KR101025666B1 - Method and apparatus of extracting finger vein feature point - Google Patents

Abstract

A finger vein feature point extraction method and apparatus are provided. Acquire a finger image from a finger passed through infrared rays, obtain a finger vein image by removing a background from the obtained finger image, calculate a radiance at each pixel of the finger vein image, Pixels whose computed radioactivity is a local maxima are extracted as bifurcation of the finger vein. By calculating the radiance and extracting the bifuracation of the finger vein as a feature point, a feature point-based finger vein recognition technique with high identification accuracy with little risk of theft and forgery can be achieved.

Finger vein, recognition, bifurcation, radiation

Description

Method and apparatus for extracting finger vein feature points {METHOD AND APPARATUS OF EXTRACTING FINGER VEIN FEATURE POINT}

The present invention relates to a method and apparatus for extracting finger vein feature points, and more particularly, to a method and apparatus for extracting finger vein feature points using a bifurcation as a feature point of the finger vein by calculating radialness. will be.

Biometric technology refers to a technology for identifying or identifying an individual using biometric information. To date, biometric technology has identified individuals using fingerprints, faces, irises or voices.

Fingerprint recognition technology is small in size and can be mass produced at low cost. In addition, it shows relatively good performance in terms of accuracy and speed of recognition. However, fingerprints are exposed outside of the body and are easy to be stolen and may be deformed by sweat, which can affect recognition performance. There are also hygiene problems.

Iris recognition technology is the most reliable biometric technology in that there is little possibility of recognition error and forgery is difficult. However, most people are reluctant to shine light directly on their eyes for iris recognition, and iris recognition devices are large and expensive.

Facial or voice recognition technology has the advantage of being the most natural form of biometric information for identity recognition. However, in terms of accuracy of recognition, it has a disadvantage that accuracy is lower than other technologies.

Finger vein recognition technology refers to a technique for identifying an individual using finger vein pattern images acquired through infrared. Human veins are located underneath the skin and have complex shapes, and studies have shown that vein patterns vary from person to person and do not change with age. Finger vein recognition technology has the following advantages.

1) Because veins exist inside the body, there is little risk of theft or forgery, and the condition of the finger surface does not affect recognition.

2) Infrared can be used to configure a non-contact device with high user convenience.

3) Because finger vein patterns can be found stable and relatively clear, low resolution cameras can be used to obtain finger vein images that are small and have simple image processing.

In other words, finger vein recognition technology can provide balanced performance in terms of security and convenience. Due to the above advantages, the finger vein recognition technology is in the spotlight recently compared to the fingerprint, iris, face and voice recognition technology.

Biometric technologies can be broadly classified into image-based technology and feature point-based technology. Image-based technology uses the acquired image itself for matching. The feature point based technique is a method of extracting feature points from an acquired image and using the feature points for matching. However, finger vein recognition using image-based technology has the following problems.

1) Since a large sized acquired image needs to be stored, it cannot be used for a personal identification system based on smart cards. There is also the possibility that the stored image is stolen.

2) Collecting and storing biometric information can cause human rights violations or personal information leakage.

3) The distortion of the original image is difficult to correct easily.

On the other hand, the feature-based technology can overcome most of the problems of the image-based technology, and the fingerprint recognition technology currently commercialized uses the feature-based technology. However, finger vein recognition technology still exists image-based technology, there is no example using a feature-based technology.

There is a need for a finger vein feature point extraction method for applying feature point based techniques to finger vein recognition techniques.

An object of the present invention is to provide a method and apparatus for extracting the bifurcation of the finger vein as a feature point of the finger vein.

In one aspect, a method for extracting finger vein feature points is provided. The method obtains a finger image from a finger passed through infrared rays, obtains a finger vein image by removing a background from the acquired finger image, and calculates a radiance at each pixel of the finger vein image. And extracting a pixel at which the calculated radioactivity becomes a local maxima as a bifurcation of the finger vein. Computing the radioactivity may include calculating an orientation field at each pixel of the finger vein image, calculating a blood cell distance (BCD) using the calculated orientation field, and As shown in the equation, the ratio of the number of pixels within a certain blood cell distance and the number of pixels within a certain physical distance may include calculating the radiation degree. In addition, the direction field may calculate a Hessian matrix to determine the direction of the eigenvector corresponding to the minimum eigenvalue of the Hessian matrix. The radiation rate may be calculated as a ratio of the number of pixels within a certain distance from the gradient field complex filtered by the following equation in each pixel.

