A method for enhancing and segmenting hand vein infrared images based on AGC and Frangi

Abstract

The present invention relates to the medical technology field of hand vein image enhancement and segmentation, and discloses a hand vein infrared image enhancement and segmentation method based on AGC and Frangi. The method first collects hand vein images under infrared light irradiation, then guides and filters the initial image to sharpen the image edge, and then stretches the image contrast through AGC, then performs multi-scale Gaussian filtering and noise reduction on the processed image, and then uses the Frangi algorithm to identify blood vessels on the processed image to extract the tubular blood vessel contour, and finally performs local binary segmentation and open operation noise reduction on the image in turn to obtain the segmented blood vessel shape with obvious visual effect. The present invention innovatively combines the AGC algorithm with the Frangi algorithm, and the combination of the two can effectively reduce the breakpoints generated after Frangi processing, thereby improving the overall image enhancement and segmentation effect.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to an enhancement and segmentation method for hand vein infrared images based on AGC and Frangi. Background Art

[0002] Venous infrared image processing technology has a wide range of applications in the medical field. It can be used to observe the patient's vein image, help medical staff better perform venous puncture, and improve puncture efficiency.

[0003] In the prior art, there are mainly the following methods for enhancing venous infrared images: methods for improving image contrast, methods for enhancing image texture features, and methods based on image fusion. Methods for improving image contrast are mainly various improved histogram equalization methods; methods for enhancing image texture features are mainly based on a series of filtering technologies; and methods based on image fusion are mainly to fuse multiple images processed by different methods. Since images processed by different methods have obvious effects on certain features, image fusion can be used to superimpose the areas with obvious enhancement effects of each image, while reducing the noise part of a single image, thereby obtaining an image with better effect than each initial image.

[0004] The inventors of this application discovered during the implementation of the present invention that the vein image enhancement method in the prior art has the following technical problems:

[0005] It is limited to processing the vein image of a certain part, and the processing effects of vein images of different parts vary greatly, and is accompanied by the problem of large noise. Summary of the invention

[0006] The present invention provides a segmentation method for hand vein infrared images based on AGC and Frangi, which is used to solve or at least partially solve the technical problems existing in the prior art that the processing effects of vein images of different parts are quite different and accompanied by relatively large noise.

[0007] The present invention discloses a method for enhancing and segmenting hand vein infrared images based on AGC and Frangi, comprising:

[0008] S1: Collect the hand vein image under infrared light as the initial image;

[0009] S2: Perform guided filtering on the initial image to obtain the enhanced image P1;

[0010] S3: Applying the AGC algorithm to perform contrast stretching on the enhanced image P1 to obtain an adaptively processed image P2, wherein the AGC algorithm is an adaptive gamma correction algorithm;

[0011] S4: performing multi-scale Gaussian filtering denoising on the adaptively processed image P2 to obtain a denoised image P3;

[0012] S5: using the Frangi algorithm to extract blood vessel features from the image P3 after noise reduction, and extracting the tubular blood vessel contour as the image P4 after Frangi algorithm processing;

[0013] S6: performing local threshold binarization segmentation on the image P4 processed by the Frangi algorithm to obtain a segmented image P5;

[0014] S7: Perform an opening operation on the segmented image P5 to reduce noise, and obtain a final image P6.

[0015] In one embodiment, S2 includes:

[0016] Extract the ROI region from the initial image and use the extracted image as the input image;

[0017] Guided filtering is performed on the input image to obtain the enhanced image P1.

[0018] In one embodiment, when the AGC algorithm is used to perform contrast stretching on the enhanced image in step S3, a judgment standard for image contrast is formulated by obtaining the mean and variance of image pixels, and different gamma values ​​are selected according to the judgment standard for image contrast to perform contrast stretching on the enhanced image P1.

[0019] In one embodiment, step S4 includes:

[0020] The three channels of the adaptively processed image P2 are subjected to weighted multi-scale Gaussian filtering.

[0021] The images of the three channels after filtering are fused to obtain the denoised image P3.

[0022] In one embodiment, when performing weighted multi-scale Gaussian filtering on the three channels of the adaptively processed image P2, three Gaussian kernels are selected to perform weighted processing on the image, and the σ values ​​of the three Gaussian kernels are selected as 15, 80, and 200, respectively, and the corresponding weights are 1 / 2, 1 / 4, and 1 / 4, respectively.

