JPH1055439A - Pattern collating device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve rate of recognition by performing amplitude compressing processing by extracting an amplitude component from data respectively consisting of registered Fourier N-dimensional pattern data and collation Fourier N-dimensional pattern data. SOLUTION: At a control part 20-1, a registered pattern is first collated with a collation pattern based on the strength of a correlative component for every data consisting of the N-dimensional pattern data in a correlative component area. When these patterns are not coincident as a result of this collation, the amplitude component is extracted from the respective data consisting of the registered Fourier N-dimensional pattern data, and amplitude compressing processing is performed to this component. Besides, the amplitude component is extracted from respective data consisting of the collation Fourier N- dimensional pattern data, and amplitude compressing processing is performed to this component. A correlative distance between the registered pattern and a collation pattern is found from these amplitude compression processed data and based on this found correlative distance, the registered pattern is collated with the collation data.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an N-dimensional pattern [for example, voice (one-dimensional), fingerprint, retina, face (two-dimensional), three-dimensional image (three-dimensional)] based on spatial frequency characteristics.
And a pattern matching device for matching.

[0002]

2. Description of the Related Art In recent years, in fields requiring personal recognition such as access control to computer rooms and important machine rooms, access control to computer terminals and financial terminals of banks, etc.
A voice collation device and a fingerprint collation device are being adopted in place of conventional passwords and ID cards.

[0003] The present applicant has previously filed Japanese Patent Application No. 7-108526.
We proposed a "pattern matching device". In this pattern matching device, two-dimensional discrete Fourier transform is performed on the image data (two-dimensional pattern data) of the matching fingerprint to create matching Fourier image data. Then, the matching Fourier image data is combined with the registered Fourier image data of the registered fingerprint created by performing the same processing, and the combined Fourier image data is subjected to amplitude suppression processing (log processing). Apply a two-dimensional discrete Fourier transform.
Then, from the predetermined correlation component area appearing in the synthesized Fourier image data subjected to the two-dimensional discrete Fourier transform, upper n pixels having a higher intensity of the correlation component are extracted, and the intensity of the correlation component of the extracted n pixels is extracted. The average of is the correlation value,
Compare with threshold. If the correlation value is higher than the threshold value, it is determined that the registered fingerprint matches the collation fingerprint.

[0004]

However, in this pattern matching apparatus, fingerprint matching is performed using a correlation value obtained from pixels in a correlation component area.
The recognition rate cannot be said to be sufficient, and a further increase in the recognition rate is desired.

The present invention has been made to solve such a problem, and an object of the present invention is to utilize a discrimination ability of an amplitude component to increase a recognition rate. Is to provide.

[0006]

In order to achieve the above object, a first invention (an invention according to claim 1) provides an N-dimensional discrete Fourier transform to an N-dimensional pattern data of a registered pattern to register the data. Create Fourier N-dimensional pattern data, apply N-dimensional discrete Fourier transform to the N-dimensional pattern data of the matching pattern to create matching Fourier N-dimensional pattern data, and register registered Fourier N-dimensional pattern data and matching Fourier N-dimensional pattern data. Are synthesized, and the resultant Fourier N-dimensional pattern data obtained is subjected to amplitude suppression processing and then subjected to one of N-dimensional discrete Fourier transform and N-dimensional discrete inverse Fourier transform. Of N-dimensional pattern data of the correlation component area appearing in the synthesized Fourier N-dimensional pattern data obtained Matches with a registration pattern and the collation pattern on the basis of the intensity of the correlation component of each data s,
If the registered pattern and the matching pattern are not determined to be the same, an amplitude component is extracted from the individual data constituting the registered Fourier N-dimensional pattern data and subjected to amplitude compression processing. The amplitude component is extracted from the individual data constituting the data, subjected to amplitude compression processing, and the correlation distance between the registered pattern and the verification pattern is obtained from the data subjected to the amplitude compression processing. The registered pattern and the matching pattern are compared based on the registered pattern.

According to the present invention, first, the registered pattern and the collation pattern are collated based on the intensity of the correlation component for each data constituting the N-dimensional pattern data of the correlation component area. As a result of this collation, if they do not match, an amplitude component is extracted from each data constituting the registered Fourier N-dimensional pattern data, and an amplitude compression process such as a log process or a √ process is applied thereto. Amplitude components are extracted from individual data constituting the N-dimensional pattern data, subjected to amplitude compression processing such as log processing and √ processing, and a registered pattern and a matching pattern are obtained from the data subjected to the amplitude compression processing. (For example, a Euclidean distance) is obtained, and the registered pattern is compared with the matching pattern based on the obtained correlation distance.

