A method and system for extracting ROI from palm veins and palm prints based on multi-feature point fusion

Abstract

The application discloses a palm vein and palm print ROI extraction method and system based on multi-feature point fusion, which comprises the following steps: acquiring a gray image including palm print and palm vein, and converting the gray image into a binary image; calculating the center of mass of the palm region and extracting the palm boundary line based on the binary image; fitting the palm boundary point and the palm center point smooth Euclidean distance function, clustering the obtained candidate palm valley points, and obtaining palm valley point coordinates; obtaining finger gap valley points according to the palm valley points and the palm center; rotating and correcting to obtain a fingertip-up image; calculating the distance of all valley points to the center, obtaining the ROI extraction side length, and obtaining the ROI extraction four vertex coordinates. The application can extract palm vein and palm print ROIs meeting national standards based on multiple key points such as palm centers and finger valley points, so as to improve the accuracy and robustness of ROI extraction, adapt to different light conditions and palm posture changes, and improve the overall performance of a palm print and palm vein recognition system.

Description

Technical Field

[0001] This invention relates to the field of biometric recognition technology, specifically to a method and system for extracting palm vein / palmprint ROI based on multi-feature point fusion. It can automatically extract the Region of Interest (ROI) of palm veins / palmprints that conforms to the national industry standard (2024-1254T-SJ) from palm images, and is applicable to scenarios such as palmprint / palm vein image acquisition and preprocessing, identity authentication, and biometric comparison. Background Technology

[0002] Palmprint and palm vein recognition systems typically involve steps such as image acquisition, preprocessing, feature extraction, and matching. One of the key steps in the image preprocessing stage is the extraction of the Region of Interest (ROI). The accuracy of ROI extraction directly impacts the effectiveness of subsequent feature extraction and matching.

[0003] Existing patent CN119027652A discloses an improved method for extracting Regions of Interest (ROIs) in palm print and palm vein recognition, including: dynamic threshold binarization, morphological operations for boundary extraction, key point determination, frequency domain processing to remove high-frequency noise, determination of palm centroid, and ROI region truncation (see...). Figure 1 ).

[0004] Existing patent CN119048729A discloses a method for extracting a palm ROI, including the following steps: using a key point detection model to detect key points on the palm, the key points including the index finger root joint P1, the middle finger root joint P2, the ring finger root joint P3, and the little finger root joint P4; using a preset parameter k according to the formula for dividing a line segment by a fixed ratio to calculate P7, P8, P9, and P10; sequentially connecting points P7, P8, P9, P10, and P7, the resulting geometric shape is output as the ROI (see...). Figure 2 ).

[0005] Existing patent CN113963158A discloses a deep learning-based hand recognition method, comprising the following steps: obtaining hand key points in the image to be extracted using a trained hand key point detection model, wherein the hand key points include the palm and wrist, thumb joint, index finger joint, middle finger joint, ring finger joint, and little finger joint; determining the hand type and hand tilt information of the image to be extracted based on the hand key points, wherein the hand type includes left hand and right hand; performing rotation calibration on the image to be extracted based on the hand type and hand tilt information; and determining the region of interest in the image to be extracted based on the hand key points (see...). Figure 3 ).

[0006] The existing literature *Laser & Infrared*, Vol. 42, No. 7: 816-820, discloses a novel method for locating the Region of Interest (ROI) in palm print images. This method utilizes an improved adaptive Harris algorithm to extract contour feature points from the palm edge and the convex and concave areas of the fingers. It uses the mean value of feature points belonging to the same feature group as candidate corner points, eliminates neighboring corner points, and locates three key corner points at the base of the fingers based on the magnitude of the angle change of the candidate corner points along the contour line. This effectively extracts the ROI from the palm print image (see...). Figure 4 ).

[0007] Referring to the draft for comments on the national electronics industry standard "Technical Specification for Multimodal Palmprint and Palm Vein Recognition System for Biometric Recognition," regarding the requirement for "palm ROI ratio," existing technologies for extracting the ROI of fingers fail to meet the standard requirements. This requirement is shown in Table 1.

[0008] Table 1

[0009] Table A.1 Sample Quality Requirements

[0010]

[0011] The main shortcomings of existing methods when following the above standards are categorized as follows:

[0012] 1. ROI extraction results are missing key regions.

