JP2006209363A - Collating method and device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To highly accurately collate a reference image of a fingerprint or the like with an authentication subject image with a small memory capacity and low computational complexity. <P>SOLUTION: An imaging part takes a reference image and a subject image for collation. An arithmetic part calculates from the taken reference image features y11 to 14 characterizing directions of lines in the reference image in a plurality of directions, and calculates from the pickup subject image features y11 to 14 characterizing directions of lines in the subject image in a plurality of directions. A collating part collates the plurality of features y11 to 14 calculated in a plurality of directions in the subject image with the plurality of features y11 to 14 calculated in a plurality of directions in the reference image. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a verification method and apparatus that can be used when performing personal authentication using biometric information such as fingerprints and irises.

  In recent years, fingerprint authentication systems have been installed in mobile devices such as mobile phones. Unlike desktop PCs and large-scale systems, portable devices are limited in memory and CPU performance, so an authentication method that can be implemented with a small amount of memory and an inexpensive CPU is required.

  Conventional fingerprint authentication methods are roughly classified into (a) minutia method, (b) pattern matching method, (c) chip matching method, and (d) frequency analysis method. (A) In the minutiae method, fingerprint points are identified by extracting characteristic points such as ridge ends and branches from the fingerprint image, that is, minutiae, and comparing the information of these points between the two fingerprint images. And determine whether or not the person is the person.

  (B) The pattern matching method performs fingerprint comparison by directly comparing image patterns between two fingerprint images, and determines whether or not the person is the person. (C) The chip matching method has an image of a small area around a feature point, that is, a chip image as registration data, and performs fingerprint collation using this chip image. (D) The frequency analysis method performs frequency analysis on a line obtained by slicing a fingerprint image, and compares fingerprint components by comparing the distribution of frequency components in the direction perpendicular to the line direction between two fingerprint images. I do.

Patent Document 1 extracts a feature vector and its quality index from a fingerprint image, etc., assigns reliability information obtained using an error distribution of the feature vector to a feature amount, and performs fingerprint matching using these features. Disclose technology.
Japanese Patent Laid-Open No. 10-177650

  However, (a) the minutiae method and (c) the chip matching method require preprocessing such as connecting cut portions in the captured image, and the amount of calculation increases. (B) Since the pattern matching method has to store the data of the entire image, the memory capacity becomes large especially when data of a large number of people is registered. (D) Since the frequency analysis method requires frequency conversion, the amount of calculation increases. The above-mentioned Patent Document 1 also requires a statistical analysis, which increases the amount of calculation.

  The present invention has been made in view of these problems, and an object of the present invention is to provide a collation method and apparatus capable of performing collation with high accuracy with a small memory capacity and a small amount of calculation.

  In order to solve the above-described problem, a collation method according to an aspect of the present invention characterizes, from a reference image for collation, a feature amount that characterizes the direction of a line in the reference image, or a reference image as a single physical quantity. The method includes a step of calculating feature amounts for a plurality of directions and a step of recording a plurality of feature amounts calculated for a plurality of directions. The “line” may be a ridge or valley line of a fingerprint. The “feature amount characterizing the direction of the line” may be a value calculated based on the gradient vector of each pixel. The “single physical quantity” may be a vector quantity or a scalar quantity, and may be an image density switching mode such as a fringe switching count. According to this aspect, reference data with high accuracy can be registered with a small memory capacity.

  A feature amount that characterizes the direction of a line in the target image or a feature amount that characterizes the target image as a single physical quantity is calculated for a plurality of directions from the target image to be collated, and is calculated for a plurality of directions of the target image. There may be provided a step of collating a plurality of feature amounts with a plurality of feature amounts calculated for a plurality of directions of the reference image. According to this aspect, since a plurality of feature quantities are collated, it is possible to perform collation with high accuracy with a small memory capacity and a small calculation amount.

  The step of collating may be performed while changing the correspondence between the feature amounts to be collated. According to this aspect, the rotational deviation between the reference image and the target image can be detected, and the collation accuracy can be improved.

  Another aspect of the present invention is also a verification method. The method includes calculating, from at least one direction, a feature quantity that characterizes the direction of a line in the reference image or a feature quantity that characterizes the reference image as a single physical quantity from a reference image for collation; Calculating a feature quantity characterizing the direction of a line in the target image or a feature quantity characterizing the target image as a single physical quantity from at least one direction, a feature quantity of the reference image, and a target Collating with the feature quantity of the image. In the step of calculating from the reference image and the step of calculating from the target image, one of the reference image and the target image is calculated for one direction, and the other is calculated for a plurality of directions. Then, at least one or more of the feature quantities calculated for a plurality of directions are collated. According to this aspect, it is possible to detect a rotational shift by collating the feature amounts in a one-to-many manner, and it is possible to perform collation with high accuracy with a small memory capacity and a small calculation amount.

  Yet another embodiment of the present invention is also a collation method. This method includes a step of dividing a reference image for collation into a plurality of regions in a plurality of directions, a feature amount characterizing the direction of a line in the region for each divided region, or a single physical amount of the region. And a step of generating a feature amount group for each direction and a step of recording a plurality of feature amount groups calculated for a plurality of directions. According to this aspect, the “feature group” may be a function of coordinate axes along each direction. According to this aspect, reference data with high accuracy can be registered with a small memory capacity.

  A step of dividing the target image for collation into a plurality of regions in a plurality of directions, and a feature amount characterizing the direction of a line in the region for each divided region, or a feature characterizing the region as a single physical amount Calculating a quantity, generating a feature quantity group for each direction, recording a plurality of feature quantity groups calculated for a plurality of directions, a plurality of feature quantity groups calculated for a plurality of directions of the target image, and a reference image And a step of collating the feature amount groups calculated for the plurality of directions. According to this aspect, since a plurality of feature quantity groups are collated with each other, collation with high accuracy can be performed with a small memory capacity and a small calculation amount.

  The step of collating may be performed while changing the correspondence relationship of the feature amount groups to be collated. According to this aspect, the rotational deviation between the reference image and the target image can be detected, and the collation accuracy can be improved.

