JP2007202912A - Collating device for image including striped pattern - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a collation device for collation of an image including a striped pattern, capable of achieving high collation precision by creating a noise-eliminated image for use while maintaining information of characteristic points to be used for collation. <P>SOLUTION: The collation device 1 comprises an image input part 3 to obtain an image including a striped pattern, a frequency spectrum analyzing part 120 to determine the frequency spectrum of the image, a frequency component selecting part 140 to select frequency components of which absolute value of amplitude is over a prescribed threshold value, an image evaluation part 150 to check if the selected frequency component meets the image quality suitable to collation in accordance with prescribed conditions, an image recomposing part 190 to recompose an image based on the selected frequency component in a case where it is determined that the prescribed condition is met, and a collation part 14 to carry out collation using the recomposed image. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a collation device for an image including a striped pattern, and more particularly to a collation device that collates after improving the image quality of a fingerprint image or a palm print image.

  In recent years, there has been a growing interest in security in entry / exit management, access to information equipment, withdrawal of deposits and savings, and the like. Under such circumstances, biometrics authentication (biometric authentication) has attracted particular attention as a personal authentication method that can provide higher security with more certainty than conventional key and password authentication.

  In fingerprint authentication, which is one of such biometric authentications, feature information is extracted from a multi-valued image representing a fingerprint pattern obtained from a fingerprint sensor in shades, and the feature information and pre-registered feature information are extracted. A method is known in which the degree of coincidence is obtained by comparison and personal authentication is performed based on the degree of coincidence.

  As one of the methods for extracting feature information from a fingerprint image, the multi-valued image obtained from the fingerprint sensor is subjected to processing such as noise removal and unevenness correction to correct the image quality, and then the binarized image is used to correct the fingerprint. There is a method for obtaining branch points and end points of ridges as feature points.

  However, noise occurs in the fingerprint image due to various conditions at the time of fingerprint image shooting, for example, the positional relationship between the finger placement on the fingerprint reader and the illumination light source, and the state of the finger such as dryness. Therefore, even though the ridge line is originally a continuous line, it may be observed on the fingerprint image as if it is interrupted or two adjacent ridge lines are connected in the middle. . Such discontinuity or connection caused by noise is extracted as a pseudo feature point and causes the accuracy of fingerprint collation to be reduced. Therefore, in order to improve the accuracy of fingerprint collation, it is important to remove noise from the fingerprint image as much as possible before extracting feature points.

  For this purpose, the image processing apparatus described in Patent Document 1 divides an input image into small regions and performs a Fourier transform for each small region to obtain a frequency spectrum. Then, the image processing apparatus searches for a frequency component having the maximum amplitude for each small region, and performs a filter process that decreases the amplitude as the frequency component is separated from the frequency component having the maximum amplitude as a center. . Thereafter, the image processing apparatus attenuates noise by leaving only the frequency component corresponding to the pitch of the ridges of the fingerprint as it is by performing inverse Fourier transform for each small region and synthesizing the filtered image. An image can be obtained.

  However, if the ridge end point or branch point is included in the small area, the frequency component due to the end point or branch point is usually different from the frequency component having the maximum amplitude. If the filter processing is performed to uniformly reduce the frequency component with the maximum amplitude, the information on the end points and branch points of the ridges may be lost.

JP 2005-115548 A

  In view of the above problems, the object of the present invention is to generate and use an image from which noise has been removed while maintaining information on feature points used for matching for an image including a striped pattern, thereby achieving high matching accuracy. It is in providing the collation apparatus obtained.

  Another object of the present invention is to accurately extract information on feature points used for matching, such as fingerprint ridge end points and branch points, for fingerprint images and palmprint images, and high matching. An object of the present invention is to provide a fingerprint or palmprint image collation device capable of obtaining accuracy.

In view of the above problems, an image collation apparatus for collating an image including a striped pattern according to the present invention includes an image input unit that acquires an image, a frequency spectrum analysis unit that obtains a frequency spectrum by performing frequency conversion on the image, and a frequency A frequency component selection unit that selects a frequency component having an amplitude whose absolute value is equal to or greater than a predetermined threshold from the spectrum, and whether or not the frequency component selected by the frequency component selection unit satisfies image quality suitable for collation according to a predetermined condition. An image reconstructing unit for reconstructing an image by performing an inverse transform of the frequency transform based on the selected frequency component when it is determined that the image quality evaluating unit satisfies a predetermined condition, And a collation unit that collates using the reconstructed image. With such a configuration, even when noise is included in an image, matching can be performed after appropriate noise removal from the image, and high matching accuracy can be achieved.
In the present invention, a striped pattern refers to a pattern in which pixel values fluctuate periodically in a certain direction. However, it does not matter whether the cycle itself fluctuates or not.

The collation device according to the present invention further includes a frequency component selection control unit that adjusts the predetermined threshold value. When the image quality evaluation unit determines that the predetermined condition is not satisfied, the frequency component selection control unit sets the predetermined threshold value. It is preferable that the frequency component selection unit performs selection of the frequency component again using the changed predetermined threshold. When it is determined that the selection of the frequency component is inappropriate, it is possible to select an appropriate frequency component by changing the conditions again and reselecting the frequency component.
In the verification device according to the present invention, when the predetermined condition is not satisfied, the frequency component selection control unit preferably sets a larger value than before changing the predetermined threshold.

  Further, in the collation device according to the present invention, the image quality evaluation unit includes a direction line calculation unit that specifies the stripe direction of the stripe pattern included in the image based on the distribution of the selected frequency component, and the selected frequency component is a stripe. When there is a concentration degree calculation means for calculating an index indicating the degree of concentration in the direction, and the index indicates a concentration that is greater than or equal to a predetermined value, satisfies a predetermined condition, and indicates a concentration that is less than a predetermined value It is preferable to have an evaluation means for evaluating that the predetermined condition is not satisfied. The striped direction of the striped pattern means a direction orthogonal to the striped pattern, that is, a direction in which the change in shade of the image is the shortest cycle. In addition, an index indicating the degree of concentration of the selected frequency component in the fringe direction is used as an average distance between each frequency component and a straight line on the frequency plane corresponding to the fringe direction, and the number of frequency components existing in the vicinity of the straight line. And the ratio of the number of frequency components existing at positions away from the straight line, or a combination thereof. With such a configuration, an image including a stripe pattern is subjected to image quality evaluation based on the presence of a strong frequency component in the stripe direction, and therefore it is possible to accurately determine whether or not the frequency component is appropriately selected. it can.

The collation device according to the present invention further includes frequencies that are not included in a range within a predetermined frequency from a straight line on a frequency plane corresponding to the fringe direction with respect to a frequency component determined by the image quality evaluation unit to satisfy a predetermined condition. It is preferable to have a frequency filter unit that performs band-pass filter processing that attenuates components, and the image reconstruction unit reconstructs an image based on the frequency components that have undergone band-pass filter processing.
By performing the bandpass filter process, it is possible to remove frequency components that are not related to the object included in the image, and to efficiently remove noise from the image. The predetermined frequency range is preferably set so as to include a frequency fluctuation amount corresponding to an angle at which the stripe direction of the stripe pattern changes based on the structure of the object included in the image. Since the fluctuation component of the pixel value can be left in consideration of the fluctuation in the fringe direction, the risk of losing information on the object by mistake can be reduced.

  Furthermore, the collation device according to the present invention includes an image dividing unit that divides an image into a plurality of regions, and an image combining unit that combines the images divided into the plurality of regions into one image, The image input to the image input unit is divided into multiple regions that can be regarded as a plurality of straight lines with stripe patterns substantially parallel, and the frequency spectrum analysis unit obtains the frequency spectrum for each of the multiple regions and selects the frequency component The unit selects a frequency component whose absolute value of the amplitude is equal to or greater than a predetermined threshold for each of the plurality of regions, and the image quality evaluation unit determines whether the selected frequency component satisfies a predetermined condition for each of the plurality of regions. If the image reconstruction unit determines that the selected frequency component satisfies the predetermined condition for each of the plurality of regions, the image reconstruction unit reconstructs the image based on the selected frequency component. And image combining unit, it is preferable to synthesize the reconstructed image into one. With such a configuration, even if the stripe direction of the stripe pattern included in the image changes depending on the location, the range in which the stripe direction faces the same direction is defined as each region, and each region changes in the stripe direction. Can be treated as an image that does not contain. Therefore, noise can be appropriately removed from the image.

  According to the present invention, it is possible to provide a collation device that can obtain high collation accuracy by generating and using an image from which noise is removed while maintaining information on feature points used for collation for an image including a striped pattern. It became possible.

  Further, according to the present invention, it is possible to accurately extract feature point information used for matching, such as fingerprint ridge end points and branch points, for fingerprint images and palmprint images, and to achieve high matching accuracy. It has become possible to provide a fingerprint or palmprint image collation device obtained.

Hereinafter, a fingerprint authentication apparatus which is an example of a collation apparatus according to the present invention will be described with reference to the drawings.
The fingerprint authentication device, which is an example of the collation device according to the present invention, creates a feature point list obtained by extracting the end points and branch points of the fingerprint ridges as feature points from the acquired fingerprint image, and is stored in advance. It is an apparatus that performs authentication by collating with a feature point list of a person's fingerprint.

