A palm vein recognition method and vein recognition terminal based on feature calibration
Through feature calibration technology, the impact of palm vein image flip imaging on recognition accuracy is solved, and higher palm vein recognition accuracy and speed are achieved. The palm vein image is converted into the same five-finger orientation through rotation and flip calibration, extracting and fusing vein features are extracted and fused To improve recognition effect.
Patent Information
- Application Number
- CN202411008329.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-07-26
AI Technical Summary
In the existing palm vein recognition technology, the flipped imaging of palm vein images of the same palm under different cameras or the same device will affect the recognition accuracy, resulting in a decrease in the recognition accuracy of the recognition authentication system.
Using a feature calibration method, the palmar vein images are converted into the same five-finger orientation through rotation calibration and flip calibration. The palmar vein feature calibration model and the area of interest detection model are preset, and the vein features of the finger vein area and the palmar vein area are extracted to perform the characteristics. Fusion and match with the palmar vein feature library.
It improves the accuracy and speed of palm vein recognition and authentication, reduces the difficulty of model training and reasoning, and enhances the richness of palm vein characteristics information.
Smart Images

Figure CN118552989B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the technical field of palm vein recognition, and more specifically, to a palm vein recognition method and a vein recognition terminal based on feature calibration. Background Art
[0002] In the application of palm vein recognition authentication, the palm vein images of the same palm obtained by different cameras across devices or in the same device will present different flipped images, as shown in Figure 1 This phenomenon will affect the detection and recognition of palm veins, and further affect the recognition accuracy of the entire recognition and authentication system.
[0003] To solve this problem, the common methods in the existing deep learning technology field are to solve it through the generalization ability of the model, and the more training sets there are, the higher the detection accuracy of the model, and the more complex the training sets are, the stronger the generalization ability is. Therefore, when training the detection model, the image will be enhanced, for example, through a series of random transformation methods such as rotation, mirroring, and cropping to increase the complexity of the training set, and the number of training sets can be increased exponentially.
[0004] However, in order to improve the generalization ability of the palm vein deep learning model during the training process, the model parameters need to pay attention to features of different dimensions and need to match the features before and after flipping, which increases the difficulty of training and reasoning of the palm vein deep learning model. However, in order to make the model converge, only the vein features in the palm area can be detected, while the rich vein features on the fingers are discarded, which limits the accuracy of palm vein recognition and authentication. Summary of the invention
[0005] The purpose of the embodiments of the present application is to provide a palm vein recognition method and a vein recognition terminal based on feature calibration, so as to solve the technical problem in the prior art that different flipped images in the palm vein recognition and authentication process affect the recognition and authentication accuracy.
[0006] To achieve the above purpose, the technical solution adopted in this application is: to provide a palm vein recognition method based on feature calibration, comprising the steps of:
[0007] A palm vein feature calibration model is preset to convert the palm vein image into the same five-finger orientation through rotation calibration and flip calibration;
[0008] A palm vein region of interest detection model is preset, and the palm vein images of the training set are first input into the palm vein feature calibration model for calibration, and then the finger vein areas and palm vein areas are annotated on the palm vein images of the training set, and used to train the palm vein region of interest detection model;
[0009] The user's palm vein image is collected in real time. The palm vein image is first input into the palm vein feature calibration model for calibration, and then the finger vein area and palm vein area are detected by the palm vein region of interest detection model;
[0010] The vein features of the finger vein area and the palm vein area are extracted for feature fusion, and then matched with the palm vein feature library.
[0011] In another embodiment of the present application, the rotation calibration method comprises the steps of:
[0012] Detect the end point B (x2, y2) of the interdigital gap between the index finger and the middle finger and the end point C (x3, y3) of the interdigital gap between the ring finger and the little finger from the palm vein image;
[0013] Connect point B and point C, and calculate the angle θ between the line connecting the two points and the horizontal line. The calculation formula for angle θ is:
[0014]
[0015] The palm vein image is rotated counterclockwise by an angle θ, where θ∈(-90°, 90°).
