WO2018196371A1 - 一种三维手指静脉识别方法及系统 - Google Patents

一种三维手指静脉识别方法及系统 Download PDF

Info

Publication number
WO2018196371A1
WO2018196371A1 PCT/CN2017/113362 CN2017113362W WO2018196371A1 WO 2018196371 A1 WO2018196371 A1 WO 2018196371A1 CN 2017113362 W CN2017113362 W CN 2017113362W WO 2018196371 A1 WO2018196371 A1 WO 2018196371A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
dimensional
finger
camera
finger vein
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2017/113362
Other languages
English (en)
French (fr)
Inventor
康文雄
刘宏达
邓飞其
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
South China University of Technology SCUT
Original Assignee
South China University of Technology SCUT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by South China University of Technology SCUT filed Critical South China University of Technology SCUT
Priority to US16/478,959 priority Critical patent/US11017205B2/en
Publication of WO2018196371A1 publication Critical patent/WO2018196371A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/12—Fingerprints or palmprints
    • G06V40/1365—Matching; Classification
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/64—Three-dimensional [3D] objects
    • G06V20/647—Three-dimensional [3D] objects by matching two-dimensional images to three-dimensional objects
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/10—Image acquisition
    • G06V10/12—Details of acquisition arrangements; Constructional details thereof
    • G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
    • G06V10/147—Details of sensors, e.g. sensor lenses
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/60—Type of objects
    • G06V20/64—Three-dimensional [3D] objects
    • G06V20/653—Three-dimensional [3D] objects by matching three-dimensional models, e.g. conformal mapping of Riemann surfaces
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/107—Static hand or arm
    • G06V40/11—Hand-related biometrics; Hand pose recognition
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/12—Fingerprints or palmprints
    • G06V40/1347—Preprocessing; Feature extraction
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/12—Fingerprints or palmprints
    • G06V40/1347—Preprocessing; Feature extraction
    • G06V40/1359—Extracting features related to ridge properties; Determining the fingerprint type, e.g. whorl or loop
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/12—Fingerprints or palmprints
    • G06V40/1365—Matching; Classification
    • G06V40/1376—Matching features related to ridge properties or fingerprint texture
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/10—Image acquisition
    • G06V10/16—Image acquisition using multiple overlapping images; Image stitching
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/12—Fingerprints or palmprints
    • G06V40/1382—Detecting the live character of the finger, i.e. distinguishing from a fake or cadaver finger
    • G06V40/1388—Detecting the live character of the finger, i.e. distinguishing from a fake or cadaver finger using image processing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/14—Vascular patterns

