A lightweight finger vein image recognition method

Abstract

The present invention relates to the technical field of biometric feature processing and identification methods, and discloses a lightweight finger vein image recognition method for identifying the characteristics of specific finger veins, comprising the following steps: S1, acquiring a user's finger vein image; S2, calculating the finger's offset angle relative to the horizontal midline using an edge detection method, fitting the image midline to the horizontal midline through image rotation, and obtaining the required horizontal finger vein grayscale image by region-of-interest (ROI); S3, performing image enhancement by dividing the grayscale image into a certain number of grayscale squares of equal size and calculating the grayscale value histogram within the squares for equalization; S4, performing bilinear interpolation on the enhanced grayscale image under lightweight optimization, and then performing adaptive local threshold binarization to obtain a filtered finger vein image in the grayscale image. The present invention improves the security and efficiency of identity authentication and reduces the memory required, processor performance requirements, and production costs of the program in actual use.

Description

Technical Field

[0001] The present invention relates to the technical field of biometric processing and recognition methods, and in particular to a lightweight finger vein image recognition method. Background Art

[0002] In the era of digitalization and informatization, industries are increasingly demanding the security of individual identity verification and the high confidentiality of personal information. Biometric recognition utilizes individual identity characteristics as a marker for authentication, resolving a range of security and accuracy issues compared to traditional authentication methods. First-generation biometric technologies such as fingerprint, facial, and iris recognition have been successfully applied in various scenarios. However, these technologies all have certain drawbacks, such as the susceptibility of fingerprints to theft, the instability of facial recognition, and the inconvenience of iris recognition, all of which have varying degrees of impact on the final recognition effect.

[0003] Finger vein recognition technology extracts vein features from near-infrared images and uses a series of image processing methods to match them with samples in a feature library to achieve identity authentication. Due to the inherent characteristics of vein images and their liveness, it is more difficult to forge and more efficient than fingerprint recognition, the first generation of biometric recognition technology. Summary of the Invention

[0004] The purpose of the present invention is to address the shortcomings of the prior art and to propose a lightweight finger vein image recognition method that can effectively improve the security and efficiency of identity authentication.

[0005] A lightweight finger vein image recognition method for identifying features of specific finger veins includes the following steps:

[0006] S1, collecting user's finger vein image;

[0007] S2. Calculate the finger's relative horizontal midline offset angle using edge detection, rotate the image to fit the image midline to the horizontal midline, and cut the ROI into the region of interest (ROI) to obtain the horizontal finger vein grayscale image required for identification.

[0008] S3, by cutting the grayscale image into a certain number of grayscale squares of the same size and calculating the grayscale value histogram in the squares for equalization to perform image enhancement;

[0009] S4, performing bilinear interpolation under lightweight optimization on the enhanced grayscale image, and then performing adaptive local threshold binarization to obtain the filtered finger vein image in the grayscale image, to avoid global changes caused by inconsistent light and dark contrast;

[0010] S5. Processing the edge image of the finger vein using median filtering and Gabor filtering, and eliminating image noise to obtain a binary texture image of the finger vein;

[0011] S6. Match the finger vein image with the template image through texture-based feature extraction. If the matching degree between the two is higher than a certain threshold, the recognition is passed.

[0012] Furthermore, in S2, obtaining the horizontal finger vein grayscale image to be identified specifically includes:

[0013] S21, by comparing the difference between the background image pixels and the finger vein image pixels, iterating with the edge of the background image as the starting point to find the edge of the finger vein image, and continuously looping to obtain the finger vein contour matrix;

[0014] S22, calculating the edge tilt angle relative to the horizontal midline by using a matrix;

[0015] S23, rotating the entire image by this angle to obtain the desired image;

[0016] S24. Based on this image, the horizontal midline is shifted up and down again to find the image edge and cut the ROI area.

[0017] Furthermore, in S3, the image enhancement process specifically includes: firstly, dividing the entire image into m*n small blocks in equal proportion according to the size of the original finger vein image, and calculating the grayscale value histogram for each small block.