FIG. 1 is a functional block diagram of a fingerprint authentication device 1 which is an example of a collation device according to the present invention. As shown in FIG. 1, the fingerprint authentication device 1 includes an operation / display unit 2, a fingerprint input unit 3, a storage unit 4, an image collation unit 5, and an output unit 6. Then, the fingerprint authentication device 1 displays operation guidance and information for an operator (a person to be verified) who performs fingerprint authentication in the operation / display unit 2, and from the fingers placed on the fingerprint input unit 3 according to the guidance. The value fingerprint image M is read. Then, the read multi-value fingerprint image M is sent to the image matching unit 5. The image matching unit 5 performs noise removal and binarization on the received multi-value fingerprint image M, extracts feature points, and creates a feature point list (hereinafter referred to as input feature point list) F _i . . Then, the image collation unit 5 collates the input feature point list F _i with a feature point list (hereinafter referred to as a registered feature point list) F _r registered in advance in the storage unit 4 for the operator. The collation result is output to an external device through the output unit 6. For example, an electric lock locking / unlocking control device is connected to the output unit 6. Then, if the collation result received from the output unit 6 is a signal indicating that the authentication is successful, the electric lock locking / unlocking control device unlocks the electric lock, and conversely, a signal indicating that the authentication has failed. If so, keep the electric lock locked.

  The fingerprint authentication device 1 analyzes the frequency component of the read multi-value fingerprint image M for each small region such that the ridges can be regarded as a plurality of straight lines that are substantially parallel to each other, and the frequency representing the ridge stripes By extracting only the component and the frequency component close to the frequency component and reconstructing the image, noise is removed while maintaining the information on the end points and branch points of the ridges. The fingerprint authentication device 1 extracts feature points based on the multi-value fingerprint image M ′ from which noise has been removed and uses it for matching, thereby suppressing the extraction of pseudo feature points and feature point extraction failures and high matching. Accuracy can be achieved.

Hereinafter, each part will be described in detail.
The operation / display unit 2 is used by an operator of the fingerprint authentication apparatus 1, that is, a person to be verified, to input an identification number IN and select an operation to be performed (for example, entering a specific room). . The operation / display unit 2 displays an operation guidance or gives a voice instruction, and includes a touch panel display and a speaker. Further, instead of the touch panel, an input device such as a keyboard or a mouse and a simple display device such as a liquid crystal display may be used. Note that the speaker may be omitted when the guidance voice instruction is not performed.
Data such as the identification number IN input by the operation / display unit 2 is used for specifying registered fingerprint data such as the registered feature point list _Fr to be called by the image matching unit 5.

  The fingerprint input unit 3 functions as an image input unit, and acquires a fingerprint image used for collation processing. The fingerprint input unit 3 includes a fingerprint sensor in which a CCD camera that images a placed finger and converts it into a digital signal is modularized, a placement sensor that detects placement of the finger on the fingerprint sensor, and illuminates the finger during imaging LED having a finger guide member for causing the operator to correctly recognize the placement position of the finger. In the present embodiment, a finger internal characteristic detection type optical sensor is used as the fingerprint sensor. However, the fingerprint sensor that can be used in the present invention is not limited to this. For example, as a fingerprint sensor, an optical sensor such as a total reflection optical type, an optical path separation optical type, a surface projection irregular reflection type, a non-optical sensor such as a capacitance type, an electric field type, a pressure sensitive type, an ultrasonic type, etc. May be used.

  In the fingerprint input unit 3, when the placement sensor detects that a finger is placed on the fingerprint sensor, the fingerprint sensor images the finger and generates a multi-value fingerprint image M. In the present embodiment, each pixel of the multi-value fingerprint image M is represented by 256 gradations of 0-255, the fingerprint ridge region has a low pixel value (or black), and the fingerprint valley region has a high pixel value. (Or white) is expressed. The pixel value here corresponds to a luminance value in each pixel on the image, and a pixel with a low pixel value is expressed with low luminance and a pixel with a high pixel value is expressed with high luminance. Further, the multi-value fingerprint image M is acquired with a resolution of 500 dpi. However, the multi-value fingerprint image usable in the present invention is not limited to the above. The resolution may be different as long as the ridge stripe structure, particularly the branch point and end point can be confirmed on the image, and the number of gradations is not limited to 256.

The storage unit 4 stores an operator identification number IN registered in advance and a registered feature point list _Fr in association with each other, and includes a nonvolatile memory such as a flash memory, a magnetic recording medium such as a hard disk, a CD-ROM, An optical recording medium such as a DVD-R / W is included.
The registered feature point list F _r is a feature point list created from a fingerprint image of a person to be registered that has been registered in advance, and is a list of the positions of feature points such as ridge branch points and end points. The registered feature point list F _r is created by the same method as the creation of the input feature point list F _i by the image matching unit 5 described later. Details of the feature point list and the creation method will be described later.
The storage unit 4 also stores programs used by the image matching unit 5, various setting files, parameters, and the like. These are read by the image collation unit 5 as necessary at a predetermined timing, such as when the fingerprint authentication device 1 is activated.

The image collation unit 5 performs fingerprint collation based on the multi-valued fingerprint image M acquired from the fingerprint input unit 3 and the registered feature point list F _r acquired from the storage unit 4, and includes a central processing unit (CPU), It is composed of a numerical processor, a semiconductor memory such as ROM or RAM, and the like. The image collation unit 5 executes a predetermined operation according to the program read from the storage unit 4. Furthermore, the image collation unit 5 is connected to the operation / display unit 2, the fingerprint input unit 3, the storage unit 4, and the output unit 6, and performs control by outputting predetermined control signals to these units. Therefore, the image collation unit 5 includes an image correction unit 10, a binarization unit 11, a thinning unit 12, a feature extraction unit 13, a collation unit 14, and a control unit 15 that controls each unit.

When the operator identification number IN is input from the operation / display unit 2, the image collation unit 5 searches whether or not the identification number IN (hereinafter referred to as an input identification number) is registered in the storage unit 4. If the matching identification number IN is registered, the image matching unit 5 reads the registered feature point list F _r associated with the input identification number IN from the storage unit 4. Further, the image correction unit 10 performs noise removal processing on the multi-value fingerprint image M acquired from the fingerprint input unit 3. The binarization unit 11 generates a binary fingerprint image B binarized into a ridge region and a valley region based on the multi-value fingerprint image M ′ subjected to the noise removal processing. Further thinning unit 12 performs the thinning of the ridge region with respect to the binary fingerprint image B, and generates a thinned binary fingerprint image B _T. The feature extraction unit 13, the branch point of the ridge from thinning the binary fingerprint image B _T, such as the feature point extracted based on the endpoint, create an input feature point list F _i. When the input feature point list F _i is created, the matching unit 14 performs matching based on the input feature point list F _i and the registered feature point list F _r . The collation unit 14 compares both feature point lists and calculates the similarity S _i . If the similarity S _i is greater than a predetermined threshold Thd _v , the collation unit 14 determines that the authentication has succeeded, and determines that the authentication has failed if the similarity S _i is equal to or less than the threshold Thd _v . Finally, the image collation unit 5 displays a collation result on whether or not the authentication is successful on the operation / display unit 2, and sends a signal indicating the collation result to an external device such as an electric lock control device via the output unit 6. Output.

Hereinafter, each part of the image collation part 5 is demonstrated in detail.
The control unit 15 controls each part of the operation / display unit 2, the fingerprint input unit 3, the storage unit 4, and the output unit 6 according to the program read from the storage unit 4 and the input signal from the operation / display unit 2. Deliver various data such as images. The control unit 15 also controls processing by the image correction unit 10, the binarization unit 11, the thinning unit 12, the feature extraction unit 13, and the verification unit 14 in the image verification unit 5.

  The image correction unit 10 performs noise removal processing on the multi-value fingerprint image M. For this purpose, the image correction unit 10 analyzes the frequency component of the multi-value fingerprint image M by dividing the multi-value fingerprint image M into small regions that can be regarded as a plurality of straight lines that are substantially parallel to each other. The direction is orthogonal to the same. Then, the image correcting unit 10 extracts only the frequency component representing the ridge stripe along the stripe direction and a specific frequency component close to the frequency component, and reconstructs the image, whereby the ridge end points and branch points are obtained. The noise is removed while maintaining the information.

FIG. 2 shows a functional block diagram of the image correction unit 10.
The image correction unit 10 includes an image division unit 110, a frequency spectrum analysis unit 120, a spectrum storage unit 130, a frequency component selection unit 140, an image quality evaluation unit 150, a frequency component selection control unit 160, a selection parameter storage unit 170, and a frequency filter unit 180. The image reconstruction unit 190 and the fingerprint image composition unit 200 are included. Then, the image dividing unit 110 divides the multi-value fingerprint image M into a plurality of small areas. The frequency spectrum analysis unit 120 calculates a frequency spectrum image for each small region of the multi-value fingerprint image M by performing Fourier transform for each small region. The spectrum storage unit 130 temporarily stores the calculated frequency spectrum image for each small region. The frequency component selection unit 140 selects only frequency components having an amplitude of a certain level or more for each frequency spectrum image, and generates a selected spectrum image. The image quality evaluation unit 150 identifies the ridge stripe direction based on the selected spectrum image. Then, the image quality evaluation unit 150 evaluates whether or not the frequency component selection unit 140 is appropriately selected based on an index representing the degree of concentration of the selected frequency component with respect to the stripe direction. When the image quality evaluation unit 150 determines that the selection of the frequency component is inappropriate, the frequency component selection control unit 160 adjusts the frequency component selection criterion and causes the frequency component selection unit 140 to generate the selected spectrum image again. Further, the frequency component selection criterion is stored in the selection parameter storage unit 170 to be used for re-adjustment. On the other hand, when the image quality evaluation unit 150 determines that the selection of the frequency component is appropriate, the frequency filter unit 180 determines the frequency characteristics of the ridge end points and branch points for each selected spectrum image based on the ridge stripe direction. A band-pass filter process is performed so as to leave a frequency component in a predetermined frequency band determined in consideration of the above and attenuate others. The image reconstruction unit 190 reconstructs a fingerprint image for each small region by performing inverse Fourier transform on each selected spectrum image that has been bandpass filtered by the frequency filter unit 180. Finally, the fingerprint image synthesizing unit 200 synthesizes the fingerprint images for each small area restored by the image reconstruction unit 190 to obtain a multi-value fingerprint image M ′ from which noise has been removed.