Definitions

  • the present invention relates to the field of vein recognition technology, and in particular, to a three-dimensional finger vein recognition method and system.
  • Biometrics is a technique that uses one or more human physiological characteristics (such as fingerprints, faces, irises, veins, etc.) or behavioral features (such as gait, signature, etc.) for identity authentication.
  • human physiological characteristics such as fingerprints, faces, irises, veins, etc.
  • behavioral features such as gait, signature, etc.
  • finger vein recognition It is a biometric identification technology that uses the texture information of blood vessels under the skin of the finger as an individual identity verification.
  • the unique advantages of finger vein recognition include: (1) using the internal physiological characteristics of the living body, it is difficult to forge, and has high safety. (2) Non-contact or weak contact measurement can be achieved with good affinity. (3) It is not easily affected by finger surface damage or oil stains and sweat, and has strong anti-interference. After development in recent years, the development and application of finger vein recognition has shown a rapid and prolific trend and has shown great market potential.
  • the vein recognition technology generally uses a monocular camera to perform two-dimensional imaging of a single-sided finger in the near-infrared band. In this imaging mode, only the vein image on one side of the finger is acquired, and thus the vein characteristic information included is limited. In addition, due to factors such as the position of the finger when the finger vein image is collected, the finger placement posture (such as rotation), uneven exposure, etc., it is difficult to locate a stable and reliable ROI, so that subsequent feature extraction and matching on the rotation deviation and The displacement deviation is sensitive.
  • a few existing studies include: using an elliptical model to expand the acquired two-dimensional finger vein image to achieve standardization of the finger vein image, and then intercept the effective area. Matching; using a circular model to expand a two-dimensional finger vein image; or, using a three-dimensional model, the key is still to use an elliptical model to normalize finger vein images in six different poses and then match. No matter which physical model is used, it has improved the difference between the vein images taken by the same finger in different postures to some extent.
  • the corresponding texture area is less, which is not conducive to matching; on the other hand, the image quality of the vein image in the edge area is generally affected by imaging factors, which is generally poor, and also affects the recognition result.
  • Another method is a three-dimensional imaging method based on multi-view geometry, but this scheme is difficult to find or even find matching feature points in 3D reconstruction, so it is difficult to calculate the depth information of all vein textures. In addition, this method collects Vein texture is also only one-sided, so there is still a problem of limited feature information.
  • the main object of the present invention is to overcome the shortcomings and deficiencies of the prior art, and to provide a three-dimensional finger vein recognition method, which can obtain a better recognition effect and is highly robust to various postures such as finger rotation and tilt.
  • Another object of the present invention is to provide a three-dimensional finger vein recognition system based on the above method, which can achieve a better recognition effect and is highly robust to various postures such as finger rotation and tilt.
  • a three-dimensional finger vein recognition method includes the following steps:
  • Approximate the finger profile as a circle S divide the three-dimensional finger into several sections, calculate the profile of each section, and approximate the finger with a circle with different positions of different radii; then press all the contours
  • the axial direction of the finger is connected in series to obtain an approximate three-dimensional model of the finger;
  • the two-dimensional image textures captured by the three cameras are mapped into the three-dimensional finger model; the images captured by the three cameras have two overlapping parts and non-overlapping parts. To determine the overlapping area, and then to treat the overlapping area and the non-overlapping area separately; obtain a three-dimensional finger vein image;
  • the three-dimensional finger is equally divided into 300 sections.
  • the method of calculating the profile of each profile is:
  • the xOy coordinate system is established according to the projection centers C 1 , C 2 , C 3 of the three cameras: the intersection of the vertical bisectors of the three points is used as the coordinate origin O, and the mid-perpendicular line of C 2 C 3 is used as the y-axis to establish the flute.
  • the Carl coordinate system xOy transforms the pixel coordinates on the image into the xOy coordinate system:
  • the distance from the imaging plane of the origin O to i camera of the coordinate system is L 0 , and the coordinates of one point on the image are A(x 0 , y 0 ), which are transformed into the imaging I i in the xOy coordinate system.
  • the coordinates are A'(x, y ):
  • the circle parameters x c , y c , r c are solved by the optimization problem solving algorithm.
  • the circle parameter can be obtained by solving the algorithm by the gradient descent method.
  • each two straight lines have an intersection point, and the intersection of L i and L j is P ij (x ij , y ij ), where:
  • the distances from P ij to L ui and L uj are r ij and r ji , respectively, which can be calculated from the distance formula of the point to the line. Therefore, the final center coordinate S(x c , y c ) and the radius r c can be Calculated by the following formula:
  • correction is needed, so that the internal parameters of the three cameras after correction are the same, the main line of sight is on the same plane, and the main line of sight is the direction from the projection center to the origin of the system coordinate system.
  • the three cameras are individually calibrated, using matlab calibration toolbox, the parameters in the three cameras are K 1 , K 2 , K 3 ;
  • H i represents a conversion matrix of the uncorrected image coordinates p i to the corrected image coordinates p i ';
  • R i represents the rotation matrix of the i camera coordinate system relative to the system coordinate system. Before and after the conversion, the camera position remains unchanged, but the converted rotation matrix is transformed into: The coordinate of C i is (x i , y i , 0), and O is the origin of the system coordinate system;
  • the corrected image coordinates p i ' can be obtained by substituting into the correction conversion formula.
  • the method of mapping the two-dimensional image texture captured by the three cameras into the three-dimensional finger model is:
  • the three-dimensional finger model coordinates are converted to image coordinates; taking the camera i as an example, for each circular model, the imaging corresponding to one camera is a vertical line segment, that is, their ordinates are the same in the image, Set to v, as for the determination of v, it should be the same as the ordinate of the corresponding edge point when the circular model is created.
  • the abscissa u is concerned; this is the case of converting the two-dimensional coordinates into one-dimensional coordinates. Then you can get the following coordinate conversion formula:
  • u is the abscissa on the corresponding image
  • v a is the proportional coefficient
  • its role is to set the ordinate of the calculation result to 1
  • ⁇ i represents the angle between the orientation of the i-camera and the positive direction of the horizontal axis in the system coordinate system.
  • L 0 is the distance from the imaging plane center of the i-camera to the system coordinate system
  • u m is the abscissa of the imaging center
  • (x', y ') is the point on the circle, which satisfies the following equation:
  • the next step is to determine the arc corresponding to the camera.
  • the clockwise direction is the positive direction, and the starting end and the end of the area range are defined.
  • the determination of the arc is determined according to the orientation of the camera placed in the actual system.
  • the range should correspond to
  • mapping For an area photographed by only a single camera, only the image captured by the camera is used for mapping, that is, the image coordinates corresponding to the points on the arc are calculated according to the coordinate conversion formula, and the gray value of the pixel in the coordinate is used as the point on the arc. Pixel gray value;
  • the Gaussian image fusion method is used to determine the gray value of the voxel point by:
  • the No. 2 camera shooting area is The No. 3 camera shooting area is They have overlapping areas for For the point (x', y') on the overlapping area, use the formula to calculate the image coordinates corresponding to the camera No. 2 and the camera No. 3, respectively, and take the gray value I 2 of the two image coordinate points (u 2 , v 2 ) and I 3 (u 3 , v 3 ), and finally calculate the gray value of the point (x', y') on the overlapping area by the following formula:
  • I(x', y') ⁇ 2 I 2 (u 2 , v 2 ) + ⁇ 3 I 3 (u 3 , v 3 )
  • ⁇ is the Gaussian weight
  • the weight value is calculated according to the formula g(p), and if it belongs to the starting end, it is calculated according to the formula 1-g(p), where:
  • the image taken by any two cameras is processed in a similar way to solve the texture mapping problem of a single circular model; for all approximate circles in the three-dimensional finger model, the gray values of the three-dimensional coordinate points are respectively calculated according to the above method, and finally the three-dimensional finger vein image is obtained. .
  • the two three-dimensional finger vein texture images that need to be matched are first aligned and expanded into a two-dimensional image, and the three-dimensional finger vein texture is matched by feature extraction and matching of the two-dimensional image.
  • the steps to expand into a two-dimensional image are:
  • mapping a three-dimensional finger vein texture onto a cylinder of a given size is equivalent to normalizing each circle into a fixed-size circle, starting from the first point in the horizontal direction for a circular model. Perform equal-angle sampling, co-sampling N points, and assign each point to the same angle point on the circle corresponding to the cylinder:
  • the side of the cylinder is cut along a bus bar according to the following formula 1 and expanded into a two-dimensional image of M ⁇ N size.
  • ⁇ represents the azimuth of the bus.
  • the cylinder corresponding to the template image and the cylinder corresponding to the image to be matched are expanded into a two-dimensional image, if the position of the busbar does not correspond, the two images to be expanded will have an offset problem, so as to ensure After the sample of the same type of object is expanded into a two-dimensional image, the corresponding position can be kept basically unchanged.
  • the following method is proposed to align the three-dimensional finger vein image, that is, the value of ⁇ is determined:
  • the purpose is to find a matching image block in the three-dimensional image to be matched, thereby realizing alignment of the three-dimensional image to be matched with the template three-dimensional image;
  • the above formula corr refers to calculating the correlation between the two images, taking k ⁇ 0 corresponding to the maximum value of the correlation as the rotation angle of the sample to be matched;
  • the three-dimensional finger vein image to be matched is expanded into a two-dimensional texture image according to Formula 1, which is aligned with the template two-dimensional image.
  • the feature extraction and matching steps of the two-dimensional image are:
  • the LBP operator was used to extract the structural features of the vein texture, and then the Manhattan distance calculation method was used for feature matching:
  • the single-value method is used, as follows:
  • a system based on the above three-dimensional finger vein recognition method comprises two parts of an upper and a lower position machine;
  • the lower position machine comprises a light source control module and an image acquisition module, and
  • the light source control module comprises a microprocessor, a drive circuit and a near-infrared light source, which are sequentially connected, and an image
  • the acquisition module includes a camera for acquiring an image;
  • the upper computer is equipped with software for realizing three-dimensional reconstruction and recognition of the finger vein, and the microprocessor of the upper computer and the light source control module realizes the transmission of the brightness control signal of the light source, and transmits the video data with the camera;
  • the lower computer is responsible for the control of the intensity of the near-infrared light source and the collection and transmission of the image information.
  • the upper computer is responsible for processing, three-dimensional reconstruction, feature extraction and recognition of the collected digital images.
  • the microprocessor of the upper computer and the light source control module realizes the transmission of the light source brightness control signal through the serial port, and the light source control module uses the microprocessor to control the brightness of the near-infrared light source, so that the three side light sources alternately illuminate the finger.
  • the upper computer and the camera transmit video data through a USB method, and the camera adopts near infrared USB. camera.
  • three cameras are mounted at three vertex positions of an equilateral triangle, the main lines of sight of the three cameras are on the same plane and the two and two intersect at 120°, and corresponding LED near-infrared light groups are disposed directly opposite each camera. .
  • the lower computer generates a PWM signal through a timer to control the brightness of the near-infrared LED, and uses the serial port to interrupt receiving the control command sent by the upper computer.
  • the present invention has the following advantages and beneficial effects:
  • the invention adopts three cameras to photograph the finger vein image in an omnidirectional angle, and then reconstructs the three-dimensional finger vein and extracts and matches the three-dimensional finger vein image feature, and the finger vein image taken by the three cameras is simpler than a single camera.
  • the finger vein image under the inner skin of the finger is more characteristic information, and more information means more accurate judgment;
  • the three-dimensional finger vein recognition method proposed by the present invention can be compared with the traditional two-dimensional finger vein recognition. Get more vein texture features, get better recognition, and also effectively overcome the problem of cognitive performance degradation caused by finger posture changes.
  • FIG. 1 is a view showing an image pickup apparatus of a three-dimensional finger vein recognition system in the first embodiment.
  • FIG. 2 is a hardware configuration diagram of a three-dimensional finger vein recognition system in the first embodiment.
  • 3 is a software flow chart of the three-dimensional finger vein recognition system in the first embodiment.
  • Figure 4 is a schematic diagram of the principle of the pinhole model.
  • Fig. 5 is a schematic diagram of an approximate imaging model in the second embodiment.
  • Fig. 6 is a schematic view showing a circle approximation in the second embodiment.
  • Fig. 7 is a schematic diagram showing the establishment of a system coordinate system in the second embodiment.
  • Figure 8 is a schematic view of a boundary calibrator in Example 2.
  • Embodiment 9 is a schematic view showing a photographing range of a single camera in Embodiment 2.
  • FIG. 10 is a schematic diagram of texture mapping in Embodiment 2.
  • FIG. 11 is a schematic diagram of a three-dimensional finger vein feature extraction and feature matching process in Embodiment 2.
  • Figure 12 is a three-dimensional finger vein diagram in the second embodiment.
  • Figure 13 is a plot of error rate for 2D and 3D methods under LBP.
  • Fig. 14 is a plot of error rate such as 2D and 3D methods under HOG.
  • a three-dimensional finger vein recognition system is composed of hardware and software, wherein the hardware part comprises a near-infrared image acquisition device, and the software part comprises a PC-side finger vein three-dimensional reconstruction and recognition algorithm; wherein, the hardware device It is mainly responsible for the control of the intensity of the near-infrared light source and the transmission of image information.
  • the software part is responsible for processing, three-dimensional reconstruction, feature extraction and recognition of the acquired digital image. Unlike the existing two-dimensional finger vein recognition system, the system uses three cameras to obtain all vein images under the finger epidermis, and to match by extracting three-dimensional finger vein texture features.
  • the system hardware includes two parts, the upper and lower position machines.
  • the lower position machine is composed of two modules: a light source control module and an image acquisition module.
  • the former uses a microcontroller to control the brightness of the near-infrared light source, and realizes three side light sources to illuminate the finger in turn, and the latter uses a 850 nm near-infrared USB camera (model KS302D). Capture images.
  • the upper computer is the PC end, the PC end and the microcontroller realize the transmission of the light source brightness control signal through the serial port, and the camera transmits the video data through the USB mode.
  • the structure diagram of the three-dimensional finger vein image acquisition device is shown in FIG.
  • the system software part is the core of the system, mainly responsible for system flow control, image processing, 3D reconstruction and matching.
  • the middle and lower machines use Keil to develop the single-chip control program
  • the upper computer uses VS2013+OpenCV+OpneGL to develop software that integrates image processing, 3D reconstruction, and feature extraction and matching.
  • the lower computer generates the PWM signal through the timer to control the brightness of the near-infrared LED, and uses the serial port to interrupt the control command sent by the host computer.