[0018] Furthermore, in S4, the filtered finger vein image is specifically implemented by the following steps:

[0019] S41, connecting the m*n local images by using a lightweight bilinear interpolation method in the finger vein image through the original coordinate values ​​to achieve a smooth transition edge effect;

[0020] S42 , calculating the grayscale histogram of the smoothed finger vein image again, and performing binarization processing on the finger vein image according to the adaptive threshold binarization.

[0021] Furthermore, in S41, the lightweight bilinear interpolation method is specifically implemented as follows: the calculation of a large number of floating-point numbers that would originally be generated in the bilinear interpolation is performed by scaling the coordinate values, so as to achieve the effect that the implementation process in the algorithm is all integer and the algorithm accuracy is not reduced.

[0022] Furthermore, in S42, the adaptive threshold binarization is specifically implemented as follows: the adaptive threshold idea of ​​maximum inter-class variance, wherein the grayscale distribution of the foreground and background will produce two grayscale peaks in the grayscale histogram, and a threshold is selected from the peaks of the foreground and background. The finger vein image is overall identified by this threshold and finally the binarization effect is achieved.

[0023] Furthermore, in said S5, the process of acquiring the binary texture image of the finger vein includes: for the binary finger vein image generated in step S4, there are some pixels and small areas that are identified as foreground, and median filtering and Gabor filtering are performed to smooth the image edges and eliminate some noise pixels.

[0024] Furthermore, in S6, the matching between the finger vein image and the template image is specifically implemented by performing zero padding on the edges of the finger vein image to expand the image size, and using the template to traverse the finger vein image pixel by pixel to calculate the matching.

[0025] The technical solution of the present invention has the following advantages over the prior art: before performing image enhancement on the finger vein image, the present invention needs to dynamically identify the finger vein angle and make corresponding adjustments to the overall angle; in the bilinear interpolation link, the bilinear interpolation is correspondingly optimized to adapt to the portability of the lightweight environment enhancement method; finally, the accuracy of finger vein recognition is increased through adaptive template matching, which can effectively improve the security and efficiency of identity authentication; and through the image matching method and pre-modulation of Gabor parameters, the memory required for the program in actual use and the processor performance requirements are effectively reduced, thereby reducing the actual production cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a flow chart of a lightweight finger vein image recognition method according to the present invention;

[0027] Figure 2 The finger vein image obtained by the present invention is collected;

[0028] Figure 3 This is an example template image obtained after being processed by the recognition method of the present invention. DETAILED DESCRIPTION

[0029] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0030] According to the embodiment of this application, refer to Figure 1 The present invention proposes a lightweight finger vein image recognition method, comprising the following steps:

[0031] S1. Collect the user's finger vein image, e.g. Figure 2 , the specific steps are:

[0032] S11. The finger is limited by a fixed limiting device, and the deviation angle of the finger vein is reduced by physical limiting, thereby achieving the purpose of reducing the calculation amount of the entire invention method;

[0033] S12. Input the obtained finger vein image into the algorithm.

[0034] S2. Calculate the finger's offset angle relative to the horizontal midline using an edge detection method, rotate the image to fit the image midline to the horizontal midline, and obtain the required horizontal finger vein grayscale image by region cutting (ROI). The specific steps of the method are as follows:

[0035] S21. Perform edge detection on the finger vein image. Because the grayscale values ​​of the background and the finger vein grayscale image differ significantly, identify the image from its corners and set the boundary where the grayscale gradient changes as the edge of the finger vein image. Save the edge coordinate information in memory for use in the next step.

[0036] S22, rotating the finger vein image, the specific process is: by calculating the offset angle of the image edge based on the left information in step S21, correcting the offset angle by rotating the image, and overlaying the corrected image on the original image;

[0037] S23, ROI area cutting specific process is: the rotated image obtained in step S22 is once again subjected to edge detection in step S21, the redundant part of the image is cropped using the saved edge coordinate values, and only the finger vein image coverage area is retained, thereby reducing the overall method calculation amount by using a smaller overall pixel amount method.