[0013] The ROI regions extracted by the above methods are all images inside the palm, excluding the gaps between the fingers and the left and right edges of the palm. If only the side length of the square is extracted by enlarging the ROI, due to the accuracy of palm localization, the extraction result cannot guarantee that it includes the above key areas and also meets the requirement that the ROI accounts for less than or equal to 80% of the palm.

[0014] 2. ROI extraction methods are not adaptable enough to finger postures.

[0015] The above methods all rely on the accurate and effective localization of the finger valley points (i.e., the indentations at the junctions between fingers). However, in real-world applications, users' finger postures vary when swiping their palms (e.g., fingers partially or fully closed, wearing rings, etc.), leading to inaccurate localization of finger valley points. Consequently, the ROI region extraction based on this method has a significant deviation and cannot guarantee compliance with the national standard requirements for ROI extraction regions.

[0016] 3. The ROI extraction results do not include finger direction information.

[0017] The above algorithms only produce square or rectangular images after extracting the Region of Interest (ROI), without including finger orientation information from the palm print acquisition. This lack of information provides a basis for ensuring a consistent orientation of the ROI images during subsequent palmprint and vein recognition applications. Palmprint and vein recognition is sensitive to ROI orientation (e.g., if the palm and fingers are pointing upwards during registration, but the palm is horizontal during verification, resulting in a large angle of deflection between the obtained ROI images, the palmprint and vein recognition will reject the image). Summary of the Invention

[0018] To address the shortcomings of existing technologies, this invention provides a method and system for palm vein and palmprint ROI extraction based on multi-feature point fusion. It extracts palm vein and palmprint ROIs that meet national standards based on multiple key points such as the palm center and finger valley points, thereby improving the accuracy and robustness of ROI extraction and adapting to different lighting conditions and changes in palm posture, thus improving the overall performance of the palm vein and palmprint recognition system.

[0019] This invention discloses a method for extracting the ROI of palm veins and palm prints based on multi-feature point fusion, comprising:

[0020] Step 1: Obtain a grayscale image including palm prints and veins, convert the grayscale image into a binary image, and separate the palm region from the background;

[0021] Step 2: Calculate the centroid of the palm region and extract the palm boundary line based on the binary image from Step 1;

[0022] Step 3: Fit the smooth Euclidean distance function between the palm boundary points and the palm centroid, calculate the second derivative of the function at each node, and select the nodes with derivatives greater than zero as candidate palm valley points; use the DBSCAN algorithm to cluster the candidate palm valley points to obtain the coordinates of the palm valley points.

[0023] Step 4: Based on the palm valley points and the palm centroid, with the palm centroid as the origin, calculate the first principal component vector of PCA to obtain the palm extension direction; and select the three nodes closest to the first principal component vector of PCA from the palm valley point vector to obtain the finger gap valley points between the index finger, middle finger, ring finger, and little finger.

[0024] Step 5: Using the center of the palm as the center, the angle between the vector connecting the center of the palm and the valley point between the middle and ring fingers and the vertical upward direction of the image is the rotation angle. After correction, the image with the fingertips pointing upward is obtained.

[0025] Step 6: Calculate the distance from all valley points to the center, exclude a preset number of outliers, and obtain the average distance; based on the average distance, obtain the ROI extraction edge length; with the center of the palm as the center point, obtain the coordinates of the four vertices of the ROI extraction based on the ROI extraction edge length.

[0026] As a further improvement of the present invention, step 1 specifically includes:

[0027] (1) Read the grayscale image of palm prints and veins in grayscale mode;

[0028] (2) Use Gaussian kernel smoothing to process the grayscale image of palm prints and veins;

[0029] (3) Calculate the grayscale histogram of the grayscale image. Based on the shape of the grayscale histogram, dynamically select a threshold T to convert the grayscale image into a binary image and separate the palm region from the background. The foreground pixels are set to 1 and the background pixels are set to 0.