  • the upper computer first collects the image and controls the lower computer to adjust the brightness of the light source, and then cuts and goes to the original image.
  • a series of preprocessing operations such as noise and contour extraction; then the proposed 3D reconstruction algorithm is used to construct the 3D model of the finger and map the vein texture to the 3D finger model; finally, the matching function is implemented according to the feature extraction method.
  • the system software flow chart is shown in Figure 3.
  • the three-camera image is used to capture the finger vein image in an omnidirectional angle, and then the three-dimensional finger vein reconstruction and the three-dimensional finger vein image feature extraction and matching are performed.
  • the vertical cross-sectional view of the finger axis approximates an ellipse with an eccentricity close to 1, but the relative mathematical equation of the circle has fewer parameters and is easier to calculate, so the finger is
  • the section view is approximated as a circle; in this case, the fingers can be approximated by a circle with different positions of different radii.
  • the camera used in this system is almost undistorted, its imaging model can be approximated by the pinhole model, as shown in Figure 4.
  • the three cameras use the same type of CMOS camera with the same parameters.
  • the center axis of the finger is located at the center of the three cameras, and the axis of the finger is approximately perpendicular to the plane formed by the projection centers of the three cameras. . Therefore, a method of constructing a three-dimensional model of a finger according to a finger contour is proposed, and the problem of unevenness of pixel points sampled by the camera in the axial direction of the finger can be ignored, and the stereoscopic structure shown in FIG. 5 is approximated to approximate expression.
  • the geometric model with a circular approximation is shown in Figure 6, and the size and position of the circle are solved by the following two steps.
  • Step 1 Transform the pixel coordinates on the image into the xOy coordinate system as shown in FIG. 6.
  • the relative positions of the projection centers C 1 , C 2 , C 3 of the three cameras are known, with the intersection of the vertical bisectors of the three points as the coordinate origin O, and the mid-perpendicular line of C 2 C 3 as the y-axis.
  • the Cartesian coordinate system xOy is established.
  • the distance from the imaging plane of the origin O to i camera of the coordinate system is L 0 (the L 0 corresponding to the three cameras is the same), and the coordinates of one point on the image are A (x 0 , y 0 ), transform it into the image I i in the xOy coordinate system, and the transformed coordinates are A'(x, y):
  • ⁇ i represents the angle between the ray OC i and the positive direction of the x-axis
  • i 1, 2, 3, the same below
  • y m represents half the height of the image in the radial direction of the finger.
  • Step 2 Find the circle parameters.
  • the angle bisector of Lau and L bi is the straight line that passes the center of the circle. Therefore, the coordinates of the center of the circle can be obtained by calculating the intersection point of the angle bisector, and then the radius of the circle is calculated.
  • the specific calculation method is as follows:
  • each two straight lines have an intersection point, and the intersection of L i and L j is P ij (x ij , y ij ), where:
  • the distances from P ij to L ui and L uj are r ij and r ji , respectively, which can be calculated from the distance formula of the point to the line. Therefore, the final center coordinate S(x c , y c ) and the radius r c can be Calculated by the following formulas (10)-(12):
  • the circle has been used to approximate the outline of a finger axial section, and the three-dimensional finger is equally divided into several (300 in the text) sections, the contour of each section is calculated, and then all the contours are oriented in the direction of the finger axis.
  • an approximate three-dimensional model of the finger can be obtained.
  • the hardware structure Due to factors such as precise positioning methods, the hardware structure is often not guaranteed to be consistent with the ideal model, so it is generally necessary to do some correction processing on the system.
  • the purpose of the correction is to make the internal parameters of the three cameras corrected to be the same, the main line of sight is on the same plane, and the main line of sight is the direction from the projection center to the origin of the system coordinate system.
  • the correction algorithm proposed here consists of the following three steps:
  • the three cameras are individually calibrated, using matlab calibration toolbox, the parameters in the three cameras are K 1 , K 2 , K 3 ;
  • ⁇ is the rotation angle around the z-axis
  • ⁇ is the rotation angle around the y- axis
  • ⁇ is the rotation angle around the axis
  • the xyz axis is the coordinate axis of the system coordinate system.
  • the first camera coordinate system C 1 x'y'z' is used as a reference.
  • the t 12, t 13 to determine A ', B', then their midpoint D can be obtained, so that the translation vector t 23 through the point where the point D, so there is a translation vector t 23 after the starting point A "and the end point B", referred A'A "is a midpoint C 2, B'B" is a midpoint C 3.
  • the plane C 1 C 2 C 3 composed of the projection centers of the three cameras is the xy plane
  • the intersection of the angle bisector of C 1 C 2 C 3 is the coordinate origin O
  • the angle bisector of C 2 C 3 is x.
  • the axis establishes a coordinate system.
  • the translation matrix t s and the rotation matrix R s from C 1 x'y'z' to the coordinate system Oxyz are easily obtained, and the system coordinate system can be established.
  • H i represents a conversion matrix of the uncorrected image coordinates p i to the corrected image coordinates p i ′;
  • R i represents the rotation matrix of the camera coordinate system of i with respect to the system coordinate system.
  • the position of the camera remains unchanged, but the converted rotation matrix is transformed into:
  • the coordinates of the point C i are (x i , y i ), and O is the origin of the system coordinate system, so that there is a projection formula after the conversion:
  • the system model has been equivalent to the ideal model, and the 3D reconstruction algorithm of the ideal model can be used to construct the 3D model of the finger.
  • the 3D reconstruction algorithm of the ideal model can be used to construct the 3D model of the finger.
  • the next step is to map the two-dimensional image texture captured by the three cameras into the three-dimensional finger model. It can be found from Fig. 6 that the images captured by the three cameras have both overlapping parts and non-existing parts. The overlapping parts are different for the two cases. Therefore, it is necessary to first determine the overlapping area, and then perform different processing on the overlapping area and the non-overlapping part. From partial to overall thinking, you only need to study the texture mapping method under a single circular model, and then you can generalize to all circular models to complete the texture mapping of this three-dimensional finger.
  • u is the abscissa on the corresponding image
  • v a is the proportional coefficient
  • its role is to set the ordinate of the calculation result to 1
  • ⁇ i represents the angle between the orientation of the i-camera and the positive direction of the horizontal axis in the system coordinate system.
  • L 0 is the distance from the imaging plane center of the i-camera to the system coordinate system
  • u m is the abscissa of the imaging center
  • (x', y') is the point on the circle, which satisfies the following equation (19):
  • the next step is to determine the arc corresponding to the camera.
  • the clockwise direction is defined as the positive direction, and the starting end and the end of the area range are defined.
  • the determination of the arc is determined according to the orientation of the camera placed in the actual system, for example, in FIG.
  • the range of camera No. 3 should correspond to
  • the overlapping area of the images taken by the two cameras is very obvious. For example, suppose the range of the No. 2 camera (the position of which refers to Figure 6) is Then, the overlapping area shot by the No. 2 camera and the No. 3 camera corresponds to
  • the area photographed by the No. 2 camera and the No. 3 camera is still taken as an example, and the No. 2 camera shooting area is The No. 3 camera shooting area is They have overlapping areas for For areas where only a single camera is shot (eg It belongs to the area where only the No. 2 camera is photographed. Only the image taken by the camera is used for mapping. That is, the image coordinates corresponding to the points on the arc are calculated according to formula (18), and the gray value of the pixel in the coordinate is used as the arc.
  • the Gaussian image fusion method is used to determine the gray value of the three-dimensional pixel point by: the point on the overlapping area Calculate the image coordinates corresponding to the camera No. 2 and the camera No. 3 using equation (18), and take the gray values I 2 (u 2 , v 2 ) and I 3 (u 3 , v 3 ) of the two image coordinate points. Finally, the gray value of the point (x', y') on the overlapping area is calculated by the following formula (20):
  • I(x',y') ⁇ 2 I 2 (u 2 ,v 2 )+ ⁇ 3 I 3 (u 3 ,v 3 ) (20)
  • ⁇ is the Gaussian weight
  • the weight value is calculated according to the formula g(p), and if it belongs to the starting end, it is calculated according to the formula 1-g(p), where:
  • 3 ⁇ overlap part length.
  • the image taken by any two cameras is processed in a similar way to solve the texture mapping problem of a single circular model.
  • the gray values of the three-dimensional coordinate points are respectively calculated according to the above method, and finally a three-dimensional finger vein image is obtained.
  • the feature extraction of 3D images can be carried out in the following two ways.
  • One idea is to extract feature points in 3D coordinates to form feature vectors. If there is literature, the concept of 2D detail points is extended to 3D in the fingerprint recognition feature extraction algorithm. Two sample matching scores are obtained by calculating the distance of the detail points in the two samples.
  • Another way is to convert the three-dimensional image into a two-dimensional image through special transformation, and then use the existing two-dimensional image feature extraction method for feature extraction. If there is literature, the 3D fingerprint image is mapped onto a cylinder, and then expanded into a 2D image for further processing.
  • the method of matching the samples is to define a distance measure function between the feature descriptors to calculate the distance value between the samples.
  • the distance calculation methods are: Manhattan distance, Euclidean distance, Mahalanobis distance, etc., and then used The single ⁇ value method, the nearest neighbor method, etc. give the final matching decision.
  • a flow chart of a method for extracting and matching three-dimensional finger vein features proposed in the embodiment.
  • a three-dimensional image of the finger vein is generated by three three-dimensional images of the collected template samples and three two-dimensional images of the samples to be matched according to the above-described three-dimensional finger vein reconstruction method. Since there is not too much constraint on finger placement, there is generally a certain rotational deviation around the axis between the two samples.
  • the method adopted in this embodiment is: first aligning two three-dimensional finger vein texture images that need to be matched, and respectively expanding into a two-dimensional image, and then performing three-dimensional finger vein texture by feature extraction and matching of the two-dimensional image. Match.
  • the details are as follows.
  • the cylinder can be expanded into a rectangle after being cut along a bus bar.
  • the three-dimensional structure of the human finger approximates the cylinder model. If the three-dimensional finger epidermis vein is to be expanded into a two-dimensional image, the three-dimensional finger vein can be first mapped to a cylinder. On, then expand into a two-dimensional image.
  • mapping a three-dimensional finger vein texture onto a cylinder of a given size is equivalent to normalizing each circle into a fixed-size circle, starting from the first point in the horizontal direction for a circular model. Perform equal-angle sampling, co-sampling N points, and assign each point to the same angle point on the circle corresponding to the cylinder:
  • ⁇ 2 ⁇ /N
  • r is the radius of a given cylinder
  • r c is the radius of the circular model in the three-dimensional finger vein reconstruction
  • (x c , y c ) is the circular model Center position. All vein models are processed to map the vein texture to the side of the cylinder. Next, the side of the cylinder is cut along a bus bar according to formula (23) and expanded into a two-dimensional image of M ⁇ N size.
  • ⁇ represents the azimuth of the busbar, and the meaning of the other symbols is the same as equation (22). If the cylinder corresponding to the template image and the cylinder corresponding to the image to be matched are expanded into a two-dimensional image, if the position of the busbar does not correspond, the two images that are expanded will have an offset problem, so as to ensure that they belong to the same class of objects. After the sample is expanded into a two-dimensional image, the corresponding position can be kept basically unchanged. The following method is proposed to align the three-dimensional finger vein image (ie, determine the value of ⁇ ):
  • the above formula corr refers to calculating the correlation between the two images, taking k ⁇ 0 corresponding to the maximum value of the correlation as the rotation angle of the sample to be matched.
  • the three-dimensional finger vein image to be matched is expanded into a two-dimensional texture image according to formula (23), which is aligned with the template two-dimensional image.
  • the feature extraction algorithm of the two-dimensional image can be used to extract the feature. Due to illumination changes, finger deformation, noise interference, etc., the detail features acquired from the vein image are not very stable, and the detail point feature is not robust to finger offset. Relatively speaking, the structural features of vein lines are robust to noise, light changes, finger deformation and other influencing factors. Therefore, the simple and efficient LBP operator [1] is used to extract the structural features of vein texture.
  • the calculation method of Manhattan distance with the least computational complexity is the simplest template matching method.
  • the single ⁇ value method is adopted.
  • the similarity of the two samples is considered to be small, and it is judged that they do not belong to the same type of object. Otherwise, the similarity of the two samples is considered to be high. , to determine that they belong to the same class of objects.
  • the three-dimensional finger vein imaging system is proposed in Embodiment 1 to collect the omnidirectional finger epidermal vein image, the three-dimensional model of the finger is constructed by the circle approximation method proposed in Embodiment 2, and the two-dimensional texture is mapped onto the three-dimensional finger model, and the obtained three-dimensional finger is obtained.
  • the image is shown in Figure 12. Each row in the figure is an observation of the same finger at different viewing angles.
  • the three-dimensional finger vein reconstruction effect of three different fingers is displayed.
  • the three-dimensional finger vein model reconstructed by the method of the second embodiment has a high degree of fidelity, including the size and bending degree of the finger, which is consistent with the real situation, and basically reduces the distribution of the epidermis vein of the human finger in the vein texture.
  • Figure 13 uses the LBP operator to extract texture features. It can be seen from the figure that the recognition effect using the three-dimensional finger vein texture feature is better than the traditional two-dimensional finger vein texture feature. Because the finger vein image taken by the three cameras used in the system in Embodiment 1 is more than the feature information contained in the finger vein image under the inner skin of the finger, the more information means more accurate judgment. . In addition, the HOG operator is used to extract the texture features for experiment. The results are shown in Fig. 14. It can be found that the same conclusion can still be obtained. The three-dimensional method proposed in this embodiment can obtain better matching results and further verify. The effectiveness of the method.
  • the equal error rate of matching using the three-dimensional scheme reaches 2.273 (the error rate is 1.515% when using the HOG operator), and the result is more than the existing two-dimensional finger vein image.
  • the equal error rate of matching is 4.167% (the error rate is 3.788% when using the HOG operator), which has a significant improvement effect.
  • Example 1 and Example 2 propose a brand-new three-dimensional finger vein recognition system, which uses three cameras to capture the vein image under the skin of the all-round finger.
  • a finger is fitted with a circle.
  • the method of section line is used to construct the three-dimensional model of the finger.
  • the two-dimensional finger veins taken from three angles of view are mapped into the three-dimensional finger model.
  • a three-dimensional finger vein texture feature extraction method is proposed, which realizes a three-dimensional finger vein data.
  • the identity authentication system, the experimental results show that the three-dimensional finger vein recognition system proposed by the embodiment can obtain more vein texture features than the traditional two-dimensional finger vein recognition system, and obtain better recognition effect, and can effectively overcome The problem of decreased recognition performance caused by finger gesture changes.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Theoretical Computer Science (AREA)
  • Human Computer Interaction (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Vascular Medicine (AREA)
  • Software Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
  • Collating Specific Patterns (AREA)
  • Image Analysis (AREA)

