returns to step 213 to perform the collation again.

If it is determined in step 221 that all the matching windows have not been matched, the process returns to step 216 in order to perform matching again in the matching windows.

If the images match at step 217, it is determined whether the window images having a predetermined threshold value or more match (step 218). In step 218, when the window images having a predetermined threshold value or more match, it is confirmed that the registered person is the registered person, the processing for confirming the person is performed (step 220), and the collation processing ends.

In step 218, if the window images equal to or larger than the predetermined threshold do not match, it is determined whether or not the matching has been completed for all the matching windows (step 219).
If the collation is not completed for all the collation windows in step 219, the process returns to step 216 to collate the next collation window.

In step 219, when the collation is completed for all the collation windows, it means that the window images having a predetermined threshold value or more do not coincide. Therefore, the provisional alignment point is moved to perform the collation again. Then, the process returns to step 213.

[0026]

In the fingerprint collation, first, it is necessary to align the registration window image with the input fingerprint image using the registered registration window. The accuracy of the alignment performed by the alignment window has a great influence on the fingerprint collation result.

For example, if the fingerprint image on the alignment window is distorted at the time of registration, the alignment accuracy at the time of collation will deteriorate,
The matching performance may deteriorate. In addition, if pseudo feature points (actually, there are no pattern features, but they are erroneously regarded as features due to the wet state of the fingerprint surface, scratches, etc.) are collated, Performance deteriorates.

In the conventional fingerprint collating apparatus, one is appropriately selected from the window images including the feature points and used as the alignment window image. Therefore, the selected alignment window image is
There is no guarantee that characteristic points that are suitable for matching and that are not easily affected by fingerprint distortion are included. In addition, the pseudo feature point may be selected as the alignment window.

For this reason, the conventional fingerprint collation device has a problem that it may take a long time to collate or an error may occur in collation.

The present invention has been made in view of the above points, and an object of the present invention is to provide a fingerprint collation device capable of improving the accuracy of alignment and improving the performance of fingerprint collation.

[0031]

FIG. 1 is a block diagram showing the principle of the present invention. The invention according to claim 1 is a fingerprint input means for detecting a fingerprint image of an input fingerprint, a fingerprint registration means for registering a registration window image and a matching window image cut out from the input fingerprint image at the time of fingerprint registration, and a fingerprint. A fingerprint collation means for performing collation of fingerprints by aligning the input fingerprint image and the registered window image with each other by using the alignment window image at the time of collation, and comparing the collation window image with the input fingerprint image. In the fingerprint collation device having the above, the pattern direction extracting means 1 for extracting the direction of the ridge pattern in each part of the input fingerprint image.
1, singular point extraction means 12 for extracting singular points having the overall characteristics of the ridge pattern of the input fingerprint image based on the direction of the ridge pattern in each part of the input fingerprint image, and the singular point A registration window image determining means 13 for determining a registration window image, and at the time of fingerprint registration, a registration window image including the singular point is determined and the registration window image is registered, At the time of fingerprint collation, the input fingerprint image and the above alignment window image are compared to perform alignment.

According to the second aspect of the present invention, the registration window image including the singular point is determined when the fingerprint is registered, and the registration window image is registered. A positioning window image including a singular point is determined, and the position of the singular point of the registered positioning window image and the position of the singular point of the registration window image of the input fingerprint image are matched. Align.

In the third aspect of the invention, the pattern direction extracting means 11 divides the input fingerprint image into a plurality of square window images, and obtains the ridge pattern direction in each window image as a discretized direction code. The singular point extracting means 12 extracts singular points based on the direction code of each window image.

In the invention of claim 6, the singular point extraction means 12 has a large change in output level at the position corresponding to the singular point and the input layer to which the direction code of the ridge pattern of each window image is inputted. A singular point is extracted from one or more types of fingerprint patterns by using a neural network that is taught and learned from data indicating the relationship between the direction code of each window image and the position of the singular point.

[0035]

According to the invention of claim 1, the singular points of the fingerprint image are extracted, and the window image of the registered fingerprint and the input fingerprint image are used by using the alignment window image including the singular points of the fingerprint image. Align the. Therefore, it is possible to improve the accuracy of alignment and improve the collation performance.