Meanwhile, the method may further include smoothing out the calculated radiance using a Gaussian filter. The method may further include extracting the direction of each finger vein forming the branching point from the extracted branching point.

In another aspect, an apparatus for extracting finger vein feature points is provided. The device obtains a finger vein sensor for acquiring a finger image from a finger passed through the infrared light, and obtains a finger vein image by removing a background from the acquired finger image, calculating radiation levels at each pixel of the finger vein image, And a processor that extracts, as the branch of the finger vein, the pixel at which the calculated radioactivity is a local maximum. The processor may calculate the radiation rate using a direction field or calculate the radiation rate using a gradient field.

By calculating the radiance and extracting the bifuracation of the finger vein as a feature point, a feature point-based finger vein recognition technique with high identification accuracy with little risk of theft and forgery can be achieved.

The biometric system refers to a pattern recognition system that obtains biometric information from an individual, finds a feature from the biometric information, and compares it with a template stored in a database. The biometric system can be divided into a personal verification system and a personal identification system according to its application environment.

1 is a block diagram showing an identity verification system.

The identity verification system checks the identity by comparing the biometric information of the individual obtained from the recognition device with the template stored in the database of the system. A user requesting identity verification in the system requests identity verification in the system by inputting additional information such as a name, a personal identification number (PIN) or a smart card. The system compares the stored biometric information with the inputted additional information with the biometric information obtained from the recognition device to confirm whether or not it is the user. That is, in the identity verification system, 1: 1 matching is performed between the biometric information.

2 is a block diagram illustrating a personal identification system.

Like the identity verification system, the personal identification system identifies the individual by comparing the biometric information of the individual obtained from the recognition device with the template stored in the database of the system. However, the personal identification system matches the biometric information acquired by the recognition device with N templates stored in the system by 1: N to identify whether the user who is to be identified is the user stored in the system.

Both the identity verification system and the personal identification system perform a process of obtaining biometric information and matching it with a template stored in a database. In many cases, the algorithms used in identity and personal identification systems do not differ significantly. Since the proposed invention can also be used in both systems, the following does not strictly distinguish between the identity verification system and the personal identification system.

N. Miura, A. Nagasaka, and T. Miyatake "Extraction of finger-vein patterns using maximum curvature points in image profiles," IEICE Transactions on Information and Systems, vol. E90-D, no. 8, pp. 1185-1194, 2007. discloses an example of finger vein recognition technology. It obtains a cross-sectional profile in four directions (horizontal direction, vertical direction and two oblique directions) from the acquired finger vein image. By calculating the curvature in each cross-sectional view, the center points of the veins are extracted, the center points are connected, and the connected center points are combined in four directions to finally extract the vein pattern. W. Song, Finger-vein identification system using level set curvature, MS Thesis, Seoul National University, 2008. and M. Vlachos and E. Dermatas, "Vein segmentation in infrared images using compound enhancing and crisp clustering," Lecture Notes in Computer Science, vol. 5008, pp. 393-402, May. Another example of finger vein recognition technology is disclosed in 2008.

Hereinafter, the proposed finger vein feature point extraction method will be described through examples.

3 shows an embodiment of the proposed finger vein feature point extraction method.

In operation S100, a finger image is acquired from a finger that is a target of recognition passed through infrared rays.

Infrared radiation with a wavelength of 700 nm to 1000 nm passes through most of the human tissue, while all hemoglobin (Hb) inside the blood is absorbed. Therefore, the image of the inside of the finger including the vein can be obtained by passing the finger through infrared light.

4 shows an example of the structure of the finger vein sensor. The finger vein sensor 200 is basically an optical filter 220 for filtering an infrared source 210 and an infrared ray passing through a finger, and a CCD camera capable of capturing infrared rays through the optical filter. 230). The infrared rays emitted from the infrared light source 210 pass through the finger and are collected in the CCD camera 230 through the optical filter 220.