The second invention (the invention according to claim 2) provides registered Fourier N-dimensional pattern data by performing N-dimensional discrete Fourier transform on the N-dimensional pattern data of the registered pattern and then performing amplitude suppression processing. After performing an N-dimensional discrete Fourier transform on the N-dimensional pattern data of the matching pattern and performing an amplitude suppression process, a matching Fourier N-dimensional pattern data is created, and the registered Fourier N-dimensional pattern data, the matching Fourier N-dimensional pattern data, , And an N-dimensional discrete Fourier transform and N
One of the two-dimensional discrete inverse Fourier transform is performed, and the intensity of the correlation component for each data constituting the N-dimensional pattern data of the correlation component area appearing in the synthesized Fourier N-dimensional pattern data subjected to the Fourier transform is calculated. When the registered pattern and the matching pattern are not determined to be the same based on the registered pattern, the N-dimensional pattern data of the registered pattern subjected to the N-dimensional discrete Fourier transform is formed. An amplitude component is extracted from each data and subjected to an amplitude compression process. The N component of the collation pattern subjected to the N-dimensional discrete Fourier transform is extracted.
The amplitude component is extracted from the individual data constituting the dimensional pattern data, subjected to amplitude compression processing, the correlation distance between the registered pattern and the verification pattern is obtained from the data subjected to the amplitude compression processing, and the obtained correlation is calculated. The matching between the registered pattern and the matching pattern is performed based on the distance.

According to the present invention, first, the registered pattern and the collation pattern are collated based on the intensity of the correlation component for each data constituting the N-dimensional pattern data of the correlation component area. As a result of this collation, if they do not match, each of the N-dimensional pattern data of the registered pattern subjected to the N-dimensional discrete Fourier transform (registered Fourier N-dimensional pattern data before the amplitude suppression processing is performed) Is subjected to amplitude compression processing such as log processing or √ processing, and N-dimensional pattern data (amplitude suppression processing is applied to the matching pattern subjected to N-dimensional discrete Fourier transform. The amplitude component is extracted from the individual data constituting the matching Fourier N-dimensional pattern data before being subjected to the amplitude compression processing such as log processing and √ processing, and the amplitude compression processing is performed on the data subjected to the amplitude compression processing. A correlation distance (for example, Euclidean distance) between the registered pattern and the matching pattern is obtained, and the registered pattern and the matching pattern are determined based on the obtained correlation distance. The collation of the emissions are carried out.

[0010] A third invention (an invention according to claim 3) is to perform N-dimensional discrete Fourier transform on N-dimensional pattern data of a registered pattern to create registered Fourier N-dimensional pattern data, Is subjected to an N-dimensional discrete Fourier transform to generate collation Fourier N-dimensional pattern data, an amplitude component is extracted from individual data constituting the registered Fourier N-dimensional pattern data, and subjected to an amplitude compression process. An amplitude component is extracted from individual data constituting the N-dimensional pattern data, subjected to amplitude compression processing, and a correlation distance between a registered pattern and a verification pattern is determined from the data subjected to the amplitude compression processing. The registered pattern and the matching pattern are compared based on the correlation distance. If not determined to be the same, the registered Fourier N-dimensional pattern data and the collated Fourier N-dimensional pattern data are combined, the resultant Fourier N-dimensional pattern data obtained is subjected to amplitude suppression processing, and then N-dimensional discrete One of the Fourier transform and the N-dimensional discrete inverse Fourier transform is performed, and the correlation of each data constituting the N-dimensional pattern data of the correlation component area appearing in the synthesized Fourier N-dimensional pattern data subjected to the Fourier transform is performed. The registered pattern is compared with the matching pattern based on the component strength.

According to the present invention, first, an amplitude component is extracted from each data constituting the registered Fourier N-dimensional pattern data, and subjected to amplitude compression processing such as log processing or √ processing. Amplitude components are extracted from individual data constituting the N-dimensional pattern data, subjected to amplitude compression processing such as log processing and √ processing, and a registered pattern and a matching pattern are obtained from the data subjected to the amplitude compression processing. (For example, a Euclidean distance) is obtained, and the registered pattern is compared with the matching pattern based on the obtained correlation distance. As a result of this collation, if there is no match, the N of the correlation component area
Matching between the registered pattern and the matching pattern is performed based on the intensity of the correlation component for each piece of data constituting the dimensional pattern data.