Abstract

一种三维手指静脉识别方法及系统，包括步骤：三个摄像头从三个角度拍摄手指静脉图像，获得三幅图像；依据手指轮廓线构建手指三维模型；将三个摄像头拍摄的二维图像纹理映射到三维手指模型中，对重叠区域和非重叠区域分别做不同的处理；得到三维手指静脉图像；对三维手指静脉图像进行特征提取与匹配，最终完成识别。所述方法能获得更好的手指静脉识别效果，对手指旋转、倾斜等多种姿态具有较高鲁棒性。

Description

一种三维手指静脉识别方法及系统 技术领域
本发明涉及静脉识别技术领域，特别涉及一种三维手指静脉识别方法及系统。
背景技术
随着信息时代的发展，身份认证的重要性开始进一步凸显，传统的身份认证方式已经难以满足新时代人们对身份认证提出的便捷性，防伪性和高安全性的诸多需求，此时，以指纹识别和人脸识别为代表的生物特征识别技术开始逐步得到深入研究和广泛应用。生物特征识别技术是利用一种或多种人类的生理特征(如指纹、人脸、虹膜、静脉等)或行为特征(如步态、签名等)进行身份认证的一种技术。其中，手指静脉识别技术以其独特的优势开始在身份认证领域占得重要的一席地位，它是一种利用手指表皮下的血管的纹路信息作为个体身份验证的生物特征识别技术。与其它生物特征识别技术相比，指静脉识别具备的独特优势包括：(1)利用的是活体的内生理特性，难以伪造，具备高安全性。(2)可实现非接触或弱接触测量，具备良好的亲和性。(3)不易受手指表面伤痕或油污、汗水影响，具备较强的抗干扰性。经过近些年的发展，指静脉识别的研发和应用呈现出迅速繁荣的趋势，并显示出巨大的市场潜力。
现阶段的静脉识别技术一般采用单目相机对单侧手指进行近红外波段的二维成像，这种成像方式下仅仅获取了手指单侧的静脉图像，因此包含的静脉特征信息有限。此外，由于受采集指静脉图像时手指放置的位置、手指摆放姿态(如旋转)、曝光不均匀等等因素影响，难以定位稳定可靠的ROI，从而后续进行的特征提取和匹配对旋转偏差和位移偏差比较敏感。
然而，针对不同姿态下手指静脉识别的研究相对较少，现有的少量研究包括：采用椭圆模型对采集的二维手指静脉图像进行拓展，从而实现指静脉图像的标准化，然后再截取有效区域进行匹配；采用圆模型将二维手指静脉图像扩展；或者，采用三维模型的方法，其关键仍然是使用椭圆模型，将六种不同姿态下的手指静脉图像标准化，再做匹配。不管是用哪种物理模型，都在一定程度上改善了同一手指在不同姿态下拍摄的静脉图像之间存在较大差异 的情况，但仍然存在的问题是：一方面，对应的纹理区域变少了，不利于匹配；另一方面，边缘区域静脉图像质量受成像因素的影响一般会比较差，同样影响到识别结果。还有一种方法是基于多视图几何的三维成像方法，但这种方案在三维重建时难以找到甚至找不到匹配的特征点，因而难以计算全部静脉纹理的深度信息，此外，这种方法采集的静脉纹理也是只有单侧的，因此仍然存在特征信息有限的问题。
发明内容
本发明的主要目的在于克服现有技术的缺点与不足，提供一种三维手指静脉识别方法，能获得更好的识别效果，对手指旋转、倾斜等多种姿态具有较高鲁棒性。
本发明的另一目的在于提供一种基于上述方法的三维手指静脉识别系统，能实现更好的识别效果，对手指旋转、倾斜等多种姿态具有较高鲁棒性。
本发明的目的通过以下的技术方案实现：
一种三维手指静脉识别方法，包括以下步骤：
S1、三个摄像头在等分角度下从三个角度拍摄手指静脉图像，获得三幅图像；
S2、依据手指轮廓线构建手指三维模型：
将手指剖面图近似视为一个圆S，将三维手指等距离分割成若干个剖面，计算每个剖面的轮廓，用多个不同半径不同位置的圆来对手指近似建模；再将所有轮廓按手指中轴方向串接起来，即可获得近似的手指三维模型；
S3、建立完成三维手指模型之后，接下来将三个摄像头拍摄的二维图像纹理映射到三维手指模型中；三个摄像头拍摄的图像既有两两重叠的部分，又有非重叠的部分，先要确定重叠区域，然后对重叠区域和非重叠区域分别做不同的处理；得到三维手指静脉图像；
S4、对三维手指静脉图像进行特征提取与匹配，最终完成识别。
优选的，将三维手指等距离分割成300个剖面。
优选的，计算每个剖面的轮廓的方法是：
1)根据三个摄像头的投影中心C1，C2，C3建立xOy坐标系：以这三点的垂直平分线交点作 为坐标原点O，以C2C3的中垂线作为y轴建立笛卡尔坐标系xOy；将图像上的像素坐标变换到xOy坐标系中：
坐标系的原点O到i号摄像头的成像平面的距离为L0，设图像上一点坐标为A(x0，y0)，将其变换至xOy坐标系中的成像Ii上，变换后的坐标为A′(x，y)：
Figure PCTCN2017113362-appb-000001
其中θi表示射线OCi与x轴正向的夹角，i＝1，2，3，ym表示手指径向方向图像高度的一半；
这里，只需要转换手指上下边缘点的图像坐标到xOy坐标上，记i号摄像头拍摄的图像的手指上边缘点Pui转换后的坐标为
Figure PCTCN2017113362-appb-000002
下边缘点Pbi转换后的坐标点为
Figure PCTCN2017113362-appb-000003
2)求圆S参数；
每个摄像头的投影中心坐标是已知的，记为Ci(xi，yi)，求出直线
Figure PCTCN2017113362-appb-000004
的和直线
Figure PCTCN2017113362-appb-000005
的方程：
Lui：y＝kuix+bui
Lbi：y＝kbix+bbi
设圆S的圆心坐标为(xc，yc)，半径为rc，那么圆心到各直线的距离为：
Figure PCTCN2017113362-appb-000006
Figure PCTCN2017113362-appb-000007
最优圆方程应该满足下列优化方程：
Figure PCTCN2017113362-appb-000008
通过优化问题求解算法求解出圆参数xc，yc，rc。
进一步的，通过梯度下降法求解算法即可求出圆参数。
进一步的，近似求解出圆参数的方法为：
1)计算Lui与Lbi的角平分线；
这两条直线的角平分线方程为Li：y＝kix+bi，有方程组：
Figure PCTCN2017113362-appb-000009
由先验知识可知：|k1|＞1，k2＞0，k3＜0，这样，对于每一个摄像头对应的方程组都有唯一的解；
2)求圆心坐标及半径
由上一步求得的三条角平分线方程，每两条直线有一个交点，记Li与Lj的交点为Pij(xij，yij)，其中：
xij＝-(bi-bj)/(ki-kj)
yij＝kixij+bi
记Pij到Lui和Luj的距离分别为rij、rji，可以通过点到直线的距离公式计算得到，因此，最终的圆心坐标S(xc，yc)以及半径rc，可以通过以下公式计算得到：
Figure PCTCN2017113362-appb-000010
Figure PCTCN2017113362-appb-000011
Figure PCTCN2017113362-appb-000012
优选的，在计算每个剖面的轮廓之前，需要进行矫正，使矫正后三个摄像头的内参数相同，主视线在同一平面上，且主视线均为投影中心到系统坐标系原点的方向。
进一步的，校正的方法是：
三个摄像头单独定标，使用matlab calibration toolbox，三个摄像头内参数分别为K1，K2，K3；
设定矫正转换公式：
pi′＝Hipi
其中i＝1，2，3，Hi表示从未矫正的图像坐标pi到矫正后的图像坐标pi′的转换矩阵；其中，
Figure PCTCN2017113362-appb-000013
Ri表示i号摄像头坐标系相对于系统坐标系的旋转矩阵，转换前后，摄像头位置保持不变，但转换后的旋转矩阵变换为：
Figure PCTCN2017113362-appb-000014
Ci点坐标为(xi，yi，0)，O为系统坐标系原点；
代入矫正转换公式中即可得到矫正后的图像坐标pi′。
优选的，将三个摄像头拍摄的二维图像纹理映射到三维手指模型中的方法是：
1)重叠区域计算
首先，将三维手指模型坐标转化到图像坐标；以摄像头i为例，对每个圆模型，对应于一个摄像头的成像是竖直的一条线段，即是在图像中它们的纵坐标是相同的，设为v，至于对v的确定，应与建立圆模型时对应的边缘点所处位置纵坐标相同，如此，只关心横坐标u；这就是将二维的坐标转换为一维坐标的情况，于是可以得到如下坐标转换公式：
Figure PCTCN2017113362-appb-000015
其中u就是对应图像上的横坐标，va是比例系数，它的作用是将计算结果纵坐标置为1，θi表示系统坐标系中i号摄像头所在方位与横轴正方向的夹角，L0是i号摄像头成像平面中心到系统坐标系的距离，um是成像中心的横坐标，(x′，y′)是圆上的点，它满足下面的方程：
Figure PCTCN2017113362-appb-000016
设定在圆模型下i号摄像头拍摄的范围为Areai，圆弧Areai的边界点则对应于拍摄的图像中手指的边缘点，所以，也即对应于目标函数max(min)F＝u的圆上坐标点，边界点Ui和Bi；以上述两个公式为约束条件，求出边界点坐标Ui和Bi；这样，每个摄像头拍摄范围的边界便可求得；
接下来是确定摄像头对应的圆弧，这里规定以顺时针方向为正方向，定义区域范围的起 始端和末端，圆弧的确定根据实际系统中摄像头摆放的方位来确定，i号摄像头拍摄的范围应对应于
Figure PCTCN2017113362-appb-000017
计算出每个摄像头拍摄的范围以后，即可得到两两摄像头拍摄的图像的重叠区域
Figure PCTCN2017113362-appb-000018
2)纹理映射
对于只有单个摄像头拍摄的区域，只用该摄像头拍摄的图像做映射，即根据坐标转换公式计算圆弧上的点对应的图像坐标，以该坐标下的像素灰度值作为该圆弧上的点的像素灰度值；
对于两个摄像头重叠的区域
Figure PCTCN2017113362-appb-000019
则利用高斯图像融合的方法确定三维像素点的灰度值，方法是：
对2号摄像头和3号摄像头拍摄的区域，2号摄像头拍摄区域为
Figure PCTCN2017113362-appb-000020
3号摄像头拍摄区域为
Figure PCTCN2017113362-appb-000021
它们有重叠区域为
Figure PCTCN2017113362-appb-000022
对重叠区域上的点(x′，y′)，利用公式计算它分别对应于2号摄像头和3号摄像头下的图像坐标，取两个图像坐标点的灰度值I2(u2，v2)和I3(u3，v3)，最后利用下面公式计算重叠区域上的点(x′，y′)的灰度值：
I(x′，y′)＝ω2I2(u2，v2)+ω3I3(u3，v3)
其中ω是高斯权重，如果重叠部分属于本区域的末端，则权重值按公式g(p)计算，如果属于起始端，则按公式1-g(p)计算，其中：
Figure PCTCN2017113362-appb-000023
3σ＝重叠部分长度；
按类似方法处理任两个摄像头拍摄的图像，解决单个圆模型的纹理映射问题；对三维手指模型中的所有近似圆，按照上述方法分别计算三维坐标点的灰度值，最终得到三维手指静脉图像。
优选的，先将需要匹配的两个三维手指静脉纹理图像进行对齐，并分别展开成二维图像，再用二维图像的特征提取和匹配完成三维手指静脉纹理的匹配。
具体的，展开成二维图像的步骤是：
首先，将三维手指静脉纹理映射到一个给定大小的圆柱体上，相当于将每个圆归一化成一个固定大小的圆，针对于一个圆模型，从水平方向上的第一个点开始，进行等角度采样，共采样N个点，将每一个点对应到圆柱体对应位置圆上的相同角度的点上：
Figure PCTCN2017113362-appb-000024
其中，α＝2π/N，
Figure PCTCN2017113362-appb-000025
N，z＝1，2，...，M是z轴坐标，r是给定圆柱体的半径，rc是三维手指静脉重建中圆模型的半径，(xc，yc)是该圆模型的圆心位置；对所有圆模型进行处理后即可将静脉纹理映射到圆柱体侧面；
按下面的公式1沿一条母线将圆柱体侧面剪开，并展开成一幅M×N大小的二维图像。
Figure PCTCN2017113362-appb-000026
其中β表示母线所在的方位角。
进一步的，对于模板图像对应的圆柱体和待匹配图像对应的圆柱体，将它们展开成二维图像时母线所在的位置若不对应，则展开的两幅图像将存在偏移问题，为确保属于同一类对象的样本展开成二维图像之后能保持对应位置基本不变，提出下面的方法对三维手指静脉图像进行对齐，即确定β的值：
用β＝kβ0，代入上面的公式，其中β0是搜索步长，取π/15，0≤k＜2π/β0，k为整数；在待匹配三维图像展开成的二维图像中截取大小为M0×N0的二维图像
Figure PCTCN2017113362-appb-000027
取M0＝1/3×M，N0＝1/3×N；
在此，规定模板三维手指图像的旋转角度β＝0，也即规定它展开成二维图像的母线位置为基准方位，取三维手指静脉模板图像展开成二维图像后的一个子图像块：
Figure PCTCN2017113362-appb-000028
得到一幅M0×N0的二维图像Itemplate，目的是在待匹配的三维图像中寻找一块与之匹配的图像块，从而实现待匹配三维图像与模板三维图像对齐；