[0038] S3. Image enhancement is performed by cutting the grayscale image into a certain number of grayscale squares of the same size and calculating the grayscale value histogram in the squares for equalization. The specific implementation process is as follows:

[0039] S31, cutting the grayscale finger vein image into a certain number of grayscale squares of the same size. The specific process is: setting a preset value m*n according to the aspect ratio of the original finger vein image, and cutting the entire finger vein image into m*n vein sub-images;

[0040] The value of m*n is an empirical value, and the recognition effect of the entire method is further debugged by selecting grid ranges of different sizes.

[0041] S32 performs histogram equalization on each grayscale vein sub-image. The specific process is as follows: first, the grayscale values ​​of the grayscale vein sub-image are counted to obtain a grayscale histogram; second, the grayscale distribution probability of the original grayscale vein sub-image is calculated by the total number of pixels in the entire image, thereby obtaining the grayscale cumulative distribution probability; finally, the grayscale cumulative distribution probability of each grayscale value is mapped to obtain the final corresponding mapped grayscale value, and then this mapping relationship is applied to the entire grayscale vein sub-image.

[0042] Since the above process operates on grayscale vein sub-images, it will cause excessive non-smoothness between sub-images, so the next step is to smooth the edges of the sub-images.

[0043] S4. The enhanced grayscale image is first subjected to bilinear interpolation under lightweight optimization and then binarized by adaptive local thresholding to obtain the vein image filtered in the grayscale image to avoid global changes caused by inconsistent light and dark contrast. The specific process is as follows:

[0044] S41. Bilinear interpolation under lightweight optimization of grayscale images, the specific implementation process is as follows: Since the original bilinear interpolation is performed by multiplying pixels diagonally opposite to each other, the pixel values ​​in the operation are all floating-point numbers. In actual operation scenarios, floating-point calculations not only increase the amount of calculations compared to integer operations but also limit the operation speed. In the present invention, by limiting the operation unit to integer, the original 0-255 is expanded by 2048 times. This not only prevents the pixel operation accuracy from decreasing, but also prevents single-point pixels from exceeding the maximum integer range, and also improves the calculation speed of the overall vein submap;

[0045] S42. The specific process of performing adaptive local threshold binarization on the grayscale vein sub-image is as follows: based on the premise that the grayscale value distribution of the grayscale image background and foreground is quite different, the grayscale histogram of the current grayscale vein image is first calculated using the maximum inter-class difference method, and an adaptive threshold is selected based on the inter-class difference between the foreground and background. Finally, this threshold is applied to the entire image, and the image is binarized based on the threshold.

[0046] Although image enhancement and adaptive threshold binarization have created a clear boundary between the vein and the image, some edge glitches and pixel noise may occur at the edge of the vein or during shooting. The next step is to eliminate these factors that affect the judgment.

[0047] S5. Use median filtering and Gabor filtering to process the vein edge and eliminate image noise to obtain a binary texture image of the vein. The specific process is as follows: Median filtering is specifically: for the pixel noise generated in the predecessor process and the influencing factors generated in the shooting process, first median filtering is performed on the finger vein image, and the entire finger vein image is filtered through the filter; Gabor filtering is specifically: according to the Gabor filter production formula, filter parameters are generated, and the vein image edge is smoothed by parameter debugging. The final finger vein image processing result is as follows Figure 3 As shown;

[0048] S6. Match the vein image with the template image through texture-based feature extraction. If the matching degree between the two is higher than a certain threshold, the recognition is passed. The specific process is as follows: first, the finger vein image is padded with zeros to 1 / 2 of the original size to prevent horizontal deviation of the finger vein; second, the templates are taken out from the template library in sequence, and the template vein image is used to overlay the zero-padded finger vein image as a sub-image. Each time it is overlaid, the gamma value is calculated by the following formula (where f(x, y) represents the pixel value of the vein image sub-image, and t(u, v) represents the pixel value of the template image) and stored in the memory. After each gamma value is calculated, the template image is shifted right by k pixels and the gamma value is calculated again.