The apparatus may further include a video capture unit and a light source controller. The image capturing unit and the light source controller may serve to adjust the intensity of the light source to prevent saturation of the finally obtained finger image.

Referring back to FIG. 3, a background vein is removed from the acquired finger image in step S110 to obtain a finger vein image.

The finger image may include a background such as a circuit board that is not required for finger vein recognition. In general, an edge detector may be used to remove a background, but only a necessary part of an image may be obtained. However, since the background may also include a sharp boundary, it is difficult to remove the background through the edge detector. However, since there is no boundary inside the finger, it can be assumed that the boundary closest to the center in the image as the boundary of the finger.

5 illustrates one embodiment of a background removal method.

In operation S300, edge pixels of the entire image are extracted from the M × N size image.

In step S310, the center point c _v = N / 2 of the vertical line is initialized, and the index m = M of the vertical line is initialized.

In step S320, the upper direction boundary pixel u _m and the lower direction boundary pixel l _m nearest to the center point c _v are extracted at the m-th vertical line. If m = M or | u _{m +1} -u _m | <γ _e, u _m is determined as the inner upper edge pixel, and m = M or | l _{m +1} -l _m | <γ _{If e,} then lm is determined to be the inner lower edge pixel. γ _e is a constant, which is used to prevent unwanted border pixels from being determined as the border pixels of the finger. The center point of the vertical line c _v = (u _m + l _m ) / 2 is newly determined and the above steps are repeated while decreasing m by one. If m = 1 or u _m -l _m <γ _s , stop repeating.

In step S330, U and L, which are quadratic polynomial curves that best match the inner upper boundary pixel and the inner lower boundary pixel, are found.

In operation S340, the pixel in the upper region than the U and the pixel in the lower region than the L are removed to finally remove the background.

In addition to the above embodiment, only the image of the background without a finger may be separately obtained to obtain a difference from the image including the finger to remove the background. In addition, a noise control algorithm may be added according to the quality of the finger image.

Referring back to FIG. 3, in step S120, an orientation field is calculated at each pixel of the acquired finger vein image.

In this embodiment, a Hessian-based method is used among differential geometry-based methods to extract blood vessels from a finger vein image. Hessian is known for its excellent performance in capturing the features of cylindrical structures and extracting points of interest. In addition, since Hessian uses second-order partial derivatives, it is not affected by changes in signal intensity that can be approximated by first-order differential equations. In addition, the Hessian-based method is fast because it does not require iteration.

The Hessian matrix may be represented by Equation 1.

In Equation 1, L represents a brightness component at (x, y). The Hessian matrix has an eigenvalue, and the first eigen value may be referred to as λ ₁ and the second eigen value to be λ ₂ (where | λ ₁ | ≥ | λ ₂ |). In addition, an eigenvector corresponding to λ ₁ may be referred to as v ₁ , and an eigenvector corresponding to λ ₂ may be referred to as v ₂ . The eigenvalue analysis can extract the principal curvature and the local second-order structure of the image. In this case, λ ₁ , λ ₂ , v ₁ , and v ₂ may represent the maximum curvature, the minimum curvature, the direction of the maximum curvature, and the direction of the minimum curvature, respectively. Since the direction field means the direction of the eigenvector v ₂ corresponding to the minimum eigenvalue λ ₂ , it can be defined as φ _v2 .

In operation S130, a blood cell distance (BCD) is calculated using the calculated direction field in each pixel of the acquired finger vein image.

6 is a graphical representation of the concept of blood cell distance. Blood cell distance refers to the difficulty of movement in the concept of distance, assuming that blood cells in a blood vessel move under force. In other words, it is a new distance measuring method that calculates a short distance if the blood cells can move smoothly and a long distance if it cannot move. Referring to FIG. 6, when a force is applied to a blood cell in a direction of a blood vessel, the blood cell may move easily (400; first position). On the other hand, when a force is applied to the blood cells in the vertical direction of the blood vessel, the blood cells cannot move easily (410; second position). Therefore, even if the first position 400 and the second position 410 are physically at the same distance in terms of blood cells, the blood cell distance to the first position 400 is shorter than the blood cell distance to the second position 410. can do.