然后计算所有k对应的二维图像与模板块Itemplate的相似程度：
Figure PCTCN2017113362-appb-000029
上式corr指计算两幅图像的相关性，取相关性最大值对应的kβ0作为待匹配样本的旋转角度；
按此方法计算出母线的方向角以后，再按照公式1将待匹配的三维手指静脉图像展开成二维纹理图像，它与模板二维图像是对齐的。
具体的，二维图像的特征提取和匹配步骤是：
采用LBP算子提取静脉纹理的结构特征，然后采用曼哈顿距离计算方法进行特征匹配：
记模板样本的纹理特征向量为F1＝[f1 (1)，f1 (2)，…，fi (m)]，待匹配样本的纹理特征向量为F2＝[f2 (1)，f2 (2)，…，f2 (m)]，这两个纹理特征的曼哈顿距离为：
Figure PCTCN2017113362-appb-000030
在决策阶段，采用单阂值法，如下式：
Figure PCTCN2017113362-appb-000031
当计算的距离超过给定的阂值T时，认为两个样本相似度较小，判断它们不属于同一类对象；否则，认为两个样本的相似度高，判断它们属于同一类对象。
一种基于上述三维手指静脉识别方法的系统，包括上、下位机两部分；下位机包括光源控制模块和图像采集模块，光源控制模块包括依次相连的微处理器、驱动电路和近红外光源，图像采集模块包括采集图像的摄像头；上位机为装有实现手指静脉三维重建与识别的软件，上位机与光源控制模块的微处理器实现光源亮度控制信号的传递，与摄像头进行视频数据的传输；
下位机负责近红外光源强度的控制和图像信息的采集传输，上位机负责对采集到的数字图像进行处理、三维重建、特征提取与识别。
优选的，上位机与光源控制模块的微处理器通过串口实现光源亮度控制信号的传递，光源控制模块采用微处理器控制近红外光源的亮度，实现三个侧面的光源轮流照射手指。
优选的，上位机与摄像头通过USB方式进行视频数据的传输，摄像头采用近红外USB 摄像头。
优选的，三个摄像头安装在等边三角形的三个顶点位置上，三个摄像头的主视线位于同一平面上且两两相交120°，在每个摄像头正对面设置了相应的LED近红外灯组。
优选的，下位机通过定时器产生PWM信号来控制近红外LED亮度，用串口中断接收上位机发送的控制命令。
本发明与现有技术相比，具有如下优点和有益效果：
本发明采用三个摄像头在等分角度下全方位拍摄手指静脉图像，然后进行三维手指静脉的重建以及三维手指静脉图像特征提取与匹配，采用的三个摄像头拍摄的手指静脉图像比单纯的一个摄像头拍摄手指内侧表皮下的手指静脉图像所包含的特征信息更多，而更多的信息意味着更准确的判断；本发明提出的三维手指静脉识别方法比传统的二维手指静脉识别而言，能够获得更多的静脉纹理特征，得到更好的识别效果，同时还可以有效的克服手指姿态变化带来的识别性能下降问题。
附图说明
图1是实施例1中三维手指静脉识别系统的图像采集装置图。
图2是实施例1中三维手指静脉识别系统的硬件结构图。
图3是实施例1中三维手指静脉识别系统的软件流程图。
图4是针孔模型原理示意图。
图5是实施例2中近似成像模型示意图。
图6是实施例2中圆近似法示意图。
图7是实施例2中系统坐标系建立示意图。
图8是实施例2中边界标定物示意图。
图9是实施例2中单个摄像头拍摄范围示意图。
图10是实施例2中纹理映射示意图。
图11是实施例2中三维手指静脉特征提取与特征匹配过程示意图。
图12是实施例2中三维手指静脉图。
图13是LBP下2D与3D方法等误率曲线。
图14是HOG下2D与3D方法等误率曲线。
具体实施方式
下面结合实施例及附图对本发明作进一步详细的描述，但本发明的实施方式不限于此。
实施例1
一种三维手指静脉识别系统，如图1所示，它由硬件和软件两部分组成，其中硬件部分包括近红外图像采集设备，软件部分包括PC端手指静脉三维重建与识别算法；其中，硬件设备主要负责近红外光源强度的控制和图像信息的传输，软件部分负责对采集到的数字图像进行处理、三维重建、特征提取与识别。与现有的二维手指静脉识别系统不同，本系统使用了三个摄像头，以获得手指表皮下全部静脉图像，并通过提取出三维手指静脉纹理特征进行匹配。
1.1硬件设计
系统硬件包括上、下位机两部分。下位机由光源控制模块和图像采集模块两个模块组成，前者采用微控制器控制近红外光源的亮度，实现三个侧面的光源轮流照射手指，后者采用850nm近红外USB摄像头(型号为KS302D)采集图像。上位机为PC端，PC端与微控制器通过串口实现光源亮度控制信号的传递，与摄像头通过USB方式进行视频数据的传输。其中，三维手指静脉图像采集装置结构示意图如图1所示，三个摄像头安装在等边三角形的三个顶点位置上，并依次编号。三个摄像头的主视线位于同一平面上且两两相交120°。静脉造影成像采用近红外光透射方式，在每个摄像头正对面设置了相应的LED近红外灯组，系统的硬件结构框图如图2所示。
1.2软件设计
系统软件部分是系统的核心，主要负责系统流程控制，图像处理、三维重建和匹配，其 中下位机采用Keil开发单片机控制程序，上位机则采用VS2013+OpenCV+OpneGL开发集图像处理、三维重建，以及特征提取与匹配于一体的软件。下位机通过定时器产生PWM信号来控制近红外LED亮度，用串口中断接收上位机发送的控制命令；上位机首先是采集图像和控制下位机进行光源亮度的调节，然后对原始图像进行裁剪、去噪、轮廓提取等一系列的预处理操作；接着采用提出的三维重建算法构建手指三维模型，并将静脉纹理映射到三维手指模型上；最后根据特征提取方法实现匹配的功能。系统软件流程图如图3所示。
实施例2
通过实施例1中的设备，采用三个摄像头在等分角度下全方位拍摄手指静脉图像，然后进行三维手指静脉的重建以及三维手指静脉图像特征提取与匹配。
1.三维手指静脉的重建
人的手指在伸直的状态下，手指轴向的垂直剖面图近似于一个离心率接近于1的椭圆，但相对椭圆而言，圆的数学方程参数更少，比较容易计算出来，因此将手指剖面图近似视为一个圆；此时，可以用多个不同半径不同位置的圆来对手指近似建模。
由于人手指表面是比较光滑的，在每个摄像头拍摄的图像中，手指轮廓线的变化都是比较平缓的，因此在利用边缘点坐标计算圆参数时，即使三个视角的对应点存在一些偏差，对计算的结果影响也并不大。下面将详细阐述手指三维模型建立和静脉纹理映射的方法。
1.1三维手指模型的建立
1.1.1理想状态下的数学模型
首先，由于本系统采用的摄像头是几乎无畸变的，其成像模型可以用针孔模型近似，如图4所示。结合本系统的成像结构特点，三个摄像头选用同一型号同一参数的CMOS摄像头，手指放置时手指中轴位于三个摄像头的中心位置，并且手指中轴与三个摄像头投影中心所成的平面近似垂直。因此，提出依据手指轮廓线构建手指三维模型的方法，可以不考虑摄像头在手指轴向上采样的像素点不均匀的问题，而近似用如图5所示的立体结构来近似表达。这样，对于每一个剖面，用圆近似的几何模型就如图6所示，通过以下两步骤求解圆的大小和位置。
步骤一：将图像上的像素坐标变换到如图6所示的xOy坐标系中。
首先，三个摄像头的投影中心C1，C2，C3的相对位置是已知的，以这三点的垂直平分线交点作为坐标原点O，以C2C3的中垂线作为y轴建立笛卡尔坐标系xOy，坐标系的原点O到i号摄像头的成像平面的距离为L0(三个摄像头对应的L0是相同的)，设图像上一点坐标为A(x0，y0)，将其变换至xOy坐标系中的成像Ii上，变换后的坐标为A′(x，y)：
Figure PCTCN2017113362-appb-000032
其中θi表示射线OCi与x轴正向的夹角，i＝1，2，3，下同，ym表示手指径向方向图像高度的一半。
这里，只需要转换手指轮廓边缘点的图像坐标到xOy坐标上，记i号摄像头拍摄的图像的手指上边缘点Pui转换后的坐标为
Figure PCTCN2017113362-appb-000033
下边缘点Pbi转换后的坐标点为
Figure PCTCN2017113362-appb-000034
步骤二：求圆参数。
由于每个摄像头的投影中心坐标是已知的，记为Ci(xi，yi)，那么直线
Figure PCTCN2017113362-appb-000035
的和直线
Figure PCTCN2017113362-appb-000036
的方程是容易求出的，为方便叙述，统一使用斜截式，实际计算中，对1号摄像头的情况需要讨论斜率的存在性，此处不作过多叙述。
Lui：y＝kuix+bui  (2)
Lbi：y＝kbix+bbi  (3)
设圆S的圆心坐标为(xc，yc)，半径为rc，那么圆心到各直线的距离为：
Figure PCTCN2017113362-appb-000037
Figure PCTCN2017113362-appb-000038
最优圆方程应该满足下列优化方程：
Figure PCTCN2017113362-appb-000039
这个优化问题的意义在于：使求取的圆尽可能与直线Lui，Lbi相切。计算时，由于各直线方程已求出，即kui、bui是已知量，将距离公式(4)(5)代入优化公式(6)，消去dui和dbi，得到 F实际上是关于xc，yc，rc的方程，通过梯度下降法或其它优化问题求解算法即可求出圆参数，但是，这种方法计算量非常大，且未必能求出全局最优解，退而求其次，接下来给出另外一种较为简单的方法，该方法求得的结果不一定是最优解，但已满足需要。注意到，最理想情况下，Lui与Lbi的角平分线就是过圆心的直线，于是可以通过计算角平分线交点得到圆心坐标，进而计算出圆半径，具体计算方法如下：
(1)计算角平分线
这两条直线的角平分线方程为Li：y＝kix+bi，有方程组：
Figure PCTCN2017113362-appb-000040
另外，由先验知识可知：|k1|＞1，k2＞0，k3＜0，这样，对于每一个摄像头对应的方程组都有唯一的解。
(2)求圆心坐标及半径
由上一步求得的三条角平分线方程，每两条直线有一个交点，记Li与Lj的交点为Pij(xij，yij)，其中：
xij＝-(bi-bj)/(ki-kj)  (8)
yij＝kixij+bi  (9)
记Pij到Lui和Luj的距离分别为rij、rji，可以通过点到直线的距离公式计算得到，因此，最终的圆心坐标S(xc，yc)以及半径rc，可以通过以下公式(10)-(12)计算得到：
Figure PCTCN2017113362-appb-000041
Figure PCTCN2017113362-appb-000042
Figure PCTCN2017113362-appb-000043
至此，已完成用圆来近似表示一个手指轴向剖面的轮廓，将三维手指等距离分割成若干个(文中是300个)剖面，计算每个剖面的轮廓，再将所有轮廓按手指中轴方向串接起来，即可获得近似的手指三维模型。但是，由于受到摄像头制造工艺上的差异、硬件安装时缺乏超 精确的定位方法等因素的影响，硬件结构往往无法保证与理想模型保持一致，所以一般需要对系统做一些矫正处理。
1.1.2一般状态到理想状态的矫正方法
矫正的目的是使矫正后三个摄像头的内参数相同，主视线在同一平面上，且主视线均为投影中心到系统坐标系原点的方向。此处提出的矫正算法由下面三个步骤组成：
(1)摄像机标定
三个摄像头单独定标，使用matlab calibration toolbox，三个摄像头内参数分别为K1，K2，K3；
摄像头两两标定，得到相对位置参数Rij，tij，其中i＝1，2；j＝2，3；i≠j，Rij表示j号摄像头摄影中心相对于i号摄像头摄影中心的旋转矩阵，tij表示j号摄像头相对于i号摄像头的偏移矩阵。其中
Figure PCTCN2017113362-appb-000044
θ为绕z轴的旋转角度，ψ为绕y轴的旋转角度，φ为绕轴的旋转角度，xyz轴指的是系统坐标系的坐标轴。
(2)系统坐标系建立
理论上向量t12，t13，t23是共面的，但通过摄像头标定计算后的结果极大可能不是共面的，用下面的方法来建立三个摄影中心的位置关系：
如图7，先以1号摄像头坐标系C1x′y′z′为基准。由t12、t13确定A′、B′，于是可以得到它们的中点D，平移向量t23使其中点过D点，于是有平移后的向量t23的起点A″和终点B″，记A′A″的中点为C2，B′B″的中点为C3。最后，以三个摄像头投影中心所组成的平面C1C2C3为xy平面，以C1C2C3的角平分线交点为坐标原点O，以C2C3的角平分线为x轴建立坐标系，如图7所示，从C1x′y′z′到坐标系Oxyz的平移矩阵ts和旋转矩阵Rs是容易求得的，即可建立系统坐标系。
(3)矫正
设定矫正转换公式：
pi′＝Hipi  (14)
其中i＝1，2，3，Hi表示从未矫正的图像坐标pi到矫正后的图像坐标pi′的转换矩阵；
假定空间坐标为P，根据投影公式有，矫正前：
Figure PCTCN2017113362-appb-000045
其中，Ri表示i号摄像头坐标系相对于系统坐标系的旋转矩阵，转换前后，摄像头位置保持不变，但转换后的旋转矩阵变换为：
Figure PCTCN2017113362-appb-000046
其中，Ci点坐标为(xi，yi)，O为系统坐标系原点，于是，转换后有投影公式：
Figure PCTCN2017113362-appb-000047
有解：
Figure PCTCN2017113362-appb-000048
