WO2014017697A1

WO2014017697A1 - Method and device for extracting finger vein pattern using guided gabor filter

Info

Publication number: WO2014017697A1
Application number: PCT/KR2012/008410
Authority: WO
Inventors: 해산연; 박동선; 윤숙
Original assignee: 전북대학교산학협력단; 목포대학교산학협력단
Priority date: 2012-07-25
Filing date: 2012-10-16
Publication date: 2014-01-30
Also published as: KR101315646B1

Abstract

Disclosed are a method and a device for extracting a finger vein pattern using a guided Gabor filter. The device for extracting a finger vein pattern, according to the present invention, comprises: a preprocessing unit which obtains a captured image including a vein pattern of a user by using near-infrared light, and which sets a region of interest that includes a finger vein from the captured image; a guided filter unit which performs guided filtering for the captured image obtained in an image obtaining unit by using a preset guide image, and which outputs a guided image having a more improved picture quality than the captured image; and a Gabor filter unit which performs, relative to the guided image, Gabor filtering for extracting the shape of a ridge from the guided image, and which outputs a filtering image. Thus, it is possible to extract a very clear finger vein pattern at high speeds by using an image picture quality improving capability of the guided filter and a ridge extracting capability of the Gabor filter.

Description

Method and apparatus for extracting finger vein pattern using guided Gabor filter

The present invention relates to a finger vein pattern extraction method and apparatus, and more particularly to a finger vein pattern extraction method and apparatus using a guided Gabor filter.

As the importance of security increases in the modern society, various user authentication methods have been proposed and used. The user authentication method using biometrics among user authentication methods is a user authentication method that has recently been in the spotlight as a method of providing reliable security since the user's body information is used as a security medium for user authentication. Currently used biometrics authentication methods include fingerprint recognition, hand shape, face contours (eyes, nose, eyebrows, mouth, cheeks, etc.), ear shapes, voiceprints, Authentication methods using various biometrics such as the eye retina, iris, back of hand, vein pattern of finger, sign, DNA, etc. have been devised and are expanding the scope of application.

Fingerprint recognition has been in the spotlight as the most widely used biometrics technology for user authentication. However, various counterfeit methods such as counterfeit fingerprints have recently emerged. In addition, fingerprint recognition has a problem that it is difficult to recognize when the finger is wound or when the finger is dry.

Finger vein pattern identification is emerging as a new user authentication method that can replace fingerprint recognition because finger vein is more difficult to forge than fingerprint and has unique pattern for each person like fingerprint.

The finger vein, unlike the fingerprint in the human body, is less affected by the human condition, and the chance of injury is very low. In addition, since the pattern does not change from birth to death, it is suitable for user authentication.

Finger vein pattern identification is to determine the pattern of the finger vein using the characteristic that the hemoglobin in the vein absorbs infrared light and appears darker than other areas. Vein pattern identification captures vein patterns using harmless infrared light or near-infrared light with a wavelength between 0.7 μm and 1 μm. By absorbing it is represented by the shadow area.

Conventional finger vein pattern identification methods include a curvature-based method, a line tracking method, and a gabor filter method. However, the finger vein pattern identification method has difficulty in extracting the finger vein due to uneven lighting or local illumination changes caused by various tissues and bones inside the finger. It is difficult to determine the threshold for dividing a, and there is a problem in that performance decreases rapidly according to the image quality of the image.

An object of the present invention is to provide a finger vein pattern extraction method using a guided Gabor filter to enable clear finger vein pattern extraction without segmentation of the image.

Another object of the present invention to provide a finger vein pattern extraction apparatus using the finger vein pattern extraction method.

Finger vein pattern extraction apparatus according to an embodiment of the present invention for achieving the above object to obtain a capture image including the vein pattern of the user by using infrared light, and from the capture image to the region of interest including the finger vein A preprocessor to set; A guided filter unit configured to perform guided filtering on the captured image acquired by the image acquisition unit using a preset guide image to output a guided image having an improved image quality than the captured image; And a Gabor filter unit for performing Gabor filtering on the guided image to extract the shape of the ridges from the guided image and outputting the filtered image. It includes.

The finger vein pattern extracting apparatus may further include a pre-processing unit receiving the captured image from the image obtaining unit and performing a pre-processing operation on the received capture vein image to transmit the pre-processed captured image to the guided filter unit. Characterized in that.

The pre-processing unit extracts at least one boundary line from the captured image to generate an edge image consisting of a plurality of edge points set; An image rotation unit which detects the at least one boundary line direction from the edge image and rotates the captured image in a predetermined direction according to the direction of the at least one boundary line; And a region of interest setting unit which receives the captured image from the image rotating unit, sets a specific region of the captured image as a region of interest according to a preset method, and outputs the captured image as the preprocessed captured image to the guided filter. It is characterized by.

The edge image extractor may generate the edge image by filtering the captured image using one of a top-hat filter and a Sobel filter.

The image rotating unit detects the direction of the boundary line by performing a Hough transform on the edge image, and rotates the captured image according to whether the detected boundary line direction satisfies a predetermined condition.

The image rotating unit is a mathematical equation for a set of edge points {(x ₁ , y ₁ ), (x ₂ , y ₂ ) ... (x _k , y _k )} included in the edge image.

[Revision 17.12.2012 under Rule 91]

(Where x _i and y _i are the coordinates of the edge points of the edge image b in the Cartesian coordinate system, and θ and ρ are the coordinates on the polar coordinate system for representing the edge points in the form of polar coordinates.)

2 is extracted from a set of edge points transformed into a polar curve using the Hough transform equation represented by P 2, where ρ ≥ 0 and 0 ≤ θ π. Peaks ((θ ₁ , ρ ₁ ), (θ ₂ , ρ ₂ ))

[Revision 17.12.2012 under Rule 91]

Where π represents the circumference.

If satisfies, the captured image is rotated at an angle of (θ ₁ + θ ₂ ) / 2.