The blood cell distance may be defined by Equation 2.

In Equation 2, d _bc (x, y, φ _v2 ) represents the blood cell distance from the origin to (x, y) when the direction field is φ _v2 at (x, y). d _e (x, y) represents the physical (Euclidean) distance from the origin to (x, y). φ _f represents the direction of movement of blood cells, and φ _v2 represents the direction field at (x, y). If the movement direction φ _f of the blood cell is _equal to the direction field φ _v2 , the blood cell distance d _bc is equal to the physical distance d _e, and φ _f is If it is perpendicular to φ _v2 , the physical distance becomes infinity.

7 shows an example of a blood cell distance isometry curve. In (a), the direction field is homogeneous, and thus the shape of the equidistant curve can be seen to elongate along the direction field. This form may appear in areas of general blood vessels. In (b), the direction field has the shape of a quadratic curve, and thus the shape of the equidistant curve can be seen to elongate along the direction field. This form can occur in the region where two blood vessels intersect.

Referring to FIG. 3 again, a radialness is calculated using the calculated blood cell distance in step S140. Emissivity is one of the local features obtained from the direction field, and can be expressed by Equation 3.

In Equation 3, R (x, y, r) represents the radiance calculated based on the distance r at (x, y). On the other hand, N (x, y, r) and N _all (x, y, r) can be expressed by the equation (4).

In the equation 4, when the neighbor indication function h _bc (u, v, r, φ (u, v)) is the blood cell distance d _bc (u, v, φ (u, v)) ≤r Is 1, otherwise 0. Another adjacent indicating function h (u, v, r) is 1 if the physical distance d (u, v) ≦ r, and 0 otherwise.

That is, N (x, y, r) in Equation 4 represents the number of pixels within the blood cell distance r from (x, y), and N _all (x, y, r) is physically derived from (x, y). It represents the number of pixels within the distance r. Therefore, the radiation rate R (x, y, r) of Equation 3 means a ratio of the number of pixels within a certain blood cell distance and the number of pixels within a certain physical distance in each pixel.

8 is a graphical representation of the concept of radiation. Referring to FIG. 8, a circle 500 (solid line) having a radius r around (x, y) represents an area of pixels within a physical distance r. On the other hand, the shaded portion 510 (dashed line) represents the area of the pixel within the blood cell distance r. Since the number of pixels within the blood cell distance r increases along the direction field, the radiation rate increases when the direction field is radial around (x, y).

Referring back to FIG. 3, in step S150, the pixel whose calculated radiation value becomes a local maxima is extracted as a branch point of the finger vein.

The branch point may be used as a feature point when matching a recognition result with a template stored in a database. The feature point based matching method is mainly used in fingerprint recognition technology, but since the vein does not have a clear end point unlike the fingerprint, the branch point acquired using the calculated radiation can be used as the feature point.

9 shows the direction field around a point in a general vascular area and a point in a branched area. Referring to FIG. 9, more radial direction vectors are distributed around branch points than general points in blood vessels. Since the bifurcation is the point at which veins stretch out in at least three directions, it is expected that the bifurcation will be greater than the other points around it. Thus, by extracting a local maxima larger than the surrounding area as a branch point, the branch point can be used as a feature point in matching. In addition, the branch point may be extracted after smoothing out the calculated radiance using a Gaussian filter having a dispersion of σ _b before extracting the branch point.

Table 1 compares the performance of various finger vein feature extraction methods.

Referring to Table 1, the performance of the finger vein feature point extraction method was compared with the number of correct and wrong branch points. The finger vein feature extraction method based on the proposed radioactivity shows better performance than the other methods. The vesselness-based method shows almost the same performance as the proposed and proposed methods, but this method requires a lot of computation time.

Figure 10 shows another embodiment of the proposed finger vein feature point extraction method.

Extraction of finger vein feature points using radiographs is derived from physical models of blood cells and veins, and the method of calculation is theoretically complicated. Therefore, a method for quickly extracting the branch point by calculating the inclination field and approximating the radioactivity is proposed. This is possible because the slope field at the fork has a radial like the direction field.