[0030] As a further improvement of the present invention, step 2 specifically includes:

[0031] (1) Calculate the spatial moments m of the binary image pq ;

[0032]

[0033] In the formula,

[0034] I(x,y): The value of pixel (x,y);

[0035] p, q: the order of the moments;

[0036] (2) Calculate the centroid coordinates (x, y) of the palm region. c ,y c );

[0037]

[0038] (3) Use a 3×3 cross-shaped kernel to perform image morphological erosion. Subtract the eroded image from the original binary image to obtain the boundary pixels. Extract the contour from the boundary image and select the longest one as the palm boundary line.

[0039] As a further improvement of the present invention, in step 6...

[0040] Exclude outliers that are 10% of the maximum and 10% of the minimum distance;

[0041] The formula for calculating the side length (side_length) of the ROI is as follows:

[0042] side_length = k * avg_radius

[0043] In the formula,

[0044] avg_radius is the average distance, and k is the adjustment factor.

[0045] This invention also discloses a palm vein and palm print ROI extraction system based on multi-feature point fusion, comprising:

[0046] The palm image preprocessing module is used to acquire a grayscale image including palm prints and veins, convert the grayscale image into a binary image, and separate the palm region from the background.

[0047] The palm centroid and boundary extraction module is used to calculate the centroid of the palm region and extract the palm boundary line based on the binary image.

[0048] The palm valley point localization module is used to fit the smooth Euclidean distance function between the palm boundary point and the palm centroid point, calculate the second derivative of the function at each node, and the nodes with derivatives greater than zero are candidate palm valley points; the DBSCAN algorithm is used to cluster the candidate palm valley points to obtain the palm valley point coordinates.

[0049] The finger gap valley point localization module is used to calculate the first principal component vector of PCA based on the palm valley point and the palm centroid, with the palm centroid as the origin, to obtain the palm extension direction; and to select the three nodes closest to the first principal component vector of PCA from the palm valley point vector to obtain the finger gap valley points between the index finger, middle finger, ring finger, and little finger.

[0050] The rotation correction module is used to rotate the image by taking the center of the palm as the center, the angle between the vector connecting the center of the palm and the valley point between the middle and ring fingers and the vertical upward direction of the image, and then obtaining an image with the fingertips pointing upward.

[0051] The ROI extraction four-vertex coordinate determination module is used to calculate the distance from all valley points to the center, and after excluding a preset number of outliers, obtain the average distance; based on the average distance, obtain the ROI extraction side length; with the center of the palm as the center point, obtain the ROI extraction four-vertex coordinates based on the ROI extraction side length.

[0052] As a further improvement of the present invention, the palm image preprocessing module is specifically used for:

[0053] (1) Read the grayscale image of palm prints and veins in grayscale mode;

[0054] (2) Use Gaussian kernel smoothing to process the grayscale image of palm prints and veins;

[0055] (3) Calculate the grayscale histogram of the grayscale image. Based on the shape of the grayscale histogram, dynamically select a threshold T to convert the grayscale image into a binary image and separate the palm region from the background. The foreground pixels are set to 1 and the background pixels are set to 0.

[0056] As a further improvement of the present invention, the palm centroid and boundary extraction module is specifically used for:

[0057] (1) Calculate the spatial moments m of the binary image pq ;

[0058]

[0059] In the formula,

[0060] I(x,y): The value of pixel (x,y);

[0061] p, q: the order of the moments;

[0062] (2) Calculate the centroid coordinates (x, y) of the palm region. c ,y c );

[0063]

[0064] (3) Use a 3×3 cross-shaped kernel to perform image morphological erosion. Subtract the eroded image from the original binary image to obtain the boundary pixels. Extract the contour from the boundary image and select the longest one as the palm boundary line.

[0065] As a further improvement of the present invention, in the ROI extraction and four-vertex coordinate determination module...

[0066] Exclude outliers that are 10% of the maximum and 10% of the minimum distance;

[0067] The formula for calculating the side length (side_length) of the ROI is as follows:

[0068] side_length = k * avg_radius

[0069] In the formula,

[0070] avg_radius is the average distance, and k is the adjustment factor.

[0071] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0072] 1. This invention optimizes the ROI extraction algorithm to meet national standards and ensures coverage of key areas such as finger gaps and the left and right edges of the palm.

[0073] 2. This invention can dynamically adapt to different palm sizes and finger postures, extract a stable and complete central area of ​​the palm, and ensure that the ROI and palm ratio are ≤80%, including the finger gaps and palm edges.