  As described above, the image correction unit 10 selects a frequency component having an amplitude of a certain level or more for each frequency spectrum image by the frequency component selection unit 140, and further increases the frequency filter unit 180 for the selected frequency component. By leaving the frequency components included in the predetermined frequency band centered on the stripe direction of the line, it is possible to appropriately remove noise while maintaining information on the end points and branch points of the ridges. Furthermore, the image correction unit 10 is based on the fact that the frequency components should concentrate only in a predetermined direction because the ridge of the fingerprint image is a striped pattern in which a plurality of substantially parallel straight lines are arranged in a small area. It is possible to select an appropriate frequency component by evaluating whether or not the frequency component once selected is appropriate and adjusting the selection criteria and selecting the frequency component again depending on the evaluation result.

Hereinafter, each part of the image correction unit 10 will be described in detail.
The image dividing unit 110 has a plurality of small regions L _i (i = 1, 2,..., M) (where m = 1, 2, and p) of the multi-value fingerprint image M acquired from the fingerprint input unit 3. = P × q). Each small area L _i, for example each be a rectangular region, dividing the multivalued fingerprint image M in a grid pattern. Each small region _Li is sized so that the ridges can be observed as a plurality of substantially parallel straight lines in order to facilitate the identification of frequency components in the ridge stripe direction in the subsequent processing.

FIG. 3 shows a state in which a small area _Li is set on the multi-value fingerprint image M. An image 300 shown in FIG. 3 is an example of the multi-value fingerprint image M. In FIG. 3, the portion that appears relatively black is a ridge. The rectangular areas 310, 320, and 330 shown on the multi-value fingerprint image M correspond to the small areas L _i−1 , L _i , and L _{i + 1} . In FIG. 3, for the sake of clarity of explanation, a small region is illustrated only in a partial region of the multi-value fingerprint image M. However, the multi-value fingerprint image M is an arbitrary portion of any small region. The whole is divided so as to be included in the region. In the present embodiment, the resolution with respect to the multi-level fingerprint image M of 500 dpi, and the respective small areas _{L i} and 32 × 32 pixels. Also as shown in FIG. 3, each small area L _i may be set to have a region overlapping with adjacent small regions. By providing such an overlapping area, when the image of each small area is combined in the fingerprint image combining unit 200 described later, an unnatural combination that generates a discontinuous portion at the boundary with the adjacent small area is performed. Can be avoided. For example, such an overlapping region can be 5 pixels wide.
In the following description, an image of L _i in each small area is denoted as f _i (x, y). Here, x is a coordinate value in the horizontal direction, and y is a coordinate value in the vertical direction. The output value of f _i (x, y) represents the pixel value of the pixel at coordinates (x, y).

なお、周波数スペクトル解析部１２０では、式（１）にしたがった周波数スペクトル画像Ｆ_ｉ（ｕ，ｖ）の算出を、高速フーリエ変換（ＦＦＴ）アルゴリズムを用いたソフトウェアで実行することができる。また、周波数スペクトル解析部１２０は、周波数スペクトル画像Ｆ_ｉ（ｕ，ｖ）を算出するために、フーリエ変換以外の周波数変換関数を使用することができる。周波数スペクトル解析部１２０は、そのような周波数変換関数として、例えば離散コサイン変換、離散ウェーブレット変換、アダマール変換などを使用することができる。
本実施形態では、各小領域Ｌ_ｉの周波数スペクトル画像Ｆ_ｉ（ｕ，ｖ）を３２×３２画素の画像として算出した。また算出された各周波数スペクトル画像Ｆ_ｉ（ｕ，ｖ）は、スペクトル記憶部１３０及び周波数成分選択部１４０に送られる。 The frequency spectrum analysis unit 120 performs frequency conversion on the image f _i (x, y) of each small region L _i of the multi-value fingerprint image M received from the image division unit 110, and the frequency spectrum image F _i (u, v) Calculate (i = 1, 2,..., m). Here, u represents a frequency in the horizontal direction, and v represents a frequency in the vertical direction. The output value of F _i (u, v) represents the amplitude of the frequency component of the frequency (u, v). In order to calculate the frequency spectrum image F _i (u, v), the frequency spectrum analysis unit 120 can use, for example, an N × N point two-dimensional discrete Fourier transform represented by the following formula (1).

In the frequency spectrum analysis unit 120, the calculation of the frequency spectrum image F _i (u, v) according to the equation (1) can be executed by software using a fast Fourier transform (FFT) algorithm. In addition, the frequency spectrum analysis unit 120 can use a frequency conversion function other than the Fourier transform in order to calculate the frequency spectrum image F _i (u, v). The frequency spectrum analysis unit 120 can use, for example, a discrete cosine transform, a discrete wavelet transform, a Hadamard transform, or the like as such a frequency transform function.
In the present embodiment, the frequency spectrum image F _i (u, v) of each small region L _i is calculated as an image of 32 × 32 pixels. Each calculated frequency spectrum image F _i (u, v) is sent to the spectrum storage unit 130 and the frequency component selection unit 140.

The spectrum storage unit 130 is a memory area secured in the RAM of the image matching unit 5 and stores each frequency spectrum image F _i (u, v) calculated by the frequency spectrum analysis unit 120. Then, the frequency spectrum images F _i (u, v) are output to the frequency component selection unit 140 as necessary. Further, as will be described later, a selected spectrum image F _si (u, v) in which a frequency component having a predetermined amplitude or more is selected from each frequency spectrum image F _i (u, v) is also stored.
Note that the storage unit 4 may be used instead of the spectrum storage unit 130 and the spectrum storage unit 130 may be omitted.

For each frequency spectrum image F _i (u, v) acquired from the spectrum storage unit 130, the frequency component selection unit 140 selects only the frequency component whose amplitude has an absolute value greater than or equal to a predetermined value, and selects the selected spectrum image F _si. (U, v) is generated. The frequency component selection unit 140 calculates a maximum amplitude component F _imax (u, v) of the spectrum for each frequency spectrum image F _i (u, v) in order to calculate a selection criterion. A value obtained by multiplying the maximum amplitude component F _imax (u, v) by a predetermined coefficient α _i (i = 1, 2,..., M) is selected in the frequency spectrum image F _i (u, v). The threshold value is Thd _αi . The frequency component selection unit 140 acquires the coefficient α _i from the frequency component selection control unit 160. Then, the frequency component selection unit 140 leaves the frequency component having the absolute value of the amplitude equal to or larger than the selection threshold Thd _αi from each frequency spectrum image F _i (u, v) as it is, and converts the frequency component less than the selection threshold Thd _αi to a real component, Both imaginary components are 0.

If the selection threshold Thd _αi is appropriate, by selecting only frequency components whose absolute value of amplitude is equal to or greater than a predetermined value, frequency components caused by noise applied to the multi-value fingerprint image M are removed, and each small region is selected. in L _i predominant ridge structure can be (in particular the pitch of the fringe direction) selecting a frequency component based on. Therefore, the frequency component selection unit 140 repeats the selection of the frequency component using the coefficient α _i adjusted by the frequency component selection control unit 160 until the image quality evaluation result in the image quality evaluation unit 150 described later satisfies a predetermined standard.

In the calculation of the selection threshold Thd _αi , the maximum amplitude component F _imax (u, v) of each frequency spectrum image F _i (u, v) is used as the maximum amplitude component other than the DC component (u = 0, v = 0). Also good. By determining the selection threshold Thd _αi excluding the DC component, it is possible to determine the selection threshold Thd _αi that more reflects the frequency characteristics of each frequency spectrum image F _i (u, v).
Each generated selected spectrum image F _si (u, v) is sent to the spectrum storage unit 130 and the image quality evaluation unit 150.

The image quality evaluation unit 150 evaluates, for each selected spectrum image F _si (u, v), whether the frequency component selected by the frequency component selection unit 140 appropriately represents a frequency component based on the ridge structure. . In each small area L _i of the multi-level fingerprint image M, as described above, since the ridges are substantially parallel to a plurality of straight lines, the frequency components are present many along a stripe direction perpendicular to the ridge considered It is done. On the other hand, it is thought that there are few strong frequency components resulting from the structure of a ridge with respect to the direction parallel to a ridge. Therefore, if a large number of frequency components are concentrated along a specific direction, it is determined that the selection of the frequency components is appropriate. Therefore, the image quality evaluation unit 150 obtains the direction line dl where the frequency component is most concentrated by the direction line calculation means 151, and concentrates the complexity C which is an index of how much the frequency component is concentrated in the vicinity of the direction line dl. The degree is calculated by the degree calculation means 152. Then, the image quality evaluation unit 150 evaluates the complexity C by the evaluation unit 153 and evaluates whether or not the selection of the frequency component is appropriate. The image quality evaluation unit 150, the selected spectral image F _{si (u,} v) above direction line calculated for each performs complexity calculation and evaluation, each selected spectral image F _{si (u,} v) of the frequency components for each Evaluate whether the selection is appropriate.