In operation S600, a finger image is acquired from a finger that is a target of recognition passed through infrared rays. In operation S610, a background is removed from the acquired finger image to obtain a finger vein image.

In step S620, a gradient field is calculated at each pixel of the finger vein image.

If the gradient in (x, y) is called G (x, y), the normalized gradient field G _N (x, y) may be calculated by Equation 5.

In addition, the complex normalized gradient length G _C (x, y) can be calculated by Equation 6.

g _x (x, y) and g _y (x, y) represent the first and second components of the normalized gradient field G _N (x, y) of Equation 5, respectively.

In addition, a complex radial filter may be defined as shown in Equation (7).

(단, 일 때, 그 이외에는 0)

(Where, 0 otherwise)

In operation S630, an approximate radiation rate R _G (x, y, r) may be calculated using the complex normalized gradient field G _C and the complex radial filter. This is shown in equation (8).

In Equation 8, R _G (x, y, r) represents an approximate radioactivity calculated based on the distance r at (x, y). N _all (x, y, r) represents the number of pixels within the physical distance r from (x, y), and G _S (x, y) represents the complex filtered complex normalization gradient calculated by Equation (9). Indicates. That is, the approximated radiation also has a form of a ratio as shown in Equation 3 above.

In Equation 9, F (x, y) * G _C (x, y) represents a complex convolution of F (x, y) and G _C (x, y), and real ( F (x, y) * G _C (x, y)) represents the real part of F (x, y) * G _C (x, y).

In step S640, the pixel at which the calculated approximated radiation becomes a local maxima is extracted as a branch point of the finger vein. The extracted branch point is used as a feature point in matching. As a result of comparing the performance with the number of right and wrong branch points using the method of the present embodiment, 276 and 346 results, respectively. This is slightly lower than the performance of the finger vein feature point extraction method using the radioactivity calculated by the Hessian matrix and blood cell distance, but the finger vein feature point extraction method according to the present embodiment can be quickly calculated using a convolution operator. There is an advantage.

On the other hand, in the fingerprint recognition technology using the feature point, it is possible to extract not only the position of the feature point but also information on the direction of the curve at the feature point. Similarly, in the finger vein recognition technology, the direction of the finger vein can be additionally extracted from the branching point for more accurate personal identification.

The branching point of the finger vein may have at least three directions. In order to find the main direction around the branch point, the direction field around the branch point may be classified into at least three groups using K-means clustering or various tracking algorithms. By repeating this process at the extracted branch point, the direction of the finger vein at the branch point can be extracted. 11 shows an example of a state in which the direction information of the finger vein is extracted.

12 is a block diagram of a finger vein feature point extraction apparatus implemented in an embodiment of the present invention.

The finger vein feature point extraction apparatus 700 includes a finger vein sensor 710 and a processor 720. The finger vein sensor 710 and the processor 720 implement the proposed functional function, process and / or method.

The finger vein sensor 710 acquires a finger image by capturing infrared rays passing through the finger. The finger vein sensor 710 may have the structure of FIG. 4. The processor 720 obtains a finger vein image by removing a background from the acquired finger image, calculates a radiation rate at each pixel of the finger vein image, and specifies a pixel at which the calculated radiation rate becomes a local maximum value. Extracted as bifurcation of the vein is used as a feature point.

The processor 720 may include an application-specific integrated circuit (ASIC), another chipset, a logic circuit, and / or a data processing device. When the embodiment is implemented in software, the above-described techniques may be implemented with modules (processes, functions, and so on) that perform the functions described above. The module may be executed by the processor 720.

In the exemplary system described above, the methods are described based on a flowchart as a series of steps or blocks, but the invention is not limited to the order of steps, and certain steps may occur in a different order or concurrently with other steps than those described above. Can be. In addition, those skilled in the art will appreciate that the steps shown in the flowcharts are not exclusive and that other steps may be included or one or more steps in the flowcharts may be deleted without affecting the scope of the present invention.