[0016] In another embodiment of the present application, the flip calibration method comprises the steps of:
[0017] Detect the end points A (x1, y1) of the thumb and index finger gaps from the palm vein image;
[0018] If y1>y2, it is judged that the finger is facing upward, and there is no need to vertically flip the palm vein image;
[0019] If y1≤y2, it is judged that the finger is facing downward, and the palm vein image needs to be vertically flipped to obtain a palm vein image with four fingers facing upward;
[0020] If x1<x3, it is judged that the thumb is on the left, and there is no need to flip the palm vein image horizontally;
[0021] If x1≥x3, it is judged that the thumb is on the right, and the palm vein image needs to be horizontally flipped to obtain a palm vein image with the thumb on the left.
[0022] In another embodiment of the present application, after converting the palm vein image into the same five-finger orientation through rotation calibration and flip calibration, the method further includes the following steps:
[0023] A rectangular coordinate system RC is established at the center point of the palm vein image, and the rectangular coordinate system RC divides the palm vein image into a first quadrant, a second quadrant, a third quadrant, and a fourth quadrant;
[0024] Calculate the average brightness values of the first quadrant, the second quadrant, the third quadrant, and the fourth quadrant as E1, E2, E3, and E4 respectively;
[0025] If E2<E1<E3<E4 exists, the current palm vein image is retained, otherwise the current palm vein image is filtered.
[0026] In another embodiment of the present application, after inputting the user's palm vein image into the palm vein feature calibration model for calibration, the method further includes the following steps:
[0027] The binocular camera collects the user's palm vein image and the user's palm print image at the same time;
[0028] Detecting the ground pattern, human pattern and sky pattern in the palm print based on the user's palm print image;
[0029] The least square method is used to fit the ground pattern, human pattern and sky pattern respectively;
[0030] Then, based on linear regression, the ground pattern, human pattern and sky pattern after straight line fitting are fitted into a three-pattern line;
[0031] The slope k of the line formed by the three lines is detected, and the user's palm vein image is subjected to secondary rotation calibration according to the slope k.
[0032] In another embodiment of the present application, a method for labeling finger vein areas in a training set of palm vein images includes the steps of:
[0033] The proximal interphalangeal joint area of the index finger, the proximal interphalangeal joint area of the middle finger, the proximal interphalangeal joint area of the ring finger, and the proximal interphalangeal joint area of the little finger are independently labeled.
[0034] In another embodiment of the present application, a method for detecting a finger vein region and a palm vein region by using a palm vein region of interest detection model comprises the steps of:
[0035] The calibrated palm vein image of the user is input into the palm vein region of interest detection model, and the palmar vein area, the proximal interphalangeal joint area of the index finger, the proximal interphalangeal joint area of the middle finger, the proximal interphalangeal joint area of the ring finger, and the proximal interphalangeal joint area of the little finger are obtained from the palm vein image of the user.
[0036] In another embodiment of the present application, a method for extracting vein features of a finger vein region and a palm vein region and performing feature fusion includes the steps of:
[0037] Define the number of proximal interphalangeal joints detected as n, and extract the finger vein features F Fi Palm vein characteristics P , the palm feature weight coefficient is α, then the calculation formula of the weighted fusion feature F is:
[0038] .
[0039] In another embodiment of the present application, a method for extracting vein features of a finger vein region and a palm vein region and performing feature fusion includes the steps of:
[0040] Taking the proximal interphalangeal joint area of the middle finger as the base point, the proximal interphalangeal joint area of the index finger, the proximal interphalangeal joint area of the middle finger, the proximal interphalangeal joint area of the ring finger, and the proximal interphalangeal joint area of the little finger are sequentially stitched to obtain the proximal interphalangeal joint stitching area;
[0041] The proximal interphalangeal joint area and the palmar vein area are fused according to the relative position parameters to obtain the fused vein area;
[0042] Extract vein features from fused vein regions.
[0043] The present application also provides a vein recognition terminal, which includes a front camera, a rear camera, an infrared fill light and an image processing unit. The infrared fill light is used to fill light on the user's palm. The front camera and the rear camera are used to collect the user's palm vein image in real time. The image processing unit is used to execute the palm vein recognition method based on feature calibration as described above to identify and authenticate the user's palm vein image.