A fourth invention (an invention according to claim 4) provides registered Fourier N-dimensional pattern data by performing N-dimensional discrete Fourier transform on the N-dimensional pattern data of the registered pattern and then performing amplitude suppression processing. After performing N-dimensional discrete Fourier transform on the N-dimensional pattern data of the collation pattern and performing amplitude suppression processing, collation Fourier N-dimensional pattern data is created, and N of the registered pattern subjected to N-dimensional discrete Fourier transform is obtained. An amplitude component is extracted from individual data constituting the dimensional pattern data and subjected to an amplitude compression process. The amplitude component is extracted from the individual data constituting the N-dimensional pattern data of the collation pattern subjected to the N-dimensional discrete Fourier transform. The components are extracted, subjected to amplitude compression processing, and compared with registered patterns from the data subjected to the amplitude compression processing. The correlation distance with the turn is determined, and the registered pattern and the matching pattern are compared based on the determined correlation distance. If the registered pattern and the matching pattern are not determined to be the same, the registered Fourier N-dimensional pattern data And the collated Fourier N-dimensional pattern data, and one of an N-dimensional discrete Fourier transform and an N-dimensional discrete inverse Fourier transform is applied to the synthesized Fourier N-dimensional pattern data obtained thereby. The registered pattern and the matching pattern are collated based on the intensity of the correlation component of each piece of data constituting the N-dimensional pattern data of the correlation component area appearing in the applied synthesized Fourier N-dimensional pattern data. is there.

According to the present invention, first, individual N-dimensional pattern data (registered Fourier N-dimensional pattern data before being subjected to the amplitude suppression processing) of the registered pattern subjected to the N-dimensional discrete Fourier transform is registered. Is subjected to amplitude compression processing such as log processing or √ processing, and N-dimensional pattern data (amplitude suppression processing is applied to the matching pattern subjected to N-dimensional discrete Fourier transform. The amplitude component is extracted from the individual data constituting the matching Fourier N-dimensional pattern data before being subjected to the amplitude compression processing such as log processing and √ processing, and the amplitude compression processing is performed on the data subjected to the amplitude compression processing. Correlation distance between registered pattern and matching pattern (for example, Euclidean distance)
Is obtained, and the registered pattern is compared with the matching pattern based on the obtained correlation distance. As a result of this collation, if there is no match, the collation between the registered pattern and the collation pattern is performed based on the intensity of the correlation component for each piece of data constituting the N-dimensional pattern data of the correlation component area.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail based on embodiments. FIG. 2 is a block diagram of a fingerprint matching device according to an embodiment of the present invention. In FIG. 1, reference numeral 10 denotes an operation unit, 20 denotes a control unit, and the operation unit 10 includes a numeric keypad 10-1 and a display (LCD).
A fingerprint sensor 10-3 is provided together with 10-2.
The fingerprint sensor 10-3 includes a light source 10-31 and a prism 10-
32, and a CCD camera 10-33. The control unit 20 includes a control unit 20-1 having a CPU,
ROM 20-2, RAM 20-3, hard disk (HD) 20-4, and frame memory (FM) 20-5
, An external connection unit (I / F) 20-6, and a Fourier transform unit (FFT) 20-7, and the ROM 20-2 stores a registration program and a collation program.

[Registration of Fingerprint] In this fingerprint collating apparatus, a user's fingerprint is registered as follows. That is,
Before operation, the user inputs the ID number assigned to the user using the numeric keypad 10-1 (step 301 shown in FIG. 3), and then enters the prism 1 of the fingerprint sensor 10-3.
Put your finger on 0-32. Light source 1 for prism 10-32
Light is irradiated from 0-31 and the prism 10-32
In the concave portion (valley line portion) of the fingerprint that does not contact the surface of
Light from -31 is totally reflected and reaches the CCD camera 10-33. Conversely, in the convex portion (ridge portion) of the fingerprint contacting the surface of the prism 10-32, the total reflection condition is broken, and the light source 10-31
Light from is scattered. As a result, a fingerprint pattern with contrast is obtained, in which the valley portion of the fingerprint is bright and the ridge portion is dark. The pattern of this collected fingerprint (registered fingerprint)
By the A / D conversion, the image is provided to the control unit 20 as a grayscale image (image data: two-dimensional pattern data) having 320 × 400 pixels and 256 gradations.

The control unit 20-1 stores the image data of the registered fingerprint given from the operation unit 10 in the frame memory 20-
5 (step 302), and a reduction process is performed on the captured registered fingerprint image data (step 303). This reduction processing is performed for 320 × 400 pixels, 2
Original image data of 56 gradations in the x direction (horizontal direction)
In the y direction (vertical direction), thinning is performed at a 3-pixel pitch except for the upper and lower ends by 8 pixels, except for the left and right ends, which are 32 pixels each. Thereby, the image data of the registered fingerprint becomes 6
The image data is reduced to image data of 4 × 128 pixels and 256 gradations (see FIG. 5).