  Yet another embodiment of the present invention is also a collation method. In this method, a reference image for collation is divided into a plurality of regions in at least one direction, and a feature amount or a region that characterizes the direction of a line in the region is divided into a single region for each divided region. Calculating a feature quantity to be characterized as a physical quantity, generating a feature quantity group in at least one direction, dividing a target image to be collated into a plurality of areas in at least one direction, and, for each divided area, an area Calculating a feature quantity that characterizes the direction of a line in the image, or a feature quantity that characterizes a region as a single physical quantity, generating a feature quantity group for at least one direction, a feature quantity group of a reference image, and a target image And a step of collating with the feature quantity group. In the step of calculating from the reference image and the step of calculating from the target image, one of the reference image and the target image is calculated for one direction, the other is calculated for a plurality of directions, and the step of collating is a feature amount group calculated for one direction And at least one or more of the feature quantity groups calculated for a plurality of directions. According to this aspect, it is possible to detect a rotational shift by collating the feature amount groups in a one-to-many manner, and it is possible to perform collation with high accuracy with a small memory capacity and a small calculation amount.

  In the calculating step, when the feature amount group is calculated for a plurality of directions, the reference image or the target image may be rotated to calculate the feature amount group with respect to the reference direction. The “reference direction” may be a vertical direction or a horizontal direction. According to this aspect, it is possible to calculate a feature amount group for a plurality of directions with a simple algorithm.

  In the calculating step, when calculating the feature amount group along the oblique direction, the region may be set as a set of a plurality of sub regions, and the feature amount for each sub region may be rotated. The “sub-region” may be a square region along the reference direction. When the feature amount in the “sub-region” is a value calculated based on the gradient vector of each pixel, each gradient vector may be rotated according to the angle from the reference direction in the oblique direction. When this gradient vector is rotated, it may be rotated by referring to a predetermined conversion table. According to this aspect, it is possible to calculate a feature amount group for a plurality of directions with a small memory capacity.

  The calculating step may determine an angle range from the reference direction when calculating a feature amount group in a plurality of directions according to an assumed range of a relative positional relationship between the imaging target image and the imaging element. According to this aspect, a feature amount group is calculated in a direction that is likely to be wasted, and it is not recorded or collated. Therefore, a small memory capacity, a small calculation amount, and an accurate High verification can be performed.

  Yet another embodiment of the present invention is a collation device. The apparatus includes an imaging unit that captures a reference image and a target image for collation, and a feature amount that characterizes the direction of a line in the reference image from the captured reference image, or the reference image is characterized as a single physical quantity. The feature quantity to be attached is calculated for a plurality of directions, and the feature quantity that characterizes the direction of the line in the target image from the captured target image or the feature quantity that characterizes the target image as a single physical quantity is obtained for the plurality of directions. A calculation unit that calculates, and a collation unit that collates a plurality of feature amounts calculated in a plurality of directions of the target image and a plurality of feature amounts calculated in a plurality of directions of the reference image. According to this aspect, since a plurality of feature quantities are collated, it is possible to perform collation with high accuracy with a small memory capacity and a small calculation amount.

  The collation unit may collate while changing the correspondence relationship of the feature amounts to be collated. According to this aspect, by changing the correspondence relationship, it is possible to detect a rotational shift between the reference image and the target image, and it is possible to perform highly accurate collation with a small memory capacity and a small amount of calculation.

  Yet another embodiment of the present invention is also a verification device. The apparatus includes an imaging unit that captures a reference image and a target image for collation, and a feature amount that characterizes the direction of a line in the reference image from the captured reference image, or the reference image is characterized as a single physical quantity. And at least one feature amount that characterizes the direction of a line in the target image or a feature amount that characterizes the target image as a single physical amount from the captured target image. A calculation unit that calculates the direction, a reference image feature amount, and a target image feature amount are provided. The calculation unit calculates one of the reference image and the target image for one direction and the other for a plurality of directions, and the matching unit calculates at least one of the feature amount calculated for one direction and the feature amount calculated for the plurality of directions. Match more than one. According to this aspect, it is possible to detect a rotational shift by collating the feature amounts in a one-to-many manner, and it is possible to perform collation with high accuracy with a small memory capacity and a small calculation amount.

  The calculation unit may determine an angle range from the reference direction in the case of calculating a feature amount in a plurality of directions according to an assumed range of a relative positional relationship between the imaging target image and the imaging unit. The “imaging unit” may include a guide that regulates the movement of the imaging target on the imaging region. A line sensor may also be included. According to this aspect, the feature amount is calculated in a direction that is likely to be wasted, and it is not recorded or collated. Therefore, a small amount of memory, a small amount of calculation, and high accuracy are achieved. Verification can be performed.

  It should be noted that any combination of the above-described constituent elements and a conversion of the expression of the present invention between a method, an apparatus, a system, a computer program, a recording medium, and the like are also effective as an aspect of the present invention.

  According to the present invention, it is possible to perform collation with high accuracy with a small memory capacity and a small calculation amount.

(Embodiment 1)
The first embodiment obtains one vector characterizing the direction of a ridge or valley in a line-shaped area along a line perpendicular to the reference direction on the fingerprint image. The component is calculated. Then, the distribution of the reference direction of the component is obtained, and compared with the distribution of registered data obtained by the same method, the fingerprint images are collated and authenticated.

  FIG. 1 is a diagram illustrating functional blocks of the matching device 1 according to the first embodiment. Each block shown here can be realized in hardware by various elements such as a processor and RAM, and various devices such as sensors, and in software, it can be realized by a computer program. Depicts functional blocks realized by. Accordingly, those skilled in the art will understand that these functional blocks can be realized in various forms by a combination of hardware and software.

  The collation device 1 includes an imaging unit 100 and a processing unit 200. The imaging unit 100 uses a CCD (Charge Coupled Device) or the like, images a user's finger, and outputs the image to the processing unit 200 as image data. For example, when a line sensor such as a CCD is mounted on a portable device, a fingerprint image is collected by holding a finger over the line sensor and sliding the finger vertically with respect to the line sensor. May be.