[0044] Compared with the prior art, the palm vein recognition method based on feature calibration provided by the present application calibrates the palm vein images of the training set of the palm vein region of interest detection model and the user's palm vein images using the same palm vein feature calibration model, and converts them into palm vein images with the same five-finger orientation, so that the palm vein region of interest detection model can only focus on the palm vein texture itself, and then accurately extract the finger vein area and the palm vein area, which not only improves the richness of palm vein characteristic information, but also reduces the training and reasoning difficulty of palm vein feature detection and recognition, thereby achieving the purpose of improving the accuracy of palm vein recognition and authentication.
[0045] Compared with the prior art, the feature-calibrated vein recognition terminal provided in the present application adopts a feature-calibrated palm vein recognition method to calibrate the palm vein images of the user captured by the front camera and the rear camera, simplifies the recognition parameters, and can improve the accuracy and speed of palm vein recognition authentication. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0047] Figure 1 A schematic diagram of obtaining a palm vein image through a front camera and a rear camera;
[0048] Figure 2 A schematic diagram of a flow chart of a palm vein recognition method based on feature calibration provided in an embodiment of the present application;
[0049] Figure 3 This is a schematic diagram of obtaining key points from a palm vein image according to an embodiment of the present application;
[0050] Figure 4 A schematic diagram of calibrating a palm vein image according to an embodiment of the present application;
[0051] Figure 5 A schematic diagram of dividing a palm vein image into quadrants according to an embodiment of the present application;
[0052] Figure 6 This is a schematic diagram of the process of processing a user's palm print image according to an embodiment of the present application;
[0053] Figure 7 Schematic diagram of the processing process of feature fusion of vein features in an embodiment of the present application. DETAILED DESCRIPTION
[0054] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0055] It should be noted that when an element is referred to as being "fixed to" or "disposed on" another element, it can be directly on the other element or indirectly on the other element. When an element is referred to as being "connected to" another element, it can be directly connected to the other element or indirectly connected to the other element.
[0056] It should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.
[0057] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0058] First of all, it is worth pointing out that there is a common bias in the existing deep learning technology field: the more training sets there are, the higher the detection accuracy of the model, and the more complex the training sets are, the stronger the generalization ability is. Therefore, when training the detection model, the image will be enhanced, for example, through a series of random transformation methods such as rotation, mirroring, and cropping to increase the complexity of the training set, and the number of training sets can be increased exponentially.
[0059] However, the present application provides a palm vein recognition method based on feature calibration, which not only does not perform enhancement processing on the training set, but also performs uniform calibration processing on the training set to weaken the complexity, thereby improving the accuracy of palm vein recognition by reducing the generalization ability.
[0060] Please also read Figure 2 The palm vein recognition method based on feature calibration provided in the embodiment of the present application is now described. The palm vein recognition method based on feature calibration comprises the following steps:
[0061] Step S1, presetting a palm vein feature calibration model, which is used to convert the palm vein image into the same five-finger orientation through rotation calibration and flip calibration;
[0062] Step S2, presetting a palm vein region of interest detection model, first inputting the palm vein images of the training set into the palm vein feature calibration model for calibration, then annotating the finger vein areas and the palm vein areas of the palm vein images of the training set, and using them to train the palm vein region of interest detection model;
[0063] Step S3, collecting the user's palm vein image in real time, first inputting the user's palm vein image into the palm vein feature calibration model for calibration, and then detecting the finger vein area and the palm vein area through the palm vein region of interest detection model;
[0064] Step S4, extracting vein features of the finger vein area and the palm vein area, performing feature fusion, and then matching them with the palm vein feature library.
[0065] It can be understood that in step S1, the palm vein image is calibrated by rotation based on the palm vein feature calibration model, and the palm vein image can be converted to the same rotation direction. In this embodiment, the index finger, middle finger, ring finger and little finger (hereinafter referred to as four fingers) face upward after rotation calibration, but the thumb may face left or right at this time. The thumb direction can be unified by flip calibration. In this embodiment, the thumb faces left after flip calibration.