According to the second aspect of the present invention, the position of the singular point of the registered registration window image and the position of the singular point of the registration window image determined from the input fingerprint image at the time of matching are matched. Align. Therefore, it is possible to improve the accuracy of alignment and shorten the alignment time.

In the third aspect of the present invention, the input fingerprint image is divided into a plurality of window images, and the singular points are extracted from the direction of the ridge pattern in each window image. Therefore, it is possible to accurately extract the singular point.

According to the sixth aspect of the present invention, the singular points are identified by using the neural network learned by being taught the data showing the relationship between the direction code of each window image and the position of the singular point for one or more fingerprint patterns. Extract. For this reason,
It is possible to extract singular points from various fingerprint patterns with high accuracy and in a short time.

[0039]

FIG. 2 is a block diagram of a fingerprint collation device according to an embodiment of the present invention. The input device for inputting a fingerprint is the fingerprint sensor 2
1 and an A / D converter 22 for converting a fingerprint image into digital data. Further, a ten-key pad 23 for inputting an operation instruction of the fingerprint collation device is provided.

Registration / verification unit 2 for registering and collating fingerprints
Reference numeral 4 denotes a CPU (central processing unit) 26 that controls the entire matching device such as fingerprint registration processing and fingerprint matching processing, a frame memory 27 that stores a fingerprint image input by an input device,
A binarization circuit 28 for binarizing the input fingerprint image, and a binary memory 29 for storing the binary image binarized by the binarization circuit.
Is equipped with. Further, a comparison circuit 31 for comparing the registered fingerprint image with the input fingerprint image at the time of fingerprint matching, and a matching buffer 30 for storing the input fingerprint image at matching are provided.

The numeric keypad 23 is connected to the bus 33 of the registration / collation unit 24 via the interface circuit 25. Further, a control signal according to the collation result is output via the interface circuit 32. For example, when the security of opening and closing the door is managed by this fingerprint collating device, the door lock is controlled by this control signal.

FIG. 3 shows an explanatory view of an example of the fingerprint sensor 21. In the fingerprint sensor 21, the light source L
An ED 43 and a CCD 44 that detects light corresponding to a fingerprint are mounted. Further, a transparent flat glass plate 45 against which the finger 41 is pressed, a mirror 46 provided at one end of the flat glass plate 45, and a lens 4 for condensing light reflected by the mirror 46.
7. A mirror 48 for reflecting the light from the lens 47 to the CCD 44 is provided.

When the finger 41 is pressed against the flat glass 45,
The convex portions make contact, but the concave portions do not. Flat glass 45
When light is emitted from the LED 43 to the plane on which the finger 41 is pressed through, the light is reflected and scattered on the surface and inside of the finger 41. The scattered light from the concave portion of the finger 41 once passes through the air and is incident on the flat glass plate 45, and therefore has no component that is totally reflected and propagates in the flat glass plate 45.

On the other hand, the reflected light and scattered light from the convex portion of the finger 41 are directly incident from the finger 41 into the flat glass plate 45 as spherical waves, and a part of them is totally reflected in the flat glass plate 45. And the total reflection is performed in the flat glass plate 45. This totally reflected light is reflected by the mirror 46 and the lens 4
7, incident on the CCD 44 via the mirror 48.

As described above, the CCD 44 can obtain an image signal of the ridge pattern of the finger 41.

Next, registration and verification of fingerprints in this embodiment will be described. FIG. 4 is a flowchart showing a fingerprint registration procedure according to the embodiment of the present invention. Also, FIG.
3 is a flowchart showing a fingerprint matching procedure in an embodiment of the present invention. FIG. 6 is a flow chart showing the procedure for determining the positioning window in the embodiment of the present invention. The process of determining the positioning window shown in FIG. 6 is performed by step 104 at the time of registration in FIG. 4 and step 1 at the time of matching in FIG.
It takes place at 13.

The present embodiment is characterized by the processing for determining the positioning window shown in FIG. The process of determining the alignment window in this embodiment will be described below.