  The processing unit 200 includes an image buffer 210, a calculation unit 220, a collation unit 230, and a recording unit 240. The image buffer 210 is a memory area that temporarily stores image data input from the imaging unit 100 and is also used as a work area of the calculation unit 220. The calculation unit 220 performs various calculations described later on the image data in the image buffer 210. The collation unit 230 compares the feature amount of the image data stored in the image buffer 210 to be an authentication target with the feature amount of the image data registered in the recording unit 240, and determines whether the fingerprints are the same person. Determine whether or not. The recording unit 240 registers the feature amount of a fingerprint image captured in advance. When used for a mobile phone or the like, data for one person is often registered, but when used for a gate in a room or the like, data for a plurality of persons is often registered.

  FIG. 2 is a flowchart for explaining processing for generating reference data of the collation device 1 according to the first embodiment. In this reference data, a fingerprint image of a person to be authenticated is registered in advance as a distribution of a predetermined direction component, here as a distribution of a feature amount characterizing a ridge or valley direction in a linear region. Is.

  First, the imaging unit 100 captures an image of a finger held by the user, converts the captured image into an electrical signal, and outputs the electrical signal to the processing unit 200. The processing unit 200 acquires the signal as image data and temporarily stores it in the image buffer 210 (S10). The arithmetic unit 220 converts the image data into binary data (S12). For example, a value brighter than a predetermined threshold is determined to be white, and a value darker than that is determined to be black. White is expressed as 1 or 0, and black is expressed as 0 or 1.

  Next, the calculation unit 220 divides the binarized image data for each linear region (S14). FIG. 3 is a diagram illustrating a captured fingerprint image in the first embodiment. In FIG. 3, the calculation unit 220 forms a linear region 12 with the X-axis direction as the longitudinal direction and the Y-axis direction as the short direction on the fingerprint image. In this linear region, the short direction is preferably set to one pixel or three pixels. A plurality of such linear regions are formed in the Y-axis direction, that is, the longitudinal direction of the finger, and the fingerprint image is divided into a plurality of regions.

  Next, the calculation unit 220 calculates the gradient of each pixel (S16). As a method for calculating the gradient, the method described in the document “Computer Image Processing, edited by Hideyuki Tamura, Ohm Co., Ltd. (p182-191)” can be used.

Briefly described below, in order to calculate a gradient for a digital image, it is necessary to calculate first-order partial differentials in the x and y directions.
Δ _x f (i, j) ≡ {f (i + 1, j) −f (i−1, j)} / 2 (Expression 1)
Δ _y f (i, j) ≡ {f (i, j + 1) −f (i, j−1)} / 2 (Expression 2)

In the difference operator for a digital image, the differential value of the pixel at (i, j) is a density value of a pixel in the vicinity of 3 × 3 centered at (i, j), that is, f (i ± 1, j ± 1). Defined as a linear combination. This means that the calculation for obtaining the derivative of the image can be realized by spatial filtering using a 3 × 3 weighting matrix, and various difference operators can be expressed by a 3 × 3 weighting matrix. In the following, a 3 × 3 neighborhood centered at (i, j) is represented by f (i−1, j−1) f (i, j−1) f (i + 1, j−1)
f (i-1, j) f (i, j) f (i + 1, j) (Formula 3)
f (i-1, j + 1) f (i, j + 1) f (i + 1, j + 1)
The difference operator can be described by a weighting matrix for this.

For example, the first-order partial differential operators in the x and y directions defined by Equation 1 and Equation 2 are
0 0 0 0 -1/2 0
−1/2 0 1/2 and 0 0 0 (Equation 4)
0 0 0 0 1/2 0
It is expressed. That is, in the 3 × 3 rectangular area of Equation 3 and Equation 4, the product of the value of the pixel at the corresponding position and the value of the element of the matrix are obtained, and the sum of those is calculated, and the right side of Equation 1 and Equation 2 Match.

The gradient magnitude and direction are spatially filtered using the weighting matrix of Equation 4, and after calculating partial differentials in the x and y directions defined by Equations 1 and 2,
| ∇f (i, j) | = √ {Δ x f (i, j) 2 + Δ y f (i, j) 2} ... ( Equation 5)
_{^{θ = tan -1 {Δ y f}} (i, j) / Δ x f (i, j)} ... ( Equation 6)
It can be asked.

  Note that the difference operator can be simply calculated using a Roberts operator, a Plewit operator, a Sobel operator, or the like, and noise countermeasures can be taken.

  Next, the computing unit 220 obtains a set of numerical values obtained by doubling the gradient direction obtained by Equation 6, that is, the angle of the gradient vector (S18). In this embodiment, the direction of the ridge or valley line of the fingerprint is calculated using the gradient vector, but the value of the gradient vector is not the same even if the ridge or valley line is in the same direction. . Therefore, by rotating the gradient vector so that the angle formed by the gradient vector and the coordinate axis is doubled to obtain a set of numerical values composed of the x component and the y component, a ridge or valley line in the same direction is obtained. It can be expressed by a set of unique numerical values having the same components. For example, 45 ° and 225 ° are just opposite directions, but if doubled, they become 90 ° and 450 °, indicating a unique direction. Here, a set of numerical values composed of an x component and a y component is obtained by rotating a vector according to a certain rule in a certain coordinate system. Hereinafter, this numerical value is also expressed as a vector in this specification.

  The direction of ridges or valleys in a certain area of a fingerprint image varies widely locally, so that an average is taken in a certain range as will be described later. In that case, if the angle of the gradient vector is doubled to make it a unique vector as described above, an approximate value of the direction of the ridge or valley is obtained by simply adding them and averaging them. be able to. If this is not done, adding a gradient vector pointing in the opposite direction will result in 0, and a simple addition cannot be used. In this case, a complicated calculation is required to compensate that 180 ° and 0 ° are equivalent.

  Next, the arithmetic unit 220 calculates an averaged vector by adding the vectors calculated for each pixel for each linear region. This vector becomes the feature amount of the region (S20). This feature amount is a value indicating the average of the direction of the ridge or valley of the region, and one feature amount is determined for each region.