The ROI setting unit sets an ROI with a predetermined size based on a point C on the captured image output from the image rotating unit, and the coordinate of the point is

(C _x , C _y ) = (width / 2, / (ρ ₁ + ρ ₂ ) / 2)

(Where ρ ₁ and ρ ₂ represent the polar coordinates at the two peaks.)

It is characterized by being calculated as.

The guided filter unit has a relation between the guide image and the guided image

[Revision 17.12.2012 under Rule 91]

(Where I is the guide image, G _u is a guided image filtered by a guided filter, ω _k is a window centered on pixel k, and a _k and b _k are linear coefficients. )

Represented by

The cost function for minimizing the difference between the input and the output of the guided filter unit

[Revision 17.12.2012 under Rule 91]

(Where p is an input image of the guided filter unit, and ε is a normalization parameter for preventing the linear coefficient a _k from increasing, and an edge preservation coefficient (edge) indicating the degree of edge preservation on the guided image (G _u ) preserving coefficient).

To determine the linear coefficients (a _k , b _k )

[Revision 17.12.2012 under Rule 91]

Where μ _k and δ _k ² are the mean and variance values of the guide image I in the window ω _k , respectively.

[Revision 17.12.2012 under Rule 91]

By substituting for the cost function,

[Revision 17.12.2012 under Rule 91]

It is characterized by calculating the.

The guided filter unit uses the guided image (G _u ) as a guide image (I), and an equation for the input image (p) of the guide filter.

[Revision 17.12.2012 under Rule 91]

Where W _ij (I, ω, ε) is the kernel weight.

The kernel weight is calculated by

[Revision 17.12.2012 under Rule 91]

[Revision 17.12.2012 under Rule 91]
(Where the cumulative sum of kernel weights is

to be.)

It is characterized by being calculated as.

The guided filter unit may perform the guided filtering using the captured image as the guide image.

[Revision 17.12.2012 under Rule 91]
The guided image filtered by the guided filter (

For

[Revision 17.12.2012 under Rule 91]

[Revision 17.12.2012 under Rule 91]
(Where * denotes a convolution operation in two dimensions, θ _k and f ₀ denote the direction of the Gabor filter and the center frequency of the Gabor filter, respectively,

Represents an even symmetric Gabor filter.)

Performs even symmetric Gabor filtering,

The even symmetric Gabor filter

Equation

[Revision 17.12.2012 under Rule 91]

[Revision 17.12.2012 under Rule 91]
(here,

Δ is a standard deviation (or scale) of an elliptic Gaussian envelope.

[Revision 17.12.2012 under Rule 91]
The Gabor filter unit sets the filtering direction (θ _k ) of the Gabor filter in eight directions,

It is characterized by that.

The finger vein pattern extracting apparatus may include: a quality estimating unit configured to receive the filtering image from the Gabor filter unit and to determine whether the quality of the vein pattern included in the filtering image is greater than or equal to a predetermined reference quality; And a feature extractor configured to receive the filtered image including the vein pattern having the reference quality or higher and extract a feature of the vein pattern included in the filtered image.

The feature extractor divides the filtered image into a plurality of small blocks, and extracts an average absolute deviation of each of the divided plurality of small blocks F _{mn as} a feature of the captured image corresponding to the filtered image. It is characterized by.

The feature extracting unit is the average absolute deviation ADD (δ ^k _mn ) for each of the plurality of small blocks F _mn .

[Revision 17.12.2012 under Rule 91]

[Revision 17.12.2012 under Rule 91]
(Where N is the number of small block (F _mn ) pixels, and μ ^k _mn is

Is the average of the size.)

It is characterized by the calculation.

Finger vein pattern extraction method according to an embodiment of the present invention for achieving the above another object is a finger vein pattern extraction apparatus including an image acquisition unit, a guided filter unit and a Gabor filter unit for obtaining a captured image using infrared light A method for extracting finger vein pattern, the method comprising: obtaining, by the image obtaining unit, the captured image including a vein pattern of a user; Outputting a guided image having improved image quality than the captured image by performing the guided filtering on the captured image by the guided filter unit; And outputting the filtered image by performing Gabor filtering on the guided image to extract the shape of the ridges from the guided image by the Gabor filter unit. It includes.

Accordingly, the method and apparatus for extracting finger vein pattern using the guided Gabor filter of the present invention protects the vein pattern by using the guided filter, and is affected by background effects such as blur, lighting, ambient conditions and changes in blood flow. By reducing the quality of the vein image and using the Gabor filter, the ridgeline of the vein pattern can be easily extracted to make clear the distinction between the vein pattern and the background. In addition, the fast operation speed enables not only to extract vein patterns in real time, but also to perform accurate image matching during user authentication.

1 illustrates a finger vein pattern extraction method using a guided Gabor filter according to an embodiment of the present invention.

2 illustrates a step of obtaining a finger vein image of FIG. 1.

3 illustrates an embodiment of the finger vein image acquisition step of FIG. 2.

4 illustrates another embodiment of the finger vein image acquisition step of FIG. 2.

5 shows an example of image quality improvement performance of the guided filter according to the guide image and the window size and the edge retention coefficient.

6 shows the finger vein pattern filtered in eight Gabor filter directions.

7 shows the performance of a Gabor filter in accordance with the present invention.

8 shows the finger vein pattern extraction performance according to the finger vein pattern extraction method of the present invention.

9 shows the matching performance of the bonded Gabor filter according to the present invention.

10 shows a finger vein pattern extraction apparatus according to an embodiment of the present invention.

11 shows an example of a user authentication system according to the present invention.

In order to fully understand the present invention, the operational advantages of the present invention, and the objects achieved by the practice of the present invention, reference should be made to the accompanying drawings which illustrate preferred embodiments of the present invention and the contents described in the accompanying drawings.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. As those skilled in the art would realize, the described embodiments may be modified in various different ways, all without departing from the spirit or scope of the present invention. In addition, in order to clearly describe the present invention, parts irrelevant to the description are omitted, and the same reference numerals in the drawings indicate the same members.