The direction line calculation means 151 calculates a straight line (hereinafter referred to as a direction line dl) for estimating the ridge stripe direction from the distribution of frequency components selected by the frequency component selection unit 140. In a certain small region L _k , when the ridge is a complete straight line and does not include any noise, the frequency component present in the corresponding selected spectrum image F _sk (u, v) is along the stripe direction of the ridge. Therefore, it exists in a straight line passing through the origin of the spectrum plane. Therefore, the direction line calculation means 151 calculates the direction line dl as a straight line passing through the origin of the spectrum plane.

ここでｎは、周波数成分選択部１４０で選択された周波数成分の個数（すなわち、選択スペクトル画像Ｆ_ｓｋ（ｕ，ｖ）で画素値が０でない画素の数）であり、ｕ_ｊ、ｖ_ｊはそれぞれある周波数成分の水平方向座標値及び垂直方向座標値を表す。あるいは、方向線算出手段１５１は、各周波数成分の振幅の大きさを考慮して、以下の式（３）で方向線ｄｌの傾きａを算出することもできる。

ただしＦ_ｓｋ（０，０）は、周波数成分のうちの直流成分を表し、Ｆ_ｓｋ（ｕ_ｊ，ｖ_ｊ）は座標（ｕ_ｊ，ｖ_ｊ）の周波数成分の振幅を表す。
得られた方向線ｄｌ（傾きａ）は、集中度算出手段１５２に送られる。 FIG. 4 shows how the direction line dl is calculated for the frequency component distribution. In FIG. 4, 400 is a selected spectrum image F _sk (u, v) of _interest , and each point 410 on the selected spectrum image F _sk (u, v) 400 is a frequency component selected by the frequency component selection unit 140. It is. A straight line 420 is a direction line dl to be calculated. As shown in FIG. 4, the direction line calculation means 151, such that the sum of the distance d _j from the frequency components up direction line dl is minimum, and calculates the direction line dl using the least squares method. In this case, the inclination a of the direction line dl of the selected spectrum image F _sk (u, v) of interest is calculated by the following equation (2).

Here, n is the number of frequency components selected by the frequency component selection unit 140 (that is, the number of pixels whose pixel values are not 0 in the selected spectrum image F _sk (u, v)), and u _j and v _j are Each represents a horizontal coordinate value and a vertical coordinate value of a certain frequency component. Alternatively, the direction line calculation unit 151 can calculate the inclination a of the direction line dl by the following equation (3) in consideration of the amplitude of each frequency component.

Here, F _sk (0,0) represents a direct current component of the frequency components, and F _sk (u _j , v _j ) represents the amplitude of the frequency component at coordinates (u _j , v _j ).
The obtained direction line dl (inclination a) is sent to the concentration degree calculation means 152.

  The degree-of-concentration calculating unit 152 obtains an index (complexity C) indicating the degree of concentration of the frequency component in the direction line dl obtained by the direction line calculating unit 151. The obtained index is used as an image quality evaluation value in an evaluation unit 153 described later.

The degree-of-concentration calculation means 152 is complex for each of two regions R ₁ and R ₂ bounded by a certain distance β from the normal line nl of the direction line dl (that is, a line representing a frequency change in the ridge direction). Indexes D ₁ and D ₂ used for calculating the degree C are calculated. Here, the region R ₁ is a region separated from the normal line nl by a distance β or more, and is used for calculating an index D ₁ or the like that indicates how much frequency components spread regardless of the direction line dl. On the other hand, the region R ₂ is used for calculating an index D ₂ or the like that indicates how much a periodic component exists in a direction orthogonal to the direction line dl.

FIG. 5A shows an outline of the regions R ₁ and R ₂ to be set. In FIG. 5A, reference numeral 500 denotes a selected spectrum image F _sk (u, v) of _interest. Each point 510 on the selected spectrum image F _sk (u, v) 500 is selected by the frequency component selection unit 140. Frequency component. The straight line 520 is a direction line dl, and the straight line 530 is a normal line nl. Here, β is preferably set to a value corresponding to about 1.2 lp / mm to 1.9 lp / mm. In this case, in the present embodiment, the resolution of the multi-value fingerprint image M is 500 dpi, and the selected spectrum image F _sk (u, v) is composed of 32 × 32 pixels. Therefore, the selected spectrum image F _sk (u, The value of β for v) is about 2.0 to 3.0.

ただしＮ_１は、領域Ｒ_１に存在する選択された周波数成分５１０の数である。Ｎ_１が大きいほど、選択された周波数成分５１０が表す構造は単純な周期構造から乖離する。またＤ_１が大きいほど、着目する選択スペクトル画像Ｆ_ｓｋ（ｕ，ｖ）に対応する小領域Ｌ_ｋの画像中に方向線ｄｌと大きく異なる方向に沿った画素値の変動が含まれることを示すため、、方向線ｄｌは隆線の縞方向を表していない可能性が高くなる。 In the region R _1, it is calculated as an indicator the sum D ₁ of the distance of the distance d _j from each frequency component 510 selected by the frequency component selector 140 to direction line dl. That _{D 1} is expressed by the following equation (4).

Where N ₁ is the number of selected frequency components 510 present in region R ₁ . As N ₁ increases, the structure represented by the selected frequency component 510 deviates from a simple periodic structure. Further, the larger D ₁ is, the more the variation of the pixel value along the direction significantly different from the direction line dl is included in the image of the small region L _k corresponding to the selected spectrum image F _sk (u, v) of _interest. Therefore, there is a high possibility that the direction line dl does not represent the stripe direction of the ridge.

ただしＮ_２は、領域Ｒ_２に存在する選択された周波数成分５１０の数である。Ｄ_２、Ｎ_２が大きいほど、着目する選択スペクトル画像Ｆ_ｓｋ（ｕ，ｖ）に対応する小領域Ｌ_ｋの画像中に、方向線ｄｌの方向とは略直交する方向に高周波の画素値の変動が含まれることになる。そのため、Ｄ_２、Ｎ_２が大きくなるほど、方向線ｄｌは隆線の縞方向を表していない可能性が高くなる。 On the other hand, even in a region R _2, it is calculated as an indicator of the sum D ₂ of the distance of the distance d _i from each frequency component 510 selected by the frequency component selector 140 to the direction line. That _{D 2} is represented by the following formula (5).

Where N ₂ is the number of selected frequency components 510 present in region R ₂ . The larger the D ₂ and N ₂ are, the higher the pixel value of the high-frequency pixel value in the direction substantially perpendicular to the direction of the direction line dl in the image of the small region L _k corresponding to the selected spectrum image F _sk (u, v) of _interest . Variations will be included. Therefore, as D ₂ and N ₂ increase, there is a higher possibility that the direction line dl does not represent the stripe direction of the ridge.

The degree-of-concentration calculation means 152 may classify the regions R ₁ and R ₂ as shown in FIG. 5B in order to obtain D ₁ , D ₂ , N ₁ , and N ₂ . In addition, the number attached | subjected in FIG.5 (b) shows the same thing as Fig.5 (a). In FIG. 5B, the boundary between the regions R ₁ and R ₂ is two straight lines having an angle φ that passes through the origin of the frequency plane and forms a normal line nl of the direction line dl. The region R ₂ is a region sandwiched between these two straight lines with the normal line nl as the center, while the region R ₁ is a region other than the region R ₂ . In this case, the angle φ can be set to about 15 ° to 30 °.

ここでκ、λは定数であり、多値指紋画像Ｍの解像度、選択スペクトル画像画像Ｆ_ｓｋ（ｕ，ｖ）の大きさに応じて最適化される。また、周波数成分選択部１４０の係数α_ｋの値が大きくなるにつれて、κ、λも大きくなるように変更してもよい。係数α_ｋが大きくなるほど選択される周波数成分の数が減少するため、単なる周波数成分数の減少だけで複雑度Ｃが低下することを防止するためである。本実施形態では、多値指紋画像Ｍの解像度が５００ｄｐｉであり、選択スペクトル画像Ｆ_ｓｋ（ｕ，ｖ）が３２×３２画素であるのに対して、係数α_ｋが０．１〜０．２程度の値であれば、κ＝０．１０〜０．５０、λ＝０．５〜１．５程度に設定することが好ましい。 When the degree-of-concentration calculating unit 152 obtains D ₁ , D ₂ , N ₁ , and N ₂ , the degree of complexity C is calculated using them. The complexity C is calculated by the following equation (6).

Here, κ and λ are constants, and are optimized according to the resolution of the multi-value fingerprint image M and the size of the selected spectrum image image F _sk (u, v). Further, κ and λ may be changed so as to increase as the value of the coefficient α _k of the frequency component selection unit 140 increases. This is because the number of frequency components to be selected decreases as the coefficient α _k increases, so that the complexity C is prevented from being reduced only by the decrease in the number of frequency components. In the present embodiment, the resolution of the multi-value fingerprint image M is 500 dpi, the selected spectrum image F _sk (u, v) is 32 × 32 pixels, and the coefficient α _k is 0.1 to 0.2. If it is a value of about, it is preferable to set κ = 0.10 to 0.50 and λ = about 0.5 to 1.5.

式（７）におけるκ、λは、式（６）と同様の値である。また集中度算出手段１５２は、複雑度Ｃの代わりに、別の値を方向線ｄｌへの集中度を表す指標とすることもできる。簡単な例としては、所定の値Ｑ_ｍａｘから複雑度Ｃを引いた値Ｑ（＝Ｑ_ｍａｘ−Ｃ）を指標することもできる。このＱ値では、選択された周波数成分が方向線ｄｌの近傍に集中するほど大きな値を示す。さらに、集中度算出手段１５２は、上記のように領域を二つに分けて指標を求めることなく、単純に各周波数成分と方向線ｄｌの距離の平均二乗誤差を求めて方向線ｄｌへの集中度を表す指標として用いてもよい。またさらに、集中度算出手段１５２は、方向線ｄｌから所定の距離以内にある周波数成分の個数と、方向線ｄｌからその所定の距離より離れたところにある周波数成分の個数との比を、方向線ｄｌへの集中度を表す指標として用いてもよい。この場合、所定の距離を、例えば上記のβに設定することができる。
集中度算出手段１５２は、算出した指標（複雑度Ｃ）を評価手段１５３へ送る。 The complexity C is an index representing how much the selected frequency component is concentrated on the direction line dl. The complexity C indicates that the lower the value, the more the selected frequency component is concentrated in the vicinity of the direction line dl.
The formula for obtaining the complexity C is not limited to the above. Instead, the following formula may be used.