[0066] In step S2, the palm vein images of the training set are calibrated using the palm vein feature calibration model in step S1, and the palm vein images of the training set with the same five-finger orientation can be obtained. In this way, the left palm vein images and the right palm vein images, the front camera palm vein images and the rear camera palm vein images can all be uniformly calibrated.
[0067] When the palm vein images in the training set are used again to train the palm vein region of interest detection model, the region of interest detection model only needs to focus on the palm vein texture itself, which reduces the difficulty of training and reasoning the palm vein region of interest detection model. In this way, the trained palm vein region of interest detection model can detect both the finger vein area and the palm vein area at the same time.
[0068] It is worth noting that, affected by the degree of opening and closing of the fingers and the differences in the directions of the five fingers, the palm vein recognition model of the middle finger in the prior art has to take into account generalization, and the detection error of the finger part is large. Therefore, the palm vein features of the middle finger in the prior art usually only include the palm vein area.
[0069] In step S3, when the user is authenticating, the real-time collected palm vein image of the user is first input into the palm vein feature calibration model for calibration, and the palm vein image of the user with the same five fingers facing can be obtained. In this way, the palm vein region of interest detection model trained based on step S2 can accurately detect the finger vein area and the palm vein area at the same time because it only focuses on the palm vein texture itself, thereby improving the richness of palm vein characteristic information.
[0070] In addition, the palm vein region of interest detection model of the present application has limited generalization ability, so only clear and complete palm images can detect both the finger vein area and the palm vein area, which is equivalent to automatically filtering images with low quality scores. In the process of real-time acquisition of user palm vein images, the recognition efficiency is improved.
[0071] In step S4, since the finger vein area and the palm vein area are independent of each other, the vein features of the finger vein area and the palm vein area are extracted and fused to achieve the purpose of reducing feature redundancy. After matching the fused vein features with the palm vein feature library calibrated based on step S3, the accuracy of palm vein recognition and authentication is improved.
[0072] Compared with the prior art, the palm vein recognition method based on feature calibration provided by the present application calibrates the palm vein images of the training set of the palm vein region of interest detection model and the user's palm vein images using the same palm vein feature calibration model, and converts them into palm vein images with the same five-finger orientation, so that the palm vein region of interest detection model can only focus on the palm vein texture itself, and then accurately extract the finger vein area and the palm vein area, which not only improves the richness of palm vein characteristic information, but also reduces the training and reasoning difficulty of palm vein feature detection and recognition, thereby achieving the purpose of improving the accuracy of palm vein recognition and authentication.
[0073] In another embodiment of the present application, please refer to Figure 3 to Figure 4 In step S1, the rotation calibration method comprises the steps of:
[0074] Detect the end point B (x2, y2) of the interdigital gap between the index finger and the middle finger and the end point C (x3, y3) of the interdigital gap between the ring finger and the little finger from the palm vein image;
[0075] Connect point B and point C, and calculate the angle θ between the line connecting the two points and the horizontal line. The calculation formula for angle θ is:
[0076]
[0077] The palm vein image is rotated counterclockwise by an angle θ, where θ∈(-90°, 90°).
[0078] It is understandable that, according to the biological characteristics of human fingers, when the four fingers are facing up or down, point B and point C are approximately on the same horizontal line, and the relative position of point B and point C will not be affected by the degree of opening and closing of the four fingers, thereby improving the accuracy of rotation calibration.
[0079] For further information, please also refer to Figure 3 to Figure 4 In step S1, the flip calibration method comprises the steps of:
[0080] Detect the end points A (x1, y1) of the thumb and index finger gaps from the palm vein image;
[0081] If y1>y2, it is judged that the finger is facing upward, and there is no need to vertically flip the palm vein image;
[0082] If y1≤y2, it is judged that the finger is facing downward, and the palm vein image needs to be vertically flipped to obtain a palm vein image with four fingers facing upward;
[0083] If x1<x3, it is judged that the thumb is on the left, and there is no need to flip the palm vein image horizontally;
[0084] If x1≥x3, it is judged that the thumb is on the right, and the palm vein image needs to be horizontally flipped to obtain a palm vein image with the thumb on the left.
