Non-contact fingerprint identification method based on ridge line repair

Abstract

The invention discloses a non-contact fingerprint identification method based on ridge line repair, and the method comprises the steps: S1, collecting a finger image in a non-contact manner, and carrying out the preprocessing of the finger image, and obtaining a fingerprint ROI image; s2, extracting coordinate information and angle information of feature points in the fingerprint ROI image to obtain a ridge graph; and S3, performing ridge line repair on the ridge line graph by using an RMO algorithm to obtain a fingerprint identification image, and the method can effectively eliminate false feature points on the basis of ridge line repair, thereby improving the feature point information acquisition accuracy and improving the fingerprint identification efficiency.

Description

Technical Field

[0001] The present invention relates to the field of fingerprint recognition, and in particular to a non-contact fingerprint recognition method based on ridge restoration. Background Art

[0002] Compared with contact sampling and recognition, contactless fingerprint recognition has greater advantages in terms of hygiene, safety and user acceptance. For example, it can effectively avoid the unfavorable factors such as the spread of COVID-19 and the nonlinear deformation caused by contact fingerprint samples. By summarizing the characteristics of contactless fingerprints, it is found that the contactless fingerprint sampling obtained by the camera sensor has the disadvantages of low contrast between ridges and valleys, and local and edge blurring of fingerprints. The researchers have also proposed some solutions for this. For example, in order to reduce the blur of the source image, the Wiener filter is used to reduce noise on the basis of filtering the image; Cb+Otsu is used to extract the finger area, and then the image is enhanced by combining high-frequency emphasis filtering and iterative adaptive histogram equalization, and finally the image is enhanced twice with a simplified Gabor filter.

[0003] The above method has a certain effect on improving the overall clarity of the fingerprint image, but it is still difficult to efficiently extract the fingerprint ridges and feature points. On the other hand, the classic fingerprint recognition method is based on matching the endpoints and forks corresponding to the fingerprint ridges. However, due to the ambiguity of non-contact fingerprints, the ridge valley map obtained after fingerprint sample preprocessing has redundant and disordered ridge endpoints and forks, which in turn affects the recognition effect or even fails to recognize. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the non-contact fingerprint recognition method based on ridge restoration provided by the present invention solves the problem that redundant and disordered ridge endpoints and forks exist in the existing non-contact fingerprint recognition method, thereby affecting the recognition effect or even the recognition failure.

[0005] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a non-contact fingerprint recognition method based on ridge restoration, comprising:

[0006] S1, collecting finger images in a non-contact manner and preprocessing them to obtain fingerprint ROI images;

[0007] S2, extracting the coordinate information and angle information of the feature points in the fingerprint ROI image to obtain a ridge map;

[0008] S3. Use the RMO algorithm to repair the ridge line of the ridge line map to obtain a fingerprint recognition image.

[0009] Further: S1 includes:

[0010] S11, collecting finger images in a non-contact manner, using the Cr+Otsu method to perform foreground and background segmentation on the finger images, and obtaining a mask image of only the foreground finger;

[0011] S12, performing image enhancement on the mask image of only the foreground finger to obtain an image of the finger after image enhancement;

[0012] S13, performing adaptive threshold segmentation on the finger image after image enhancement to obtain a fingerprint ROI image.

[0013] Further: In S11, the expression for foreground-background segmentation is:

[0014]

[0015] in, represents the standard deviation of the forewarp and background; M f and M b Respectively represent the global mean, foreground mean and background mean; P f and P b Represent the foreground ratio and background ratio respectively.

[0016] Further: In S12, the expression of image enhancement is:

[0017] I (i,j) =M (i,j) +G×(f (i,j) -M (i,j) )-L

[0018]

[0019] Where f(i,j) and I(i,j) represent the pixel values ​​of the coordinate point (i,j) before and after enhancement, respectively; L and G represent the brightness adjustment value and the contrast enhancement coefficient, respectively; M(i,j) and σ(i,j) represent the mean and standard deviation of the (2n+1)×(2m+1) local area with the coordinate point (i,j), respectively; n and m are both natural numbers.