The control unit 20-1 sends the reduced registered fingerprint image data (see FIG. 1A) to the Fourier transforming unit 20-7, and the two-dimensional discrete Fourier transform is added to the registered fingerprint image data. Perform a transform (DFT) (step 30)
4). As a result, the image data of the registered fingerprint shown in FIG. 1A becomes Fourier image data (registered Fourier image data) as shown in FIG. Control unit 20
-1, the Fourier image data is stored as a file in the hard disk 20-4 in correspondence with the ID number as original image data of the registered fingerprint (step 305).

The two-dimensional discrete Fourier transform is described in, for example, "Introduction to Computer Image Processing, edited by Japan Industrial Technology Center, published by Soken Shuppan Co., Ltd., pp. 44-45 (Document 1)". .

[Fingerprint Verification] In this fingerprint verification apparatus, verification of a user's fingerprint is performed as follows. That is, during operation, the user inputs the ID number assigned to the user using the numeric keypad 10-1 (step 401 shown in FIG. 4), and then enters the prism 10 of the fingerprint sensor 10-3.
Put your finger on -32. Thus, in the same manner as in the case of fingerprint registration, the pattern of the collected fingerprint (collation fingerprint) is 3
As a grayscale image (image data: two-dimensional pattern data) of 20 × 400 pixels and 256 gradations, the control unit 2
Given to 0.

When the ID number is given through the numeric keypad 10-1, the control unit 20-1
The Fourier image data of the registered fingerprint corresponding to the ID number is read from the registered fingerprint filed in Step 4 (Step 402). Further, the control unit 20-1 fetches the image data of the collation fingerprint given from the operation unit 10 via the frame memory 20-5 (step 403),
A reduction process similar to that performed in step 303 is performed on the captured image data of the verification fingerprint (step 40).
4). As a result, the image data of the verification fingerprint is 64 × 1
The image data is reduced to image data of 28 pixels and 256 gradations.

Then, the control unit 20-1 sends the reduced image data of the collation fingerprint (see FIG. 1E) to the Fourier transform unit 20-7, and adds the two-dimensional discrete Fourier transform to the image data of the collation fingerprint. Perform a transform (DFT) (Step 40)
5). Thus, the image data of the collation fingerprint shown in FIG. 1E becomes Fourier image data (collation Fourier image data) as shown in FIG.

Next, the control unit 20-1 executes step 405.
Fourier image data of the collation fingerprint obtained in
Is combined with the Fourier image data of the registered fingerprint read in step (step 406) to obtain combined Fourier image data.

Here, the synthesized Fourier image data is expressed as A · e ^j θ where the Fourier image data of the collation fingerprint is A · e ^j θ and A · B when the Fourier image data of the registered fingerprint is B · e ^j φ.
-It is expressed by e ^{j (} θ ^- φ ⁾ . However, A, B, θ, and φ are all functions of the frequency (Fourier) space (u, v).

[0024] ^{and, A · B · e j (} θ - φ) ^{is, A · B · e j (} θ - φ) = A · B · cos (θ-φ) + j · A · B · sin (θ- φ) (1) where A · e ^j θ = α ₁ + jβ ₁ , B · e ^j φ =
When _{α 2 + jβ 2, A =} (α 1 2 + β 1 2) 1/2, B =
_{^{(Α 2 2 + β 2 2}} ) 1/2, θ = tan -1 (β 1 / α 1), φ =
tan ^-1 (β ₂ / α ₂ ). By calculating this equation (1), synthetic Fourier image data is obtained.

A.B.e ^{j (} θ ^- φ ⁾ = A.B.e ^j
θ ・ e- ^j φ = A ・ e ^j θ ・ B ・ e- ^j φ = (α ₁ + j
β ₁ ) · (α ₂ −jβ ₂ ) = (α ₁ · α ₂ + β ₁ ·
The combined Fourier image data may be obtained as β ₂ ) + j (α ₂ β ₁ −α ₁ β ₂ ).

After obtaining the synthesized Fourier image data in this way, the control unit 20-1 performs an amplitude suppression process (step 407). In this embodiment, log processing is performed as amplitude suppression processing. That is, an arithmetic equation of synthesized Fourier image data described above ^{A · B · e j (θ} -
φ ⁾ is taken as log (A · B) · ej ⁽ θ ⁻ φ ⁾ , so that the amplitude A · B is log (A · B)
(A · B> log (A · B)).

FIG. 1D shows the synthesized Fourier image data after the amplitude suppression processing. In the synthesized Fourier image data subjected to the amplitude suppression processing, the influence of the illuminance difference between when the registered fingerprint is collected and when the verification fingerprint is collected is reduced. That is, by performing the amplitude suppression processing, the spectrum intensity of each pixel is suppressed, there is no outstanding value, and more information becomes effective. In addition, by performing amplitude suppression processing, feature points (end points, branch points) and ridge features (vortex, branch), which are personal information, are more emphasized in the fingerprint information, and ridges, which are general fingerprint information, are emphasized. The overall flow and direction can be suppressed.