[0023] In one embodiment, step S5 includes:

[0024] S5.1: solving the eigenvalues ​​λ1 and λ2 of the Hessian matrix according to the pixel points of the image P3 after the noise reduction processing, wherein the eigenvalue and eigenvector of the Hessian matrix represent the magnitude and direction of the curvature of the image along a certain direction at the point, respectively;

[0025] S5.2: According to the obtained eigenvalues, P3 is divided into three categories: background, isolated points and blood vessels, wherein the extracted blood vessel part is processed into a tubular structure, that is, a tubular blood vessel contour, as the image P4 processed by the Frangi algorithm.

[0026] In one embodiment, step S5.2 comprises:

[0027] When λ1 and λ2 in any direction meet the first preset condition, the grayscale distribution is uniform and classified as the background area;

[0028] When λ1 and λ2 in any direction meet the second preset condition, they are classified as isolated points;

[0029] When λ1 and λ2 in any direction meet the third preset condition, it is classified as a blood vessel area.

[0030] Compared with the prior art, the advantages and beneficial technical effects of the present invention are as follows:

[0031] After collecting the hand vein image under infrared light irradiation, the present invention performs guided filtering on it to sharpen the image edge, and then stretches the image contrast through AGC (adaptive gamma correction), and then performs multi-scale Gaussian filtering noise reduction on the processed image, and then uses the Frangi algorithm to identify blood vessels on the processed image to extract the tubular blood vessel contour, and finally performs local binary segmentation and open operation noise reduction on the image in sequence to obtain the segmented blood vessel shape with obvious visual effect. The present invention innovatively combines the AGC algorithm with the Frangi algorithm, and the combination of the two can effectively reduce the breakpoints generated after the Frangi algorithm is processed. After the Frangi algorithm is processed, there are many breakpoints in the image, so before the Frangi algorithm is processed, the image needs to be pre-processed. In addition to the guided filtering and multi-scale Gaussian filtering used in the present invention, the AGC algorithm is used to more obviously reduce the breakpoints generated after the Frangi processing, thereby improving the processing effect of vein images of different parts and reducing noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 Flow chart of the method for enhancing and segmenting hand vein infrared images based on AGC and Frangi in an embodiment of the present invention;

[0034] Figure 2 4 is a flowchart of an algorithm for multi-scale Gaussian noise reduction processing in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The purpose of the present invention is to provide a segmentation method for hand vein infrared images based on AGC and Frangi, so as to improve the technical problems existing in the prior art that the processing effects of vein images of different parts are quite different and accompanied by large noise.

[0036] The main concepts and innovations of the present invention are as follows:

[0037] The AGC algorithm is innovatively combined with the Frangi algorithm, and the combination of the two can effectively reduce the breakpoints generated after Frangi processing. After the Frangi algorithm is processed, there are many breakpoints in the image. Therefore, before the Frangi algorithm is processed, the image needs to be preprocessed. In addition to the guided filtering and multi-scale Gaussian filtering used in the present invention, the AGC algorithm can be used to more significantly reduce the breakpoints generated after Frangi processing.

[0038] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] The embodiment of the present invention provides a method for enhancing and segmenting hand vein infrared images based on AGC and Frangi, including:

[0040] S1: Collect the hand vein image under infrared light as the initial image;

[0041] S2: Perform guided filtering on the initial image to obtain the enhanced image P1;

[0042] S3: Applying the AGC algorithm to perform contrast stretching on the enhanced image P1 to obtain an adaptively processed image P2, wherein the AGC algorithm is an adaptive gamma correction algorithm;

[0043] S4: performing multi-scale Gaussian filtering denoising on the adaptively processed image P2 to obtain a denoised image P3;

[0044] S5: using the Frangi algorithm to extract blood vessel features from the image P3 after noise reduction, and extracting the tubular blood vessel contour as the image P4 after Frangi algorithm processing;

[0045] S6: performing local threshold binarization segmentation on the image P4 processed by the Frangi algorithm to obtain a segmented image P5;

[0046] S7: Perform an opening operation on the segmented image P5 to reduce noise, and obtain a final image P6.

[0047] See also Figure 1 , which is a flow chart of the enhancement and segmentation method of hand vein infrared image based on AGC and Frangi.

[0048] Specifically, step S1 utilizes infrared light to illuminate the skin surface, collects the infrared radiation image reflected or transmitted by the skin through an infrared camera device, and uses the collected image as an initial vein infrared image (initial image).

[0049] In step S2, after extracting the ROI area from the initial image, the image is used as the input image for guided filtering.