[0020] Further: In S13, the expression of adaptive threshold segmentation is:

[0021]

[0022] Where I(x,y) and I'(x,y) represent the grayscale values ​​of the pixel with coordinates (x,y) before and after adaptive mean threshold processing, respectively; Σsize_k represents the sum of the grayscale values ​​of the pixels within the k×k range of the pixel with coordinates (x,y), and C represents a constant.

[0023] Further: In S2, the coordinate information of the feature points in the fingerprint ROI image is obtained by the cross number, and its expression is:

[0024]

[0025] Among them, C N (p) represents the crossover value of pixel point p; p(i) represents the neighborhood pixel value of point p; mod represents the remainder operation.

[0026] Further: S3 includes:

[0027] S31, obtain the contour information and limit information of the ridge line graph, and calculate the contour perimeter L outline ;

[0028] S32, determine the contour perimeter L outline Is it less than the threshold Thr? outline :

[0029] If yes, then go to S33;

[0030] If not, proceed to S34;

[0031] S33, filling the outline of the ridge line graph, and then thinning the ridge line graph and entering S34;

[0032] S34, obtain the length L of the ridge line where the ridge line endpoint is located m_ridge , and remove the length L of the ridge line where the ridge line endpoint is located in the feature point set m_ridge Smaller than the contour circumference L outline The ridgeline endpoints of

[0033] S35. Calculate the Euclidean distance D between any two feature points mm , and remove the Euclidean distance D from the feature point set mm Smaller than the contour perimeter L outline The corresponding two feature points;

[0034] S36. Use the remaining feature points in the feature point set to mark the ridges after ridge repair to obtain a fingerprint recognition image.

[0035] The beneficial effects of the present invention are:

[0036] 1. It can effectively remove pseudo feature points on the basis of ridge repair, thereby improving the accuracy of feature point information acquisition and fingerprint recognition efficiency;

[0037] 2. RMO algorithm is used to effectively remove some chaotic ridges and ridge holes by locating the ridge contours and short branch ridges without affecting the original shape of the fingerprint ridges. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 The flowchart of the contactless fingerprint recognition method based on ridge restoration.

[0039] Figure 2 Schematic diagram of the angle information of feature points in the fingerprint ROI image.

[0040] Figure 3 This is a comparison chart of ridge repair. DETAILED DESCRIPTION

[0041] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0042] like Figure 1 As shown, in one embodiment of the present invention, a non-contact fingerprint recognition method based on ridge restoration is provided, comprising:

[0043] S1, collecting finger images in a non-contact manner and preprocessing them to obtain fingerprint ROI images;

[0044] S2, extracting the coordinate information and angle information of the feature points in the fingerprint ROI image to obtain a ridge map;

[0045] S3. Use the RMO algorithm to repair the ridge line of the ridge line map to obtain a fingerprint recognition image.

[0046] The S1 includes:

[0047] S11, collecting finger images in a non-contact manner, using the Cr+Otsu method to perform foreground and background segmentation on the finger images, and obtaining a mask image of only the foreground finger;

[0048] S12, performing image enhancement on the mask image of only the foreground finger to obtain an image of the finger after image enhancement;

[0049] S13, performing adaptive threshold segmentation on the finger image after image enhancement to obtain a fingerprint ROI image.

[0050] Since human skin color is clustered in certain color channel intervals, in the finger image YcCrb pattern diagram, the skin color in the Cr channel component is clustered in the interval [132,144], and the skin color Cr value landing point and the background Cr value landing point are far away and have no intersection, so the two can be cleanly separated. In S11, the expression of foreground and background segmentation is:

[0051]

[0052] in, represents the standard deviation of the forewarp and background; M f and M b Respectively represent the global mean, foreground mean and background mean; P f and P b Represent the foreground ratio and background ratio respectively.