FIG. 7 is an explanatory diagram of a fingerprint pattern model. There are various patterns in the fingerprint pattern.
Shows a typical example of this fingerprint pattern. Figure 7
7A and 7C are spiral fingerprint patterns, which are shown in FIG.
(B) and (D) are flow fingerprint patterns.

There is always one or more singular points in the fingerprint pattern. The singular point refers to a portion where no other pattern exists in the closed curve pattern in the spiral fingerprint or a turning point of a pattern of the flow fingerprint that does not sandwich another ridge pattern. In FIG. 7, singular points P _{1 to} P ₅ are shown in each pattern. In the spiral fingerprint pattern shown in FIG. 7C, there are two singular points P ₃ and P ₄ .

This singularity is not the local characteristic of the fingerprint but the general characteristic of the fingerprint and a very distinctive characteristic. Therefore, by using the window image including the singular point as the alignment window, it is possible to improve the alignment accuracy and improve the matching performance.

In the fingerprint collation apparatus of this embodiment, this singular point is extracted in the positioning window determination process to determine the positioning window including the singular point.

Next, a method of detecting a singular point will be described. The above singular point detection is impossible with the conventional small window method used for detecting characteristic points. The singular point of the fingerprint is closely related to the direction of the ridge pattern in each part around the singular point. Therefore, the position of the singular point can be obtained by examining the direction of the ridge pattern of each part around the singular point.

FIG. 8 is an explanatory view of the relationship between the ridge direction and the singular point of the fingerprint. Here, the case where the curvature of the ridge pattern is focused will be described. The fingerprint image 51 shown in FIG. 8 is an example of a flow fingerprint pattern, and a singular point P _C exists near the center.
As shown in FIG. 8, the singular point of the fingerprint is often the point where the curvature of the ridge pattern is maximized. Further, even when the curvature is not the maximum, it is a peculiar point of the curvature of the ridge pattern.

The position of the singular point of the fingerprint is closely related to the curvature of the ridge pattern around the singular point. Further, the change in the curvature of the ridge pattern is almost continuous depending on the place. For example, if the ridge spacing is substantially constant, it can be considered that the change in the curvature between two distant points continuously changes on the line between the two points. Therefore, the position of the singular point can be obtained by obtaining the curvature of the ridge pattern around the singular point from the direction of the ridge pattern in each part around the singular point.

In the present embodiment, a plurality of window images in the vicinity of a region where a singular point is considered to be present are taken, and for each window image,
Examine the direction of the ridge pattern. A singular point is obtained from the direction of the ridge pattern of each window image.

As the direction of the ridge pattern, for example, the normal line or tangent line of the ridge can be used. Here, the normal of the ridge is used. In the example of FIG. 8, the normal V _K of the window k in the image, the normal V _L of the image of the window L, the normal V _M of the image window M is shown. The window including the singular point P _C to be obtained is the window C.

Next, the procedure for determining the positioning window will be described with reference to the flowchart in FIG. Note that FIG.
Then, the pattern direction extraction means comprises steps 131 to 134, and the singular point extraction means comprises step 134.
Then, the registration window image determination means performs step 13
It is 5.

First, for the fingerprint image to be registered, a plurality of directional windows for determining the alignment window are set. FIG. 9 shows an explanatory view of the direction window and the direction code. As shown in FIG. 9A, a direction window including a plurality of vertical and horizontal windows is set for the fingerprint image 61. In the example of FIG. 9, 6 × 6 directional windows are set. This directional window has a size sufficiently small with respect to the rate at which the direction of the ridge pattern changes depending on the location. The direction windows are sequentially numbered from 1 (step 131).

Next, the direction of the ridge pattern is extracted in each direction window. The ridge normal is used as the direction of the ridge pattern (step 132). Next, step 13
The direction of the ridge pattern in each direction window obtained in 2, that is, the normal direction is encoded. Various methods are conceivable for encoding the normal direction, but here, the normal direction is replaced with a direction code corresponding to the angle. Figure 9
(B) shows a method of encoding in the normal direction. Figure 9
As shown in (B), the normal direction is set to 8
Convert to street direction code.