至此，系统模型已与理想模型等价，便可以运用理想模型下的三维重建算法构建手指的三维模型了。需要指出的是，对于边缘点对应位置的确定，考虑到在一般情况下，成年人的手指粗细程度相差不会非常大，所以，采用一个跟普通人的手指粗细程度相当的圆柱体作为标定物，如图8所示，这个标定物标记了几个圆，以圆在三个摄像头中的成像位置标记三个视图对应点的位置。
2.2静脉纹理映射
建立完成三维手指模型之后，接下来就是将三个摄像头拍摄的二维图像纹理映射到三维手指模型中，由图6可以发现，三个摄像头拍摄的图像既有两两重叠的部分，又有非重叠的部分，而对这两种情况处理方法是有差异的。所以，先要确定重叠区域，然后对重叠区域和非重叠部分分别做不同的处理。从部分到整体的思维，只需研究单个圆模型下的纹理映射方法，即可推广到所有圆模型，完成这个三维手指的纹理映射。
(1)重叠区域计算
首先，将系统坐标转化到图像坐标。如图9所示，以3号摄像头为例，对每个圆模型，对应于一个摄像头的成像是竖直的一条线段，即是在图像中它们的纵坐标是相同的，设为v，至于对v的确定，应与建立圆模型时对应的边缘点所处位置纵坐标相同，如此，只关心横坐 标u。这就是将二维的坐标转换为一维坐标的情况，于是可以得到如下公式(18)：
Figure PCTCN2017113362-appb-000049
其中u就是对应图像上的横坐标，va是比例系数，它的作用是将计算结果纵坐标置为1，θi表示系统坐标系中i号摄像头所在方位与横轴正方向的夹角，L0是i号摄像头成像平面中心到系统坐标系的距离，um是成像中心的横坐标，(x′，y′)是圆上的点，它满足下面的方程(19)：
Figure PCTCN2017113362-appb-000050
设定在圆模型下i号摄像头拍摄的范围为Areai，其边界点(即圆弧Areai的边界点)则对应于拍摄的图像中手指的边缘点，所以，也即对应于目标函数max(min)F＝u的圆上坐标点(边界点Ui和Bi)。以公式(18)和公式(19)为约束条件，利用梯度下降法是容易求出边界点坐标Ui和Bi的。也可以直接根据直线与圆相切的方法求出这两个坐标点。这样，每个摄像头拍摄范围的边界便可求得。
接下来是确定摄像头对应的圆弧，这里规定以顺时针方向为正方向，定义区域范围的起始端和末端，圆弧的确定根据实际系统中摄像头摆放的方位来确定，例如在图9中3号摄像头拍摄的范围应对应于
Figure PCTCN2017113362-appb-000051
计算出每个摄像头拍摄的范围以后，两两摄像头拍摄的图像的重叠区域就非常明显了，例如：假设2号摄像头(其位置参考图6)对应的范围为
Figure PCTCN2017113362-appb-000052
那么，2号摄像头和3号摄像头拍摄的重叠区域就对应于
Figure PCTCN2017113362-appb-000053
(2)纹理映射
接下来讨论如何将二维图像的纹理映射到三维手指模型中，同样的，只需讨论一个圆模型下的处理方法。
如图10所示，仍以2号摄像头和3号摄像头拍摄的区域为例，2号摄像头拍摄区域为
Figure PCTCN2017113362-appb-000054
3号摄像头拍摄区域为
Figure PCTCN2017113362-appb-000055
它们有重叠区域为
Figure PCTCN2017113362-appb-000056
对于只有单个摄像头拍摄的区域(如
Figure PCTCN2017113362-appb-000057
属于只有2号摄像头拍摄的区域)，只用该摄像头拍摄的图像做映射，即根据公式(18)计算圆弧上的点对应的图像坐标，以该坐标下的像素灰度值作为该圆弧上的点的像素灰度值； 对于两个摄像头重叠的区域
Figure PCTCN2017113362-appb-000058
则利用高斯图像融合的方法确定三维像素点的灰度值，方法是：对重叠区域上的点
Figure PCTCN2017113362-appb-000059
利用公式(18)计算它分别对应于2号摄像头和3号摄像头下的图像坐标，取两个图像坐标点的灰度值I2(u2，v2)和I3(u3，v3)，最后利用下面公式(20)计算重叠区域上的点(x′，y′)的灰度值：
I(x′，y′)＝ω2I2(u2，v2)+ω3I3(u3，v3)  (20)
其中ω是高斯权重，如果重叠部分属于本区域的末端，则权重值按公式g(p)计算，如果属于起始端，则按公式1-g(p)计算，其中：
Figure PCTCN2017113362-appb-000060
3σ＝重叠部分长度。按类似方法处理任两个摄像头拍摄的图像，解决单个圆模型的纹理映射问题。对三维手指模型中的所有近似圆，按照上述方法分别计算三维坐标点的灰度值，最终得到三维手指静脉图像。
3.三维手指静脉图像特征提取与匹配
基于三维的生物特征识别，在人脸和指纹识别领域中有不少的研究。三维图像的特征提取可以用以下两种思路进行，一种思路是在三维坐标下提取特征点，构成特征向量，如有文献将指纹识别特征提取算法中将二维细节点概念扩展到三维，并通过计算两个样本中细节点问的距离得到两个样本匹配分数；另一种思路，将三维图像再经过特殊变换转成二维图像，然后用已有的二维图像特征提取方法进行特征提取，如有文献将三维指纹图像映射到一个圆柱体上，再展开成二维图像后进行下一步处理。在获取特征向量以后，对样本匹配的方法是定义一种特征描述子之间的距离度量函数计算样本间的距离值，距离计算方法有：曼哈顿距离、欧式距离、马氏距离等等，然后用单阂值法、最近邻法等给出最终匹配决策。
如图11所示，为实施例提出的三维手指静脉特征提取与匹配的方法流程图。首先对采集的模板样本的三张二维图像和待匹配样本的三张二维图像按上述的三维手指静脉重建方法生成手指静脉的三维图像。由于对手指放置并未作过多约束，因此，一般而言两个样本之间可能存在一定的绕轴的旋转偏差。针对这个问题，本实施例采用的方法是：先将需要匹配的两个三维手指静脉纹理图像进行对齐，并分别展开成二维图像，再用二维图像的特征提取和匹配完成三维手指静脉纹理的匹配。具体描述如下。
3.1特征提取
众所周知，圆柱体沿一条母线剪开之后能展开成矩形，人体手指三维结构近似于圆柱体模型，如果要将三维手指表皮静脉展开成一幅二维图像，可以先将三维手指静脉映射到一个圆柱体上，然后再展开成二维图像。
首先，将三维手指静脉纹理映射到一个给定大小的圆柱体上，相当于将每个圆归一化成一个固定大小的圆，针对于一个圆模型，从水平方向上的第一个点开始，进行等角度采样，共采样N个点，将每一个点对应到圆柱体对应位置圆上的相同角度的点上：
Figure PCTCN2017113362-appb-000061
其中，α＝2π/N，
Figure PCTCN2017113362-appb-000062
z＝1，2，...，M是z轴坐标，r是给定圆柱体的半径，rc是三维手指静脉重建中圆模型的半径，(xc，yc)是该圆模型的圆心位置。对所有圆模型进行处理后即可将静脉纹理映射到圆柱体侧面。紧接着，按公式(23)沿一条母线将圆柱体侧面剪开，并展开成一幅M×N大小的二维图像。
Figure PCTCN2017113362-appb-000063
其中β表示母线所在的方位角，其它符号的意义与公式(22)相同。对于模板图像对应的圆柱体和待匹配图像对应的圆柱体，将它们展开成二维图像时母线所在的位置若不对应，则展开的两幅图像将存在偏移问题，为确保属于同一类对象的样本展开成二维图像之后能保持对应位置基本不变，提出下面的方法对三维手指静脉图像进行对齐(即确定β的值)：
在此，规定模板三维手指图像的旋转角度β＝0，也即规定它展开成二维图像的母线位置为基准方位，取三维手指静脉模板图像展开成二维图像后的一个子图像块：
Figure PCTCN2017113362-appb-000064
得到一幅M0×N0的二维图像Itemplate，取M0＝1/3×M，N0＝1/3×N，目的是在待匹配的三维图像中寻找一块与之匹配的图像块，从而实现待匹配三维图像与模板三维图像对齐。用β＝kβ0(其中β0是搜索步长，取π/15，0≤k＜2π/β0，k为整数)代入公式(23)，同样在待匹配三维图像展开成的二维图像中截取大小为M0×N0的二维图像
Figure PCTCN2017113362-appb-000065
然后计算所有k对应的二维图像与模板块Itemolate的相似程度：
Figure PCTCN2017113362-appb-000066
上式corr指计算两幅图像的相关性，取相关性最大值对应的kβ0作为待匹配样本的旋转角度。按此方法计算出母线的方向角以后，再按照公式(23)将待匹配的三维手指静脉图像展开成二维纹理图像，它与模板二维图像是对齐的。
将三维手指静脉图像展开成二维图像之后，即可利用二维图像的特征提取算法提取特征。由于光照变化、手指形变、噪声干扰等，从静脉图像获取到的细节特征并非很稳定，且细节点特征对手指偏移的鲁棒性不高。相对而言静脉纹路的结构特征，对于噪声干扰、光照变化、手指形变等影响因素具有较强的鲁棒性，因此采用简单高效的LBP算子[1]提取静脉纹理的结构特征。
3.2特征匹配
在模式识别领域，计算两个样本的距离度量方式有很多种，这里采用计算复杂度最小的曼哈顿距离计算方法，也是最简单的一种模板匹配法。记模板样本的纹理特征向量为F1＝[f1 (1)，f1 (2)，…，f1 (m)]，待匹配样本的纹理特征向量为F2＝[f2 (1)，f2 (2)，…，f2 (m)]，这两个纹理特征的曼哈顿距离为：
Figure PCTCN2017113362-appb-000067
在决策阶段，采用单阂值法，当计算的距离超过给定的阂值T时，认为两个样本相似度较小，判断它们不属于同一类对象，否则，认为两个样本的相似度高，判断它们属于同一类对象。
Figure PCTCN2017113362-appb-000068
采用实施例1提出三维手指静脉成像系统采集全方位的手指表皮静脉图像，以实施例2提出的圆近似法构建手指三维模型，并将二维的纹理映射到三维手指模型上，得到的三维手指图像如图12所示，图中每一行是同一根手指在不同视角下的观察图，这里共展示三根不同手指的三维手指静脉重建效果。可以看到，实施例2方法重建出的三维手指静脉模型逼真程度很高，包括手指的大小、弯曲程度等都与真实情况相符，在静脉纹理方面也基本还原了人体手指表皮静脉分布情况。
征集了11名志愿者采集手指静脉图像数据，为保证采集方式的友好性，不对志愿者做过多要求，志愿者在采集数据时可采用比较自然的状态将手指置于装置上。对每个志愿者采集 了左右手食指和中指共四根手指的数据，每根手指采集图像四次，总共有44根手指的三个视角的图像，作为实验的数据。然后，采用本实施例提出的方法，进行三维手指静脉图像的特征提取和匹配试验。为了说明本方法的有效性，将其与二维静脉图像识别结果进行比较。为了保证比较的公平性和有效性，选择对1号摄像头采集的二维静脉图像进行二维静脉图像识别实验，因1号摄像头位于手指内侧表皮下，这与传统二维手指静脉成像系统中摄像头的放置位置一致，实验结果如图13所示。
图13采用的是LBP算子提取纹理特征，从图中可以发现，使用三维手指静脉纹理特征的识别效果要比传统的使用二维手指静脉纹理特征的效果要好。因为实施例1中的系统采用的三个摄像头拍摄的手指静脉图像比单纯的一个摄像头拍摄手指内侧表皮下的手指静脉图像所包含的特征信息更多，而更多的信息意味着更准确的判断。另外，还采用HOG算子提取纹理特征进行实验，结果如图14所示，可以发现，仍然可以得出相同的结论，用本实施例提出的三维的方法可以获得更好的匹配结果，进一步验证了本方法的有效性。
在等误率方面，如表1所示，使用三维的方案进行匹配的等误率达到2.273(使用HOG算子时等误率达到1.515％)，其结果比现有的采用二维手指静脉图像进行匹配的等误率4.167％(使用HOG算子时等误率为3.788％)具有明显的提升效果。
表1 不同方法的等误率
LBP HOG
2D纹理 4.167 3.788
3D纹理 2.273 1.515
综上，实施例1和实施例2提出了一种全新的三维手指静脉识别系统，采用了三个摄像头拍摄全方位手指表皮下的静脉图像，根据硬件结构特点提出了一种用圆拟合手指剖面线的方法构建手指三维模型，同时将三个视角下拍摄的二维手指静脉映射到三维手指模型中，提出了一种三维手指静脉纹理特征提取方法，完整实现了一个基于三维手指静脉数据的身份认证系统，实验结果表明实施例提出的三维手指静脉识别系统比传统的二维手指静脉识别系统而言，能够获得更多的静脉纹理特征，得到更好的识别效果，同时还可以有效的克服手指姿态变化带来的识别性能下降问题。
参考文献：
[1]Ojala T，
Figure PCTCN2017113362-appb-000069
M，Harwood D.A comparative study of texture measures with classification based on featured distributions[J].Pattern recognition，1996，29(1)：51-59.
上述实施例为本发明较佳的实施方式，但本发明的实施方式并不受上述实施例的限制，其他的任何未背离本发明的精神实质与原理下所作的改变、修饰、替代、组合、简化，均应为等效的置换方式，都包含在本发明的保护范围之内。