[0074] 3. This invention can identify the finger direction and uniformly correct the vertical upward direction of the palm, ensuring the consistency of the ROI extraction results.

[0075] This invention can be widely applied to palmprint and palm vein acquisition terminals, mobile identification devices, and security identity verification systems, significantly improving image quality control capabilities and system adaptability in the preprocessing stage of palmprint and palm vein recognition. Attached Figure Description

[0076] Figure 1 A schematic diagram showing the palm centroid and ROI region extracted using the existing extraction method 1;

[0077] Figure 2 This is a schematic diagram of ROI region extraction using the existing extraction method 2;

[0078] Figure 3 This is a schematic diagram of ROI region extraction using the existing extraction method 3;

[0079] Figure 4 This is a schematic diagram of ROI region extraction using the existing extraction method 4;

[0080] Figure 5 This is the binary image disclosed in this invention;

[0081] Figure 6 The centroid and boundary line of the palm region disclosed in this invention;

[0082] Figure 7 The smoothed Euclidean distance function between the palm boundary point and the palm centroid point is disclosed in this invention.

[0083] Figure 8 This is a schematic diagram illustrating the determination of candidate palm valley points as disclosed in this invention.

[0084] Figure 9 The coordinates of the palm valley point disclosed in this invention;

[0085] Figure 10 This is a schematic diagram showing the location of the palm valley point on the palm as disclosed in this invention;

[0086] Figure 11 The finger crease valley point disclosed in this invention;

[0087] Figure 12 The coordinates of the four vertices and the extracted ROI region are extracted for the ROI disclosed in this invention;

[0088] Figure 13 This is a flowchart of the palm vein and palm print ROI extraction method based on multi-feature point fusion disclosed in this invention;

[0089] Figure 14 This is a framework diagram of the palm vein and palm print ROI extraction system based on multi-feature point fusion disclosed in this invention.

[0090] Figures 15a-15cThis is a schematic diagram of palm valley point localization, finger gap valley point localization, and ROI extraction disclosed in Embodiment 1 of the present invention;

[0091] Figures 16a-16c This is a schematic diagram of palm valley point localization, finger gap valley point localization, and ROI extraction disclosed in Embodiment 3 of the present invention;

[0092] Figures 17a-17c This is a schematic diagram of palm valley point localization, finger gap valley point localization, and ROI extraction disclosed in Embodiment 3 of the present invention;

[0093] Figures 18a-18c This is a schematic diagram of palm valley point localization, finger gap valley point localization, and ROI extraction disclosed in Embodiment 4 of the present invention;

[0094] Figures 19a-19c This is a schematic diagram of palm valley point localization, finger gap valley point localization, and ROI extraction disclosed in Embodiment 5 of the present invention;

[0095] Figures 20a-20c This is a schematic diagram of palm valley point localization, finger gap valley point localization, and ROI extraction disclosed in Embodiment 6 of the present invention; Detailed Implementation

[0096] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0097] The present invention will now be described in further detail with reference to the accompanying drawings:

[0098] like Figure 13 As shown, this invention provides a method and system for palm vein and palmprint ROI extraction based on multi-feature point fusion. It employs multi-feature point localization (including localization of the palm edge, valley points between fingers, and the center of the palm) to determine the center point for palm ROI extraction, extracting the side length of the square to ensure compliance with national standards for ROI proportion. It also fuses information from multiple feature points to reduce over-reliance on the accuracy of individual feature point localization. Specifically, it includes:

[0099] Step 1: Hand image preprocessing: Obtain a grayscale image including palm prints and veins, convert the grayscale image into a binary image, and separate the palm region from the background.

[0100] Specifically, it includes:

[0101] (1) Read the palm print and palm vein grayscale image in grayscale mode (single channel) to reduce the subsequent computational complexity;

[0102] (2) Use a Gaussian kernel (generally (3,3) is selected as the kernel size) to smooth the grayscale image of palm prints and veins to suppress noise;

[0103] (3) Calculate the grayscale histogram of the grayscale image. Based on the shape of the grayscale histogram, dynamically select a threshold T (to maximize the inter-class variance) to convert the grayscale image into a binary image, separating the palm region from the background, such as... Figure 5 As shown; where the foreground pixel (palm part) is set to 1 and the background pixel is set to 0.