Throughout the specification, when a part is said to "include" a certain component, it means that it may further include other components, without excluding the other components unless otherwise stated. In addition, the terms “... unit”, “... unit”, “module”, “block”, etc. described in the specification mean a unit that processes at least one function or operation, which means hardware, software, or hardware. And software.

Referring to Figure 1, the finger vein pattern extraction method using a guided Gabor filter according to the present invention first obtains the finger vein image (S100). Here, the finger vein image may be obtained by using an image capture device or an image scan device that captures a finger vein image by scanning a user's finger. Can also be obtained. If necessary, a preprocessing operation may be performed on the acquired finger vein image. Detailed description of the pretreatment will be described later.

When the finger vein image is obtained, filtering is performed using the guided filter on the acquired finger vein image (S200). The guided filter improves the image quality of the finger vein image according to a preset guidance image. The guide image serves as a director to protect vein patterns and to reduce the effects of background effects such as haze, light variations, physical ambient conditions and changes in blood flow.

Thereafter, filtering is performed using the Gabor filter on the finger vein image whose image quality is improved by the guided filter (S300). The Gabor filter allows the vein pattern to be easily extracted from the vein image by the guided filter so that the vein and the background can be easily distinguished.

As described above, the finger vein pattern extraction method according to the present invention applies a Gabor filter together with a guided filter to the acquired finger vein image so that the vein pattern in the finger vein image is clearly distinguished from the background. To facilitate the recognition of finger vein patterns.

FIG. 2 shows a step of acquiring the finger vein image of FIG. 1, and FIG. 3 shows one embodiment of the step of acquiring the finger vein image of FIG. 2.

Referring to FIGS. 2 and 3, the obtaining of the finger vein image is performed. First, the step of obtaining the finger vein image acquires an input image as shown in FIG. 3A (S110). The input image a is an image including a finger vein pattern for at least one finger. The input image a may be obtained using an image capture device such as an infrared camera so that the finger vein pattern may be displayed.

When the input image (a) is obtained, it is determined whether to perform a preprocessing operation on the obtained input image (S120). If the acquired input image is an image that has been preprocessed and stored in advance for identification of the finger vein pattern, it is not necessary to perform the preprocessing. However, when the input image is an image in which the user's finger is scanned by the image scanning device for user authentication, the preprocessing is preferably performed to facilitate the extraction of the finger vein pattern. In the finger vein image extraction method using the guided Gabor filter according to the present invention, although the pretreatment is not performed, the finger vein pattern can be extracted more clearly than the conventional technique, but in order to extract the finger vein pattern more clearly, the pretreatment should be performed. . Whether or not to perform the preprocessing may be set in advance in the finger vein pattern extraction apparatus.

If it is determined that the preprocessing operation is set as a result of the determination, as a preprocessing operation, first, a finger profile is extracted from the input image a (S130). Finger contour extraction may be performed by applying a top-hat filter or a Sobel filter to the input image. The top hat filter is a filter that converts each point of an image to a binary value of 0 or 1 based on a predetermined boundary value. The Sobel filter is a nonlinear filter that finds the difference between sums of pixels at both ends in an image and then averages the horizontal and vertical directions to emphasize the boundary. Both the top-hat filter and the Sobel filter are filters commonly used for image processing to extract a boundary line from an image, and thus a detailed description thereof will be omitted. The finger outline is extracted in the form of an edge image as shown in (b) by the boundary line extracted from the input image (a). In addition, the edge image b may be regarded as being composed of a set of a plurality of edge points.

When the edge image b is extracted from the input image a, the finger outline is localized (S140). Localization of the finger outline is performed using a Hough transform. The finger contour can be estimated as a line of small curvature, and the Hough transform is used to detect the position and angle of the finger contour. Hough transform is generally a function used to extract a straight line from an image, but a function that can also extract a circle or a curve is mainly used in image processing, and thus a detailed description thereof will be omitted.

The set of edge points {(x ₁ , y ₁ ), (x ₂ , y ₂ ) ... (x _k , y _k )} extracted from the edge image (b) is equation 1, which is an equation of the Hough transform. It can be converted into a sine curve on the polar plane (θ, ρ) (where ρ ≥ 0, 0 ≤ θ ≤ π).

Equation 1

The set of edge points {(x ₁ , y ₁ ), (x ₂ , y ₂ ) ... (x _k , y _k )} is transformed into a polar coordinate system by Hough transform, accumulates, and If there is at least one peak in the set of accumulated edge points, this is evidence that there is at least one straight line corresponding to the edge image, depending on the nature of the Hough transform.

(c) represents two peaks ((θ ₁ , ρ ₁ ), (θ ₂ , ρ ₂ ) extracted from a set of cumulative edge points from which the edge image (b) is Hough transformed, and the two peaks ((θ) ₁ , ρ ₁ ), (θ ₂ , ρ ₂ )) respectively correspond to the finger contour line. (d) indicated two straight lines corresponding to two peaks ((θ ₁ , ρ ₁ ), (θ ₂ , ρ ₂ )) on the input image a in blue.

Thereafter, in consideration of the curvature of the finger, it is determined whether to rotate the input image (a) (S150). It is not necessary to rotate the input image (a), but to facilitate extraction of the finger vein pattern, and to facilitate comparison with the previously saved finger vein pattern at the time of user authentication, It is preferable to arrange so that a direction may be constant. And whether to rotate the input image (a) is determined according to whether the extracted two peaks (θ ₁ ρ ₁ ), (θ ₂ , ρ ₂ ) satisfy the equation (2).

Equation 2

Where π represents the circumference.

If the extracted two peaks ((θ ₁ , ρ ₁ ), (θ ₂ , ρ ₂ ) do not satisfy Equation 2, the input image is not rotated, but the extracted two peaks ((θ ₁ , When ρ ₁ ), θ ₂ , ρ ₂ ) satisfy Equation 2, the input image a is rotated at an angle of (θ ₁ + θ ₂ ) / 2 (S160).