Claims (10)

  1. 一种三维手指静脉识别方法，其特征在于，包括以下步骤：
    S1、三个摄像头在等分角度下从三个角度拍摄手指静脉图像，获得三幅图像；
    S2、依据手指轮廓线构建手指三维模型：
    将手指剖面图近似视为一个圆S，将三维手指等距离分割成若干个剖面，计算每个剖面的轮廓，用多个不同半径不同位置的圆来对手指近似建模；再将所有轮廓按手指中轴方向串接起来，即可获得近似的手指三维模型；
    S3、建立完成三维手指模型之后，接下来将三个摄像头拍摄的二维图像纹理映射到三维手指模型中；三个摄像头拍摄的图像既有两两重叠的部分，又有非重叠的部分，先要确定重叠区域，然后对重叠区域和非重叠区域分别做不同的处理；得到三维手指静脉图像；
    S4、对三维手指静脉图像进行特征提取与匹配，最终完成识别。
  2. 根据权利要求1所述的三维手指静脉识别方法，其特征在于，计算每个剖面的轮廓的方法是：
    1)根据三个摄像头的投影中心C1，C2，C3建立xOy坐标系：以这三点的垂直平分线交点作为坐标原点O，以C2C3的中垂线作为y轴建立笛卡尔坐标系xOy；将图像上的像素坐标变换到xOy坐标系中：
    坐标系的原点O到i号摄像头的成像平面的距离为L0，设图像上一点坐标为A(x0，y0)，将其变换至xOy坐标系中的成像Ii上，变换后的坐标为A′(x，y)：
    Figure PCTCN2017113362-appb-100001
    其中θi表示射线OCi与x轴正向的夹角，i＝1，2，3，ym表示手指径向方向图像高度的一半；
    这里，只需要转换手指上下边缘点的图像坐标到xOy坐标上，记i号摄像头拍摄的图像的手指上边缘点Pui转换后的坐标为
    Figure PCTCN2017113362-appb-100002
    下边缘点Pbi转换后的坐标点为
    Figure PCTCN2017113362-appb-100003
    2)求圆S参数；
    每个摄像头的投影中心坐标是已知的，记为Ci(xi，yi)，求出直线
    Figure PCTCN2017113362-appb-100004
    的和直线
    Figure PCTCN2017113362-appb-100005
    的方程：
    Lui：y＝kuix+bui
    Lbi：y＝kbix+bbi
    设圆S的圆心坐标为(xc，yc)，半径为rc，那么圆心到各直线的距离为：
    Figure PCTCN2017113362-appb-100006
    Figure PCTCN2017113362-appb-100007
    最优圆方程应该满足下列优化方程：
    Figure PCTCN2017113362-appb-100008
    通过优化问题求解算法或者通过梯度下降法求解算法求解出圆参数xc，yc，rc。
  3. 根据权利要求2所述的三维手指静脉识别方法，其特征在于，可采用近似方法求解出圆参数：
    1)计算Lui与Lbi的角平分线；
    这两条直线的角平分线方程为Li：y＝kix+bi，有方程组：
    Figure PCTCN2017113362-appb-100009
    由先验知识可知：|k1|＞1，k2＞0，k3＜0，这样，对于每一个摄像头对应的方程组都有唯一的解；
    2)求圆心坐标及半径
    由上一步求得的三条角平分线方程，每两条直线有一个交点，记Li与Lj的交点为Pij(xij，yij)，其中：
    xij＝-(bi-bj)/(ki-kj)
    yij＝kixij+bi
    记Pij到Lui和Luj的距离分别为rij、rji，可以通过点到直线的距离公式计算得到，因此，最终的圆心坐标S(xc，yc)以及半径rc，可以通过以下公式计算得到：
    Figure PCTCN2017113362-appb-100010
    Figure PCTCN2017113362-appb-100011
    Figure PCTCN2017113362-appb-100012
  4. 根据权利要求1所述的三维手指静脉识别方法，其特征在于，在计算每个剖面的轮廓之前，需要进行矫正，使矫正后三个摄像头的内参数相同，主视线在同一平面上，且主视线均为投影中心到系统坐标系原点的方向。
  5. 根据权利要求4所述的三维手指静脉识别方法，其特征在于，校正的方法是：
    三个摄像头单独定标，使用matlab calibration toolbox，三个摄像头内参数分别为K1，K2，K3；
    设定矫正转换公式：
    pi′＝Hipi
    其中i＝1，2，3，Hi表示从未矫正的图像坐标pi到矫正后的图像坐标pi′的转换矩阵；其中，
    Figure PCTCN2017113362-appb-100013
    Ri表示i号摄像头坐标系相对于系统坐标系的旋转矩阵，转换前后，摄像头位置保持不变，但转换后的旋转矩阵变换为：
    Figure PCTCN2017113362-appb-100014
    Ci点坐标为(xi，yi)，O为系统坐标系原点；
    代入矫正转换公式中即可得到矫正后的图像坐标pi′。
  6. 根据权利要求1所述的三维手指静脉识别方法，其特征在于，将三个摄像头拍摄的二维图像纹理映射到三维手指模型中的方法是：
    1)重叠区域计算
    首先，将三维手指模型坐标转化到图像坐标；以摄像头i为例，对每个圆模型，对应于一个摄像头的成像是竖直的一条线段，即是在图像中它们的纵坐标是相同的，设为v，至于对v的确定，应与建立圆模型时对应的边缘点所处位置纵坐标相同，如此，只关心横坐标u；这就是将二维的坐标转换为一维坐标的情况，于是可以得到如下坐标转换公式：
    Figure PCTCN2017113362-appb-100015
    其中u就是对应图像上的横坐标，va是比例系数，它的作用是将计算结果纵坐标置为1，θi表示系统坐标系中i号摄像头所在方位与横轴正方向的夹角，L0是i号摄像头成像平面中心到系统坐标系的距离，um是成像中心的横坐标，(x′，y′)是圆上的点，它满足下面的方程：
    Figure PCTCN2017113362-appb-100016
    设定在圆模型下i号摄像头拍摄的范围为Areai，圆弧Areai的边界点则对应于拍摄的图像中手指的边缘点，所以，也即对应于目标函数max(min)F＝u的圆上坐标点，边界点Ui和Bi；以上述两个公式为约束条件，求出边界点坐标Ui和Bi；这样，每个摄像头拍摄范围的边界便可求得；
    接下来是确定摄像头对应的圆弧，这里规定以顺时针方向为正方向，定义区域范围的起始端和末端，圆弧的确定根据实际系统中摄像头摆放的方位来确定，i号摄像头拍摄的范围应对应于
    Figure PCTCN2017113362-appb-100017
    计算出每个摄像头拍摄的范围以后，即可得到两两摄像头拍摄的图像的重叠区域
    Figure PCTCN2017113362-appb-100018
    2)纹理映射
    对于只有单个摄像头拍摄的区域，只用该摄像头拍摄的图像做映射，即根据坐标转换公式计算圆弧上的点对应的图像坐标，以该坐标下的像素灰度值作为该圆弧上的点的像素灰度值；
    对于两个摄像头重叠的区域
    Figure PCTCN2017113362-appb-100019
    则利用高斯图像融合的方法确定三维像素点的灰度值，方法是：
    对2号摄像头和3号摄像头拍摄的区域，2号摄像头拍摄区域为
    Figure PCTCN2017113362-appb-100020
    3号摄像头拍摄区域为
    Figure PCTCN2017113362-appb-100021
    它们有重叠区域为
    Figure PCTCN2017113362-appb-100022
    对重叠区域上的点(x′，y′)，利用公式计算它分别对应于2号摄像头和3号摄像头下的图像坐标，取两个图像坐标点的灰度值I2(u2，v2)和I3(u3，v3)，最后利用下面公式计算重叠区域上的点(x′，y′)的灰度值：
    I(x′，y′)＝ω2I2(u2，v2)+ω3I3(u3，v3)
    其中ω是高斯权重，如果重叠部分属于本区域的末端，则权重值按公式g(p)计算，如果属于起始端，则按公式1-g(p)计算，其中：
    Figure PCTCN2017113362-appb-100023
    3σ＝重叠部分长度；
    按类似方法处理任两个摄像头拍摄的图像，解决单个圆模型的纹理映射问题；对三维手指模型中的所有近似圆，按照上述方法分别计算三维坐标点的灰度值，最终得到三维手指静脉图像。
  7. 根据权利要求1所述的三维手指静脉识别方法，其特征在于，先将需要匹配的两个三维手指静脉纹理图像进行对齐，并分别展开成二维图像，再用二维图像的特征提取和匹配完成三维手指静脉纹理的匹配。
  8. 根据权利要求7所述的三维手指静脉识别方法，其特征在于，展开成二维图像的步骤是：
    首先，将三维手指静脉纹理映射到一个给定大小的圆柱体上，相当于将每个圆归一化成一个固定大小的圆，针对于一个圆模型，从水平方向上的第一个点开始，进行等角度采样，共采样N个点，将每一个点对应到圆柱体对应位置圆上的相同角度的点上：
    Figure PCTCN2017113362-appb-100024
    其中，α＝2π/N，
    Figure PCTCN2017113362-appb-100025
    z＝1，2，...，M是z轴坐标，r是给定圆柱体的半径，rc是三维手指静脉重建中圆模型的半径，(xc，yc)是该圆模型的圆心位置；对所有圆模型进行处理后即可将静脉纹理映射到圆柱体侧面；
    按下面的公式1沿一条母线将圆柱体侧面剪开，并展开成一幅M×N大小的二维图像。
    Figure PCTCN2017113362-appb-100026
    其中β表示母线所在的方位角。
  9. 根据权利要求8所述的三维手指静脉识别方法，其特征在于，对于模板图像对应的圆柱体和待匹配图像对应的圆柱体，将它们展开成二维图像时母线所在的位置若不对应，则展开的两幅图像将存在偏移问题，为确保属于同一类对象的样本展开成二维图像之后能保持对应位置基本不变，提出下面的方法对三维手指静脉图像进行对齐，即确定β的值：
    用β＝kβ0，代入上面的公式，其中β0是搜索步长，取π/15，0≤k＜2π/β0，k为整数；在待匹配三维图像展开成的二维图像中截取大小为M0×N0的二维图像
    Figure PCTCN2017113362-appb-100027
    取M0＝1/3×M，N0＝1/3×N；
    在此，规定模板三维手指图像的旋转角度β＝0，也即规定它展开成二维图像的母线位置为基准方位，取三维手指静脉模板图像展开成二维图像后的一个子图像块：
    Figure PCTCN2017113362-appb-100028
    Figure PCTCN2017113362-appb-100029
    得到一幅M0×N0的二维图像Itemplate，目的是在待匹配的三维图像中寻找一块与之匹配的图像块，从而实现待匹配三维图像与模板三维图像对齐；
    然后计算所有k对应的二维图像与模板块Itemplate的相似程度：
    Figure PCTCN2017113362-appb-100030
    上式corr指计算两幅图像的相关性，取相关性最大值对应的kβ0作为待匹配样本的旋转角度；
    按此方法计算出母线的方向角以后，再按照公式1将待匹配的三维手指静脉图像展开成二维纹理图像，它与模板二维图像是对齐的。
  10. 一种基于权利要求1所述的方法的三维手指静脉识别系统，其特征在于，包括上、下位机两部分；下位机包括光源控制模块和图像采集模块，光源控制模块包括依次相连的微处理器、驱动电路和近红外光源，图像采集模块包括采集图像的摄像头；上位机为装有实现 手指静脉三维重建与识别的软件，上位机与光源控制模块的微处理器实现光源亮度控制信号的传递，与摄像头进行视频数据的传输；
    下位机负责近红外光源强度的控制和图像信息的采集传输，上位机负责对采集到的数字图像进行处理、三维重建、特征提取与识别。
PCT/CN2017/113362 2017-04-26 2017-11-28 一种三维手指静脉识别方法及系统 Ceased WO2018196371A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/478,959 US11017205B2 (en) 2017-04-26 2017-11-28 Three-dimensional finger vein recognition method and system

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710281130.0A CN106919941B (zh) 2017-04-26 2017-04-26 一种三维手指静脉识别方法及系统
CN201710281130.0 2017-04-26

Publications (1)

Publication Number Publication Date
WO2018196371A1 true WO2018196371A1 (zh) 2018-11-01

Family

ID=59567591

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/113362 Ceased WO2018196371A1 (zh) 2017-04-26 2017-11-28 一种三维手指静脉识别方法及系统

Country Status (3)

Country Link
US (1) US11017205B2 (zh)
CN (1) CN106919941B (zh)
WO (1) WO2018196371A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111160247A (zh) * 2019-12-28 2020-05-15 广东智冠信息技术股份有限公司 一种扫描掌静脉进行立体三维建模、识别的方法
CN111178256A (zh) * 2019-12-28 2020-05-19 广东智冠信息技术股份有限公司 一种掌静脉图像的三维识别认证方法、系统及装置
CN111292230A (zh) * 2020-02-18 2020-06-16 上海交通大学 深度学习中螺旋变换数据扩增方法、系统、介质及设备
CN111639556A (zh) * 2020-05-15 2020-09-08 圣点世纪科技股份有限公司 一种非均匀插值的手指轴旋转指静脉图像纠正方法
CN112308912A (zh) * 2020-11-03 2021-02-02 长安大学 一种路面病害同源多特征图像获取系统、装置及方法
CN114595438A (zh) * 2022-03-14 2022-06-07 中国第一汽车股份有限公司 文件处理方法、装置、电子设备及存储介质
CN115778333A (zh) * 2022-11-10 2023-03-14 北京悬丝医疗科技有限公司 一种视觉定位寸、关、尺脉搏穴位的方法和装置
CN119964208A (zh) * 2024-12-12 2025-05-09 北京兆讯恒达技术有限公司 一种双目指静脉身份识别的方法及装置