[0053] In S12, the expression of image enhancement is:

[0054] I (i,j) =M (i,j) +G×(f (i,j) -M (i,j) )-L

[0055]

[0056] Where f(i,j) and I(i,j) represent the pixel values ​​of the coordinate point (i,j) before and after enhancement, respectively; L and G represent the brightness adjustment value and the contrast enhancement coefficient, respectively; M(i,j) and σ(i,j) represent the mean and standard deviation of the (2n+1)×(2m+1) local area with the coordinate point (i,j), respectively; n and m are both natural numbers.

[0057] Before obtaining the ridge map, since the ridges and valleys show different grayscale values ​​in the local area of ​​the image, the adaptive mean thresholding (AMT) segmentation method can be used for the grayscale image to obtain a fingerprint ROI similar to that obtained by contact fingerprint recognition. The segmentation principle of AMT is as described in S13, and the expression of adaptive threshold segmentation is:

[0058]

[0059] Where I(x,y) and I'(x,y) represent the grayscale values ​​of the pixel with coordinates (x,y) before and after adaptive mean threshold processing, respectively; Σsize_k represents the sum of the grayscale values ​​of the pixels within the k×k range of the pixel with coordinates (x,y), and C represents a constant.

[0060] Finally, a custom Gabor filter is used to filter and enhance the binary ridges, and the wide ridges are refined into a single-pixel ridge map. At this time, the main feature points of the fingerprint of interest (ridge endpoints and fork points) are obvious.

[0061] When acquiring feature points, the adaptive mean threshold segmentation method is first used for the grayscale image to obtain the fingerprint ROI. The primary task of extracting the feature vector of the fingerprint ROI is to obtain the angle information of the coordinate information of the feature point in the ridge map. The coordinates of the feature points are obtained through the crossing number (CN) theory, and its expression is:

[0062]

[0063] Among them, C N (p) represents the crossover value of pixel point p; p(i) represents the neighborhood pixel value of point p; mod represents the remainder operation;

[0064] When C N When the (p) value is 1, point p is the end point of the fingerprint ridge. N When the (p) value is 3, point p is the bifurcation point of the fingerprint ridge.

[0065] When obtaining the angle information of feature points, such as Figure 2 (a) Define the direction θ of the fork point as follows: f ;

[0066] like Figure 2 (b) Define the angle between the ridge endpoint and the line connecting the 20 pixel points in the opposite direction of the endpoint and the horizontal x-axis as the ridge endpoint direction θ f ;

[0067] At this point, the coordinate and direction information of each feature point in the fingerprint ROI can be obtained.

[0068] Even if the fingerprint image is enhanced to a certain extent, partial blurring and the fingerprint structure itself will still cause pseudo ridges to be present in the obtained ridge map. The difficulty lies in the sampling conditions of the fingerprint ROI and the ridge holes, pseudo ridges and dense detail point groups caused by the fingerprint itself. Figure 3 As shown in the figure, the two images are from the same finger ridge map. The figure uses a rectangular box to mark the ridge pseudo contours and some pseudo endpoints caused by the fingerprint ridge itself or light oversaturation; if they are not processed, a large number of pseudo feature points will be extracted later. However, a large number of chaotic pseudo feature points are not what the algorithm that recognizes fingerprints based on feature points wants to see (such as Figure 3 a), and eventually change the recognition result.

[0069] Based on the above problems, this embodiment proposes an RMO algorithm that can effectively repair and optimize the ridge contour and remove dense feature point groups, that is, a ridge and feature point optimization algorithm.

[0070] Without affecting the original shape of the fingerprint ridges, the ridge contours and short branch ridges can be located to effectively remove some chaotic ridges and ridge holes. The pseudo feature points can be removed by using a method based on feature point density. The image after ridge repair using the RMO algorithm is shown below: Figure 3 As shown in b.