In Equation (7), κ and λ are the same values as in Equation (6). Further, the concentration degree calculating means 152 can use another value as an index representing the degree of concentration on the direction line dl instead of the complexity C. As a simple example, a value Q (= Q _max −C) obtained by subtracting the complexity C from a predetermined value Q _max may be used as an index. This Q value shows a larger value as the selected frequency components are concentrated in the vicinity of the direction line dl. Further, the concentration degree calculation means 152 simply calculates the mean square error of the distance between each frequency component and the direction line dl without dividing the region into two areas as described above, and concentrates on the direction line dl. You may use as a parameter | index showing a degree. Furthermore, the concentration degree calculation means 152 calculates the ratio of the number of frequency components within a predetermined distance from the direction line dl and the number of frequency components located away from the direction line dl from the predetermined distance. It may be used as an index representing the degree of concentration on the line dl. In this case, the predetermined distance can be set to the above β, for example.
The concentration degree calculation unit 152 sends the calculated index (complexity C) to the evaluation unit 153.

The evaluation unit 153 has sufficient image quality for the image f _k ′ (x, y) of the small region L _k reconstructed based on the selected spectrum image F _sk (u, v) to extract feature points. Evaluate whether or not it has. For the evaluation, the evaluation unit 153 compares the complexity C acquired from the concentration degree calculation unit 152 with a predetermined threshold Thd _conf . When the complexity C is less than the threshold Thd _conf , the evaluation unit 153 is considered to concentrate the selected frequency components in the vicinity of the direction line dl so as to satisfy a certain standard. _{For sk} (u, v), it is determined that an image with sufficient image quality can be reconstructed. Then, the selected spectrum image F _sk (u, v) is stored in the spectrum storage unit 130. On the other hand, when the complexity C is equal to or greater than the threshold Thd _conf , the evaluation unit 153 determines that an image with sufficient image quality cannot be reconstructed for the selected spectrum image F _sk (u, v). In this case, the evaluation unit 153 notifies the frequency component selection control unit 160 of the determination result, and causes the coefficient α _k to be readjusted.
The image quality evaluation unit 150 performs the above image quality evaluation on all the selected spectrum images F _si (u, v) (i = 1, 2,..., M).

In the frequency component selection control unit 160, the frequency component selection unit 140 sets a coefficient α _i that is a selection criterion for the frequency component. When the frequency component selection control unit 160 determines that the image quality evaluation unit 150 does not reconstruct an image with sufficient image quality based on the selected frequency component (that is, determines that these frequency components are inappropriate). The value of the coefficient α _i is appropriately adjusted. In addition, the frequency component selection control unit 160 adjusts the coefficient α _i for each frequency spectrum image F _i (u, v). In the following, the operation of the frequency component selection control unit 160 will be described taking the frequency spectrum image F _k (u, v) of interest as an example for the sake of simplification.

In the selection of the first frequency component, the frequency component selection control unit 160 sets the coefficient α _k to a low value so that many frequency components are selected. The coefficient α _k is stored in the selection parameter storage unit 170. Then, every time the selected frequency component is determined to be inappropriate by the image quality evaluation unit 150, the coefficient α _k is acquired from the selection parameter storage unit 170 so as to gradually decrease the number of selected frequency components, The coefficient α _k is changed to a large value by a predetermined ratio. Then, the changed coefficient α _k is sent to the selection parameter storage unit 170, and the stored value of the coefficient α _k is updated. On the other hand, the frequency component selection control unit 160 sends the coefficient α _k to the frequency component selection unit 140, and selects a frequency component having an absolute value greater than or equal to the threshold Thd _αk determined based on the coefficient α _k as described above. Let it be done. The frequency component selection control unit 160 repeats the adjustment of the coefficient α _k until the image quality evaluation unit 150 determines that the selected frequency component is appropriate or reaches a predetermined number of repetitions.

In the present embodiment, the initial value of α _k is set to 0.1. The frequency component selection control unit 160 increases the coefficient α _k by 1.2 each time it is adjusted once. However, the initial value of the coefficient α _k is not limited to 0.1, and a value smaller or larger than 0.1 may be set as the initial value. Further, the ratio for increasing the coefficient α _k is not limited to 1.2, and for example, a value of about 1.1 to 1.3 can be used. In addition, the frequency component selection control unit 160 may add a constant increment, for example, a value such as 0.05 to the coefficient α _k each time adjustment is performed, or may have a predetermined value according to the number of repetitions. Good. Further the frequency component selection control unit 160, the coefficient alpha _k is a predetermined upper limit value, may be terminated to adjust the coefficient alpha _k when it reaches the example 0.3.

Furthermore, in the above description, α _k is changed only in the increasing direction. However, in a predetermined case, α _k may be changed in the decreasing direction. For example, the evaluation unit 153 of the image quality evaluation unit 150 sets the second threshold Thd _conf2 having a value smaller than the threshold Thd _conf for the complexity C. When the complexity C calculated by the concentration calculation unit 152 is _{equal to} or less than the second threshold Thd _conf2 , α _k is decreased by 0.8, conversely to the above. By configuring in this way, it is possible to excessively limit the number of frequency components to be selected, and to reduce the possibility of excessively deleting information representing the ridge structure.

The selection parameter storage unit 170 is a memory area secured in the RAM of the image matching unit 5, and stores the coefficient α _i for each frequency spectrum image F _i (u, v). The coefficient α _i is output in response to the acquisition request from the frequency component selection control unit 160, and the value of the coefficient α _i is updated in response to the update request from the frequency component selection control unit 160.
Note that the storage unit 4 may be used instead of the selection parameter storage unit 170, and the selection parameter storage unit 170 may be omitted.

The frequency filter unit 180 performs band-pass filter processing so as to leave a frequency component included in a specific frequency band for the frequency component selected by the frequency component selection unit 140 for each selected spectrum image F _si (u, v). I do. By performing this band-pass filter processing, it is possible to remove noise components without erasing information on ridge branch points and end points.

  Here, what frequency components are generated at the branch points and end points of the ridge will be described. FIG. 6 shows a schematic diagram of an image of a small area including branch points and end points of ridges. As shown in FIG. 6, generally, each ridge 610 has substantially the same thickness within a small region 600 in which several ridges are included. These ridges 610 are arranged in parallel lines at substantially the same interval. Further, the thickness of the ridge 610 hardly changes even in the vicinity of the branch point 620 of the ridge. The branched ridge 610 is also a straight line substantially parallel to the other ridges 610. Therefore, the frequency component in the ridge stripe direction (direction perpendicular to the ridge) does not significantly differ between the case where the ridge branch point 620 is included and the case where it is not included. However, near the branch point 620 of the ridge 610, the branched ridge 610 faces a slightly different direction from the other ridges 610. Therefore, when the ridge branch point 620 is included in the small region 600, a frequency component in a direction slightly different from the stripe direction of the ridge is included. Similarly, when the end point 630 of the ridge exists, the distance between the adjacent ridges 610 changes in the vicinity of the end point 630, so that the direction of the ridge 610 changes slightly. Therefore, a frequency component in a direction slightly different from the ridge stripe direction is included.

Therefore the frequency filter unit 180, based on the direction line dl obtained in image quality evaluation unit 150 sets an R ₃ having a certain spread around the that direction line dl as a frequency band to leave the frequency component. Then, the frequency filter unit 180 sets the real component and the imaginary component to 0 for the frequency components not included in the region R ₃ except for the DC component (F _sk (0, 0)). Incidentally, leave without reducing weakness the DC component, when reconstructing a fingerprint image of each small area L _i, it is possible to prevent the image edges only of the ridge is enhanced, which will be described later binary This is to prevent only the edge portion of the ridge from becoming the ridge region by the conversion process. It will be described region _{R 3} below.

FIG. 7 shows an example of a filter region for performing frequency filtering. In FIG. 7, an image 700 is an example of the selected spectrum image F _sk (u, v). Each point 710 on the image is a selected frequency component. Furthermore, the straight lines 720 and 730 are the normal line nl of the direction line dl and the direction line dl, respectively. A region R ₃ is an area to leave frequency components. Region R ₃ is no more than the distance γ from direction line dl (width 2γ on both sides), spaced above the distance δ from the origin of the frequency plane, and a range from the origin distance below ε of the frequency spectrum.