  At this time, when a plurality of white points or black points continue, a state in which the gradient cannot be defined continues. Therefore, when the number exceeds a predetermined number, the portion may be excluded from the target of the averaging process. The predetermined number of optimum values may be obtained experimentally.

  Finally, the calculation unit 220 obtains the x component and y component of the vector obtained for each region and records them as reference data in the recording unit 240 (S22). Note that the calculation unit 220 may record the x component and y component distribution of the vector after performing a smoothing process as described later.

  FIG. 4 is a diagram showing a vector V (y0) representing the feature in the linear region 12 of FIG. The linear region 12 is a region cut out at y = y0 on the coordinate plane shown in FIG. FIG. 4 shows a vector V = (Vx, Vy) representing features in the ridge or valley direction within that region. Vx (y0) and Vy (y0) represent end point coordinates on orthogonal coordinates with the start point of the vector V (y0) as the origin. As the feature amount, an average value obtained by doubling the angle of the gradient vector of each pixel described above can be used.

  FIG. 5 is a diagram illustrating a distribution of feature amounts when the image of FIG. 3 is sliced into linear regions. FIG. 4 shows a distribution of feature amounts when scanning is performed in the y direction on the coordinate plane shown in FIG. 3, that is, in a direction perpendicular to the slice direction of a linear region. The horizontal axis in FIG. 5 indicates the y axis in FIG. 3, that is, the sub-scanning direction, and the vertical axis indicates the feature amount in each region. In FIG. 5, as shown in FIG. 4, a vector characterizing each region is expressed by an x component and a y component. The calculation unit 220 can obtain the distribution of the x component and the y component of the vector characterizing each region from the fingerprint image to be registered, and record it as reference data in the recording unit 240.

  FIG. 6 is a flowchart for explaining the authentication process of the verification device 1 according to the first embodiment. First, the imaging unit 100 captures an image of a finger held by a user requesting authentication, converts the captured image into an electrical signal, and outputs the electrical signal to the processing unit 200. The processing unit 200 performs the same processing as the processing shown in steps S12 to S20 in FIG. 2 on the acquired image, and calculates the feature amount distribution of the image data as authentication target data (S30).

  Next, the computing unit 220 applies a smoothing process to the feature amount distribution (S32). For example, several points of the previous and subsequent values are averaged. The degree of smoothing depends on the application and the optimum value may be obtained experimentally.

  Next, the collation unit 230 collates the feature quantity distribution of the reference data with the feature quantity distribution of the authentication target data (S34). This collation is performed by obtaining the most matching pattern by fixing one distribution and sliding the other distribution. You can simply perform pattern matching for all patterns, but in order to reduce the amount of calculation, detect the attention points of both distributions, and only perform matching of several patterns around the matching points. May be. For example, a point that becomes the maximum value of the x component, a point that is 0, a point that becomes 0 when differentiated, a point with the steepest slope, or the like can be used as a point of interest.

Pattern matching can be performed by detecting a difference between each reference sentence and authentication target data for each point on the y-axis. For example, the calculation is performed by calculating the matching energy E by the following equation (7).
E = Σ√ {ΔVx ² + ΔVy ² } Equation 7

  Taking the square root of the value obtained by squaring the error ΔVx of the x component and the error ΔVy of the y component of both distributions, the x component and the y component are originally vector components, so that the vector size error is obtained. be able to. A value obtained by adding them in the y-axis direction is defined as a matching energy E. Therefore, the larger the energy E, the closer the image becomes, and the smaller the energy E, the closer the image. Then, the pattern having the smallest matching energy E is set as the overlapping position (S36). The pattern matching method is not limited to this, and the absolute values of the x component error ΔVx and the y component error ΔVy of both distributions may be added together. A highly accurate collation method may be experimentally obtained and used.

  FIG. 7 is a diagram depicting the feature quantity distribution of the reference data and the feature quantity distribution of the authentication target data in the first embodiment. In FIG. 7, the feature amount distribution of the reference data is indicated by a solid line, and the feature amount distribution of the authentication target data is indicated by a broken line. In the example of FIG. 7, the maximum value of the x component of both distributions is first detected. Then, pattern matching is performed at a position where the maximum values p1 coincide with each other and a position where one of the distributions is shifted from that position by several points, and the position that best matches is set as the overlapping position.

  The collation unit 230 compares the calculated matching energy E with a threshold value for determining whether or not to succeed in the preset authentication. If the matching energy E is less than the threshold value, the reference data The user of the fingerprint image is authenticated (S38). Conversely, if the matching energy E is greater than or equal to the threshold value, authentication is not performed. The above processing is performed between each reference data and authentication target data when a plurality of reference data is registered.

  As described above, according to the first embodiment, an image of biometric information such as a fingerprint is divided into predetermined areas, and values that characterize each area are used for matching processing between the reference image and the authentication target image. As a result, authentication processing can be performed with a small memory capacity. Also, the amount of calculation can be reduced, and the authentication process can be speeded up. Therefore, if it is applied to authentication processing for a portable device such as a battery drive, the circuit area can be reduced and power can be saved. Furthermore, by obtaining a feature amount for each linear region, it is suitable for collation of fingerprint images captured by a line sensor or the like. Furthermore, the feature amount of each pixel is averaged for each region, and the distribution of the averaged feature amount is smoothed, so that it is also resistant to noise. In addition, since the averaging process is performed on the linear area, it is possible to collate whether or not the fingers are the same person even if the finger slides left and right between the reference data and the authentication target data. Further, unlike the minutiae method, even a noisy fingerprint image can function effectively if registration and authentication are performed in the same state.

(Embodiment 2)
In the first embodiment, an example of dividing in one direction has been described as a method of dividing an image for obtaining a linear region. In this regard, in the second embodiment, an example of dividing in a plurality of directions will be described. For example, an image can be sliced in two directions.