[0085] It is understandable that, based on the rotation calibration, a palm vein image with four fingers facing upward or downward is obtained. In order to obtain the same orientation of five fingers, some palm vein images need horizontal calibration, some palm vein images need vertical calibration, and even some palm vein images need to be calibrated horizontally and vertically at the same time. To this end, this embodiment detects the coordinate values of point A and point C on the basis of the rotation calibration to achieve calibration with the same orientation of five fingers, thereby further improving the calibration accuracy.
[0086] For further information, please also refer to Figure 5 In step S1, after the palm vein image is converted to the same five-finger orientation through rotation calibration and flip calibration, the following steps are also included:
[0087] A rectangular coordinate system RC is established at the center point of the palm vein image, and the rectangular coordinate system RC divides the palm vein image into a first quadrant, a second quadrant, a third quadrant, and a fourth quadrant;
[0088] Calculate the average brightness values of the first quadrant, the second quadrant, the third quadrant, and the fourth quadrant as E1, E2, E3, and E4 respectively;
[0089] If E2<E1<E3<E4 exists, the current palm vein image is retained, otherwise the current palm vein image is filtered.
[0090] It can be understood that since the rotation calibration and flip calibration in this embodiment rely on the palm key points, the detection accuracy of the palm key points depends on the local detail features. In the case of insufficient image quality, the existing palm key point model will have inaccurate detection, especially the positions of point A and point C are likely to be interchanged.
[0091] Since the palm vein image is an image collected under infrared light, only the palm area has high brightness and the background area has low brightness. In the calibrated palm vein image, the four fingers face upward and the thumb faces left. In this way, this embodiment relies on the rectangular coordinate system RC to divide the palm vein image into four quadrants. Unlike the detection of palm key points, it does not rely on local details and calculates the average brightness value of the four quadrants from a global level, which can verify the accuracy of palm vein image calibration.
[0092] For further information, please also refer to Figure 6 In step S3, after inputting the user's palm vein image into the palm vein feature calibration model for calibration, the method further includes the following steps:
[0093] The binocular camera collects the user's palm vein image and the user's palm print image at the same time;
[0094] Detecting the ground pattern, human pattern and sky pattern in the palm print based on the user's palm print image;
[0095] The least square method is used to fit the ground pattern, human pattern and sky pattern respectively;
[0096] Then, based on linear regression, the ground pattern, human pattern and sky pattern after straight line fitting are fitted into a three-pattern line;
[0097] The slope k of the line formed by the three lines is detected, and the user's palm vein image is rotated and calibrated again according to the slope k.
[0098] It is understandable that since the user's palm vein image and the user's palm print image are collected simultaneously based on a binocular camera, they have the same rotation angle and finger orientation. After a large amount of palm data analysis, each person's ground pattern, human pattern and sky pattern are highly similar, and the extension directions of these three lines are close, that is, the slope range of the three lines in a normal palm is (-1, -1.3), preferably, the slope of the three lines in a normal palm is -1.2. Based on the research findings, this embodiment re-rotates and calibrates the user's palm vein image by fitting the ground pattern, human pattern and sky pattern into a three-line line based on the visible light image level, thereby further improving the accuracy of palm vein image calibration.
[0099] It is worth adding that the earth lines, human lines and sky lines should be based on the definition of palm lines in the diagnostic discipline of hand diagnosis, where hand diagnosis refers to an auxiliary means of prevention and treatment that infers the evolution of human organs through the shape, changes, and patterns of the lines on the human hand.
[0100] In another embodiment of the present application, please refer to Figure 7 In step S2, the method for marking the finger vein area of the palm vein image of the training set includes the steps of:
[0101] The proximal interphalangeal joint area of the index finger, the proximal interphalangeal joint area of the middle finger, the proximal interphalangeal joint area of the ring finger, and the proximal interphalangeal joint area of the little finger are independently labeled.
[0102] It is understandable that, unlike the palm vein area, the relative positions of the four finger veins will change due to the random degree of finger opening and closing. If the finger vein area obtained by annotating the four fingers as a whole will inevitably contain a lot of interference information. In this embodiment, the four fingers are independently annotated, which is not affected by the degree of finger opening and closing, and only the proximal interphalangeal joints of the four fingers are taken, which are the most obvious and concentrated areas of finger veins, so as to improve the detection accuracy of finger veins.