As a method of extracting the direction of the ridge pattern, for example, eight direction masks may be used and the degree of correlation may be obtained.

In the example of FIG. 9A, the code 6 corresponding to the normal direction is obtained in the direction window 1, and the code 7 corresponding to the normal direction is obtained in the direction window 2 (step 13).
3).

Next, using the direction code in each direction window obtained in step 133, a singular point, that is, the maximum curvature point is extracted and the position thereof is obtained. There are various methods for obtaining the position of the singular point from the directional code in each directional window around the singular point.

For example, data indicating the correspondence between the direction code of each part around the singular point and the singular point position is stored in advance for various fingerprint patterns, and the directional code stored in advance at the time of actual fingerprint registration. The singular point can be obtained by classifying the direction code obtained from each directional window by using the data of the correspondence relationship between the singular point position and However, the above method has a drawback that it often requires division and takes a long processing time.

In this embodiment, a neural network is used to find a singular point from the direction code of each directional window around this singular point. FIG. 10 shows an explanatory diagram of a neural network used for maximum curvature point extraction. The neural network of the present embodiment is provided with an input layer 71 for inputting a direction code, an output layer 73 in which a bit at a position corresponding to a singular point position is set, and an intermediate layer 72.

For various fingerprint patterns as shown in FIG. 7, data indicating the relationship between the direction code and the position of the singular point in each part around the singular point is obtained in advance. In the learning of the neural network, the data of the relationship between the direction code and the singular point position is taught to learn the weight of the neural network. By this learning, the neural network in FIG. 10 can correctly show the singular points for various fingerprint patterns.

As shown in FIG. 10, the input layer 71 is provided with input units corresponding to the respective directional windows by the number N of directional windows. Further, the output layer 73 is provided with the output units corresponding to the respective direction windows by the number N of the direction windows.

The direction code in each direction window obtained in step 133 is input to the input layer 71 of the neural network. As a result, in the output layer 73 of the neural network, bit 1 is set in the output unit corresponding to the maximum curvature point. In this way, the maximum curvature point, that is, the singular point can be extracted from the direction code of each window. Since a neural network is used, this singular point extraction processing is
It can be executed at a higher speed than the method based on simple case classification.

Next, the positioning window including the maximum curvature point obtained in step 134, that is, the singular point is determined.
The positioning window should be of a size that allows appropriate matching.
As the positioning window, if the direction window is set to an appropriate size, the direction window including the singular point extracted in step 134 may be used as it is (step 135).

In the output layer 73, depending on the case,
It is possible for bits to stand in multiple directional windows. In this case, the alignment can be performed using a plurality of windows or a single window.

Further, each unit of the output layer 73 is not limited to one that outputs a binary signal, but may be one that outputs a multilevel signal. In this case, the unit with the largest fluctuation in output level corresponds to the singular point position.

Next, the fingerprint registration procedure in this embodiment will be described with reference to FIG. When registering your fingerprint,
A fingerprint image is detected by the input device (step 10
1) binarize the detected fingerprint image (step 10
2) Then, the binarized fingerprint image is thinned (step 103).

Next, the registration window determination process described with reference to FIG. 6 is performed on the thinned fingerprint image (step 104).

Next, feature points (branch points or end points) are extracted from the fingerprint image subjected to the thinning process. (Step 10
5). From the fingerprint image including this feature point, step 104
A plurality of collation windows around the alignment window obtained in step 3 are set (step 106).

As described above, after setting a plurality of windows including the feature points, the window images of the alignment window and the matching window and the window positions are stored in the fingerprint dictionary while the positional relationship of each window is preserved. Register (step 107).

Next, the fingerprint collation procedure in this embodiment will be described with reference to FIG. As shown in FIG. 5, when collating fingerprints, first, a fingerprint image is detected by the input device (step 111), and the detected fingerprint image is binarized (step 112).

The binarized fingerprint image is collated using the registered window image. In order to collate the fingerprints, the registered fingerprint image and the input fingerprint image are aligned using the alignment window.