[0050] In step S3, adaptive gamma correction (AGC) is applied to the enhanced image P1 to perform contrast stretching. Compared with the traditional global correction, the AGC algorithm is more effective in increasing the contrast between the image and the background.

[0051] Step S4 performs multi-scale Gaussian filtering on image P2, separates the channels of the image, performs Gaussian filtering with convolution kernels of different sizes, and then selects the optimal weight to perform weighted fusion on the images after Gaussian filtering to obtain image P3.

[0052] Step S5 uses the Frangi algorithm to identify blood vessels on the processed image and extract the tubular blood vessel contour. After Frangi processing, some noise is converted into small tubular structures, and it is still necessary to remove the noise by performing local binary segmentation and opening operation. Therefore, through steps S6 and S7, the image is sequentially subjected to local binary segmentation and opening operation noise reduction, and the segmented blood vessel shape with obvious visual effect is obtained.

[0053] In one embodiment, S2 includes:

[0054] Extract the ROI region from the initial image and use the extracted image as the input image;

[0055] Guided filtering is performed on the input image to obtain the enhanced image P1.

[0056] Among them, the ROI area is the region of interest, and the input image is guided filtered to achieve the effect of enhancing the edge of the image.

[0057] In one embodiment, when the AGC algorithm is used to perform contrast stretching on the enhanced image in step S3, a judgment standard for image contrast is formulated by obtaining the mean and variance of image pixels, and different gamma values ​​are selected according to the judgment standard for image contrast to perform contrast stretching on the enhanced image P1.

[0058] In one embodiment, step S4 includes:

[0059] The three channels of the adaptively processed image P2 are subjected to weighted multi-scale Gaussian filtering.

[0060] The images of the three channels after filtering are fused to obtain the denoised image P3.

[0061] See also Figure 2 , is a schematic diagram of the algorithm flow of the multi-scale Gaussian noise reduction processing of the present invention.

[0062] In one embodiment, when performing weighted multi-scale Gaussian filtering on the three channels of the adaptively processed image P2, three Gaussian kernels are selected to perform weighted processing on the image, and the σ values ​​of the three Gaussian kernels are selected as 15, 80, and 200, respectively, and the corresponding weights are 1 / 2, 1 / 4, and 1 / 4, respectively.

[0063] Through the above processing, the blood vessels of the image obtained after processing are clearer. Compared with the existing multi-scale Gaussian filtering, the innovation of the present invention lies in the selection of σ value and weight. According to a large amount of experimental data, when the number of Gaussian kernels is 3, the image enhancement effect is the best. At the same time, the three σ values ​​are selected to cover the low, medium and high scale spaces, and the optimal weight fusion is performed on them. Different from the existing direct selection of the maximum enhancement value, which is easy to cause information loss, the present invention selects weight fusion to retain as much information as possible.

[0064] In one embodiment, step S5 includes:

[0065] S5.1: solving the eigenvalues ​​λ1 and λ2 of the Hessian matrix according to the pixel points of the image P3 after the noise reduction processing, wherein the eigenvalue and eigenvector of the Hessian matrix represent the magnitude and direction of the curvature of the image along a certain direction at the point, respectively;

[0066] S5.2: According to the obtained eigenvalues, P3 is divided into three categories: background, isolated points and blood vessels, wherein the extracted blood vessel part is processed into a tubular structure, that is, a tubular blood vessel contour, as the image P4 processed by the Frangi algorithm.

[0067] In one embodiment, step S5.2 comprises:

[0068] When λ1 and λ2 in any direction meet the first preset condition, the grayscale distribution is uniform and classified as the background area;

[0069] When λ1 and λ2 in any direction meet the second preset condition, they are classified as isolated points;

[0070] When λ1 and λ2 in any direction meet the third preset condition, it is classified as a blood vessel area.

[0071] Specifically, the first preset condition, the second preset condition and the third preset condition can be adjusted according to circumstances.

[0072] For example, when λ1 and λ2 are both small in any direction (|λ1|≈0, |λ2|≈0), the first preset condition is met, and the grayscale distribution is uniform, which belongs to the background area; when λ1 and λ2 are both large (|λ1|>>0, |λ2|>>0), the second preset condition is met, and it belongs to an isolated point; when λ1 and λ2 are small and large (|λ1|≈0, |λ2|>>|λ1|), the third preset condition is met, and it belongs to the vascular area. Based on this, the vascular area can be partially extracted and processed. Compared with the existing Frangi algorithm, which uses the Hessian matrix eigenvalue to enhance the image, the innovation of the present invention is mainly to pre-process the image with adaptive gamma correction AGC, and combine the Frangi algorithm with the AGC algorithm. After the image is processed by AGC, the problem of uneven image illumination is solved, and the vascular information is enhanced, which reduces the error rate when using Frangi for vascular recognition.