[0103] For further information, please also refer to Figure 7 In step S3, the method of detecting the finger vein area and the palm vein area by using the palm vein region of interest detection model comprises the steps of:
[0104] The calibrated palm vein image of the user is input into the palm vein region of interest detection model, and the palmar vein area, the proximal interphalangeal joint area of the index finger, the proximal interphalangeal joint area of the middle finger, the proximal interphalangeal joint area of the ring finger, and the proximal interphalangeal joint area of the little finger are obtained from the palm vein image of the user.
[0105] It is understandable that since the proximal interphalangeal joint area is small, some parts of the proximal interphalangeal joint area are not clear. In this embodiment, only the proximal interphalangeal joint area of the index finger with high confidence is output, which can reduce the misrecognition rate.
[0106] Furthermore, in step S4, the method for extracting vein features of the finger vein area and the palm vein area for feature fusion includes the steps of:
[0107] Define the number of proximal interphalangeal joints detected as n, and extract the finger vein features F Fi Palm vein characteristics P , the palm feature weight coefficient is α, then the calculation formula of the weighted fusion feature F is:
[0108] .
[0109] It can be understood that feature redundancy is reduced by performing multi-modal weighted fusion on the vein features of the finger vein area and the palm vein area.
[0110] For further information, please also refer to Figure 7 In step S4, another method for extracting vein features of the finger vein area and the palm vein area for feature fusion is also provided, including the steps of:
[0111] Taking the proximal interphalangeal joint area of the middle finger as the base point, the proximal interphalangeal joint area of the index finger, the proximal interphalangeal joint area of the middle finger, the proximal interphalangeal joint area of the ring finger, and the proximal interphalangeal joint area of the little finger are sequentially stitched to obtain the proximal interphalangeal joint stitching area;
[0112] The proximal interphalangeal joint area and the palmar vein area are fused according to the relative position parameters to obtain the fused vein area;
[0113] Extract vein features from fused vein regions.
[0114] It can be understood that when the degree of opening and closing of the fingers is different, the relative position of the proximal interphalangeal joint of the middle finger and the palm remains unchanged. Therefore, based on the palm characteristics, this embodiment uses the proximal interphalangeal joint area of the middle finger as the base point for image stitching to obtain the proximal interphalangeal joint stitching area. While achieving the effect of calibrating the four fingers together, it also retains the relative position characteristics of the finger vein features and the palm vein features, and finally obtains a fused vein area containing relative position features. Compared with the weighted fusion feature method, this embodiment further enriches the feature dimension and improves the accuracy of recognition and authentication.
[0115] The present application also provides a vein recognition terminal, which includes a front camera, a rear camera, an infrared fill light and an image processing unit. The infrared fill light is used to fill light on the user's palm. The front camera and the rear camera are used to collect the user's palm vein image in real time. The image processing unit is used to execute the palm vein recognition method based on feature calibration as described above to identify and authenticate the user's palm vein image.
[0116] Compared with the prior art, the feature-calibrated vein recognition terminal provided in the present application adopts a feature-calibrated palm vein recognition method to calibrate the palm vein images of the user captured by the front camera and the rear camera, simplifies the recognition parameters, and can improve the accuracy and speed of palm vein recognition authentication.
[0117] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.