[0071] The RMO algorithm performs ridge repair on the ridge map to obtain the fingerprint recognition image, comprising:

[0072] S31, obtain the contour information and limit information of the ridge line graph, and calculate the contour perimeter L outline ;

[0073] The contour information and limit information are obtained through a function in Python to obtain closed curve information, and the contour perimeter L can also be calculated. outline ;

[0074] S32, determine the contour perimeter L outline Is it less than the threshold Thr? outline :

[0075] If yes, then go to S33;

[0076] If not, proceed to S34;

[0077] The threshold Thr outline It is an empirical threshold, calculated based on the overall length and width of the image;

[0078] S33, filling the outline of the ridge line graph, and then thinning the ridge line graph and entering S34;

[0079] S34, obtain the length L of the ridge line where the ridge line endpoint is located m_ridge , and remove the length L of the ridge line where the ridge line endpoint is located in the feature point set m_ridge Smaller than the contour perimeter L outline The ridgeline endpoints of

[0080] S35. Calculate the Euclidean distance D between any two feature points mm , and remove the Euclidean distance D from the feature point set mm Smaller than the contour perimeter L outline The corresponding two feature points;

[0081] S36. Use the remaining feature points in the feature point set to mark the ridges after ridge repair to obtain a fingerprint recognition image.

[0082] In order to verify the effectiveness of the fingerprint recognition method proposed in this embodiment, experiments were conducted based on the PolyU contactless fingerprint dataset;

[0083] The dataset was obtained by sampling the fingerprints of 300 volunteers, each of whom provided 6 fingerprint images, that is, 1,800 contactless two-dimensional fingerprint samples.

[0084] The computer hardware used in this embodiment: Intel (R) i5-9400 CPU 2.90GHz;

[0085] Algorithm writing language: Python;

[0086] Two sets of experiments were conducted using the above data sets. Experiment 1 is to verify the help of the RMO algorithm proposed in this embodiment for feature point extraction and the improvement of fingerprint recognition rate. To complete Experiment 1, this embodiment designed two trial schemes.

[0087] Solution 1: 10 finger images were randomly and repeatedly selected from the data set 5 times, and feature point extraction data statistics were performed using the RMO algorithm or not. The results are shown in Table 1:

[0088] Table 1 Feature point statistics table using RMO algorithm or not

[0089]

[0090] Table 1 lists the actual results without using the RMO algorithm, the RMO algorithm, and manual observation. Comparing the first row, it can be seen that after the RMO algorithm optimizes the ridges and feature points, the number of feature points of the fingerprint is controlled. Because according to the statistics of all data sets, the average number of fingerprint ROI feature points (robust and effective cross points and endpoints) of a single finger is 22.5. In the case of no ridge and feature point optimization, the average number of fingerprint ROI feature points extracted far exceeds the statistical mean and the observed mean, which indicates that a large number of pseudo feature points are mistakenly extracted.

[0091] Secondly, compare the three T P / F P It is not difficult to find that the RMO algorithm ensures that the true feature points are extracted while the false recognition rate of pseudo feature points is relatively low. Finally, compared with the last row, the fingerprint image not processed by the RMO algorithm has a large variance in the Total feature point number due to pseudo ridges and pseudo feature points. The optimized image Total and T P The variance of the number of feature points is relatively small, indicating that the proportion of the extracted fingerprint feature points that are true feature points is very large.

[0092] Solution 2: With and without the RMO algorithm, the data set is used for recognition based on the Euclidean distance matching of feature points. The results are shown in Table 2:

[0093] Table 2 Performance comparison of using RMO algorithm and not using it

[0094]

[0095] In fact, several parameter indicators are statistically analyzed, namely, the average number of feature points, that is, the number of effective feature points obtained by the algorithm in a single fingerprint ROI; the average number of feature points M P , average computation time M T , that is, the recognition time is the average calculation time of a single fingerprint image; the false rejection rate F RR , false acceptance rate F AR , whose expressions are:

[0096]

[0097] Among them, T P represents the true class, that is, the number of positive classes identified as positive classes, F P represents the false positive class, that is, the number of negative classes identified as positive classes, T N represents the true negative class, that is, the number of negative classes that are identified as negative classes, F N Represents the false negative class, that is, the number of positive classes that are identified as negative classes.