[0073] The hand vein infrared image enhancement and segmentation method based on AGC and Frangi provided by the present invention is described in detail below with reference to specific examples.

[0074] The overall process is as follows Figure 1 As shown in Figure 1, S1, obtains a vein image under infrared light. It can be seen that the obtained image has low contrast and uneven illumination, so the image needs to be further processed.

[0075] S2, use guided filtering to sharpen the image edges and edge-preserving filtering.

[0076] S3, AGC adaptive contrast stretching.

[0077] S4, multi-scale Gaussian noise reduction processing.

[0078] S5, 2D image blood vessel recognition.

[0079] S6, binary segmentation.

[0080] S7, binary noise removal.

[0081] S8, through algorithm processing, outputs images or real-time video.

[0082] Step 1: Obtain a single hand vein image under near-infrared light, extract the ROI region of interest (ROI) of the image, and obtain an image area of ​​320x240.

[0083] Implementation effect: The acquired image contains not only the region of interest of the hand vein, but also redundant information such as background, which will reduce the accuracy of image processing and increase unnecessary calculation in subsequent algorithm processing, thereby reducing the processing speed. Extracting the ROI region of the image can effectively solve this problem.

[0084] Step 2: Perform guided filtering on the ROI area, using the input image as a guided image to achieve the effect of maintaining the edge.

[0085]

[0086] in i (x,y) is the filtered image, G i (x, y) is the guide image, k represents the pixel at a certain point in the guide image, ω k is a local square window with point k as the center and radius r, a k (x,y) and b k (x,y) is the guide image at ω k The linear coefficients within .

[0087] Result: The image obtained in step 1 has problems such as uneven illumination and low contrast, so the image needs to be further processed. Guided filtering is used to enhance the edge of the image. The guided filtering process retains the edge information of the image and suppresses the noise of the image, which has an enhancement effect on the image. The edge of the processed image is deepened, the veins are more obvious, and the image contrast is improved. However, the edges of some noise are also magnified, so further processing is required.

[0088] Step 3: Use adaptive gamma correction (AGC) to stretch the contrast. Convert the image color space to HSV, extract the H component, perform adaptive gamma correction on the H component, and then perform channel fusion. The gamma value is determined according to the characteristics of the image, that is, the image is divided into four types according to the image contrast and brightness.

[0089] Result: Since the image is unevenly illuminated, the traditional grayscale stretching method is a global correction method, which will produce over-bright exposed areas or over-dark shadow areas. In order to overcome the above shortcomings, an adaptive method is used to process the image. The blood vessel part after algorithm processing has a more obvious contrast with the background, which is convenient for subsequent processing.

[0090] Step 4: Perform multi-scale Gaussian blur on step 3, and obtain appropriate Gaussian kernel parameters and corresponding weights through testing and verification. The present invention selects three Gaussian kernels to perform weighted processing on the image. The σk values ​​of the three Gaussian kernels are selected as 15, 80, and 200, respectively, and the corresponding weights are 1 / 2, 1 / 4, and 1 / 4.

[0091] The specific formula is as follows:

[0092]

[0093] Among them, G(x,y,σ) is a Gaussian function, and x, y, and σ are the parameters of the Gaussian function.

[0094]

[0095] Where P(x,y) is the grayscale value of the output image P at the pixel point (x,y), G(x,y,15) is the Gaussian function with a σ value of 15, G(x,y,80) is the Gaussian function with a σ value of 80, G(x,y,200) is the Gaussian function with a σ value of 200, and s(x,y) is the grayscale value of the input source image s at the pixel point (x,y).

[0096] Result: After processing in step 3, a three-channel image with high contrast is obtained. Although the blood vessels are clearer, some noise in the image is also amplified. For this, a multi-scale Gaussian filtering algorithm is used to perform weighted multi-scale Gaussian filtering on the three channels respectively. The best denoising effect is found by continuous debugging, so as to obtain an image with less noise. The images of the three channels after processing are fused.