4 illustrates another embodiment of the finger vein image acquisition step of FIG. 2. In FIG. 4, unlike the input image a of FIG. 3, the input image a does not have a horizontal direction. Since two peaks ((θ ₁ , ρ ₁ ), (θ ₂ , ρ ₂ ) extracted by the Hough transform shown in (b) satisfy the condition of Equation 2, as shown in (c) Likewise, it is rotated at an angle of (θ ₁ + θ ₂ ) / 2 and converted into a rotationally corrected image.

Thereafter, a region of interest (ROI) is set in the rotation correction image c (S170). The region of interest is an area set for extracting the finger vein pattern in the finger outline, as shown in (d). In the present invention, the region of interest is set to a predetermined size around the point C = (C _x , C _y ). The coordinate of the point C is set as (C _x , C _y ) = (width / 2, / (ρ ₁ + ρ ₂ ) / 2). In the present invention, the size of the ROI is set to, for example, [256, 96].

In FIG. 4, (e) illustrates a region of interest obtained by separating the set region of interest from the rotational correction image. When the region of interest e is set, as illustrated in FIG. 1, filtering using the guided filter is performed (S200).

The key assumption of the guided filter is that the guide image (I) and the filter output (Gu) consist of a local linear model relationship. Therefore, in the present invention, it is assumed that the guided image G _u filtered by the guided filter is a linear transformation of the guide image I in the window w _k centered on the pixel k, as shown in Equation 3 below. I can express it.

[Revision 17.12.2012 under Rule 91]
Equation 3

Here, I is a guide image, G _u is a guided image filtered by the guided filter, and w _k is a window centered on the pixel k. And a _k and b _k are linear coefficients.

Since the local linear model according to the above assumption satisfies ∇G _u = a∇I, it is ensured that the guided image G _u has an edge if the guide image I has an edge. It is well known that such linear models are useful for image matting, image resolution enhancement and blur removal.

In order to determine the linear coefficients a _k and b _k , a cost function that minimizes the difference between the input and the output is calculated in Equation 4 in each window.

[Revision 17.12.2012 under Rule 91]
Equation 4

Where p is the input image of the guided filter and ε is a normalization parameter to prevent the linear coefficient a _k from growing.

Equation 4 is given according to linear regression according to "Draper, N., Smith, H, Applied Regression Analysis, 2 edn., John Wiley (1981)".

[Revision 17.12.2012 under Rule 91]
Equation 5

[Revision 17.12.2012 under Rule 91]
Equation 6

Equations 5 and 6 are equations for calculating linear coefficients a _k and b _k according to Equation 4.

Substituting Equations 5 and 6 into Equation 4, the guided image G _u is calculated as in Equation 7.

[Revision 17.12.2012 under Rule 91]
Equation 7

Of the linear coefficients a _k and b _k , the linear coefficient a _k is a linear edge retention coefficient and the linear edge retention coefficient a _k decreases as the normalization parameter ε increases. Thus, the normalization parameter ε can be seen as an edge preserving coefficient that indicates the degree of edge retention on the guided image G _u .

In FIG. 5, columns 1 of (a) and (b) represent input images, and input images of (a) and (b) are identical images. Column 2 is a guide image and (a) and (b) use different guide images. Columns 3 to 6 represent guided images whose image quality is improved by the guided filter when the window size ω and the edge retention coefficient ε are changed, respectively. As shown in FIG. 5, it can be seen that the guided filter is effective in removing blur and improving image quality of the vein image by selecting an appropriate guide image.

The relation between the guide image I, the input image p of the guide filter, and the guided image G _u may be expressed as shown in Equation (8).

[Revision 17.12.2012 under Rule 91]
Equation 8

Where W _ij (I, ω, ε) represents the kernel weight of the guided filter. Therefore, the kernel weight of Equation 8 based on Equations 3 to 7 may be expressed as Equation 9.

[Revision 17.12.2012 under Rule 91]
Equation 9

[Revision 17.12.2012 under Rule 91]
Where cumulative sum of kernel weights

Can be proved without any separate normalization process.

When the filtering operation by the guided filter is performed, filtering by the Gabor filter is performed on the guided image.

The Gabor filter makes it possible to extract the energy of the local frequency band according to a specific scale (δ) and orientation (θ _k ) from an object extracted by a useful spectral decomposition method. Therefore, it is widely used to analyze texture information by expressing features according to scale and direction of objects as vectors. Gabor filters are divided into even symmetric Gabor filters used for ridge extraction and odd symmetric Gabor filters used for edge extraction. In the present invention, the Gabor filter is applied to extract ridges on the guided image filtered by the guided filter. Therefore, a random symmetric Gabor filter is applied to the guarded image.

Therefore, the even symmetric guided gabor filter of the present invention may be represented as a combined form of an even symmetric Gabor filter and a guided filter, as shown in Equation (10).

[Revision 17.12.2012 under Rule 91]
Equation 10

[Revision 17.12.2012 under Rule 91]
Where * represents a convolution operation in two dimensions, θ _k and f ₀ represent the direction of the Gabor filter and the center frequency of the Gabor filter, respectively,

Denotes an even symmetric Gabor filter.

[Revision 17.12.2012 under Rule 91]
In the guided Gabor filter according to the present invention, it is assumed that the direction (θ _k ) of the Gabor filter is set in eight directions.

Set to. Since the direction of the Gabor filter (θ _k ) is set in eight directions, the direction of the Gabor filter (θ _k ) is 0 °, 22,5 °, 45 °, 67.5 °, 90 °, 112.5 °, 135 °, 157.5 ° Is set to. However, the number of Gabor filter directions (θ _k ) can be adjusted.

[Revision 17.12.2012 under Rule 91]
Since the generalized form of the even symmetric Gabor filter is well known,

Is expressed as in Equation (11).