In this embodiment, log processing is performed as amplitude suppression processing. However, Δ processing may be performed. In addition to log processing and √ processing,
Any processing may be used as long as the amplitude can be suppressed. If all the amplitudes are set to, for example, 1 in the amplitude suppression, that is, if only the phase is used, there is an advantage that the amount of calculation can be reduced and an amount of data can be reduced as compared with the log processing or the √ processing.

After performing the amplitude suppression processing in step 407, the control unit 20-1 sends the synthesized Fourier image data subjected to the amplitude suppression processing to the Fourier transformation unit 20-7.
A second two-dimensional discrete Fourier transform (DFT) is performed (step 408). Thus, the combined Fourier image data shown in FIG. 1D becomes the combined Fourier image data shown in FIG.

Then, the control unit 20-1 executes step 40
8. The combined Fourier image data obtained in Step 8 is fetched, the intensity (amplitude) of the correlation component of each pixel in a predetermined correlation component area is scanned from the combined Fourier image data, and a histogram of the intensity of the correlation component of each pixel is obtained. The upper n pixels (8 pixels in this embodiment) having a higher correlation component intensity are extracted from the histogram, and the average of the extracted correlation component intensity of the n pixels is obtained as a correlation value (score) (step 409). . Here, the correlation component area is shown in FIG.
With respect to the synthesized Fourier image data shown in (h), it is defined as a region S0 surrounded by a white dotted line. FIG. 6 shows a numerical example of the intensity of the correlation component of each pixel in a part of the correlation component area S0. In this figure, the values circled are the intensities of the correlation components of the top eight pixels.

Then, the control unit 20-1 executes step 40
The correlation value obtained in Step 9 is compared with a predetermined threshold value (Step 410).
It is determined that the registered fingerprint and the matching fingerprint match (step 4).
11), a message to that effect is displayed and the output for the electric lock is transmitted. If the correlation value is equal to or smaller than the threshold value, it is determined that the registered fingerprint and the verification fingerprint do not match (step 412).

Here, the threshold value to be compared with the correlation value is:
A total of 100 fingers obtained by inputting fingerprints of index fingers of 10 men and women in their 20s and 50s 10 times each as a sample are used for registration and collation, and are collated 10,000 times, and are obtained from the collation results. . In this case, a correlation value at which the other person exclusion rate becomes 100% is used as the threshold value. Note that the other person exclusion rate need not be 100%, but may be set to an arbitrary rate according to the purpose.

If it is determined in step 412 that the registered fingerprint and the verification fingerprint do not match, the control unit 20-1 extracts the amplitude as C from the individual data constituting the Fourier image data of the registered fingerprint read in step 402. (Fig. 7
Step 701 shown in FIG.
) To C ′ (step 702). Also,
The amplitude is extracted as D from the individual data constituting the Fourier image data of the collation fingerprint obtained in step 405 (step 703), and amplitude compression processing (log processing) is performed on the extracted data to obtain D '(step 704).

Then, the control unit 20-1 executes step 70
The Euclidean distance between the amplitude C ′ of each data subjected to the amplitude compression processing in step 2 and the amplitude D ′ of each data subjected to the amplitude compression processing in step 704 is expressed by U = (Σ (C′−
D ′) ² ) It is obtained as ^1/2 (step 705). Then, the obtained Euclidean distance (correlation distance) U is compared with a predetermined threshold value (step 706),
If the Euclidean distance U is equal to or less than the threshold value, it is determined that the registered fingerprint and the verification fingerprint match (step 411),
This is indicated and the output for the electric lock is transmitted.
If the Euclidean distance U is equal to or larger than the threshold value, it is determined that the registered fingerprint and the verification fingerprint do not match (step 70).
7) After displaying that effect, the process returns to step 401.

In this embodiment, step 70
Although the log processing is performed as the amplitude compression processing in 2,704, the Δ processing may be performed. Also, log
The processing is not limited to the processing and the 圧 縮 processing, and any processing may be used as long as the amplitude can be compressed. In this case, the concept of the amplitude compression process does not include a process of setting all amplitudes to a predetermined value.

Further, in this embodiment, the correlation distance between the registered fingerprint and the verification fingerprint is obtained as the Euclidean distance, but the correlation distance does not always have to be obtained as the Euclidean distance. For example, "weighted Euclidean distance", "Maharanobis distance", as shown in Document 2 ("Image Analysis Handbook", published by The University of Tokyo Press, supervised by Mikio Takagi and Hirohisa Shimoda, pages 653-665), "City distance", "chessboard distance", "Mink
"owski distance", "Chebyshev distance", etc. can be variations of the correlation distance.