[0097] Step 5: Use Frangi two-dimensional filter to extract blood vessel features, thereby distinguishing blood vessels from noise and enhancing vein information in the image.

[0098] The specific operation is to solve the Hessian matrix eigenvalues, i.e. λ1 and λ2, for the pixel points in the image. The Hessian matrix eigenvalues ​​and the corresponding eigenvectors represent the magnitude and direction of the curvature of the image along a certain direction at the point.

[0099] The Hessian matrix H of the pixel point (x, y) in the image is as follows,

[0100]

[0101] in is the second-order partial derivative of the image with respect to x, is the second-order partial derivative of the image with respect to y, is the second-order mixed partial derivative of the image with respect to x and y.

[0102] According to the linear scale space theory (LOG), the elements in the Hessian matrix can be solved, and the derivative of the function f(x,y) is equal to the convolution of the function f(x,y) and the derivative of the Gaussian function. That is,

[0103]

[0104]

[0105]

[0106] Where f(x,y) is the gray value of image f at point (x,y), where G(x,y;σ) is the Gaussian function, is the second-order partial derivative of the Gaussian function with respect to x, is the second-order partial derivative of the Gaussian function with respect to y, is the second-order mixed partial derivative of the Gaussian function with respect to x and y.

[0107] After the above processing, the Hessian matrix of the image is obtained, and then the eigenvalues ​​λ1 and λ2 of the Hessian matrix can be obtained (λ1 is selected as a smaller value, that is, λ1<λ2).

[0108] According to the obtained eigenvalues ​​λ1 and λ2, the image morphology can be roughly divided into three categories:

[0109] 1) Background, their grayscale distribution is relatively uniform. The eigenvalues ​​λ1 and λ2 in any direction are small.

[0110] 2) Isolated points, whose eigenvalues ​​λ1 and λ2 in any direction are large.

[0111] 3) Blood vessels: The characteristic values ​​λ1 and λ2 obtained at the blood vessels are one small and one large.

[0112] Construct two parameters R according to the values ​​of λ1 and λ2 B 、S:

[0113]

[0114]

[0115] Among them, λ1 and λ2 are the desired eigenvalues.

[0116] Define the response function v0(R B ,S) is as follows,

[0117]

[0118] in, v0(R B,S) is the feature enhancement value of the pixel (x, y) when the scale factor of the Gaussian function G(x, y; σ) is σ, and the parameters β and c are adjustment coefficients.

[0119] In particular, if β is very large, the suppression of isolated areas will be weakened. If β is very small, A is easily affected by Rb and tends to 0, so it is also easy to be suppressed at the bends of the blood vessels. Parameter c affects the overall smoothness of the filtered image. If c is large, the image becomes smooth. If c is very small, the filtered image becomes more volatile. According to experimental tests, the present invention selects β as 0.9 and c as 23.

[0120] The response function is affected by the σ value, so the present invention selects a series of σ values, and obtains the feature enhancement value of the pixel point (x, y) under different σ values, and selects the maximum feature enhancement value as the feature enhancement value of the pixel point (x, y), and enhances the image according to the obtained feature enhancement value. According to experimental tests, the σ values ​​selected by the present invention are 2, 4, and 6.

[0121] Effect: The image processed by this algorithm can distinguish the blood vessels from the background, forming a tubular structure, and further filter out the noise of non-vascular morphology. There are still some problems with the processed image, such as the unclear distinction between the background and the blood vessel structure in the image, and some noise is converted into smaller tubular structures, which need further processing to remove.

[0122] Step 6: Perform local threshold binarization segmentation (NiBlack) processing, select the neighborhood of rxr for the pixel point, calculate the mean and variance in the neighborhood of rxr, and combine the correction value to obtain the threshold of image segmentation. The calculation formula is as follows:

[0123] T(x,y)=m(x,y)+k*s(x,y)

[0124] Wherein, T(x, y) is the local threshold, m(x, y) is the local mean, and s(x, y) is the local variance. According to the test, the parameters in the present invention are selected as r=20 and k=-0.2.

[0125] Effect: After segmentation, the blood vessels and background are clearly black and white, which is easier to identify and has a much better effect than the original image. When traversing the image, if the domain r×r is full of background, a part of it will be determined as the target after NiBlack calculation, generating pseudo noise. In order to remove the pseudo noise, the image needs to be further processed.

[0126] Step 7: Perform an opening operation on the obtained image. The specific implementation method is to first corrode the image, select the size of the corrosion kernel as 5x5, eliminate small noise points, and then expand the image, and the size of the expansion kernel is 3x3.