[0049]

[0050] Finally, the maximum gamma value of this image is returned as the template value of the finger vein image and this template and recorded in the memory. After traversing the above steps for all templates, the maximum template value is returned, and it is judged whether the template value exceeds the threshold (by the maximum inter-class difference method). If it exceeds the threshold, the template number at this time is returned and the match is successful. If it does not exceed the threshold, the match fails and there is no such finger vein homologous image in the template library.

[0051] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A lightweight finger vein image recognition method, characterized in that: The feature recognition for a specific finger vein includes the following steps: S1, collecting user's finger vein image; S2. Calculate the finger's relative horizontal midline offset angle using edge detection, rotate the image to fit the image midline to the horizontal midline, and cut the ROI into the region of interest (ROI) to obtain the horizontal finger vein grayscale image required for identification. S3, by cutting the grayscale image into a certain number of grayscale squares of the same size and calculating the grayscale value histogram in the squares for equalization to perform image enhancement; S4, performing bilinear interpolation under lightweight optimization on the enhanced grayscale image, and then performing adaptive local threshold binarization to obtain the filtered finger vein image in the grayscale image, to avoid global changes caused by inconsistent light and dark contrast; S5. Processing the edge image of the finger vein using median filtering and Gabor filtering, and eliminating image noise to obtain a binary texture image of the finger vein; S6. Match the finger vein image with the template image through texture-based feature extraction. If the matching degree between the two is higher than a certain threshold, the recognition is passed.

2. A lightweight finger vein image recognition method according to claim 1, characterized in that: In S2, obtaining the grayscale image of the horizontal finger veins to be identified specifically includes: S21, by comparing the difference between the background image pixels and the finger vein image pixels, iterating with the edge of the background image as the starting point to find the edge of the finger vein image, and continuously looping to obtain the finger vein contour matrix; S22, calculating the edge tilt angle relative to the horizontal midline by using a matrix; S23, rotating the entire image by this angle to obtain the desired image; S24. Based on this image, the horizontal midline is shifted up and down again to find the image edge and cut the ROI area.

3. The lightweight finger vein image recognition method according to claim 1, characterized in that: In S3, the image enhancement process specifically includes: firstly, dividing the entire image into m*n small blocks in equal proportion according to the size of the original finger vein image, and calculating the grayscale value histogram for each small block.

4. A lightweight finger vein image recognition method according to claim 3, characterized in that: In S4, the steps of implementing the filtered finger vein image include: S41, connecting the m*n local images by using a lightweight bilinear interpolation method in the finger vein image through the original coordinate values ​​to achieve a smooth transition edge effect; S42 , calculating the grayscale histogram of the smoothed finger vein image again, and performing binarization processing on the finger vein image according to the adaptive threshold binarization.

5. A lightweight finger vein image recognition method according to claim 4, characterized in that: In S41, the lightweight bilinear interpolation method is specifically implemented as follows: the calculation of a large number of floating-point numbers that would originally be generated in the bilinear interpolation is performed by scaling the coordinate values, so that the implementation process in the algorithm is all integer-typed without reducing the accuracy of the algorithm.

6. A lightweight finger vein image recognition method according to claim 4, characterized in that: In the above-mentioned S42, the adaptive threshold binarization is specifically implemented as follows: the adaptive threshold concept of maximum inter-class variance, wherein the grayscale distribution of the foreground and background will produce two grayscale peaks in the grayscale histogram, and a threshold is selected from the foreground and background peaks. The finger vein image is overall identified by this threshold and finally the binarization effect is achieved.

7. The lightweight finger vein image recognition method according to claim 1, characterized in that: In step S5, the process of acquiring the binary texture image of the finger vein includes: for the binary finger vein image generated in step S4, some pixels and small areas are identified as foreground, and median filtering and Gabor filtering are performed to smooth the image edge and eliminate some noise pixels.

8. The lightweight finger vein image recognition method according to claim 1, characterized in that: In S6, the matching between the finger vein image and the template image is specifically implemented by performing zero padding on the edges of the finger vein image to expand the image size, and using the template to traverse the finger vein image pixel by pixel to calculate the matching.

Images