Also, in this embodiment, the upper n pixels having a higher intensity of the correlation component are obtained from each pixel of the correlation component area S0.
Although the pixels are extracted and the average is used as the correlation value, the sum of the intensities of the correlation components of the upper n pixels may be used as the correlation value. Further, the intensities of the correlation components of all the pixels exceeding the threshold value may be added, and the added value may be used as the correlation value, or the average of the added values may be used as the correlation value. If at least one of the intensities of the correlation components of each pixel is equal to or greater than the threshold value, it may be determined that “match”.
If the number is equal to or more than one, various judgment methods such as judging “match” can be considered.

In this embodiment, the two-dimensional discrete Fourier transform is performed in the Fourier transform unit 20-7, but may be performed in the CPU 20-1. Further, in this embodiment, the reduction processing is performed on the image data of the registered fingerprint in step 303.
The reduction process may be performed at a stage after reading the Fourier image data of the registered fingerprint (between steps 402 and 403). Further, it is not always necessary to perform the reduction processing on the image data of the registered fingerprint or the collated fingerprint, and the Fourier image data may be created using the input image data as it is. If the reduction processing is performed, the capacity of the image memory used for processing the input image data can be reduced accordingly.

In this embodiment, the two-dimensional discrete Fourier transform is performed in step 408 shown in FIG. 4, but the two-dimensional discrete inverse Fourier transform is performed instead of the two-dimensional discrete Fourier transform. You may do so. That is, instead of performing two-dimensional discrete Fourier transform on the synthesized Fourier image data subjected to the amplitude suppression processing,
A dimensional discrete inverse Fourier transform may be performed. 2
The matching accuracy between the two-dimensional discrete Fourier transform and the two-dimensional discrete inverse Fourier transform does not change quantitatively. The two-dimensional discrete inverse Fourier transform is described in the aforementioned reference 1.

In this embodiment, the two-dimensional discrete Fourier transform is performed by performing amplitude suppression processing on the synthesized Fourier image data (step 40).
7, 408), the synthesis may be performed after the amplitude suppression processing is performed on the Fourier image data of the registered fingerprint and the collation fingerprint before synthesis. That is, as shown in FIG. 8A, a step 306 for performing amplitude suppression processing is provided between steps 304 and 305 in FIG. 3, and as shown in FIG. 8B, steps 406 and 407 in FIG. May be replaced.

In this case, the Fourier image data (registered Fourier image data) of the registered fingerprint subjected to the amplitude suppression processing as shown in FIG. 1C is obtained by the amplitude suppression processing in step 306. 406 and 407
As a result, Fourier image data (collation Fourier image data) of the collation fingerprint subjected to the amplitude suppression processing as shown in FIG. Then, the Fourier image data of the registered fingerprint and the verification fingerprint subjected to the amplitude suppression processing are combined, and combined Fourier image data as shown in FIG. 1D is obtained.

At this time, the rate of suppression of the amplitude of the combined Fourier image data is smaller than the case where the amplitude suppression processing is performed on the synthesized Fourier image data (FIG. 4). Therefore,
When the amplitude suppression processing is performed on the synthesized Fourier image data (FIG. 4), the matching accuracy is improved as compared with the method of performing amplitude suppression processing on the synthesized Fourier image data (FIG. 8). In the case where the synthesized Fourier image data is obtained after performing the amplitude suppression processing (FIG. 8), the synthesized Fourier image data is not subjected to the two-dimensional discrete Fourier transform, but to the two-dimensional discrete inverse Fourier transform. You may.

In this case, steps 701 and 702
Then, the amplitude is extracted as C from the individual data constituting the registered Fourier image data before the amplitude suppression processing is performed, and is subjected to the amplitude compression processing to obtain C ′. In steps 703 and 704, the amplitude is extracted as D from the individual data constituting the collated Fourier image data before the amplitude suppression processing is performed, and the extracted data is subjected to amplitude compression processing to obtain D ′.

In this embodiment, first, the registered pattern is compared with the collation pattern based on the intensity of the correlation component. If the collation results in a mismatch, the registered pattern is collated based on the Euclidean distance. Is compared with the matching pattern.First, the matching between the registered pattern and the matching pattern is performed based on the Euclidean distance, and if there is no match, the registration is performed based on the strength of the correlation component. The pattern may be compared with the matching pattern (third invention, fourth invention).

For reference, FIG. 9 shows each image of the fingerprint collation process when the collation fingerprint is another person, corresponding to FIG. FIG.
Are the images in the fingerprint collation process when the collation fingerprint is the principal, and the correlation component area S when the collation fingerprint is the principal.
A portion where the intensity of the correlation component is high occurs at 0, but does not occur when the verification fingerprint is another person.