The values of γ, δ, and ε are optimized empirically based on the pitch of the ridges in the actual multi-value fingerprint image M, the angle formed by the branching ridge and the branching ridge. . In the present embodiment, γ is a value corresponding to 1.9 lp / mm, δ is a value corresponding to 1.2 lp / mm, and ε is a value corresponding to 4.9 lp / mm. That is, in the present embodiment, the resolution of the multi-value fingerprint image M is 500 dpi, and each selected spectrum image F _si (u, v) is composed of 32 × 32 pixels, so that γ = 3.0 and δ, respectively. = 2.0 and ε = 8.0. Note that γ, δ, and ε are not limited to the above values. It can be optimized according to the characteristics of the fingerprint sensor used in the fingerprint input unit 3. For example, γ can be set in a range corresponding to about 1.2 lp / mm to 2.4 lp / mm. Furthermore, δ can be set in a range corresponding to about 0.6 lp / mm to 1.9 lp / mm. Similarly, ε can be set in a range corresponding to about 3.7 lp / mm to 6.1 lp / mm. If the values of γ, δ, and ε are set to values that make the region R ₃ narrower than the above range, the information on the branch points and end points of the ridges is attenuated, and the feature extraction unit 13 (to be described later) accurately features. It becomes impossible to extract points. On the other hand, if the values of γ, δ, and ε are set to values that make the region R ₃ wider than the above range, the frequency component due to noise cannot be removed, and the feature extraction unit 13 extracts pseudo feature points. There is a high possibility that it will end.

These γ, δ, and ε may be changed based on other image quality information. For example, the width of the ridge may be calculated separately and changed according to the width. The width of the ridge can be determined as follows, for example. First, a process in the binarization unit 11 described later is performed on the multi-value fingerprint image M to obtain a temporary binary fingerprint image. Then, by dividing the binary fingerprint image in the small area L _i, determining a plurality of directions an average number of consecutive pixel values corresponding to the ridges in each small region L _i. Among the average continuous numbers obtained for the plurality of directions, the smallest one can be set as the width of the ridge. When the width of the ridge is thick, the frequency component due to the structure of the ridge exists on the relatively low frequency side. Therefore, the larger the width of the ridge, the smaller the γ, δ, and ε are set.

FIG. 8 shows another example of a filter region for performing frequency filtering. In FIG. 8, an image 800 is an example of a selected spectrum image F _sk (u, v). Each point 810 on the image is a selected frequency component. Further, the straight line 820 is a direction line dl. The region R ₃ is an area to leave frequency components. Region R ₃ is formed of a circumference of radius δ centered at the origin of the frequency plane, it exists in the range between the two concentric circles of radius ε about the origin of the frequency spectrum, a and direction line dl It is a range sandwiched between two straight lines having an angle ψ. In this case, ψ can be set to about 15 ° to 30 °.

The image reconstruction unit 190, for each small area L _i of the selected spectral images filtered by the frequency filter unit 180, performs inverse transformation of the frequency conversion performed in the frequency spectrum analyzer 120. The image reconstruction unit 190 reconstructs an image of each small area L _i from which noise is removed. The reconstructed image of each small area is sent to the fingerprint image composition unit 200.

Fingerprint image synthesis section 200 synthesizes the image of each small area L _i reconstructed, generating one multi-level fingerprint image M '. This multi-valued fingerprint image M ′ has noise removed, and can clearly observe end points and branch points of ridges. Note that the fingerprint image synthesizing unit 200, when synthesizing the image of each small region L _i, adjacent for the pixels included in the overlapping portion of the small region, the average values of pixels of overlapping all the small areas corresponding By calculating the pixel value, the pixel value is calculated. By performing such averaging, it is possible to prevent ridges from becoming discontinuous at the boundary between adjacent small regions.
When the generation of the multi-value fingerprint image M ′ is completed, the image correction unit 10 sends the multi-value fingerprint image M ′ to the binarization unit 10.

The binarization unit 11 generates a binary fingerprint image B binarized into a ridge region and a valley line region based on the multi-value fingerprint image M ′ subjected to noise removal processing by the image correction unit 10.
The binarization unit 11 compares the pixel value Ic of the pixel of interest C of the multi-value fingerprint image M ′ with a predetermined binarization threshold Thd _BIN . Then, the binarizing unit 11 sets the pixel value B _C of the binary fingerprint image B corresponding to the target pixel C of the multi-value fingerprint image M ′ as follows.
If I _C ≧ Thd _BIN , B _C = '255'
If I _C <Thd _BIN , B _C = '0'
Here, in the binary fingerprint image B, the pixel value “255” is a valley pixel value, and “0” is a ridge pixel value. In other words, the value of B _C is '255' indicates that the pixel is valley pixel. On the other hand, the value of B _C is '0' indicates that the pixel is ridge pixel.
The value of B _C is' 0 'and is not limited to the combination of' 255 ', any two different values that can be handled in a data format that represents a binary fingerprint image B, for example,' 0 'and' 1 'can be used.

Moreover, the binarization threshold value _{Thd BIN} can be set to, for example, average pixel value _{I A} of the multi-level fingerprint image M '. The binarization threshold Thd _BIN divides the multi-value fingerprint image M ′ into a plurality of small areas L _i (i = 1, 2,..., M), and the average pixel value I for each small area L _i. You may set to _Li . Further binarization threshold _{Thd BIN} is the average pixel value _{I A} of the multi-level fingerprint image M ', the difference _I D (= _{I A} of the average pixel value _{I A} and the average pixel value _{I Li} for each small area _{L i} It is good also as a value which added _-ILi ). In this case, each small region may be set to the same size and shape as the small region divided by the image dividing unit 110, or may be set to a separate size and shape.

The binarization unit 11 can generate the binary fingerprint image B by performing the above comparison for all the pixels of the multi-value fingerprint image M ′. The generated binary fingerprint image B is sent to the thinning unit 12.
In the following, for convenience of explanation, represented by the value of the ridge pixels in the binary fingerprint image B and the thinned binary fingerprint image B _T '0', the value of the valley pixel '255'.

The thinning unit 12 performs a thinning process on the binary fingerprint image B generated by the binarizing unit 11 so that the ridge region is thinned. The ridge region to generate a thinned binary fingerprint image B _T to be a continuous line with a width of 1 pixel. For example, the thinning unit 12 prepares a plurality of predetermined mask patterns in which a plurality of adjacent pixels have a value of “0”, sequentially applies the mask patterns, and applies to any one of the mask patterns. A method of converting the pixel value of “0” to “255” can be used. However, the thinning unit 12 is not limited to the thinning method described above, and various known methods can be used.

The feature extraction unit 13 extracts feature points based on the thinned binary fingerprint image B _T generated by the thinning unit 12, and creates an input feature point list F _i . Here, the feature points are end points, branch points, and the like of the ridge region.

FIG. 9 shows an outline of feature points in a thinned fingerprint image.
In FIG. 9, the portion shown in black represents a thinned ridge region. Further, in FIG. 9, portions 900 and 910 that are the end of the ridge region are end points of the ridge, and a portion 920 where the ridge is branched from one to two is a branch point. Since the position and number of these feature points are different for each individual, it is very useful information for collating fingerprints.

The information representing the feature point includes each element of the type of feature point (end point or branch point), position, direction, and reliability. Here endpoints P _e, for example, among the ridge pixel (i.e., pixel value '0'), and among the eight neighboring pixels adjacent, can only 1 pixel is a pixel which is ridge pixel. The branch point P _b are among the ridge pixels, of the adjacent eight neighboring pixels, 3 pixels is ridge pixel, and the ridge pixels to their 8 in neighboring pixels to the pixel not adjacent be able to. The position of the feature point is represented by coordinate values (X _t , Y _t ) on the thinned binary fingerprint image B _T of the end point _Pe or the branch point P _b . However X _t is the horizontal coordinate values on thinning the binary fingerprint image B _T, Y _t is the vertical coordinate value of the thinned binary fingerprint image B _T.

The direction θ _t of the feature point represents the direction of the ridge at the feature point. The direction theta _t of feature points, as shown in FIG. 9, represented by the angle formed, for example, in the horizontal axis and the clockwise direction of thinning the binary fingerprint image B _T. If the feature point is the end point P _e, the direction theta _t of feature points, a straight line connecting the ridge pixels adjacent to its end point P _e, as the angle between the horizontal axis of the thinning binary fingerprint image B _T Can be sought. In the case the feature point is a branch point P _b, and the branch point P _b, the line segment determined (such segment with the ridge pixels at a distance of two pixels from the branch point P _b should there three ). Next, the angle formed by each line segment is obtained. Then, a bisector of two line segments having the narrowest angle is obtained. The angle between the horizontal axis of the bisector and thinning the binary fingerprint image B _T can be a direction theta _t of the feature point. The method of obtaining the feature point direction θ _t is not limited to the above, and other known methods may be used.

The feature point reliability T is a value indicating the reliability of the existence, position, and direction of the feature point. For example, in a fingerprint image, a plurality of ridges are arranged substantially in parallel when viewed locally. Therefore, when considering a local region including a plurality of feature points, the direction θ _t of each feature point is substantially the same in that region. Therefore, the reliability T can be set as follows. First, a local region (for example, about 32 × 32 pixels) around the feature point of interest is set. Next, an average value θ _av of the direction θ _t of each feature point included in the local region is calculated. Then, the absolute value of the difference between the average value θ _av and the direction of the feature point of interest θ _t is defined as the reliability.
Note that the feature points to be extracted are not limited to those described above. For example, the positional relationship between feature points, the number of ridges or valleys existing between adjacent feature points, and the like may be extracted as feature points.

Feature extraction unit 13, for each of the feature points extracted from one thinned binary fingerprint image B _T as described above, to create a list in accordance with a predetermined format information of the feature point.

The collation unit 14 collates the input feature point list F _i extracted by the feature extraction unit 13 from the fingerprint image of the operator and the registered feature point list F _r read from the storage unit 4. As a specific method of collation, a known method for calculating the similarity between feature quantities, for example, a minutia matching method can be used. Specifically, first, for some feature points included in the input feature point list F _i , the average position difference with any one of the feature points included in the registered feature point list F _r is the smallest. Thus, the coordinate value parallel movement amount Δd and rotation correction angle Δθ are obtained. Then, for each feature point included in the input feature point list F _i, by adding the translation amount Δd and the rotational correction angle [Delta] [theta], to align the input feature point list F _i and the registered feature point list F _r .