[0104] Step 2, Hand centroid and boundary extraction: Calculate the centroid of the hand region and extract the hand boundary line based on the binary image from Step 1;

[0105] Specifically, it includes:

[0106] (1) Calculate the spatial moments of the binary image th to describe the geometric distribution characteristics of the image;

[0107]

[0108] In the formula,

[0109] I(x,y): The value of pixel (x,y) (0 or 255 in a binary image);

[0110] p, q: The order of the moment (e.g., m10 is the first moment, m00 is the zero moment).

[0111] (2) Calculate the centroid coordinates (x, y) of the palm region. c ,y c ), that is, the "center point" of the region, such as Figure 6 The green crosshair;

[0112]

[0113] (3) Image morphology erosion is performed using a 3×3 cross-shaped kernel. The eroded image is subtracted from the original binary image to obtain boundary pixels. Contours are extracted from the boundary images, and the longest one is selected as the palm boundary line (i.e., the main palm contour). Figure 6 The red palm border.

[0114] Step 3, Palm Valley Point Localization: Fit a smooth Euclidean distance function between the palm boundary points and the palm centroid (e.g., ... Figure 7 The blue curve (in the middle) is used to calculate the second derivative of the function at each node. Nodes with derivatives greater than zero are candidate palm valleys (e.g., the blue curve). Figure 8(Points corresponding to the blue line segment above the middle horizontal axis); use the DBSCAN algorithm to cluster the candidate palm valley points to obtain the coordinates of the palm valley points (e.g., ...). Figure 9 , Figure 10 ).

[0115] Step 4: Finger Gap Valley Point Location: Based on the palm valley points and the palm centroid, calculate the first principal component vector of the PCA with the palm centroid as the origin to obtain the palm extension direction (i.e., the direction of the middle three fingers); and select the three nodes closest to the first principal component vector of the PCA from the palm valley point vector (the vector pointing from the palm centroid as the origin to the valley point), to obtain the finger gap valley points between the index finger, middle finger, ring finger, and little finger (e.g., ...). Figure 11 ).

[0116] Step 5, Rotation Correction (Fingertips Up): With the center of the palm as the center, the angle between the vector connecting the center of the palm and the valley point between the middle and ring fingers and the vertical upward direction of the image is the rotation angle. After correction, the image with fingertips up is obtained.

[0117] Specifically, it includes:

[0118] (1) Calculate the direction vector from the palm to the middle finger.

[0119] (2) Normalize the length of the direction vector

[0120] (3) Calculate the angle between the direction vector and the vertical upward direction to obtain the rotation angle of the palm image.

[0121] (4) Rotate the image around the center of mass of the palm while keeping the image size constant.

[0122] Step 6: Determine the coordinates of the four vertices for ROI extraction: Calculate the distance from all valley points to the center, exclude 20% of outliers (10% maximum and 10% minimum), and obtain the average distance avg_radius. By setting the adjustment factor k (empirical value is 2.5), obtain the ROI extraction side length side_length = k * avg_radius. With the centroid of the hand as the center point, calculate the coordinates of the four vertices for ROI extraction (e.g., ...). Figure 12 ).

[0123] like Figure 14 As shown, this invention provides a palm vein and palm print ROI extraction system based on multi-feature point fusion, comprising:

[0124] A palm image preprocessing module is used to implement step 1 above;

[0125] The palm centroid and boundary extraction module is used to implement step 2 above;

[0126] The palm valley point positioning module is used to implement step 3 above;

[0127] The finger gap valley point positioning module is used to implement step 4 above;

[0128] A rotation correction module is used to implement step 5 above;

[0129] The ROI extraction and four-vertex coordinate determination module is used to implement step 6 above.

[0130] To verify the effectiveness of this method, images of the following hand poses and orientations were selected for ROI extraction based on factors such as hand posture (fingers extended or closed), hand direction (vertical upward or horizontal to the left), and whether rings were worn on the fingers.

[0131] Example 1: Extend your fingers vertically upwards.