  First, the configuration and operation of the collation device 1 according to the second embodiment are the same as those of the collation device 1 according to the first embodiment shown in FIG. FIG. 8 is a diagram illustrating an example of slicing an image in the vertical direction according to the second embodiment. Here, as an image to be collated, in addition to the above-described fingerprint image, an iris image or other images expressing biological information may be used. In FIG. 8, it is expressed as a striped pattern image for convenience. FIG. 9 is a diagram illustrating the distribution of feature amounts in each linear region in FIG. The feature amounts A and B characterizing each linear region are obtained, and the distribution in the y direction, that is, the vertical direction is shown.

  FIG. 10 is a diagram illustrating an example of slicing an image in the oblique 45 ° direction according to the second embodiment. FIG. 10 slices the same image as in FIG. 8 in a 45 ° oblique direction. FIG. 11 is a diagram showing the distribution of feature amounts in each linear region in FIG. The feature amounts C and D characterizing each linear region are obtained, and the distribution in the z direction, that is, the oblique 45 ° direction is shown. In this manner, the image is sliced in two directions, and the feature of the image is expressed by the four types of feature amounts A, B, C, and D. The process of generating reference data and the process of authenticating in the second embodiment can be performed in the same manner as in FIGS. 2 and 6 described in the first embodiment, using these feature amounts. In this case, since there are a plurality of matching energies E calculated from each direction, for example, the collation unit 230 may determine the average value to determine whether to authenticate.

  Here, the feature amount is not limited to the x component and the y component of the vector that characterizes the ridge or valley line in the linear region described in the first embodiment. For example, it may be image density, brightness, color information, a lot of local image information, or a scalar amount, a vector amount, or other numerical values obtained from these image information by differentiation or other calculations.

  Note that the image in the second embodiment may be captured by a one-dimensional sensor such as a line sensor, loaded into the image buffer 210, and sliced in two or more directions, or a two-dimensional sensor may be used. Examples of two or more directions are generally the vertical direction, the horizontal direction, the oblique 45 °, and the oblique 135 °, but are not limited thereto, and can be sliced in any direction. Furthermore, the combination of two or more directions is not limited to the vertical direction and the oblique 45 °, and can be arbitrarily set.

  As described above, according to the second embodiment, by using a plurality of different directions as slice directions of an image for obtaining a linear region, the accuracy of matching can be improved as compared with the first embodiment. . In the second embodiment, another image is not generated from a certain image as in the minutia method, and therefore, it is sufficient if the memory capacity is sufficient to store the original image. Therefore, highly accurate authentication can be performed with a small memory capacity and a small amount of calculation.

(Embodiment 3)
In the first and second embodiments, an example in which the sliced linear region has a linear shape has been described. This shape is not limited thereto, and may be a non-linear shape such as a curved shape, a closed curved shape, a circular shape, or a concentric circular shape. In the third embodiment, the matching of iris images will be described as a typical example of dividing into non-linear regions.

  First, since the configuration and operation of the collation device 1 in the third embodiment are also the same as those of the collation device 1 in the first embodiment shown in FIG. FIG. 12 is a diagram illustrating an example of dividing the iris portion of the eye into concentric regions in the third embodiment. In FIG. 12, each linear region is divided into concentric circles. FIG. 13 is a diagram showing the distribution of feature amounts in each concentric region in FIG. The distribution of the dynamic system direction r of values characterizing each region is obtained. The process of generating reference data and the process of authenticating in the third embodiment can be performed in the same manner as in FIGS. 2 and 6 described in the first embodiment, using this feature amount.

  In the case of an iris, the matching process described in FIG. 6 is simplified. The distribution of both feature amounts may be overlapped so that the boundary R0 between the pupil and the iris and the boundary R1 between the iris and the white eye portion match between the authentication target data and the reference data. However, since the size of the pupil changes depending on the brightness of the surroundings, it is necessary to expand and contract uniformly so that the distance between the boundary R0 and the boundary R1 of the authentication target data matches that of the reference data.

  As described above, according to the third embodiment, the iris image can be collated with a small memory capacity and a small calculation amount. In addition, since the feature amount is obtained by averaging the gradient vector of each pixel in each concentric region, even if the eye position is rotated with respect to the reference data, it is possible to accurately collate. This property corresponds to the property that the finger image is resistant to sliding in the horizontal direction in the fingerprint image of the first embodiment. In addition, even when the eyelid is blocked and all the portions of the iris cannot be imaged, a considerable degree of accuracy can be maintained even if collation is performed on a plurality of regions near the pupil.

  When half of the iris is used for collation, it may be divided not for concentric areas but for each semicircular area. Further, such a method of dividing in a non-linear region is not limited to an iris image but can be applied to a fingerprint image. For example, the center of the fingerprint vortex may be detected and concentrically divided from there.

(Embodiment 4)
In the second embodiment, the example of dividing in a plurality of directions has been described. That is, highly accurate collation is performed by comparing a plurality of features obtained by dividing a plurality of directions with respect to the reference image and the authentication target image. In this regard, in the fourth embodiment, when comparing a plurality of images, it is assumed that the number of features acquired in one or more directions is different for each image. Then, by comparing the features in various combinations between the images, the rotational deviation is compensated.

  First, it is assumed that either the reference image or the authentication target image has a plurality of features extracted in a plurality of directions, and the other has a feature extracted in one direction. Here, the feature may be obtained as a function of a coordinate axis indicating a reference direction for obtaining the feature, or may be a distribution of feature amount calculated based on the gradient vector as described above. . FIG. 14 is a diagram illustrating an example of extracting features in a plurality of directions of an image according to the fourth embodiment. Features are extracted for a plurality of directions with different angles. FIG. 14 shows only four directions for convenience. More specifically, only four coordinate axes y11 to 14 are set. In this regard, coordinate axes may be set for a plurality of directions, and the angle formed between adjacent coordinate axes may be reduced. The more the direction in which the features are extracted, the more accurately the rotational shift when collating the images can be detected and compensated.

  FIG. 15 is a diagram illustrating a modification example in which features are extracted in a plurality of directions of an image according to the fourth embodiment. When calculating features for a plurality of directions of an image, the angle range for setting the directions is limited. In FIG. 15, when a plurality of coordinate axes y21 to 24 having different angles are set on the image, a coordinate axis that is narrower than the vertical axis is set, and a wide-angle coordinate axis is not set. Such setting of the direction on the image is associated with the physical form of the imaging unit 100 that captures the image.