[Revision 17.12.2012 under Rule 91]
Equation 11

[Revision 17.12.2012 under Rule 91]
here,

Is the standard deviation (or scale) of the elliptic Gaussian envelope.

The filtered images in eight Gabor filter directions θ _k are stored in the guided Gabor filter storage.

6 shows the finger vein pattern filtered in eight Gabor filter directions.

Direction (θ _k) direction (θ _k) of the Gabor represents a specified filtered vein image, each filter in accordance with the Gabor filter top of the image is set to eight, respectively in Fig. 6 (a) to (h) is 0 °, 22 The case of 5 degrees, 45 degrees, 67.5 degrees, 90 degrees, 112.5 degrees, 135 degrees, and 157.5 degrees is shown.

The lower image represents average absolute deviations (AADs) corresponding to the upper image.

In FIG. 7, an image disposed at the top is an original image and an image converted by a conventional image conversion technique, (a) an original image, and (b) an image converted by a global histogram. c) represents an image converted by a local histogram blocked with a size of [32, 16]. And (d) represents an image transformed by the wavelet normalization technique, (e) represents an image filtered by the Gabor filter when the edge preservation coefficient ε = 1 ² and the window size ω = 5, and (f) represents the edge preservation Coefficient ε = 1 ² , the image filtered by the guided filter when window size ω = 15, (g) is the edge retention coefficient ε = 1 ² , window size according to the enhanced guidance image When ω = 15, it represents the image filtered by the guided filter.

The bottom image represents an image to which a Gabor filter is applied to the corresponding top image, respectively.

As shown in (e), (f), and (g) in FIG. 7, the performance of the image filtered by the guided filter is preserved when the Gabor filter is applied to the image filtered by the guided filter. Although there are variations depending on the coefficient ε and the window size ω, most of the ridges can be clearly extracted. However, the image g filtered by the guide filter using the augmented GUID image includes noise that is not suitable for performing matching of the acquired finger vein pattern image. In the present invention, a test result for a large image is used as a guide image (I), and when the edge retention coefficient ε = 1 ² and the window size ω = 15, the filtered image by the guided filter is performed by the Gabor filter. In case of filtering, it was confirmed that the optimal vein pattern could be extracted. However, the parameters of the guided filter (edge retention coefficient ε and window size ω) may be adjusted according to the acquired state of the image.

The present invention is directed to the extraction of finger vein pattern using a guided Gabor filter, but the substantially extracted finger vein pattern is generally used for user authentication, and user authentication requires image matching. For image matching, the feature of the finger vein pattern is extracted and the image matching operation is performed on the feature of the extracted finger vein pattern.

In order to facilitate feature extraction of the finger vein pattern, the filtered image may be divided into a plurality of small blocks. In the present invention, it is assumed that the filtered image is divided into 16 × 16 small blocks. In the filtered image divided into small blocks, an 8-dimensional vector based on statistical information may be constructed instead of the pixel-based vector. Thus, for a normalized finger vein image, 96 ([16 ㅧ 6]) vectors can be extracted from the blocks.

[Revision 17.12.2012 under Rule 91]
Statistics based on the block (elements of F in the mth column, nth row, where m = 1, 2… 16, n = 1, 2… 6) to estimate F _mn , which represents the block matrix of the filtered image Can be calculated. A guided Gabor filtered image corresponding to the block matrix F _mn

Average Absolute Deviation (ADD) (δ ^k _mn ) of the magnitude of) is calculated as shown in Equation 12.

[Revision 17.12.2012 under Rule 91]
Equation 12

[Revision 17.12.2012 under Rule 91]
Where N is the number of block matrix (F _mn ) pixels at, and μ ^k _mn is

It is shown in Figure 6 as the average of the size of.

[Revision 17.12.2012 under Rule 91]
Accordingly, image matching may be performed using a support vector machine (SVM) classifier, which is a kind of supervised learning that can be applied to classification and regression as one of classification algorithms. SVM classifiers according to the present invention are each for image matching

It is possible to obtain 768 [96 벡터 8] feature vectors represented by. Each feature vector V may be calculated as shown in Equation 13.

[Revision 17.12.2012 under Rule 91]
Equation 13

The calculated feature vector V is used in image matching for user authentication.

In order to verify the performance of the finger vein pattern extraction method by the guided Gabor filter according to the present invention, a test was performed using a database in which 106 open finger vein data were stored. In the database, images of the two-handed index finger, middle finger, and ring finger of each 106 people were acquired six times, and a total of 3,816 finger vein images were stored. The pixel size of each image was stored as 320 x 240.

8, three kinds of low quality images were tested by four finger vein extraction methods. In FIG. 8, (a) shows a maximum curvature method, (b) shows a wide line tracking method, and (c) shows a finger vein pattern extracted by a Gabor filter. And (d) shows the finger vein pattern extracted by the guided Gabor filter according to the present invention.

Images

1 and 2 are low contrast images at light and dark intensity, respectively, and image 3 is an image affected by illumination. As shown in FIG. 8, the bonded Gabor filter effectively extracts clear vein patterns without being affected by the thickness and brightness of the vein image. In addition, the finger vein pattern extraction method according to the other guided Gabor filter according to the present invention does not require classification or classification of images, and the pretreatment time for one image takes 0.122 seconds, which is much higher than that of other finger vein extraction methods. You can see it works quickly. That is, user authentication can be performed in real time.

In FIG. 9, the 768 feature vectors extracted above were evaluated for matching performance. To evaluate the matching performance, the finger vein data was divided into six groups for training and testing, and applied in the form of (the number of training groups, the number of test groups) to train the matching system. Cosine-like measurements and SVM (Support Vector Machines) were used as classifiers, and the performance between the Gabor and Guarded Gabor filters was evaluated. As shown in FIG. 9, it can be seen that the SVM classifier based on the guided Gabor filter according to the present invention is performed with higher accuracy than the cosine-like measurement and the SVM classifier based on the conventional Giber Peter.