[0132] According to the palm vein and palm print ROI extraction method of the present invention, the obtained palm valley point localization (red x point) is as follows: Figure 15a As shown, the finger gap valley point is located (green x point) as follows. Figure 15b As shown, ROI extraction is as follows Figure 15c As shown.

[0133] Example 2: Fingers together, pointing vertically upwards.

[0134] According to the palm vein and palm print ROI extraction method of the present invention, the obtained palm valley point localization (red x point) is as follows: Figure 16a As shown, the finger gap valley point is located (green x point) as follows. Figure 16b As shown, ROI extraction is as follows Figure 16c As shown.

[0135] Example 3: Extend your fingers horizontally to the left.

[0136] According to the palm vein and palm print ROI extraction method of the present invention, the obtained palm valley point localization (red x point) is as follows: Figure 17a As shown, the finger gap valley point is located (green x point) as follows. Figure 17b As shown, ROI extraction is as follows Figure 17c As shown.

[0137] Example 4: Fingers together, pointing horizontally to the left

[0138] According to the palm vein and palm print ROI extraction method of the present invention, the obtained palm valley point localization (red x point) is as follows: Figure 18a As shown, the finger gap valley point is located (green x point) as follows. Figure 18b As shown, ROI extraction is as follows Figure 18c As shown.

[0139] Example 5: Wearing a ring with the finger extended, facing horizontally to the left.

[0140] According to the palm vein and palm print ROI extraction method of the present invention, the obtained palm valley point localization (red x point) is as follows: Figure 19a As shown, the finger gap valley point is located (green x point) as follows. Figure 19b As shown, ROI extraction is as follows Figure 19c As shown.

[0141] Example 6: Wearing the ring with fingers together, facing horizontally to the left.

[0142] According to the palm vein and palm print ROI extraction method of the present invention, the obtained palm valley point localization (red x point) is as follows: Figure 20a As shown, the finger gap valley point is located (green x point) as follows. Figure 20b As shown, ROI extraction is as follows Figure 20c As shown.

[0143] The test results above demonstrate that the ROI extraction achieved by the method of this invention meets the national standard requirements for ROI occupancy, specifically, the palm ROI includes the finger gaps, the right side includes the right edge of the hand, and the left side includes the left edge of the index finger, with the complete palm ROI accounting for ≤80% of the image. Furthermore, the palm print and vein acquisition images undergo a uniform deflection process in the vertically upward direction of the fingers, ensuring that the orientation of the obtained ROI region images remains consistent.

[0144] The advantages of this invention are:

[0145] (1) The method for extracting the ROI of the palm has good robustness to palm posture and different angles of the palm.

[0146] (2) The ROI extraction results of the palm are uniformly corrected to the palm being extended vertically upward.

[0147] (3) The palm ROI results extracted by this method conform to the palm ROI extraction standard in the national industry standard "Technical Specification for Multimodal Palmprint and Palm Vein Recognition System for Biometric Recognition".

[0148] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A palm vein palm print ROI extraction method based on multi-feature point fusion, characterized in that, Comprise: Step 1, obtain the gray image including palm print and palm vein, convert the gray image into binary image, separate palm area from background; Step 2, calculate the centroid of palm area and extract palm boundary line based on the binary image of step 1; specifically comprising: (1) calculating spatial moments of a binary image m pq ; ； In the formula, I ( x , y ): pixel ( x , y The value of ); p , q : order of the moment; (2) calculating the centroid coordinates of the palm region x c , y c ) ； (3) use 3*3 cross-shaped structure kernel to perform image morphological erosion, subtract the eroded image from the original binary image to obtain the boundary pixels, extract the contour from the boundary image, and select the longest one as the palm boundary line; Step 3, fit the palm boundary point and palm centroid point smooth Euclidean distance function, calculate the second derivative of the function at each node, and the node with derivative greater than zero is the candidate palm valley point; use DBSCAN algorithm to cluster the candidate palm valley points to obtain the palm valley point coordinates; Step 4, according to the palm valley point and the palm centroid, take the palm centroid as the origin, calculate the first principal component vector of PCA to obtain the palm stretching direction; and select the three nodes closest to the first principal component vector of PCA from the palm valley point vector to obtain the finger gap valley points between index finger, middle finger, ring finger and little finger; Step 5, take the palm centroid as the center, the angle between the centroid and the vector connecting the valley point between middle finger and ring finger and the vertical upward direction of the image is the rotation angle, and the corrected image with fingertips upward is obtained; Step 6, calculate the distance of all valley points to the center, exclude a preset number of abnormal points, and obtain the average distance; based on the average distance, obtain the ROI extraction edge length; take the palm centroid as the center point, and obtain the ROI extraction four vertex coordinates based on the ROI extraction edge length.