  For example, if a finger is slid from top to bottom or from bottom to top with respect to a line sensor mounted horizontally, and a fingerprint image is captured, a guide for regulating the sliding of the finger is provided. The range of finger rotation deviation is regulated. If a direction is set on the image in association with this, it is not necessary to extract a useless feature for detecting a rotational deviation in a direction that is not normally possible. On the other hand, when there is a possibility that the image captured by the finger or the like is rotationally displaced in any direction as in the case of image capturing by a surface sensor or the like, as shown in FIG. It is good to extract features.

  Hereinafter, the details of the fourth embodiment will be described based on this assumption. The configuration and operation of the collation device 1 in the fourth embodiment are basically the same as those of the collation device 1 in the first embodiment shown in FIG. Hereinafter, different parts will be described. FIG. 16 is a flowchart for explaining the process of collating a plurality of images in the fourth embodiment. The flowchart shown in FIG. 16 is regarded as a subroutine of the collation process (S34) in the flowchart shown in FIG.

  First, an example will be described in which either the reference image or the authentication target image has characteristics in a plurality of directions shown in FIGS. 14 and 15 and the other has characteristics in one direction. More specifically, a case where the reference image is in a plurality of directions and the authentication target image is in one direction will be described.

  In FIG. 16, the calculation unit 220 sequentially compares the feature calculated for a predetermined direction of the authentication target image with each feature calculated for a plurality of directions of the reference image registered in advance in the recording unit 240. (S362). Here, it is not necessary to perform pattern matching on the same axis described in the flowchart of FIG. 6 for each comparison. The feature of the reference image that most closely approximates the feature of the authentication target image may be specified. For each feature, the sum of the feature values constituting the feature may be calculated and compared. The order of comparison may be any order, and various algorithms can be used. For example, the comparison target may be gradually shifted from the direction of an angle close to horizontal to the direction of an angle close to vertical alternately left and right. Furthermore, as a result of the comparison, if there is a feature with very high accuracy of matching, it is not necessary to determine that feature and compare it with the remaining features.

  The calculation unit 220 determines a combination of the feature of the authentication target image and the feature of the reference image that is closest to the feature (S364). Hereinafter, as described in the flowchart of FIG. 6, the features are compared in more detail to determine whether or not to authenticate. Further, each comparison between the feature of the authentication target image and each feature of the reference image may be processed as shown in FIG. And when the collation result of the matching degree which should permit authentication is obtained, authentication may be permitted at that stage.

  FIG. 17 is a diagram for describing an example in which one image to be collated in Embodiment 4 has a feature in a plurality of directions and the other image has a feature in one direction. In FIG. 17, one fingerprint image 10 has coordinate axes y11 to 14 set in a plurality of directions. The calculation unit 220 calculates a feature along each coordinate axis. A feature amount y11 in FIG. 17 indicates a feature with respect to the corresponding coordinate axis y11. The other fingerprint image 20 has a coordinate axis y10 set in the vertical direction. In this case, the feature y20 with respect to the coordinate axis y10 and the features y11 to 14 of the former fingerprint image 10 are compared and collated. Therefore, in FIG. 17, four comparisons and comparisons are performed.

  In addition, when the feature of the reference image is set to plural and the feature of the authentication target image is set to one, the time for extracting the feature of the authentication target image does not increase at the time of collation, and thus, while suppressing the increase in the collation time, The collation accuracy can be improved. On the other hand, when the number of features of the reference image is set to one and the number of features of the authentication target image is set, the verification accuracy can be improved while suppressing the data volume to be recorded as the reference data. The collation method as shown in FIG. 17 can be applied not only when comparing and collating between two images, but also when performing between three or more images. It suffices if at least one of the images has a feature in one direction.

  Next, an example will be described in which both the reference image and the authentication target image have characteristics in a plurality of directions shown in FIGS. Then, the relative angle is changed between the reference image and the authentication target image, and comparison and verification are performed a plurality of times. For example, comparison and collation are performed while rotating either image, such as a form in which both images are not tilted, a form in which one of them is slightly tilted, or a form in which one of them is further tilted.

  In FIG. 16, the calculation unit 220 includes a feature group including features extracted in a plurality of directions of the authentication target image and features extracted in a plurality of directions of the reference image registered in the recording unit 240 in advance. The feature groups are sequentially compared while changing the composition form of each feature constituting the feature group (S362). The computing unit 220 determines a combination of a feature group with a fixed composition form and a feature group with a composition form that most closely approximates the fixed feature group among the feature groups with the changed composition form (S364). Hereinafter, as described with reference to the flowchart of FIG. 6, the features constituting the feature group are collated in detail to determine whether or not to authenticate. In this case, as described in the second embodiment, authentication accuracy can be improved by comparing and collating features in a plurality of directions.

  FIG. 18 is a diagram for describing an example in which a plurality of images to be collated in the fourth embodiment have features in a plurality of directions. In FIG. 18, one fingerprint image 10 has coordinate axes y11 to 14 set in a plurality of directions. In the other fingerprint image 20, coordinate axes y11 to 14 are set in a plurality of directions. The calculation unit 220 extracts features along each coordinate axis. The feature group 15 is composed of the calculated features y11 to 14 for the former fingerprint image 10. For the latter fingerprint image 20 as well, the calculated features y11 to 14 are combined to form a feature group. And the composition form of this feature group is changed. Alternatively, a plurality of feature groups 21 to 24 having different composition forms may be preliminarily composed and sequentially used. In FIG. 18, four types of feature groups 21 to 24 are composed. Each feature group and the feature group 15 of the former fingerprint image 10 are compared and collated.