That is, the guided gabor filter according to the present invention makes the guided filter clear the edge contour of the vein by using a guide image. The Gabor filter clarifies the ridges in the guided image filtered by the guided filter, making it easier to distinguish between the background and the finger vein pattern. Finally, the vein profile is extracted from eight directions.

Advantages of the guided Gabor filter according to the present invention do not require image segmentation, easy haze removal of the image, easy to protect the edge and slope of the finger vein. It also enables high speed operation and obtains clear finger vein contours with low noise. In addition, the image quality deterioration is low even for low quality images.

The finger vein pattern extracting apparatus 10 according to the present invention extracts the finger vein pattern using the guided Gabor filter. Referring to FIG. 10, the finger vein pattern extraction apparatus 10 includes an image acquisition unit 110, a preprocessor 120, a guided filter unit 130, a Gabor filter unit 140, a quality estimation unit 150, and features. An extraction unit 160 is provided.

The image acquisition unit 110 may capture a vein image of the user, and may be implemented as an image capturing means such as a camera including an image sensor (or an infrared sensor). In particular, in the present invention, the vein image should be obtained, and since the vein is darkened by absorbing infrared light as described above, it is preferable that the vein is implemented by an infrared lamp, an infrared camera, or an infrared sensor. In addition, in some cases, when the finger vein pattern extraction apparatus 10 receives an image captured from the outside, the image acquisition unit 110 may be omitted.

The preprocessor 120 performs a preprocessing operation on the vein image acquired by the image acquirer 110. As described above, the preprocessor 120 may perform operations such as image rotation and ROI setting as preprocessing operations to facilitate image matching. The preprocessing unit 120 is a configuration necessary for ease of image matching, and may be omitted in some cases. In addition, in the above, the image acquisition unit 110 and the preprocessor 120 are divided and illustrated for convenience of description, but the preprocessor 120 may be implemented by being included in the image acquisition unit 110.

The guided filter unit 130 receives the vein image from the preprocessor if the preprocessor 120 is present, and receives the vein image from the image acquisition unit 110 if the preprocessor 120 does not exist. Perform a filtering operation. Guided filtering uses a guide image to protect vein patterns in the vein image and improves the quality of the vein image by reducing the effects of background effects such as blur, lighting, ambient conditions, and changes in blood flow.

The Gabor filter 140 receives the guided filtered image from the guided filter unit 130, and easily extracts the ridge line of the vein pattern from the guided image to clarify the distinction between the vein pattern and the background.

The quality estimator 150 receives the filtered vein image from the Gabor filter 140 and determines whether the quality of the received vein image is greater than or equal to a predetermined reference quality, and transmits the vein image to the feature extractor 160 if the quality is higher than the reference quality. do. Since the quality estimator 150 estimates the quality of an image, there are various known methods, and thus, a detailed description thereof will be omitted.

The feature extractor 160 receives the vein image from the quality estimator 150 and extracts a feature of the received vein image.

Referring to FIG. 11, the user authentication system according to the present invention includes a vein pattern extraction device 10, an interface device 20, and a database 30. Since the vein pattern extraction apparatus 10 is shown in FIG. 10, a detailed description thereof will be omitted.

The interface device 20 may be implemented as at least one of a display device and an acoustic device. If the vein image received by the quality estimator 150 of the vein pattern extraction apparatus 10 is determined to be an image having a lower quality than the reference quality, The interface device 20 requests the user to rescan the vein image through the image acquisition unit 110.

The database 30 stores the vein image obtained by the vein pattern extraction apparatus 10 when the user is registered and filtered to extract features. Subsequently, upon requesting a user's seat, the vein pattern extracting apparatus 10 searches for the stored vein image matching the characteristic of the obtained vein image and notifies the interface apparatus 10 of the matching search result.

The interface device 10 displays the matching search result to the user.

In addition, since the user authentication system is generally used for security, an additional locking device may be provided. The locking device may be configured to be unlocked if a matching search result from the database 30 determines that a matching vein pattern exists.

Although it has been described as extracting the finger vein pattern, this is because a finger is used as a representative example of the user identification method. However, the present invention can also be used for images other than fingers. That is, in the present invention, the input image may be not only a finger image but also an image of another body part such as a back of a hand or a wrist, not a hand or a wrist. That is, the present invention is not limited to the finger vein pattern but may be applied to an image including all kinds of vein patterns that can be identified by the user.

The method according to the invention can be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of the recording medium include a ROM, a RAM, a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, and the like, and also include a carrier wave (for example, transmission through the Internet). The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

Although the present invention has been described with reference to the embodiments shown in the drawings, this is merely exemplary, and it will be understood by those skilled in the art that various modifications and equivalent other embodiments are possible.

Therefore, the true technical protection scope of the present invention will be defined by the technical spirit of the appended claims.

Claims

A pre-processing unit which acquires a captured image including a vein pattern of a user using infrared light and sets a region of interest including a finger vein from the captured image;

A guided filter unit configured to perform guided filtering on the captured image acquired by the image acquisition unit using a preset guide image to output a guided image having an improved image quality than the captured image; And

A Gabor filter unit for performing Gabor filtering on the guided image to extract the shape of the ridge line from the guided image and outputting a filtered image; Finger vein pattern extraction apparatus comprising a.
[Revision 17.12.2012 under Rule 91]

The method of claim 1, wherein the guided filter unit

The relationship between the guide image and the guided image

Equation

(Where I is the guide image, G u is a guided image filtered by a guided filter, ω k is a window centered on pixel k, and a k and b k are linear coefficients. )

Represented by

Cost function for minimizing the difference between the input and the output of the guided filter unit

Equation

(Where p is an input image of the guided filter unit, and ε is a normalization parameter for preventing the linear coefficient a k from increasing, and an edge preservation coefficient (edge) indicating the degree of edge preservation on the guided image (G u ) preserving coefficient).