Therefore, for the input fingerprint image,
A process similar to the registration window determination process at registration shown in FIG. 6 is performed to determine the registration window of the input fingerprint image. Positioning can be performed by matching the singular point position of the registration window of the input fingerprint image with the singular point position of the registration window of the registered fingerprint image. Since the singular point is a general feature of the fingerprint image, the alignment can be performed with high accuracy. Further, unlike the conventional method shown in FIG. 12, since it is not necessary to scan the alignment window, the alignment time can be shortened.

At the time of matching, the registered registration window is scanned with respect to the input fingerprint image in the same manner as in the conventional method without performing the process of determining the registration window of the input fingerprint image. You may make a match. Even in this case, since the alignment window includes a singular point, alignment can be performed with higher accuracy than the conventional method.

In this embodiment, since the alignment accuracy is high, it is not necessary to perform the alignment again after performing the above alignment (step 113).

Next, at the alignment point obtained in step 113, the matching process is performed using a plurality of matching windows around the singular point.

In pattern matching using the matching window,
By grasping the amount of deviation between the position of the alignment point and the position of the registration window at the time of registration, the collation window is collated with the input fingerprint image at a position displaced by this displacement amount.

Incidentally, at the time of collation by this collation window,
In order to deal with the distortion of the fingerprint due to the softness of human skin, the matching window is two-dimensionally scanned by a slight amount, the matching position of the image is searched, and the pattern matching is performed (step 114).

It is judged whether or not the images match by the pattern matching of the matching window (step 115).

If the images do not match in step 115, it is determined whether matching has been completed for all matching windows (step 120). In step 120,
When the collation is completed for all the collation windows, it means that the window images that are equal to or larger than the predetermined threshold value do not coincide. Then, the verification process is completed (step 119).

If the collation is not completed for all the collation windows in step 120, the process returns to step 114 to collate the next collation window again.

When the images match at step 115, it is determined whether window images having a predetermined threshold value or more match (step 116). In step 116, when the window images having the predetermined threshold value or more match, it is confirmed that the person is the registered person, the processing for confirming the person is performed (step 118), and the collation processing ends.

If the window images of the predetermined threshold value or more do not match in step 116, it is determined whether the matching has been completed for all the matching windows (step 117).
If the matching is not completed for all the matching windows in step 117, the process returns to step 114 to perform the matching again for the next matching window.

In step 117, when the collation is completed for all the collating windows, it means that the window images having a predetermined threshold value or more do not coincide, so it is judged that the person is not the registered person. Processing other than that of the person himself / herself is performed (step 119), and the collation processing ends.

As described above, in the present embodiment, it is possible to determine the registration window including the singular point for any fingerprint pattern, and use the registration window including the singular point for the registered fingerprint. The window image and the input fingerprint image are aligned. Therefore, compared with the conventional fingerprint matching device,
It is possible to improve the alignment accuracy during fingerprint matching and improve the fingerprint matching performance.

[0090]

As described above, according to the invention of claim 1,
The singularity of the fingerprint image is extracted, and the registration window image including the singularity of the fingerprint image is used to align the registered fingerprint window image with the input fingerprint image. It has the features of improving accuracy and collation performance.

According to the second aspect of the present invention, the position of the singular point of the registered alignment window image matches the position of the singular point of the alignment window image determined from the input fingerprint image at the time of matching. Since the alignment is performed by doing so, the alignment accuracy can be improved and the alignment time can be shortened.

According to the third aspect of the invention, the input fingerprint image is divided into a plurality of window images, and the singular points are extracted from the direction of the ridge pattern in each window image. Singular points used for image determination can be extracted.

According to the invention of claim 6, a singular point is obtained by using a neural network learned by being taught the data showing the relationship between the direction code of each window image and the position of the singular point for one or more fingerprint patterns. Therefore, it is possible to accurately extract the singular point used for determining the alignment window image for various fingerprint patterns in a short time.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of the present invention.

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

FIG. 3 is an explanatory diagram of an example of a fingerprint sensor.