  Specifically, the features y11 to 14 constituting the feature group 15 of the former fingerprint image 10 are compared with the features y11 to 14 constituting the latter feature group 21 corresponding thereto. Next, with the former fixed, the features y11 to 14 constituting the feature group 22 having a different composition form from the latter are compared and verified with the corresponding features y11 to 14 respectively. Hereinafter, the latter will be compared and collated with feature groups having different composition forms. This means that both fingerprint images 10 and 20 are compared and verified while changing the corresponding features, and the situation where the latter fingerprint image 20 is rotated and compared with the former fingerprint image is simulated. can do. In this way, it is possible to accurately detect rotational deviations by extracting features in a plurality of directions of the reference image and the authentication target image and comparing and collating them while pretending to be a state in which one of the images is rotated. it can. Therefore, collation accuracy can be improved.

  Next, a specific method for extracting features in a plurality of directions of an image as shown in FIGS. 14 and 15 will be described. FIG. 19 is a diagram for explaining an example of extracting features in a plurality of directions by rotating an image. The calculation unit 220 extracts features with respect to a predetermined direction of image data input from the imaging unit 100 and temporarily stored in the image buffer 210. Next, the image data is rotated by a predetermined angle, and features are extracted with respect to a predetermined direction of the image data. By repeating this, it is possible to extract features in a plurality of directions of the image. In FIG. 19, the image data is rotated clockwise by a predetermined angle to extract features in the vertical direction of the image data in each state.

  Next, another method for extracting features in a plurality of directions of an image will be described. FIG. 20 is a diagram for explaining an example of extracting features in a plurality of directions without rotating an image. With respect to the vertical and horizontal directions, one linear region can be easily determined and features can be extracted. FIG. 20 illustrates a method for extracting features in an oblique direction without rotating an image. First, when extracting a feature in an oblique direction, the arithmetic unit 220 needs to set a linear region in a direction perpendicular to the direction. This linear area is simulated by a combination of predetermined rectangular areas L along the vertical and horizontal axes of the image. As described above, when the feature amount is calculated using the gradient vector of each pixel, the processing described below can be performed by setting a square area of 9 pixels at least in a matrix.

  The calculation unit 220 rotates the feature amount of each square region L according to the angle between the direction for obtaining the feature set in the image and the vertical or horizontal direction. As a result, it is possible to perform a calculation in a direction perpendicular to the direction in which the feature is substantially obtained. And the feature-value of each square area | region L is added together about the square area | region L which imitates the same linear area | region. For example, when using a gradient vector, each pixel vector in each square region L is calculated, and the vector is rotated according to the angle. The gradient vector obtained in this way is added to a square area that simulates the same linear area. In this case, the rotation amount of the gradient vector may be described as a table for each angle to be compensated in advance in the recording unit 240.

  By such processing, it is possible to extract features in a plurality of directions without rotating the image. That is, it is possible to simulate the same situation as when the image is rotated. Further, since it is not necessary to rotate the image data, a memory area necessary for the image data is not required, and this does not cause a pressure on the system.

  As described above, according to the fourth embodiment, image matching can be performed with high accuracy with a small memory capacity and a small calculation amount. For example, even when the fingerprint of the same person is used, the direction of the finger is different when registering reference data and when authentication is requested, and there is a high possibility of authentication failure if both images have a rotational shift. In this regard, according to the fourth embodiment, since the rotation deviation is corrected and the images are collated, if the fingerprints of the same person are used, it can be determined that the authentication is successful.

  The present invention has been described above based on the embodiment. It is to be understood by those skilled in the art that these embodiments are exemplifications, and various modifications can be made to the combination of each component, and such modifications are also within the scope of the present invention.

  In the above-described embodiment, for each linear region, a vector calculated from the gradient vector of each pixel is used as a single physical quantity that characterizes the region. In this regard, as the single physical quantity, the number of times of switching the image density may be used for each region. For example, the image may be binarized into a black and white image, and the number of times that the black and white are switched may be counted. The count value is the number of stripes such as fingerprints. Where the stripes are vertical, the density is high, and a lot of switching occurs within a certain distance, that is, per unit length. This may be done in the x and y directions. According to this, collation is possible with a simple calculation and a small memory capacity.

  In the above-described embodiment, binary data is used. However, multi-tone data such as 256 values may be used. Also in that case, the technique described in the above-mentioned document “Computer Image Processing, edited by Hideyuki Tamura, Ohm Co., Ltd., (p182-191)” can be used, and the gradient vector of each pixel can be calculated. According to this, although the amount of calculation increases compared with a black-and-white image, highly accurate collation can be performed.

  Furthermore, in the above-described collation process, a region that is not striped may be omitted from each linear region. For example, if a vector cannot be detected for a certain line length more than a set value, white continues for a set value or more, black is determined to be a black area that continues for more than a set value, black and white switching count When the value is less than the set value, the processing is performed by omitting the area. Thereby, the amount of calculation at the time of collation can be reduced.

It is a figure which shows the functional block of the collation apparatus in Embodiment 1. FIG. 6 is a flowchart for explaining a process of generating reference data for the collation device according to the first embodiment. It is a figure which shows the fingerprint image imaged in Embodiment 1. FIG. It is a figure which shows the vector V (y0) expressing the characteristic in the linear area | region of FIG. It is a figure which shows distribution of the feature-value at the time of slicing the image of FIG. 3 in each linear area | region. 5 is a flowchart for explaining an authentication process of the collation device according to the first embodiment. FIG. 3 is a diagram in which the feature quantity distribution of reference data and the feature quantity distribution of authentication target data are overlaid in the first embodiment. It is a figure which shows the example which slices the image in Embodiment 2 to the orthogonal | vertical direction. It is a figure which shows distribution of the feature-value in each linear area | region of FIG. It is a figure which shows the example which slices the image in Embodiment 2 in the diagonal 45 degree direction. It is a figure which shows distribution of the feature-value in each linear area | region of FIG. It is a figure which shows the example which divides | segments the iris part of the eye in Embodiment 3 into a concentric area | region. It is a figure which shows distribution of the feature-value in each concentric area | region of FIG. FIG. 10 is a diagram illustrating an example of extracting features with respect to a plurality of directions of an image according to a fourth embodiment. FIG. 10 is a diagram illustrating a modification example in which features are extracted with respect to a plurality of directions of an image in the fourth embodiment. 10 is a flowchart for explaining a process of collating a plurality of images in the fourth embodiment. FIG. 10 is a diagram for describing an example in which one image to be a collation target in Embodiment 4 has characteristics in a plurality of directions and the other image has characteristics in one direction. It is a figure for demonstrating the example in which the some image which should be made into the collation target in Embodiment 4 has the characteristic with respect to several directions. It is a figure for demonstrating the example which rotates an image and extracts a characteristic with respect to several directions. It is a figure for demonstrating the example which extracts a characteristic with respect to several directions, without rotating an image.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 collation apparatus, 100 imaging part, 200 process part, 210 image buffer, 220 calculating part, 230 collation part, 240 recording part.