And calculate

To determine the linear coefficients a k , b k

Equation

(Where μ k and δ k 2 are the mean and variance values of the guide image I in the window w k , respectively)

Equation

Is substituted into the cost function,

Equation

Finger vein pattern extraction apparatus, characterized in that for calculating.
[Revision 17.12.2012 under Rule 91]
The method of claim 2, wherein the guided filter unit
Equation for the guided image (G u ) to the guide image (I), the input image of the guide filter (p)

Where W ij (I, ω, ε) is the kernel weight.
And calculate
The kernel weight is
Equation

(Where the cumulative sum of kernel weights is
to be.)

Finger vein pattern extraction device, characterized in that calculated as.
The method of claim 3, wherein the guided filter unit

The finger vein pattern extraction apparatus, wherein the guided filtering is performed using the captured image as the guide image.
[Revision 17.12.2012 under Rule 91]
The method of claim 3, wherein the Gabor filter unit
The guided image filtered by the guided filter (
For

(Where * denotes a convolution operation in two dimensions, θ k and f 0 denote the direction of the Gabor filter and the center frequency of the Gabor filter, respectively,
Represents an even symmetric Gabor filter.)

Performs even symmetric Gabor filtering,
The even symmetric Gabor filter
Equation

(here,
And δ is calculated as a standard deviation (or scale) of an elliptic Gaussian envelope.
[Revision 17.12.2012 under Rule 91]
The method of claim 5, wherein the Gabor filter unit
Set the Gabor filter's filtering direction (θ k ) in eight directions,
Finger vein pattern extraction device, characterized in that.
The method of claim 1, wherein the pretreatment unit

An edge image extracting unit extracting at least one boundary line from the captured image to generate an edge image composed of a plurality of edge points;

An image rotation unit which detects the at least one boundary line direction from the edge image and rotates the captured image in a predetermined direction according to the direction of the at least one boundary line; And

A region of interest setter configured to receive the captured image from the image rotating unit, set a specific region of the captured image as a region of interest according to a preset method, and output the preprocessed captured image to the guided filter as a preprocessed capture image; Finger vein pattern extraction device characterized in.
The method of claim 7, wherein the edge image extraction unit

The finger vein pattern extraction apparatus, characterized in that for generating the edge image by filtering the captured image using one of a top-hat filter or a Sobel filter.
The method of claim 7, wherein the image rotating unit

The finger vein pattern extraction apparatus of claim 1, wherein the edge image is detected by performing a Hough transform on the edge image, and the captured image is rotated according to whether the detected boundary line direction satisfies a predetermined condition.
[Revision 17.12.2012 under Rule 91]

The method of claim 9, wherein the image rotating unit

Equation for the set of edge points {(x 1 , y 1 ), (x 2 , y 2 ) ... (x k , y k )} included in the edge image

(Where x i and y i are the coordinates of the edge points of the edge image b in the Cartesian coordinate system, and θ and ρ are the coordinates on the polar coordinate system for representing the edge points in the form of polar coordinates.)

Converting into a sinusoidal curve on the polar plane (θ, ρ) (where ρ ≥ 0, 0 ≤ θ ≤ π) using the Hough transformation equation

Two peaks ((θ 1 , ρ 1 ), (θ 2 , ρ 2 ) extracted from a set of edge points transformed into polar coordinates are

Where π represents the circumference.

If satisfied, the finger vein pattern extraction apparatus, characterized in that for rotating the captured image at an angle of (θ 1 + θ 2 ) / 2.
The apparatus of claim 10, wherein the ROI setting unit

The ROI is set to a predetermined size based on the point C on the captured image output from the image rotating unit, and the coordinate of the point is

(C x , C y ) = (width / 2, / (ρ 1 + ρ 2 ) / 2)

(Where ρ 1 and ρ 2 represent the polar coordinates at the two peaks.)

Finger vein pattern extraction device, characterized in that calculated as.
According to claim 1, wherein the finger vein pattern extraction device

A quality estimating unit configured to receive the filtered image from the Gabor filter unit and determine whether the quality of the vein pattern included in the filtered image is equal to or greater than a predetermined reference quality; And

And a feature extractor configured to receive the filtered image including the vein pattern having the reference quality or higher and extract a feature of the vein pattern included in the filtered image.
The method of claim 12, wherein the feature extraction unit

The filtering image is divided into a plurality of small blocks, and an average absolute deviation of each of the divided plurality of small blocks F mn is extracted as a feature of the captured image corresponding to the filtering image. Finger vein pattern extraction device.
[Revision under Rule 91 01.02.2013]
The mean absolute deviation ADD (δ k mn ) for each of the plurality of small blocks F mn is

(Where N is the number of small block (F mn ) pixels at and μ k mn is
Is the average of the size.)
Finger vein pattern extraction apparatus, characterized in that the calculation.
The finger vein pattern extraction apparatus according to claim 13;

A user interface implemented as at least one of a display device and a sound device for requesting an image capture operation from the user to obtain the captured image including the vein pattern; And

Storing a feature of the captured image acquired by the feature extracting unit, and performing image matching by comparing a pre-stored feature of the captured image with a feature of the captured image received by the finger vein pattern extraction apparatus; User authentication system comprising a; database.
In the finger vein pattern extraction method of the finger vein pattern extraction chapter comprising an image acquisition unit, a guided filter unit and a Gabor filter unit for obtaining a captured image using infrared light,

Acquiring, by the image acquisition unit, the captured image including the vein pattern of the user;

Outputting a guided image having improved image quality than the captured image by performing the guided filtering on the captured image by the guided filter unit; And

Outputting a filtered image by performing Gabor filtering on the guided image to extract the shape of the ridges from the guided image by the Gabor filter unit; Finger vein pattern extraction method comprising a.
[Revision 17.12.2012 under Rule 91]

The method of claim 16, wherein the outputting the guided image

The relationship between the guide image and the guided image

Equation

(Where I is the guide image, G u is a guided image filtered by a guided filter, ω k is a window centered on pixel k, and a k and b k are linear coefficients. )

Calculated as;

Cost function for minimizing the difference between the input and the output of the guided filter unit

Equation

(Where p is an input image of the guided filter unit, and ε is a normalization parameter for preventing the linear coefficient a k from increasing, and an edge preservation coefficient (edge) indicating the degree of edge preservation on the guided image (G u ) preserving coefficient).