2. The palm vein palm print ROI extraction method of claim 1, wherein, The step 1 specifically comprises: (1) read the palm print and palm vein gray image in gray mode; (2) smooth the palm print and palm vein gray image using Gaussian kernel; (3) calculate the gray histogram of the gray image, dynamically select a threshold T according to the shape of the gray histogram, convert the gray image into binary image, and separate the palm area from the background; wherein the foreground pixels are set to 1 and the background pixels are set to 0.

3. The palm vein palm print ROI extraction method of claim 1, wherein, In the step 6, exclude 10% of the abnormal points with the maximum distance and the minimum distance.

4. The palm vein palm print ROI extraction method of claim 1, wherein, In the step 6, the calculation formula of the ROI extraction edge length side_length is: side_length=k*avg_radius In the formula, avg_radius is the average distance, and k is the adjustment coefficient.

5. A palm vein palm print ROI extraction system based on multi-feature point fusion, characterized in that, Comprise: A palm image preprocessing module for obtaining a gray image including palm print and palm vein, converting the gray image into a binary image, and separating the palm area from the background; A palm centroid and boundary extraction module for calculating the centroid of the palm area and extracting the palm boundary line based on the binary image; specifically comprising: (1) calculating spatial moments of a binary image m pq ; ； In the formula, I ( x , y ): pixel ( x , y The value of ); p , q : order of the moment; (2) calculating the centroid coordinates of the palm region x c , y c ) ； (3) use 3*3 cross-shaped structure kernel to perform image morphological erosion, subtract the eroded image from the original binary image to obtain the boundary pixels, extract the contour from the boundary image, and select the longest one as the palm boundary line; The palm valley point positioning module is configured to fit a smooth Euclidean distance function of palm boundary points and palm centroid points, calculate second-order derivatives of the function at each node, and determine that a node with a derivative greater than zero is a candidate palm valley point; and use a DBSCAN algorithm to cluster the candidate palm valley points to obtain palm valley point coordinates. The finger gap valley point positioning module is configured to calculate a first principal component vector of PCA based on the palm valley points and the palm centroid, taking the palm centroid as the origin, to obtain a palm stretching direction; and select three nodes closest to the first principal component vector of PCA from the palm valley point vector to obtain finger gap valley points between the index finger, the middle finger, the ring finger, and the little finger. The rotation correction module is configured to take the palm centroid as the center, and correct the image with the index finger and the ring finger valley point connecting vector and the vertical upward direction of the image as the rotation angle, to obtain an image with the fingertips upward. The ROI extraction four vertex coordinate determination module is configured to calculate distances from all valley points to the center, and obtain an average distance after excluding a preset number of abnormal points; based on the average distance, obtain an ROI extraction side length; and take the palm centroid as the center point, and obtain ROI extraction four vertex coordinates based on the ROI extraction side length.

6. The palm vein palm print ROI extraction system of claim 5, wherein, The palm image preprocessing module is specifically configured to: (1) read a palm print and vein grayscale image in a grayscale mode; (2) smooth the palm print and vein grayscale image using a Gaussian kernel; (3) calculate a grayscale histogram of the grayscale image, dynamically select a threshold T according to a shape of the grayscale histogram, convert the grayscale image into a binary image, and separate a palm region from a background; wherein a foreground pixel is set as 1, and a background pixel is set as 0.

7. The palm vein palm print ROI extraction system of claim 5, wherein, In the ROI extraction four vertex coordinate determination module, 10% of abnormal points with the maximum distance and the minimum distance are excluded.

8. The palm vein palm print ROI extraction system of claim 5, wherein, In the ROI extraction four vertex coordinate determination module, a calculation formula of the ROI extraction side length side_length is: side_length=k*avg_radius In the formula, avg_radius is the average distance, and k is an adjustment coefficient.

Images