Next, for each feature point included in the input feature point list F _i , among the feature points included in the registered feature point list F _r , the distance and direction from the feature point of the same type and the closest position The score is calculated based on the absolute value and the reliability of the difference. This score is set to be higher as the distance is shorter. Similarly, the score is set higher as the absolute value of the direction difference is smaller and the reliability is higher.

When the score for each feature point is calculated, the collation unit 14 calculates the sum of the scores as the similarity S _i . The collation unit 14 determines that the authentication is successful if the calculated similarity S _i exceeds a predetermined collation threshold Thd _V (that is, the operator's fingerprint is a registered fingerprint). In contrast, if the calculated similarity S _i is equal to or smaller than a predetermined matching threshold Thd _V , it is determined that the authentication has failed (that is, the operator's fingerprint is a registered fingerprint). It does not match).

The output unit 6 is an interface for connecting the fingerprint authentication device 1 to an external device (for example, an electric lock locking / unlocking control device) and inputting / outputting signals, and includes Ethernet (registered trademark), USB, SCSI, RS- It is composed of communication ports, electronic circuits, driver software, and the like that comply with standards such as 232C.
The output unit 6 outputs a signal indicating the collation result of the image collation unit 5 to an external device.

Next, the operation of the fingerprint authentication device 1 which is an example of the collation device according to the present invention will be described.
FIG. 10 is a flowchart showing the fingerprint authentication operation in the fingerprint authentication apparatus 1. The operation of the fingerprint authentication device 1 described below is executed by the control unit 15 of the image matching unit 5 in accordance with a program read by the image matching unit 5.

  First, when the fingerprint authentication apparatus 1 detects the approach of the operator by using a near infrared sensor (not shown), for example, a predetermined guidance is displayed on the operation / display unit 2, and the fingerprint authentication work is performed on the operator according to the guidance. At the same time, the fingerprint authentication operation is started.

  First, the fingerprint authentication device 1 acquires the operator's input identification number IN from the operation / display unit 2 (step S101). If the placement sensor of the fingerprint input unit 3 detects a finger before the operator inputs the identification number IN, the fingerprint authentication device 1 prompts the input of the identification number through the operation / display unit 2. Is displayed. Next, the fingerprint authentication device 1 searches whether or not the acquired input identification number IN is registered in the storage unit 4 (step S102). When the input identification number IN is registered in the storage unit 4, the fingerprint authentication device 1 operates the operation / display unit 2 so that the finger is placed at a predetermined position of the fingerprint input unit 3 in order to obtain a fingerprint image of the operator. Guidance to the operator is performed through (step S103). On the other hand, if the acquired input identification number IN is not registered in step S102, the fingerprint authentication device 1 displays an error message in which the identification number is not registered through the operation / display unit 2 (step S104). Then, control is returned to the beginning.

After step S103, when the placement sensor of the fingerprint input unit 3 detects a finger, the fingerprint input unit 3 acquires a multi-value fingerprint image M of the operator (step S105). The acquired multi-value fingerprint image M is sent to the image matching unit 5. When the image collation unit 5 acquires the multi-value fingerprint image M, the image correction unit 10 of the image collation unit 5 performs a noise removal process on the multi-value fingerprint image M (step S106). The procedure for executing the noise removal process will be described later. Thereafter, the binarization unit 11 of the image collation unit 5 generates a binary fingerprint image B based on the noise-removed multi-value fingerprint image M ′ (step S107). The binarization unit 11 binarizes the ridge region and the valley region in the multi-value fingerprint image M ′. When the binary fingerprint image B is generated, thinning unit 12 of the image matching unit 5, ridge region to generate thinned thinned binary fingerprint image B _T (step S108).

In step S108, the thinning the binary fingerprint image B _T is generated, feature extraction unit 13 of the image matching unit 5, the end point P _e of ridge region from the thinning binary fingerprint image B _T, the branch point P Feature points such as _b are extracted to create an input feature point list F _i (step S109). When the input feature point list F _i is created, the image matching unit 5 reads the registered feature point list F _r associated with the input identification number IN from the storage unit 4, and the matching unit 14 uses the input feature point list F _r. The similarity S _i between _i and the registered feature point list F _r is calculated (step S110).
Then, the collation unit 14 compares the similarity S _i calculated in step S110 with a predetermined collation threshold Thd _V (step S111). In step S111, when the similarity S _i is equal to or smaller than the verification threshold Thd _V , the image verification unit 5 determines that the authentication has failed. Then, the fingerprint authentication device 1 displays a message indicating that the authentication has failed on the operation / display unit 2 (step S112). Then, in order to acquire the operator's fingerprint image again, the control is returned to before step S103.

On the other hand, when the similarity S _i is larger than the matching threshold value Thd _V , the matching unit 14 of the image matching unit 5 determines that the authentication is successful. Then, the fingerprint authentication device 1 displays a message indicating that the authentication has been successful on the operation / display unit 2 and transmits a signal indicating the verification result to the external device through the output unit 6 as necessary (step S113). Then, the fingerprint authentication device 1 ends the operation.

  If the authentication fails for a predetermined number of times, the fingerprint authentication device 1 may display a warning message to the operator through the operation / display unit 2. Further, in such a case, the fingerprint authentication device 1 may notify the monitoring center (not shown) or the like of the identification information and warning information of the fingerprint authentication device 1 through the output unit 6.

  Next, the operation of noise removal processing in the fingerprint authentication device 1 which is an example of the collation device according to the present invention will be described with reference to the flowcharts shown in FIGS. Note that the noise removal process is executed exclusively by the image correction unit 10 of the fingerprint authentication device 1.

As shown in FIG. 11, first, the multi-value fingerprint image M is sent to the image correction unit 10, and the noise removal process is started. When the noise removal process is started, the image dividing unit 110 divides the multi-value fingerprint image M into a plurality of small regions L _i (i = 1, 2,..., M) (step S201). Next, the frequency spectrum analysis unit 120 performs frequency transformation on each of the small areas _{L i,} the frequency spectral images _F i (u, v) of each small area _{L i} to create a (step S202). Then, each created frequency spectrum image F _i (u, v) is stored in the spectrum storage unit 130 and sent to the frequency component selection unit 140.

Next, the frequency component selection unit 140 firstly selects one frequency spectrum image F _k (u, v) (k = 1, 2,..., M) from among the frequency spectrum images F _i (u, v). (Step S203). The frequency component selection unit 140 then uses the amplitude coefficient α _k (k = 1, 2,..., M) as a reference for determining the frequency component to be selected for the focused frequency spectrum image F _k (u, v). ) Is acquired from the frequency component selection control unit 160, and a threshold Thd _αk is calculated. Then, a selected spectrum image Fs _k (u, v) is created by selecting only frequency components having an absolute value greater than or equal to the threshold Thd _αk for the frequency spectrum image F _k (u, v) (step S204).

When the selected spectrum image Fs _k (u, v) is created, the image quality evaluation unit 150 determines whether or not the selection of the frequency component is appropriate. For this purpose, the direction line calculation unit 151 of the image quality evaluation unit 150 calculates the direction line dl based on the selected frequency component for the selected spectrum image Fs _k (u, v) (step S205). Next, the concentration degree calculation unit 152 of the image quality evaluation unit 150 calculates the complexity C, which is an index of how much the selected frequency component is concentrated with respect to the direction line dl (step S206). Then, the evaluation unit 153 of the image quality evaluation unit 150 compares the complexity C with a predetermined threshold value Thd _conf and determines whether or not the selected frequency component is appropriate (step S207). If the complexity C is equal to or greater than the threshold Thd _conf, it is determined that the selected frequency component is inappropriate, and the frequency component selection control unit 160 adjusts the coefficient α _{k serving} as a reference for frequency component selection (step S208). . The adjusted coefficient α _k is stored in the selection parameter storage unit 170 and is referred to when the adjustment is performed again. Then, control is transferred before step S204, and the processing of step S204 to step S207 is repeated. On the other hand, if the complexity C is less than the threshold Thd _conf in step S207, it is determined that the selected frequency component is appropriate, and the selected spectrum image Fs _k (u, v) is temporarily stored in the spectrum storage unit 130. (Step S209).

As shown in FIG. 12, after step S209, the image correction unit 10 determines whether or not the processing of steps S204 to S209 has been completed for all frequency spectrum images (step S210). When the image correction unit 10 determines that any of the frequency spectrum images (for example, F _{i + 1} (u, v)) has not been subjected to the processes in steps S204 to S209, the image correction unit 10 pays attention to the frequency spectrum image next. The frequency spectrum image F _k (u, v) is set (S211). And control is transferred before step S204 and the process of step S204-step S210 is repeated. On the other hand, in step S210, the image correction unit 10 finishes the processes of steps S204 to S209 for all the frequency spectrum images, and selects the selected spectrum image Fs _i (u, v) corresponding to each frequency spectrum image F _i (u, v). If it is confirmed that v) is created, the selected spectrum images Fs _i (u, v) are sent to the frequency filter unit 180. Then, the frequency filter unit 180 performs a band-pass filter process that leaves only frequency components in a predetermined frequency band for each selected spectrum image Fs _i (u, v) (step S212). Then, the image reconstruction unit 190 performs the inverse transform process of the frequency transform process of the frequency spectrum analysis unit 120 on each selected spectrum image Fs _i (u, v) on which the bandpass filter process has been performed. The image f _i ′ (x, y) in the region L _i is reconstructed (step S213). Finally, the fingerprint image synthesizing unit 200 synthesizes the reconstructed images f _i ′ (x, y) of the small regions L _i to generate a multi-value fingerprint image M ′ subjected to noise processing (step S1). S214). Then, the image correction unit 10 erases each frequency spectrum image F _i (u, v) and each selected spectrum image Fs _i (u, v) stored in the spectrum storage unit 130 and generates the generated multi-value fingerprint. The image M ′ is output and the process ends.