Calculated as;

The linear coefficients a k and b k are each

(Where μ k and δ k 2 are the mean and variance values of the guide image I in the window ω k , respectively)

Equation

Calculated as; And

Substituting the linear coefficient (a k , b k ) into the cost function, the guided image is

Steps calculated as; finger vein pattern extraction method comprising the.
[Revision 17.12.2012 under Rule 91]
18. The method of claim 17, wherein outputting the guided image
The guided image is a guide image (I), the equation for the input image (p) of the guide filter

Where W ij (I, ω, ε) is the kernel weight.
Calculated as; And
The kernel weight is
Equation

(Where the cumulative sum of kernel weights is
to be.)

Steps calculated as; finger vein pattern extraction method characterized in that it further comprises.
19. The method of claim 18, wherein outputting the guided image

The finger vein pattern extraction method, characterized in that for performing the guided filtering using the captured image as the guide image.
[Revision 17.12.2012 under Rule 91]
19. The method of claim 18, wherein outputting the filtered image
The guided image filtered by the guided filter (
For

(Where * denotes a convolution operation in two dimensions, θ k and f 0 denote the direction of the Gabor filter and the center frequency of the Gabor filter, respectively,
Represents an even symmetric Gabor filter.)

Performing even symmetric Gabor filtering according to
The even symmetric Gabor filter
Equation

(here,
And δ is calculated as a standard deviation (or scale) of an elliptic Gaussian envelope.
[Revision 17.12.2012 under Rule 91]
The method of claim 20, wherein outputting the filtered image
Set the Gabor filter's filtering direction (θ k ) in eight directions,
Finger vein pattern extraction method characterized in that.
The method of claim 16, wherein the finger vein pattern extraction method

And a preprocessing step of the image obtaining unit receiving the captured image from the image obtaining unit and performing a preprocessing operation on the received capture vein image to transmit the preprocessed captured image to the guided filter unit. Finger vein pattern extraction method.
The method of claim 22 wherein the pretreatment step

Extracting at least one boundary line from the captured image to generate an edge image composed of a plurality of edge points;

Detecting the at least one boundary line direction, and rotating and outputting the captured image in a predetermined direction according to the direction of the at least one boundary line; And

Receiving the captured image from the image rotating unit, setting a specific region of the captured image as a region of interest according to a preset method, and outputting the region of interest as the preprocessed captured image to the guided filter; Finger vein pattern extraction method characterized in that.
The method of claim 23, wherein generating the edge image

The finger vein pattern extraction method of claim 1, wherein the captured image is filtered using one of a top-hat filter and a sobel filter.
The method of claim 23, wherein the rotating and outputting

And performing a Hough transform on the edge image to detect the direction of the boundary line and rotate the captured image according to whether the detected boundary line direction satisfies a predetermined condition.
[Revision 17.12.2012 under Rule 91]

The method of claim 25, wherein the rotating and outputting

Equation for the set of edge points {(x 1 , y 1 ), (x 2 , y 2 ) ... (x k , y k )} included in the edge image

(Where x i and y i are the coordinates of the edge points of the edge image b in the Cartesian coordinate system, and θ and ρ are the coordinates on the polar coordinate system for representing the edge points in the form of polar coordinates.)

Converting to a sinusoidal curve on the polar plane (θ, ρ), where ρ ≧ 0, 0 ≤ θ ≤ π, using the Hough transformation equation represented by; And

Two peaks ((θ 1, ρ 1 ), (θ 2 , ρ 2 ) extracted from a set of edge points transformed into polar coordinates are

Where π represents the circumference.

If satisfied, rotating the captured image at an angle of (θ 1 + θ 2 ) / 2; finger vein pattern extraction method comprising a.
27. The method of claim 26, wherein outputting the region of interest as the preprocessed captured image to the guided filter

Setting a region of interest to a predetermined size based on a point C on the captured image; Including,

The coordinate of the point is

(C x , C y ) = (width / 2, / (ρ 1 + ρ 2 ) / 2)

(Where ρ 1 and ρ 2 represent the polar coordinates at the two peaks.)

Finger vein pattern extraction method characterized in that it is calculated as.
The method of claim 16, wherein the finger vein pattern extraction method

The finger vein pattern extraction apparatus further includes a quality estimation unit and a feature extraction unit,

Receiving, by the quality estimating unit, the filtered image and determining whether the quality of the vein pattern included in the filtered image is equal to or greater than a predetermined reference quality; And

And receiving the filtered image including the vein pattern equal to or greater than the reference quality determined in the determining whether the reference quality is equal to or greater than the reference quality, and extracting a feature of the vein pattern included in the filtered image. Finger vein pattern extraction method.
29. The method of claim 28, wherein extracting the feature

Dividing the filtered image into a plurality of small blocks;

Calculating an average absolute deviation for each of the separated plurality of small blocks (F mn ); And

And extracting the average absolute deviation as a feature of the captured image corresponding to the filtered image.
[Revision 17.12.2012 under Rule 91]
30. The method of claim 29, wherein calculating the mean absolute deviation is
The average absolute deviation ADD (δ k mn ) for each of the plurality of small blocks F mn is

(Where N is the number of small block (F mn ) pixels at and μ k mn is
Is the average of the size.)

Finger vein pattern extraction method, characterized in that the calculation.
31. A recording medium having recorded thereon program instructions for implementing the finger vein pattern extraction method according to any one of claims 16 to 30.