[0127] Result: After the opening operation, the noise in the image is significantly reduced, and the size of the blood vessels remains almost unchanged, achieving the purpose of optimizing the image.

[0128] Step 8: Output the processed image to facilitate subsequent comparison.

[0129] The present invention provides an enhancement and segmentation method for hand vein infrared images based on AGC and Frangi, which innovatively combines the AGC algorithm with the Frangi algorithm. The combination of the two can effectively reduce the breakpoints generated after Frangi processing. After the Frangi algorithm is processed, there are many breakpoints in the image. Therefore, before the Frangi algorithm is processed, the image needs to be preprocessed. In addition to the guided filtering and multi-scale Gaussian filtering used in the present invention, the AGC algorithm can be used to more significantly reduce the breakpoints generated after Frangi processing.

[0130] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of patent protection of the present invention. Under the enlightenment of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested for the present invention shall be based on the attached claims.

Claims

1. A method for enhancing and segmenting hand vein infrared images based on AGC and Frangi, characterized in that: include: S1: Collect the hand vein image under infrared light as the initial image; S2: Perform guided filtering on the initial image to obtain the enhanced image P1; S3: Applying the AGC algorithm to perform contrast stretching on the enhanced image P1 to obtain an adaptively processed image P2, wherein the AGC algorithm is an adaptive gamma correction algorithm; S4: performing multi-scale Gaussian filtering denoising on the adaptively processed image P2 to obtain a denoised image P3; S5: using the Frangi algorithm to extract blood vessel features from the image P3 after noise reduction, and extracting the tubular blood vessel contour as the image P4 after Frangi algorithm processing; S6: performing local threshold binarization segmentation on the image P4 processed by the Frangi algorithm to obtain a segmented image P5; S7: Perform an opening operation on the segmented image P5 to reduce noise, and obtain a final image P6.

2. The method for enhancing and segmenting hand vein infrared images based on AGC and Frangi as claimed in claim 1, characterized in that S2 include: Extract the ROI region from the initial image and use the extracted image as the input image; Guided filtering is performed on the input image to obtain the enhanced image P1.

3. The method for enhancing and segmenting hand vein infrared images based on AGC and Frangi as claimed in claim 1, characterized in that: When the AGC algorithm is used to perform contrast stretching on the enhanced image in step S3, the image contrast judgment standard is formulated by obtaining the mean and variance of the image pixels, and different gamma values ​​are selected according to the image contrast judgment standard to perform contrast stretching on the enhanced image P1.

4. The method for enhancing and segmenting hand vein infrared images based on AGC and Frangi as claimed in claim 1, characterized in that: Step S4 includes: The three channels of the adaptively processed image P2 are subjected to weighted multi-scale Gaussian filtering. The images of the three channels after filtering are fused to obtain the denoised image P3.

5. The method for enhancing and segmenting hand vein infrared images based on AGC and Frangi as claimed in claim 4, characterized in that: When performing weighted multi-scale Gaussian filtering on the three channels of the adaptively processed image P2, three Gaussian kernels are selected to perform weighted processing on the image, and the σ values ​​of the three Gaussian kernels are selected as 15, 80, and 200, respectively, and the corresponding weights are 1 / 2, 1 / 4, and 1 / 4, respectively.

6. The method for enhancing and segmenting hand vein infrared images based on AGC and Frangi as claimed in claim 1, characterized in that: Step S5 includes: S5.1: solving the eigenvalues ​​λ1 and λ2 of the Hessian matrix according to the pixel points of the image P3 after the noise reduction processing, wherein the eigenvalue and eigenvector of the Hessian matrix represent the magnitude and direction of the curvature of the image along a certain direction at the point, respectively; S5.2: According to the obtained eigenvalues, P3 is divided into three categories: background, isolated points and blood vessels, wherein the extracted blood vessel part is processed into a tubular structure, that is, a tubular blood vessel contour, as the image P4 processed by the Frangi algorithm.

7. The method for enhancing and segmenting hand vein infrared images based on AGC and Frangi as claimed in claim 6, characterized in that: Step S5.2 includes: When λ1 and λ2 in any direction meet the first preset condition, the grayscale distribution is uniform and classified as the background area; When λ1 and λ2 in any direction meet the second preset condition, they are classified as isolated points; When λ1 and λ2 in any direction meet the third preset condition, it is classified as a blood vessel area.

Images