In this embodiment, the case where fingerprint collation is performed has been described as an example. However, the present invention can be similarly applied to the case where voiceprint collation is performed, and can be handled as image data regardless of fingerprints and voiceprints. It can be used for collation of various two-dimensional patterns that can be performed. Further, the present invention is not limited to the two-dimensional pattern collation, and the same can be applied to the N-dimensional pattern collation such as a one-dimensional pattern or a three-dimensional pattern.

In this embodiment, the two-dimensional pattern is obtained as an image. However, it is not always necessary to obtain the two-dimensional pattern as an image. For example, a vibration detector is two-dimensionally arranged at each location, and a two-dimensional pattern (seismic wave) obtained by the two-dimensionally arranged vibration detector is used as a collation pattern, which is collated with a pattern registered in advance. You may do so. In addition, a flow measuring device is two-dimensionally arranged at each part, and a two-dimensional pattern (flow distribution) obtained by the two-dimensionally arranged flow measuring device is set as a collation pattern, and collated with a pre-registered pattern. You may make it.

[0048]

As apparent from the above description, according to the present invention, in the first and second inventions, first, the correlation component of each individual data constituting the N-dimensional pattern data of the correlation component area is obtained. Matching between the registered pattern and the matching pattern is performed based on the strength, and if the matching result indicates a mismatch, a correlation distance (for example, a Euclidean distance) between the registered pattern and the matching pattern is obtained. The comparison between the registered pattern and the matching pattern is performed based on the calculated correlation distance.
In the present invention, first, a correlation distance (for example, a Euclidean distance) between the registered pattern and the matching pattern is obtained, and the registered pattern is compared with the matching pattern based on the obtained correlation distance. In the case of non-coincidence, the matching between the registered pattern and the matching pattern is performed based on the strength of the correlation component for each piece of data constituting the N-dimensional pattern data in the correlation component area. The recognition rate can be further improved by utilizing the existing discrimination ability.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating a fingerprint matching process in a fingerprint matching device according to the present invention.

FIG. 2 is a block diagram of the fingerprint matching device.

FIG. 3 is a flowchart for explaining a fingerprint registration operation in the fingerprint collation apparatus.

FIG. 4 is a flowchart for explaining a fingerprint collation operation (collation operation using a correlation value) in the fingerprint collation device.

FIG. 5 is a diagram illustrating a reduction process on image data.

FIG. 6 is a diagram illustrating a numerical example of the intensity of the correlation component of each pixel in a part of the correlation component area.

FIG. 7 is a flowchart for explaining a fingerprint matching operation (matching operation using a correlation distance) in the fingerprint matching device.

FIG. 8 is a flowchart for explaining another example of the fingerprint registration operation and the fingerprint collation operation.

FIG. 9 is a diagram showing each image in a fingerprint matching process when the matching fingerprint is another person, corresponding to FIG. 1;

[Explanation of symbols]

10 operation section, 20 control section, 10-1 numeric keypad, 10-2 display (LCD), 10-3 ...
Fingerprint sensor, 10-31: light source, 10-32, prism, 10-33: CCD camera, 20-1: control unit, 2
0-2 ROM, 20-3 RAM, 20-4 Hard disk (HD), 20-5 Frame memory (F
M), 20-6: external connection unit (I / F), 20-7: Fourier transform unit (FFT).

Claims (4)