[0098] Compared with the data in Table 2, the RMO algorithm can effectively reduce the computation time. RR and F AR They are greatly reduced, which proves that the RMO algorithm has a significant improvement effect on the fingerprint recognition algorithm based on the feature point recognition method.

[0099] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A non-contact fingerprint recognition method based on ridge restoration, characterized in that: include: S1, collecting finger images in a non-contact manner and preprocessing them to obtain fingerprint ROI images; S2, extracting the coordinate information and angle information of the feature points in the fingerprint ROI image to obtain a ridge map; S3. Use the RMO algorithm to repair the ridge line of the ridge line map to obtain a fingerprint recognition image.

2. The non-contact fingerprint recognition method based on ridge restoration according to claim 1, characterized in that: The S1 includes: S11, collecting finger images in a non-contact manner, using the Cr+Otsu method to perform foreground and background segmentation on the finger images, and obtaining a mask image of only the foreground finger; S12, performing image enhancement on the mask image of only the foreground finger to obtain an image of the finger after image enhancement; S13, performing adaptive threshold segmentation on the finger image after image enhancement to obtain a fingerprint ROI image.

3. The non-contact fingerprint recognition method based on ridge restoration according to claim 2, characterized in that: In S11, the expression for foreground-background segmentation is: in, represents the standard deviation of the forewarp and background; M f and M b Respectively represent the global mean, foreground mean and background mean; P f and P b Represent the foreground ratio and background ratio respectively.

4. The non-contact fingerprint recognition method based on ridge restoration according to claim 2, characterized in that: In S12, the expression of image enhancement is: I (i,j) =M (i,j) +G×(f (i,j) -M (i,j) )-L Where f(i,j) and I(i,j) represent the pixel values ​​of the coordinate point (i,j) before and after enhancement, respectively; L and G represent the brightness adjustment value and the contrast enhancement coefficient, respectively; M(i,j) and σ(i,j) represent the mean and standard deviation of the (2n+1)×(2m+1) local area with the coordinate point (i,j), respectively; n and m are both natural numbers.

5. The non-contact fingerprint recognition method based on ridge restoration according to claim 2, characterized in that: In S13, the expression of adaptive threshold segmentation is: Where I(x,y) and I'(x,y) represent the grayscale values ​​of the pixel with coordinates (x,y) before and after adaptive mean threshold processing, respectively; Σsize_k represents the sum of the grayscale values ​​of the pixels within the k×k range of the pixel with coordinates (x,y), and C represents a constant.

6. The non-contact fingerprint recognition method based on ridge restoration according to claim 1, characterized in that: In S2, the coordinate information of the feature points in the fingerprint ROI image is obtained by the cross number, and its expression is: Among them, C N (p) represents the crossover value of pixel point p; p(i) represents the neighborhood pixel value of point p; mod represents the remainder operation.

7. The non-contact fingerprint recognition method based on ridge restoration according to claim 1, characterized in that: The S3 includes: S31, obtain the contour information and limit information of the ridge line graph, and calculate the contour perimeter L outline ; S32, determine the contour perimeter L outline Is it less than the threshold Thr? outline : If yes, then go to S33; If not, proceed to S34; S33, filling the outline of the ridge line graph, and then thinning the ridge line graph and entering S34; S34, obtain the length L of the ridge line where the ridge line endpoint is located m_ridge , and remove the length L of the ridge line where the ridge line endpoint is located in the feature point set m_ridge Smaller than the contour circumference L outline The ridgeline endpoints of S35. Calculate the Euclidean distance D between any two feature points mm , and remove the Euclidean distance D from the feature point set mm Smaller than the contour perimeter L outline The corresponding two feature points; S36. Use the remaining feature points in the feature point set to mark the ridges after ridge repair to obtain a fingerprint recognition image.

Images