Families Citing this family (50)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106919941B (zh) * 2017-04-26 2018-10-09 华南理工大学 一种三维手指静脉识别方法及系统
CN107563366A (zh) * 2017-07-26 2018-01-09 安徽讯飞爱途旅游电子商务有限公司 一种定位方法及装置、电子设备
CN109791606A (zh) * 2017-08-01 2019-05-21 南京东屋信息科技有限公司 手指静脉图像采集装置
CN107731272A (zh) * 2017-09-29 2018-02-23 苏州安斯特灵智能科技有限公司 一种门诊药品安全发放流程优化方法
CN107729826A (zh) * 2017-09-29 2018-02-23 苏州安斯特灵智能科技有限公司 药品包装图像识别方法及其装置
CN108090415A (zh) * 2017-11-24 2018-05-29 江西智梦圆电子商务有限公司 一种基于人脸和掌静脉识别的身份比对方法
CN108319890A (zh) * 2017-12-01 2018-07-24 中国电子科技集团公司电子科学研究院 基于多视角图像处理的指静脉识别方法、设备及存储介质
CN109190517B (zh) * 2018-08-14 2022-05-10 北京凌云天润智能科技有限公司 一种手指静脉图像边缘提取及方向矫正方法
CN109190554A (zh) * 2018-08-30 2019-01-11 深圳大学 一种基于指纹和指静脉的3d识别系统及方法
CN109543535B (zh) * 2018-10-23 2021-12-21 华南理工大学 三维指静脉特征提取方法及其匹配方法
CN109543591A (zh) * 2018-11-19 2019-03-29 珠海格力电器股份有限公司 一种三维手指静脉采集的方法及设备
CN109598235B (zh) * 2018-12-04 2022-05-17 云天瀚科技发展有限公司 一种指静脉图像认证方法及装置
CN109740561A (zh) * 2019-01-11 2019-05-10 重庆工商大学 基于单目相机的三维手指静脉成像系统
CN110032936B (zh) * 2019-03-08 2022-08-09 吉林大学 无参数手背静脉最大圆形区域提取方法
CN109977926A (zh) * 2019-04-23 2019-07-05 珠海格力电器股份有限公司 一种指静脉识别装置、智能门锁及智能开锁方法
DE102020112352A1 (de) * 2019-05-30 2020-12-03 Panasonic I-Pro Sensing Solutions Co., Ltd. Kamera und Roboter-System
CN110197156B (zh) * 2019-05-30 2021-08-17 清华大学 基于深度学习的单图像的人手动作与形状重建方法及装置
CN110276408B (zh) * 2019-06-27 2022-11-22 腾讯科技（深圳）有限公司 3d图像的分类方法、装置、设备及存储介质
CN110399914B (zh) * 2019-07-18 2021-10-01 北京四维博大科技有限公司 陶瓷指纹识别方法及系统
CN111009007B (zh) * 2019-11-20 2023-07-14 广州光达创新科技有限公司 一种指部多特征全面三维重建方法
CN111209851B (zh) * 2020-01-04 2023-06-20 圣点世纪科技股份有限公司 一种基于指腹静脉和指背静脉深度融合的指静脉识别方法
CN111292410B (zh) * 2020-01-19 2022-04-12 华中科技大学同济医学院附属协和医院 一种静脉显影照相装置及其三维全景模型的生成方法
CN111275130B (zh) * 2020-02-18 2023-09-08 上海交通大学 基于多模态的深度学习预测方法、系统、介质及设备
CN111310688B (zh) * 2020-02-25 2023-04-21 重庆大学 一种基于多角度成像的手指静脉识别方法
CN111597880B (zh) * 2020-04-03 2023-06-30 浙江工业大学 一种基于单幅条纹投影的多视角三维指纹获取方法
CN111428693B (zh) * 2020-04-27 2023-11-14 洛阳师范学院 一种基于转轴手柄的手指静脉三维图像提取装置及方法
CN112101332B (zh) * 2020-11-23 2021-02-19 北京圣点云信息技术有限公司 一种基于3d指静脉的特征提取和比对方法及装置
CN112560590A (zh) * 2020-11-30 2021-03-26 安徽澄小光智能科技有限公司 利用腕静脉主动识别身份的智能穿戴设备
CN112528836A (zh) * 2020-12-08 2021-03-19 熵基科技股份有限公司 一种掌静脉信息采集方法、装置、设备及存储介质
CN112784758A (zh) * 2021-01-25 2021-05-11 北京航空航天大学合肥创新研究院(北京航空航天大学合肥研究生院) 一种具有指静脉识别装置的微型设备
CN113012271B (zh) * 2021-03-23 2022-05-24 华南理工大学 一种基于uv贴图的手指三维模型纹理映射方法
CN113052072B (zh) * 2021-03-25 2024-05-10 浙江工业大学 一种基于昇腾处理器的指节纹和指静脉身份识别装置
CN113421218B (zh) * 2021-04-16 2024-02-23 深圳大学 一种血管网络分支点的提取方法
CN112819124B (zh) * 2021-04-19 2021-06-22 国网浙江省电力有限公司 一种财务单据凭证防伪标签制作使用方法
CN113076921B (zh) * 2021-04-21 2022-11-18 华南理工大学 一种三维指部生物特征模型的多光谱纹理同步映射方法
US12518580B2 (en) 2021-04-21 2026-01-06 Maxsafes Security Llc Door lock with remote vein reader
CN113298934B (zh) * 2021-05-26 2022-07-08 重庆邮电大学 一种基于双向匹配的单目视觉图像三维重建方法及系统
CN113762071B (zh) * 2021-07-26 2024-09-13 深圳市鑫保泰技术有限公司 指静脉识别装置
CN113554667B (zh) * 2021-07-27 2023-12-12 上海海瞩智能科技有限公司 一种基于图像识别的三维位移检测方法及装置
CN113888640B (zh) * 2021-09-07 2024-02-02 浙江大学 一种适用于无人机云台相机的改进标定方法
TWI802123B (zh) * 2021-12-02 2023-05-11 財團法人工業技術研究院 3d靜脈圖譜的辨識方法及其識別裝置
CN114067231B (zh) 2022-01-14 2022-05-10 成都飞机工业（集团）有限责任公司 一种基于机器视觉学习识别的零件加工特征识别方法
CN115049726B (zh) * 2022-06-09 2025-01-24 安徽工程大学 基于视觉定位的pcb焊接方法及系统
CN115797991B (zh) * 2022-07-12 2025-10-10 北京交通大学 一种根据人脸侧面图像生成可识别的人脸图像的方法
CN115457612A (zh) * 2022-08-31 2022-12-09 重庆文理学院 一种手指静脉识别方法及系统
JP2024062300A (ja) * 2022-10-24 2024-05-09 キヤノン株式会社 画像処理装置、画像処理方法及びコンピュータプログラム
CN116712157B (zh) * 2023-01-16 2024-03-08 天津市鹰泰利安康医疗科技有限责任公司 一种用于心脏脉管内不可逆穿孔系统
CN116630348B (zh) * 2023-06-06 2025-09-16 重庆邮电大学 一种基于红外成像的自动皮下静脉采血方法
CN117531834A (zh) * 2023-08-18 2024-02-09 三明学院 一种基于视觉定位的钢线圈头部坐标转换方法
CN118951877B (zh) * 2024-07-29 2025-09-23 珠海格力电器股份有限公司 加工位置的定位方法、加工方法及加工系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101980243A (zh) * 2010-10-15 2011-02-23 中国人民解放军国防科学技术大学 基于双目视觉的手指静脉三维识别方法及装置
CN102609697A (zh) * 2012-04-14 2012-07-25 中国人民解放军国防科学技术大学 一种用于手指静脉三维特征识别的藤蔓模型建模方法
CN104966082A (zh) * 2015-06-11 2015-10-07 解山娟 基于多摄像头的手指静脉采集装置
CN106919941A (zh) * 2017-04-26 2017-07-04 华南理工大学 一种三维手指静脉识别方法及系统

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102004041115A1 (de) * 2004-08-24 2006-03-09 Tbs Holding Ag Verfahren und Anordnung zur Erfassung biometrischer Daten
JP5006062B2 (ja) * 2007-02-05 2012-08-22 オリンパス株式会社 バーチャルスライド作成装置、バーチャルスライド作成方法およびバーチャルスライド作成プログラム
US9479768B2 (en) * 2009-06-09 2016-10-25 Bartholomew Garibaldi Yukich Systems and methods for creating three-dimensional image media
TW201705031A (zh) * 2015-07-22 2017-02-01 Egalax_Empia Tech Inc 生物特徵辨識裝置
US20190026453A1 (en) * 2016-04-15 2019-01-24 All It Top Co., Ltd. Polyhedral three-dimensional imaging device for simultaneously authenticating fingerprint and finger veins

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101980243A (zh) * 2010-10-15 2011-02-23 中国人民解放军国防科学技术大学 基于双目视觉的手指静脉三维识别方法及装置
CN102609697A (zh) * 2012-04-14 2012-07-25 中国人民解放军国防科学技术大学 一种用于手指静脉三维特征识别的藤蔓模型建模方法
CN104966082A (zh) * 2015-06-11 2015-10-07 解山娟 基于多摄像头的手指静脉采集装置
CN106919941A (zh) * 2017-04-26 2017-07-04 华南理工大学 一种三维手指静脉识别方法及系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
LIU, TONG ET AL.: "A liana model for 3D finger vein representation", BIOELECTRONICS AND BIOINFORMATICS (ISBB), 2015 INTERNATIONAL SYMPOSIUM, 3 December 2015 (2015-12-03), XP032824082 *

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111160247A (zh) * 2019-12-28 2020-05-15 广东智冠信息技术股份有限公司 一种扫描掌静脉进行立体三维建模、识别的方法
CN111178256A (zh) * 2019-12-28 2020-05-19 广东智冠信息技术股份有限公司 一种掌静脉图像的三维识别认证方法、系统及装置
CN111160247B (zh) * 2019-12-28 2023-05-12 智冠一掌通科技(深圳)有限公司 一种扫描掌静脉进行立体三维建模、识别的方法
CN111178256B (zh) * 2019-12-28 2023-09-29 智冠一掌通科技(深圳)有限公司 一种掌静脉图像的三维识别认证方法、系统及装置
CN111292230A (zh) * 2020-02-18 2020-06-16 上海交通大学 深度学习中螺旋变换数据扩增方法、系统、介质及设备
CN111292230B (zh) * 2020-02-18 2023-04-28 上海交通大学 深度学习中螺旋变换数据扩增方法、系统、介质及设备
CN111639556A (zh) * 2020-05-15 2020-09-08 圣点世纪科技股份有限公司 一种非均匀插值的手指轴旋转指静脉图像纠正方法
CN112308912A (zh) * 2020-11-03 2021-02-02 长安大学 一种路面病害同源多特征图像获取系统、装置及方法
CN112308912B (zh) * 2020-11-03 2023-09-15 长安大学 一种路面病害同源多特征图像获取系统、装置及方法
CN114595438A (zh) * 2022-03-14 2022-06-07 中国第一汽车股份有限公司 文件处理方法、装置、电子设备及存储介质
CN115778333A (zh) * 2022-11-10 2023-03-14 北京悬丝医疗科技有限公司 一种视觉定位寸、关、尺脉搏穴位的方法和装置
CN119964208A (zh) * 2024-12-12 2025-05-09 北京兆讯恒达技术有限公司 一种双目指静脉身份识别的方法及装置

Also Published As

Publication number Publication date
US20190347461A1 (en) 2019-11-14
CN106919941A (zh) 2017-07-04
US11017205B2 (en) 2021-05-25
CN106919941B (zh) 2018-10-09

Similar Documents

Publication Publication Date Title
US11017205B2 (en) Three-dimensional finger vein recognition method and system
CN108921100B (zh) 一种基于可见光图像与红外图像融合的人脸识别方法及系统
Woodard et al. Finger surface as a biometric identifier
Zuo et al. On a methodology for robust segmentation of nonideal iris images
WO2017059591A1 (zh) 手指静脉识别方法及装置
CN104933389B (zh) 一种基于指静脉的身份识别方法和装置
CN112232109A (zh) 一种活体人脸检测方法及系统
CN109190554A (zh) 一种基于指纹和指静脉的3d识别系统及方法
WO2020083407A1 (zh) 三维指静脉特征提取方法及其匹配方法
AU2015317344A1 (en) Mobility empowered biometric appliance a tool for real-time verification of identity through fingerprints
Cui et al. Monocular 3D fingerprint reconstruction and unwarping
Puhan et al. Efficient segmentation technique for noisy frontal view iris images using Fourier spectral density
CN108052912A (zh) 一种基于矩傅里叶描述子的立体面像识别方法
CN101770566A (zh) 快速三维人耳识别方法
Parziale et al. Advanced technologies for touchless fingerprint recognition
CN110119724A (zh) 一种指静脉识别方法
Zhang et al. Sweat gland extraction from optical coherence tomography using convolutional neural network
Li et al. Three dimensional palmprint recognition
Tang et al. Contactless fingerprint image enhancement algorithm based on Hessian matrix and STFT
Bastias et al. A method for 3D iris reconstruction from multiple 2D near-infrared images
Doroz et al. An accurate fingerprint reference point determination method based on curvature estimation of separated ridges
Burge et al. Multispectral iris fusion for enhancement, interoperability, and cross wavelength matching
TW201447625A (zh) 自適應加伯濾波器的手掌靜脈識別方法
Wang et al. Automatic fundus images mosaic based on SIFT feature
CN111626181A (zh) 人脸识别大数据分析方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17908077

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS (EPO FORM 1205A DATED 20.01.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 17908077

Country of ref document: EP

Kind code of ref document: A1