[Claims] A registered Fourier pattern data generating means for performing N-dimensional discrete Fourier transform on N-dimensional pattern data of a registered pattern to generate registered Fourier N-dimensional pattern data; Pattern Fourier pattern data creating means for creating a matching Fourier N-dimensional pattern data by performing a dynamic Fourier transform, and synthesizing the registered Fourier N-dimensional pattern data and the matching Fourier N-dimensional pattern data to obtain a combined Fourier N Pattern processing means for performing one of N-dimensional discrete Fourier transform and N-dimensional discrete inverse Fourier transform after performing amplitude suppression processing on the dimensional pattern data; and synthesizing which is subjected to Fourier transform by the pattern processing means. Appears in Fourier N-dimensional pattern data First pattern matching means for matching the registered pattern with the matching pattern based on the intensity of the correlation component for each piece of data constituting the N-dimensional pattern data of the correlation component area; and the first pattern matching means. If the registered pattern is not determined to be the same as the collation pattern, an amplitude component is extracted from the individual data constituting the registered Fourier N-dimensional pattern data, and subjected to an amplitude compression process. An amplitude component is extracted from individual data constituting the N-dimensional pattern data, subjected to amplitude compression processing, and a correlation distance between the registered pattern and the verification pattern is obtained from the data subjected to the amplitude compression processing. Second pattern matching means for matching the registered pattern with a matching pattern based on the obtained correlation distance. A pattern matching device characterized by the following. 2. Registered Fourier pattern data creating means for creating registered Fourier N-dimensional pattern data by applying N-dimensional discrete Fourier transform to N-dimensional pattern data of the registered pattern and then performing amplitude suppression processing; Matching Fourier pattern data creating means for creating matching Fourier N-dimensional pattern data by performing N-dimensional discrete Fourier transformation on the N-dimensional pattern data and then performing amplitude suppression processing; the registered Fourier N-dimensional pattern data and the matching Fourier Pattern processing means for synthesizing the N-dimensional pattern data and performing one of N-dimensional discrete Fourier transform and N-dimensional discrete inverse Fourier transform on the synthesized Fourier N-dimensional pattern data obtained therefrom; By means of the Fourier transform First pattern matching means for matching the registered pattern with the matching pattern based on the intensity of the correlation component for each piece of data constituting the N-dimensional pattern data of the correlation component area appearing in the composed Fourier N-dimensional pattern data If the first pattern matching means does not determine that the registered pattern and the matching pattern are the same,
An amplitude component is extracted from individual data constituting the N-dimensional pattern data of the registered pattern subjected to the dimensional discrete Fourier transform, subjected to an amplitude compression process, and subjected to the N-dimensional discrete Fourier transform. An amplitude component is extracted from individual data constituting the N-dimensional pattern data of the collation pattern, subjected to an amplitude compression process, and a correlation distance between the registration pattern and the collation pattern is calculated from the data subjected to the amplitude compression process. And a second pattern matching means for comparing the registered pattern with a matching pattern based on the determined correlation distance. 3. A registered Fourier pattern data generating means for performing N-dimensional discrete Fourier transform on N-dimensional pattern data of a registered pattern to generate registered Fourier N-dimensional pattern data; Matching Fourier pattern data creating means for creating a matching Fourier N-dimensional pattern data by performing a dynamic Fourier transform, and extracting an amplitude component from each data constituting the registered Fourier N-dimensional pattern data and performing an amplitude compression process on the extracted amplitude component. Further, an amplitude component is extracted from individual data constituting the matching Fourier N-dimensional pattern data, subjected to amplitude compression processing, and a correlation between the registered pattern and the matching pattern is obtained from the data subjected to the amplitude compression processing. A distance is obtained, and the registered pattern is referred to based on the obtained correlation distance. A first pattern matching unit that performs matching with a pattern; and if the first pattern matching unit does not determine that the registered pattern and the matching pattern are the same, the registered Fourier N-dimensional pattern data and the matching Fourier N-dimensional pattern data is synthesized, the resultant Fourier N-dimensional pattern data obtained is subjected to amplitude suppression processing, and then N-dimensional discrete Fourier transform and N
One of the two-dimensional discrete inverse Fourier transform is performed, and the intensity of the correlation component for each data constituting the N-dimensional pattern data of the correlation component area appearing in the synthesized Fourier N-dimensional pattern data subjected to the Fourier transform is calculated. A second step of comparing the registered pattern with the matching pattern based on
And a pattern matching unit. 4. Registered Fourier pattern data creating means for creating registered Fourier N-dimensional pattern data by performing N-dimensional discrete Fourier transform on the N-dimensional pattern data of the registered pattern and then performing amplitude suppression processing; Matching Fourier pattern data creating means for creating matching Fourier N-dimensional pattern data by performing N-dimensional discrete Fourier transform on the N-dimensional pattern data and then performing amplitude suppression processing; An amplitude component is extracted from individual data constituting the N-dimensional pattern data of the registered pattern, subjected to an amplitude compression process, and the N-dimensional of the collation pattern subjected to the N-dimensional discrete Fourier transform is obtained.
An amplitude component is extracted from individual data constituting the dimensional pattern data, subjected to an amplitude compression process, and a correlation distance between the registered pattern and the verification pattern is obtained from the data subjected to the amplitude compression process. A first pattern matching unit that compares the registered pattern with a matching pattern based on the correlation distance, and if the registered pattern and the matching pattern are not determined to be the same by the first pattern matching unit, The registered Fourier N-dimensional pattern data and the matching Fourier N-dimensional pattern data are combined, and any one of an N-dimensional discrete Fourier transform and an N-dimensional discrete inverse Fourier transform is performed on the combined Fourier N-dimensional pattern data obtained thereby. One is applied, and appears in the synthesized Fourier N-dimensional pattern data subjected to the Fourier transform. A second pattern matching unit that matches the registered pattern with the matching pattern based on the intensity of the correlation component for each piece of data constituting the N-dimensional pattern data of the correlation component area. Pattern matching device.