Claims (15)

Calculating a feature quantity characterizing the direction of a line in the reference image from a reference image for matching, or a feature quantity characterizing the reference image as a single physical quantity for a plurality of directions;
Recording a plurality of feature quantities calculated for the plurality of directions;
A collation method comprising: Calculating a feature amount that characterizes the direction of a line in the target image or a feature amount that characterizes the target image as a single physical amount for a plurality of directions from the target image to be verified;
Collating a plurality of feature amounts calculated for a plurality of directions of the target image with a plurality of feature amounts calculated for a plurality of directions of the reference image;
The collation method according to claim 1, further comprising:   The collation method according to claim 2, wherein the collating step performs collation while changing a correspondence relationship of feature amounts to be collated. Calculating, from at least one direction, a feature amount that characterizes the direction of a line in the reference image or a feature amount that characterizes the reference image as a single physical amount from a reference image for matching;
Calculating, from at least one direction, a feature quantity that characterizes the direction of a line in the target image or a feature quantity that characterizes the target image as a single physical quantity from the target image to be collated;
Collating the feature quantity of the reference image with the feature quantity of the target image,
The step of calculating from the reference image and the step of calculating from the target image calculate one of the reference image and the target image in one direction and the other in a plurality of directions,
The collating method is characterized by collating a feature amount calculated for one direction with at least one of feature amounts calculated for a plurality of directions. Dividing a reference image for verification into a plurality of regions in a plurality of directions;
For each divided area, calculating a feature quantity characterizing the direction of a line in the area, or a feature quantity characterizing the area as a single physical quantity, and generating a feature quantity group for each direction;
Recording a plurality of feature amount groups calculated for the plurality of directions;
A collation method comprising: Dividing the target image for verification into a plurality of regions in a plurality of directions;
For each divided area, calculating a feature quantity characterizing the direction of a line in the area, or a feature quantity characterizing the area as a single physical quantity, and generating a feature quantity group for each direction;
Recording a plurality of feature amount groups calculated for the plurality of directions;
Collating a plurality of feature amount groups calculated for a plurality of directions of the target image with a feature amount group calculated for a plurality of directions of the reference image;
The collation method according to claim 5, further comprising:   The collation method according to claim 6, wherein the collating step performs collation while changing a correspondence relationship between the feature amount groups to be collated. Dividing a reference image for verification into a plurality of regions in at least one direction;
Calculating a feature quantity characterizing the direction of a line in the area for each divided area, or a feature quantity characterizing the area as a single physical quantity, and generating a feature quantity group in at least one direction;
Dividing the verification target image into a plurality of regions in at least one direction;
Calculating a feature quantity characterizing the direction of a line in the area for each divided area, or a feature quantity characterizing the area as a single physical quantity, and generating a feature quantity group for at least one direction;
Collating the feature quantity group of the reference image with the feature quantity group of the target image,
The step of calculating from the reference image and the step of calculating from the target image calculate one of the reference image and the target image in one direction and the other in a plurality of directions,
The collating step is characterized by collating a feature amount group calculated for one direction with at least one of feature amount groups calculated for a plurality of directions.   9. The feature calculation method according to claim 5, wherein, in the step of calculating, when calculating a feature amount group for a plurality of directions, the reference image or the target image is rotated to calculate a feature amount group with respect to a reference direction. The verification method described in Crab.   The calculating step includes calculating the feature amount group along an oblique direction by setting the region as a set of a plurality of sub-regions and rotating the feature amount for each sub-region. The collation method according to any one of 5 to 9.   The calculating step includes determining an angle range from a reference direction when calculating the feature amount group in a plurality of directions according to an assumed range of a relative positional relationship between the imaging target image and the imaging element. The collation method according to any one of claims 5 to 10. An imaging unit that captures a reference image and a target image for verification;
A feature quantity that characterizes the direction of a line in the reference image or a feature quantity that characterizes the reference image as a single physical quantity is calculated for a plurality of directions from the captured reference image, and the target is calculated from the captured target image. A calculation unit that calculates a characteristic amount that characterizes the direction of a line in the image or a characteristic amount that characterizes the target image as a single physical quantity for a plurality of directions;
A collation unit for collating a plurality of feature amounts calculated for a plurality of directions of the target image with a plurality of feature amounts calculated for a plurality of directions of the reference image;
A collation device comprising:   The collation apparatus according to claim 12, wherein the collation unit performs collation while changing a correspondence relationship of feature amounts to be collated. An imaging unit that captures a reference image and a target image for verification;
A feature amount that characterizes the direction of a line in the reference image or a feature amount that characterizes the reference image as a single physical quantity is calculated in at least one direction from the captured reference image, and from the captured target image, A calculation unit that calculates a feature amount that characterizes the direction of a line in the target image or a feature amount that characterizes the target image as a single physical quantity in at least one direction;
A matching unit that compares the feature amount of the reference image and the feature amount of the target image,
The calculation unit calculates one of the reference image and the target image for one direction and the other for a plurality of directions.
The collation unit is configured to collate a feature amount calculated for one direction with at least one of feature amounts calculated for a plurality of directions.   The calculation unit determines an angle range from a reference direction when calculating a feature value in a plurality of directions according to an assumed range of a relative positional relationship between an imaging target image and the imaging unit. Item 15. The verification device according to any one of Items 12 to 14.

Images