FIG. 4 is a flowchart showing a fingerprint registration procedure according to an embodiment of the present invention.

FIG. 5 is a flowchart showing a fingerprint matching procedure according to an embodiment of the present invention.

FIG. 6 is a flowchart showing a procedure for determining an alignment window in an embodiment of the present invention.

FIG. 7 is an explanatory diagram of a fingerprint pattern model.

FIG. 8 is an explanatory diagram of a relationship between a ridge direction and a singular point of a fingerprint.

FIG. 9 is an explanatory diagram of a direction window and a direction code.

FIG. 10 is an explanatory diagram of a neural network used for maximum curvature point extraction.

FIG. 11 is a flowchart showing a fingerprint registration procedure in a conventional moving window method.

FIG. 12 is a flowchart showing a fingerprint matching procedure in the conventional moving window method.

FIG. 13 is an explanatory diagram of extraction of fingerprint feature points and windows.

FIG. 14 is an explanatory diagram of pattern matching using a window image of a fingerprint dictionary and an alignment window.

[Explanation of symbols]

 11 Pattern Direction Extraction Means 12 Singularity Extraction Means 13 Positioning Window Image Determining Means 21 Fingerprint Sensor 22 A / D Converter 23 Numeric Keypad 24 Registration / Collision Section 25, 32 Interface Circuit 26 CPU 27 Frame Memory 28 Binarization Circuit 29 2 Value memory 30 Collation buffer 31 Comparison circuit 33 Bus 42 Circuit board 43 LED 44 CCD 45 Flat glass plate 46, 48 Mirror 47 Lens 71 Input layer 72 Intermediate layer 73 Output layer

Claims (6)

[Claims] 1. A fingerprint input unit for detecting a fingerprint image of an input fingerprint, a fingerprint registration unit for registering a registration window image and a matching window image cut out from the input fingerprint image at the time of fingerprint registration, and a fingerprint matching process. And a fingerprint collating means for collating the fingerprint by comparing the input window image with the input fingerprint image after aligning the input fingerprint image with the registered window image using the alignment window image. In the fingerprint collation device, pattern direction extraction means (11) for extracting the direction of the ridge pattern in each part of the input fingerprint image, and the ridge pattern of the input fingerprint image based on the direction of the ridge pattern in each part of the input fingerprint image. Singular point extracting means (12) for extracting a singular point having the overall characteristics of the line pattern, and positioning window image determining means (1 for determining a positioning window image including the singular point. 3) and, when registering the fingerprint, determining the registration window image including the singular point and registering the registration window image, and at the time of fingerprint matching, the input fingerprint image and the registration window image. A fingerprint collation device characterized by performing position alignment by comparing with. 2. A registration window image including the singular point is determined at the time of fingerprint registration and the registration window image is registered, and at the time of fingerprint matching, a position including the singular point of the input fingerprint image is detected. A registration window image is determined, and the registration is performed by matching the position of the singular point of the registered registration window image with the position of the singular point of the registration window image of the input fingerprint image. Claim 1 characterized by the above-mentioned.
The fingerprint matching device described. 3. The pattern direction extracting means (11) comprises:
The input fingerprint image is divided into a plurality of square window images, and the ridge pattern direction in each window image is obtained as a discretized direction code, and the singular point extraction means (12) is the direction code of each window image. The fingerprint collation device according to claim 1 or 2, wherein the singular point is extracted based on 4. The pattern direction extraction means (11) comprises:
4. The fingerprint collation device according to claim 3, wherein an average value of the ridge pattern in the tangential direction is obtained as a direction code. 5. The pattern direction extraction means (11) comprises:
4. The fingerprint collation apparatus according to claim 3, wherein an average value in the normal direction of the ridge pattern is obtained as a direction code. 6. The singular point extraction means (12) includes an input layer to which a direction code of a ridge pattern of each window image is input, and an output layer whose output level largely changes at a position corresponding to the singular point. And extracting a singular point for one or more fingerprint patterns by using a neural network learned by being taught the data indicating the relationship between the direction code of each window image and the position of the singular point. The fingerprint collation device according to claim 3.