As described above, the image correction unit 10 selects a frequency component having an amplitude of a certain level or more for each frequency spectrum image F _i (u, v), and the selected frequency component. Furthermore, by leaving the frequency component included in a predetermined frequency band centered on the ridge stripe direction in the frequency filter unit 180, it is possible to appropriately remove noise while leaving information on the end points and branch points of the ridges. it can. Furthermore, the image correction unit 10 is based on the fact that the frequency components should be concentrated only in a predetermined direction because the ridges of the fingerprint image become a striped pattern in which a plurality of substantially parallel straight lines are arranged in a small region. It is possible to select an appropriate frequency component by evaluating whether or not the frequency component once selected is appropriate, and by selecting the frequency component again after adjusting the selection criterion depending on the evaluation result.

As described above, the fingerprint authentication device 1 which is an example of a collation device according to the present invention creates a multi-value fingerprint image M ′ from which noise has been removed while leaving information on ridge branch points and end points. The fingerprint authentication device 1, extracted from the multi-level fingerprint image M 'is created from a thinned binary fingerprint image B _T by performing the matching by extracting feature points of the extracted failure or pseudo feature points of the feature point Therefore, high collation accuracy can be achieved.

  In the above-described embodiment, the ridge region of the fingerprint has a low pixel value (or black) and the valley region of the fingerprint has a high (or white) pixel value. The present invention is applied even when the fingerprint ridge region has a high pixel value (or white) and the fingerprint valley region has a low pixel value (or black), as in the case of using the optical method. be able to. In this case, in the above description, the pixel values can be handled in the same manner by reversing the magnitude of the pixel values (for example, the expression “˜” is replaced with “˜”).

  Further, in the above-described embodiment, the fingerprint collation is determined based on the feature point-based similarity, but the collation apparatus of the present invention can also be applied when fingerprint collation is performed by another method. For example, the present invention can be applied when fingerprint matching is performed based on pattern matching between fingerprint images. In this case, a binary fingerprint image of the person to be verified is registered in advance as feature information instead of the feature point list. Then, a binary fingerprint image separated into ridges and valleys from the multi-value fingerprint image M ′ is generated, and the generated binary fingerprint image is subjected to pattern matching with the previously registered binary fingerprint image of the person to be verified. If the degree of coincidence exceeds a predetermined threshold value, the authentication can be determined to be successful. Alternatively, the multi-value fingerprint image M ′ itself may be used for pattern matching instead of the binary fingerprint image.

Furthermore, in the image correction unit 10, multiple values generated due to the difference between the pressing force at the center of the finger and the pressing force at the peripheral part when the finger is placed on the fingerprint input unit 3 before performing the noise removal processing. The density unevenness of the fingerprint image M may be corrected to make the density of the entire fingerprint image uniform.
For example, in the unevenness correction, the multi-value fingerprint image M is divided into partial areas of about 32 × 32 pixels, and the average pixel value and the variance of the pixel values are calculated for each partial area. Then, the pixel values are converted so that the average pixel value and the variance of the pixel values are substantially equal to the average pixel value and the variance of the pixel values of the entire multi-value fingerprint image M (Image Analysis Handbook p.478, Mikio Takagi, (See Tokyo University Press).

  In the above description, the collation apparatus according to the present invention has been described as a fingerprint image authentication apparatus. However, the present invention can also be applied to an image collation apparatus that includes a stripe pattern other than a fingerprint image. For example, the present invention can be applied to a palmprint image collation device. Further, unlike the fingerprint image, when the entire image includes a substantially parallel linear stripe pattern, the entire image is divided into small areas as described above, and the frequency analysis / frequency is performed. You may make it perform each process of a component extraction, image quality evaluation, and a band pass filter. When division into small regions is not performed in this way, when performing frequency conversion, processing focusing on a specific portion in an image may be performed using a window function such as a Hanning window or a Hamming window. .

  As described above, various modifications can be made within the scope of the present invention according to the embodiment to be implemented.

It is a functional block diagram of a fingerprint authentication device which is an example of a collation device according to the present invention. It is a functional block diagram of the image correction part of a fingerprint authentication device. It is a figure which shows the outline of the setting of the small area | region with respect to a multi-value fingerprint image. It is a figure which shows the mode of calculation of the direction line dl with respect to distribution of a frequency component. (A) is a figure which shows an example of the area | region which calculates the 2 parameter | index of complexity C, (b) is a figure which shows another example of the area | region which calculates the 2 parameter | index of complexity C. It is the schematic of the image of the small area | region containing the branch point of a ridgeline. It is the schematic which shows an example of the filter area | region for performing frequency filtering. It is the schematic which shows another example of the filter area | region for performing frequency filtering. It is a figure which shows the outline of the feature point in the thinned fingerprint image. It is a flowchart which shows the operation | movement of a fingerprint authentication process in the fingerprint authentication apparatus which is an example of the collation apparatus which concerns on this invention. It is a flowchart which shows the operation | movement of a noise removal process in the fingerprint authentication apparatus which is an example of the collation apparatus which concerns on this invention. It is a flowchart which shows the operation | movement of a noise removal process in the fingerprint authentication apparatus which is an example of the collation apparatus which concerns on this invention.

Explanation of symbols

1 Fingerprint Authentication Device 2 Operation / Display Unit 3 Fingerprint Input Unit (Image Input Unit)
DESCRIPTION OF SYMBOLS 4 Storage part 5 Image collation part 6 Output part 10 Image correction part 11 Binarization part 12 Thinning part 13 Feature extraction part 14 Collation part 15 Control part 110 Image division part 120 Frequency spectrum analysis part 130 Spectrum storage part 140 Frequency component selection Unit 150 image quality evaluation unit 151 direction line calculation unit 152 concentration degree calculation unit 153 evaluation unit 160 frequency component selection control unit 170 selection parameter storage unit 180 frequency filter unit 190 image reconstruction unit 200 fingerprint image synthesis unit 300 multi-value fingerprint image 310, 320, 330 Small region of multi-value fingerprint image 400, 500, 700, 800 Selected spectral image 410, 510, 710, 810 Frequency component 420, 520, 720, 820 Direction line 530, 730 Normal of direction line 600 Multi-value fingerprint Small area of image 610 Ridge 620 Branch Point 630 End point 900, 910 Feature point (end point)
920 feature point (branch point)

Claims (6)

An image matching device for matching an image including a stripe pattern,
An image input unit for acquiring the image;
A frequency spectrum analysis unit that obtains a frequency spectrum by converting the frequency of the image;
From the frequency spectrum, a frequency component selection unit that selects a frequency component whose absolute value of amplitude is a predetermined threshold value or more,
An image quality evaluation unit that examines whether the frequency component selected by the frequency component selection unit satisfies a quality of an image suitable for collation according to a predetermined condition;
If the image quality evaluation unit determines that the predetermined condition is satisfied, an image reconstructing unit that reconstructs the image by performing an inverse transform of the frequency transform based on the selected frequency component;
A matching unit for matching using the reconstructed image;
The collation apparatus characterized by having. A frequency component selection control unit for adjusting the predetermined threshold;
When the image quality evaluation unit determines that the predetermined condition is not satisfied, the frequency component selection control unit changes the predetermined threshold value, and the frequency component selection unit uses the changed predetermined threshold value to perform frequency again. The collation apparatus according to claim 1, wherein a component is selected.   The collation apparatus according to claim 2, wherein when the predetermined condition is not satisfied, the frequency component selection control unit sets the predetermined threshold value to a larger value than before the change. The image quality evaluation unit
Direction line calculation means for specifying a stripe direction of a stripe pattern included in the image based on the distribution of the selected frequency components;
Concentration calculating means for calculating an index representing the degree to which the selected frequency component is concentrated in the fringe direction,
2. An evaluation unit that evaluates that the predetermined condition is satisfied when the index indicates a concentration greater than or equal to a predetermined value, and that the predetermined condition is not satisfied when the index indicates a concentration less than a predetermined value. The collation apparatus as described in any one of -3. Further, a bandpass that attenuates a frequency component that is not included in a range within a predetermined frequency from a straight line on a frequency plane corresponding to the fringe direction with respect to the frequency component that the image quality evaluation unit determines to satisfy the predetermined condition. A frequency filter unit for performing a filtering process;
The collation apparatus according to any one of claims 1 to 4, wherein the image reconstruction unit reconstructs an image based on a frequency component subjected to the bandpass filter processing. And an image dividing unit that divides the image into a plurality of regions, and an image combining unit that combines the images divided into the plurality of regions into one image,
The image dividing unit divides the image input to the image input unit into a plurality of regions that can be regarded as a plurality of straight lines with stripe patterns substantially parallel,
The frequency spectrum analysis unit obtains a frequency spectrum for each of the plurality of regions,
The frequency component selection unit selects, for each of the plurality of regions, a frequency component whose absolute value of amplitude is a predetermined threshold or more,
The image quality evaluation unit examines whether or not the selected frequency component satisfies a predetermined condition for each of the plurality of regions,
When the image reconstruction unit determines that the selected frequency component satisfies the predetermined condition for each of the plurality of regions, the image reconstruction unit reconstructs an image based on the selected frequency component,
The collation apparatus according to any one of claims 1 to 5, wherein the image synthesis unit synthesizes the reconstructed images into one.

Images