The above-described embodiments include examples of various aspects. Although not all possible combinations may be described in order to represent the various aspects, one of ordinary skill in the art will recognize that other combinations are possible. Accordingly, it is intended that the invention include all alternatives, modifications and variations that fall within the scope of the following claims.

1 is a block diagram showing an identity verification system.

2 is a block diagram illustrating a personal identification system.

3 shows an embodiment of the proposed finger vein feature point extraction method.

4 shows an example of the structure of the finger vein sensor.

5 illustrates one embodiment of a background removal method.

6 is a graphical representation of the concept of blood cell distance.

7 shows an example of a blood cell distance isometry curve.

8 is a graphical representation of the concept of radiation.

9 shows the direction field around a point in a general vascular area and a point in a branched area.

Figure 10 shows another embodiment of the proposed finger vein feature point extraction method.

11 shows an example of a state in which the direction information of the finger vein is extracted.

12 is a block diagram of a finger vein feature point extraction apparatus implemented in an embodiment of the present invention.

Claims (10)

Obtain a finger image from a finger passed through infrared light, Obtaining a finger vein image by removing a background from the acquired finger image, Calculate a radiance at each pixel of the finger vein image, And a finger vein feature point extraction method comprising extracting a pixel at which the calculated radioactivity is a local maxima as a bifurcation of the finger vein. The method of claim 1, Calculating the radioactivity is Calculate an orientation field at each pixel of the finger vein image, The blood cell distance (BCD) is calculated using the calculated direction field, Finger vein feature point extraction method comprising the step of calculating the ratio of the number of pixels within a certain blood cell distance and the number of pixels within a certain physical distance in each pixel as shown below. However, R (x, y, r) represents the radiance calculated based on the distance r at (x, y). N (x, y, r) represents the number of pixels within blood cell distance r from (x, y), and N _all (x, y, r) represents the number of pixels within physical distance r from (x, y) Indicates.

The method of claim 2, The direction field is a finger vein feature extraction method characterized in that it is determined in the direction of the eigenvector corresponding to the minimum eigenvalue of the Hessian matrix by calculating the Hessian matrix by the following equation. However, L represents the brightness component in (x, y).

The method of claim 2, The blood cell distance is finger vein feature point extraction method characterized in that calculated by the following formula. However, d _bc (x, y, φ _v2 ) represents the blood cell distance from the origin to (x, y) when the direction field is φ _v2 at (x, y). d _e (x, y) represents the physical (Euclidean) distance from the origin to (x, y). φ _f represents the direction of movement of blood cells, and φ _v2 represents the direction field at (x, y).

The method of claim 1, The radiation rate is finger vein feature point extraction method characterized in that calculated by the ratio of the number of pixels within a certain distance and the complex field (gradient field) complex filtered by the following formula in each pixel. However, R _G (x, y, r) represents an approximate radioactivity calculated based on the distance r at (x, y). N _all (x, y, r) represents the number of pixels within physical distance r from (x, y), and G _S (x, y) represents the complex filtered complex normalized gradient.

The method of claim 5, The complex filtered gradient is finger vein feature point extraction method characterized in that it is calculated by the following formula. Where G _S (x, y) represents a complex filtered complex normalized gradient, F (x, y) represents a complex radial filter, and G _C (x, y) represents a complex normalized gradient at (x, y) . F (x, y) * G _C (x, y) represents the complex convolution of F (x, y) and G _C (x, y), real ( F (x, y) * G _C (x, y)) represents the real part of F (x, y) * G _C (x, y).

The method of claim 1, A finger vein feature point extraction method further comprising smoothing out the calculated radiance using a Gaussian filter. The method of claim 1, Finger vein feature point extraction method further comprises extracting the direction of each finger vein forming the branch point from the extracted branch point. Finger vein sensor for obtaining a finger image from the finger passed in the infrared; And Obtaining a finger vein image by removing a background from the acquired finger image, Calculate the emissivity at each pixel of the finger vein image, And a processor for extracting a pixel at which the calculated radioactivity becomes a local maximum value as a branch point of the finger vein. The method of claim 9, The processor is a finger vein feature extraction apparatus, characterized in that for calculating the radioactivity using a direction field or using the gradient field.

Images