Claims
1. A palm vein recognition method based on feature calibration, characterized in that: include: A palm vein feature calibration model is preset to convert the palm vein image into the same five-finger orientation through rotation calibration and flip calibration; A palm vein region of interest detection model is preset, and the palm vein images of the training set are first input into the palm vein feature calibration model for calibration, and then the finger vein areas and palm vein areas are annotated on the palm vein images of the training set, and used to train the palm vein region of interest detection model; The user's palm vein image is collected in real time. The palm vein image is first input into the palm vein feature calibration model for calibration, and then the finger vein area and palm vein area are detected by the palm vein region of interest detection model; Extract the vein features of the finger vein area and the palm vein area for feature fusion, and then match them with the palm vein feature library; The method for labeling the finger vein region of the palm vein image of the training set comprises the steps of: independently labeling the proximal interphalangeal joint region of the index finger, the proximal interphalangeal joint region of the middle finger, the proximal interphalangeal joint region of the ring finger, and the proximal interphalangeal joint region of the little finger; The method for detecting the finger vein area and the palm vein area by using a palm vein region of interest detection model comprises the steps of: inputting a calibrated user palm vein image into the palm vein region of interest detection model, acquiring the palm vein area, the proximal interphalangeal joint area of the index finger, the proximal interphalangeal joint area of the middle finger, the proximal interphalangeal joint area of the ring finger, and the proximal interphalangeal joint area of the little finger from the user palm vein image; The method for extracting vein features of a finger vein region and a palm vein region and performing feature fusion comprises the steps of: taking a proximal interphalangeal joint region of a middle finger as a base point, sequentially performing image splicing on a proximal interphalangeal joint region of an index finger, a proximal interphalangeal joint region of a middle finger, a proximal interphalangeal joint region of a ring finger, and a proximal interphalangeal joint region of a little finger to obtain a proximal interphalangeal joint splicing region; fusing the proximal interphalangeal joint splicing region with a palm vein region according to a relative position parameter to obtain a fused vein region; and extracting vein features of the fused vein region.
2. The palm vein recognition method based on feature calibration as claimed in claim 1, characterized in that: The rotation calibration method comprises the steps of: Detect the end point B (x2, y2) of the interdigital gap between the index finger and the middle finger and the end point C (x3, y3) of the interdigital gap between the ring finger and the little finger from the palm vein image; Connect point B and point C, and calculate the angle θ between the line connecting the two points and the horizontal line. The calculation formula for angle θ is: The palm vein image is rotated counterclockwise by an angle θ, where θ∈(-90°, 90°).
3. The palm vein recognition method based on feature calibration as claimed in claim 2, characterized in that: The flip calibration method comprises the steps of: Detect the end points A (x1, y1) of the thumb and index finger gaps from the palm vein image; If y1>y2, it is judged that the finger is facing upward, and there is no need to vertically flip the palm vein image; If y1≤y2, it is judged that the finger is facing downward, and the palm vein image needs to be vertically flipped to obtain a palm vein image with four fingers facing upward; If x1<x3, it is judged that the thumb is on the left, and there is no need to flip the palm vein image horizontally; If x1≥x3, it is judged that the thumb is on the right, and the palm vein image needs to be horizontally flipped to obtain a palm vein image with the thumb on the left.
4. The palm vein recognition method based on feature calibration as claimed in claim 3, characterized in that: After the palm vein image is converted to the same five-finger orientation through rotation calibration and flip calibration, the following steps are also included: A rectangular coordinate system RC is established at the center point of the palm vein image, and the rectangular coordinate system RC divides the palm vein image into a first quadrant, a second quadrant, a third quadrant, and a fourth quadrant; Calculate the average brightness values of the first quadrant, the second quadrant, the third quadrant, and the fourth quadrant as E1, E2, E3, and E4 respectively; If E2<E1<E3<E4 exists, the current palm vein image is retained, otherwise the current palm vein image is filtered.
5. The palm vein recognition method based on feature calibration as claimed in claim 3, characterized in that: After inputting the user's palm vein image into the palm vein feature calibration model for calibration, the following steps are also included: The binocular camera collects the user's palm vein image and the user's palm print image at the same time; Detecting the ground pattern, human pattern and sky pattern in the palm print based on the user's palm print image; The least square method is used to fit the ground pattern, human pattern and sky pattern respectively; Then, based on linear regression, the ground pattern, human pattern and sky pattern after straight line fitting are fitted into a three-pattern line; The slope k of the line formed by the three lines is detected, and the user's palm vein image is subjected to secondary rotation calibration according to the slope k.
6. A vein recognition terminal, characterized in that: The invention comprises a front camera, a rear camera, an infrared fill light and an image processing unit, wherein the infrared fill light is used to fill light on the palm of the user, the front camera and the rear camera are used to collect the palm vein image of the user in real time, and the image processing unit is used to execute the palm vein recognition method based on feature calibration as described in any one of claims 1 to 5 to identify and authenticate the palm vein image of the user.
