A multi-view three-dimensional fingerprint rough registration method

Abstract

A multi-perspective three-dimensional fingerprint coarse registration method includes the following steps: 1) reading two sets of three-dimensional fingerprint data from different perspectives, respectively serving as the source fingerprint and target fingerprint. Each set of data contains a frame of point cloud data and a corresponding two-dimensional fingerprint image taken simultaneously. The two-dimensional fingerprint image is enhanced and refined to obtain a binary fingerprint image with distinct ridges and valleys. 2) feature points are extracted from the two binary fingerprint images and feature descriptions are performed. 3) fuzzy matching is performed on the two sets of feature point information to obtain initial fuzzy matching pairs. 4) double matching verification is performed using the geometric position relationship between different feature point pairs to filter out erroneous matching point pairs. 5) based on the matched two-dimensional feature point pairs, the corresponding three-dimensional point cloud feature point pairs of the source three-dimensional fingerprint point cloud data and the target three-dimensional fingerprint point cloud data are found, coarse registration is performed, and an initial transformation matrix is obtained. The present invention can extract matching feature point pairs from non-contact two-dimensional fingerprint images, and achieve multi-perspective three-dimensional fingerprint point cloud coarse registration with higher accuracy and robustness.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud registration, and in particular to a multi-view three-dimensional fingerprint coarse registration method. Background Art

[0002] Biometric technology has experienced rapid development in recent years. Compared to traditional identity authentication methods, biometric-based authentication offers greater security and convenience. In recent years, biometric technologies, including palm print recognition, hand shape recognition, fingerprint recognition, iris recognition, face recognition, signature recognition, and vein recognition, have emerged and gained widespread application. Fingerprint recognition is the most widely used and widely accepted technology. Compared to other biometrics, fingerprints offer advantages such as ubiquity, uniqueness, permanence, collectability, feasibility, and reliability.

[0003] In recent years, optical 3D measurement technology has made significant progress in the field of computer vision. Its application to 3D finger measurement can address the shortcomings of traditional 2D acquisition methods, such as fingerprint deformation and information loss caused by fingerprint pressure. Structured light projection measurement uses structured light source-assisted reconstruction to collect 3D information about the finger surface. Encoded light is projected onto the finger, causing the structured light to deform, and the 3D fingerprint information implied by these deformations is calculated. The measurement system structure is relatively simple, low-cost, and easy to maintain. Limited by the field of view, scanning range, and occlusion issues of the 3D measurement system itself, the finger measurement process often needs to be performed from multiple perspectives. 3D fingerprint registration transforms 3D fingerprint data from different perspectives into a unified coordinate system, aligning the 3D data and thus obtaining a more complete 3D model of the fingerprint.

[0004] Three-dimensional fingerprint registration includes coarse registration and fine registration: Coarse registration is a registration method that quickly estimates a rough point cloud registration matrix without knowing any initial relative positions of the two sets of three-dimensional fingerprint data. Fine registration is a registration method that uses the initial transformation matrix obtained from coarse registration to calculate a more accurate solution through methods such as the iterative closest point algorithm (ICP algorithm). Existing fingerprint coarse registration methods perform three-dimensional mapping of the fingerprint point cloud data, mapping the three-dimensional fingerprint to a two-dimensional coordinate system, and then performing feature point matching. However, the mapping process introduces errors, resulting in mapping errors or distortion, leading to low coarse registration accuracy or even registration failure. Summary of the Invention

[0005] In order to overcome the above-mentioned shortcomings of the existing fingerprint coarse registration technology, the present invention provides a multi-view 3D fingerprint coarse registration method with strong robustness, high efficiency and high precision, which solves the problem of incomplete 3D fingerprint data.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A multi-view three-dimensional fingerprint registration method includes the following steps:

[0008] 1) Read two sets of 3D fingerprint data from different perspectives, one as the source fingerprint and the other as the target fingerprint. Each set of data contains a frame of point cloud data and a corresponding 2D fingerprint image captured simultaneously. Preprocess the two 2D fingerprint images.

[0009] 2) Extract feature points from the two preprocessed fingerprint images and perform feature description.

[0010] 3) Perform fuzzy matching on the two sets of feature point information to obtain an initial fuzzy matching pair.

[0011] 4) Use the geometric position relationship between different feature point pairs to perform double matching verification on the fuzzy matching pairs and filter out incorrect matching point pairs.

[0012] 5) According to the matched two-dimensional feature point pairs, the corresponding three-dimensional point cloud feature point pairs of the source three-dimensional fingerprint point cloud data and the target three-dimensional fingerprint point cloud data are found, and a rough registration is performed to obtain the initial transformation matrix.

[0013] Furthermore, the pretreatment in step 1) includes the following steps:

[0014] (11) Reading two two-dimensional fingerprint images and enhancing them to obtain a binary fingerprint image with distinct ridges and valleys and a corresponding fingerprint foreground mask image;

[0015] (12) Due to the influence of shooting, the edge area of the enhanced fingerprint image has poor quality, so the edge area of the foreground mask image is corroded by corrosion operation, and finally the mask image is used to filter and enhance the fingerprint edge area;

[0016] (13) Calculating the directional maps of the two binary fingerprint images and refining them to obtain the source fingerprint refinement map and the target fingerprint refinement map;

[0017] Furthermore, the step 2) includes the following steps:

[0018] (21) Extracting endpoints, bifurcation points, and core points from the source fingerprint refinement map and the target fingerprint refinement map as feature points, obtaining the two-dimensional coordinates of the feature points, and calculating the direction field of the feature points based on the direction map of the fingerprint image and the two-dimensional coordinates of the feature points. The obtained initial feature point information is (x, y, T, θ), where (x, y) is the coordinate of the feature point, T is the feature point type, including endpoints, bifurcation points, and core points, and θ is the direction of the feature point;

[0019] (22) The feature point set extracted from the source two-dimensional fingerprint M is F M ={f m1 ,f m2 ,…,f mm}, the feature point set extracted from the target two-dimensional fingerprint N is F N ={f n1 ,f n2 ,…,f nn};

[0020] (23) For the feature point set F M and F N For each detail point except the core point, calculate the number of ridges between the detail point and the core point in the feature set. The specific calculation steps are as follows:

[0021] (231) The coordinates of the core point C are (x C ,y C ), the coordinates of feature point P are (x P ,y P ), calculate the slope k and intercept b of the straight line passing through points C and P according to the following equations:

[0022]

[0023] (232) Calculate the horizontal distance d between the core point C and the feature point P respectively x and vertical distance d y , when d x >d y When the line from the core point C to the feature point P passes through the pixel point, the x value is recorded, and the corresponding y value is calculated according to the slope k and the intercept b; conversely, the y value of the pixel point is recorded, and the corresponding x value is calculated according to the slope k and the intercept b, and finally the pixel point set through which the line from the core point C to the feature point P passes is determined;

[0024] (233) Combined with the two-dimensional fingerprint image information, the grayscale value of each pixel in the pixel set is recorded. The grayscale value of the pixel located on the ridge line in the image is 1, and the grayscale value of the pixel located on the valley line is 0. Each pixel is traversed. If the grayscale value of the pixel is 1 and the grayscale value of the next pixel is 0, the point is considered to be a mutation point from the ridge line to the valley line. In order to increase the accuracy of calculating the number of ridges, the threshold of the width between ridges is set to 4. If the distance between the mutation point and the previous mutation point is greater than the threshold of the width between ridges, the number of ridges crossed is increased by one.

[0025] (234) The calculated number of ridge crossings is recorded in the feature point information, and the feature point information is expanded to (x, y, T, θ, N), where N represents the number of ridge crossings between the detail point and the core point;

[0026] (24) For the feature point set F M and F NFor each detail point except the core point, calculate the direction difference between the detail point and the eight neighboring blocks of the block it belongs to. Take the neighboring block pointed by the direction of the detail point as the starting block, and calculate the direction difference between each neighboring block and the block where the detail point is located in clockwise order. Finally, obtain eight direction differences θ1-θ8 and record them in the feature point information. The feature point information is expanded to (x, y, T, θ, N, θ1, θ2, θ3, θ4, θ5, θ6, θ7, θ8).

[0027] Furthermore, step 3) includes the following steps:

[0028] (31) For the feature point set F M For each detail point except the core point Traverse the detail points f in the feature point set F N in n , and find the feature point that matches this detail point. The matching process is as follows:

[0029] (311) Calculate the difference S1 in the number of ridge crossings between f m and the corresponding core point. The calculation formula is as follows: n

[0030]

[0031] (312) Calculate the Euclidean distance S2 of the direction differences of the eight neighboring blocks of f m and f n . The calculation formula is as follows:

[0032]

[0033] (313) Set the threshold threshold_ridge of the difference in the number of ridge crossings to 2 and the threshold threshold_ori of the Euclidean distance of the direction difference to 0.3 according to the empirical value. For the feature point f m in the source fingerprint image, select all feature points that can satisfy S1 < threshold_ridge from F N to form a set M1. Calculate the feature point with the smallest Euclidean distance S2 between M1 and f m . If S2 < threshold_ori is satisfied at this time, it is considered that this feature point matches f m , and save it in the matching set matched Record the coordinates of the two matched feature points f m and f n as well as the corresponding S1 and S2;

[0034] (32) The source fingerprint core point and the target fingerprint core point are by default the feature points that match each other. Therefore, the information of the two fingerprint core points is added to the set matched to obtain the initial feature point set F. M and F N The set of matching feature point pairs in :

[0035] matched={m1,m2,m3…m i} (4)

[0036] in:

[0037]

[0038] (33) Perform a preliminary screening of feature point pairs, and for each matching point pair in the set matched, find out whether there is the same point pair in the entire set. A matching pair of coordinates; if it exists, prove that F N There are characteristic points and F M Match multiple feature points in the , and calculate the Euclidean distance S of the direction difference between all matching points in the repeated matching i2 The three smallest matching point pairs are used as the fuzzy matching point pairs for preliminary correct matching, and the remaining feature point pairs are deleted. The matching effect after screening is as follows: Figure 7 shown.

[0039] (34) In summary, the initial fuzzy matching point pair set matched of the source two-dimensional fingerprint and the target two-dimensional fingerprint is obtained.

[0040] Furthermore, the step 4) includes the following steps:

[0041] (41) Further matching verification is performed using the mutual position relationship between the initial matching point pairs. A score is defined for each pair of feature points and initialized to 0;

[0042] (42) Feature point pairs and (p≠q) is two pairs of matching feature points. The first level of verification is performed on the matching point pairs. Calculate and The distance between them is denoted as d1. and The distance between them is recorded as d2; calculate the feature points Direction and point and The angle between the lines is recorded as Feature Points Direction and point and The angle between the lines is recorded as Calculate feature points Direction and point and The angle between the lines is recorded as Feature Points Direction and point and The angle between the lines is recorded as

[0043] (43) Set the threshold T d ,T α , when both the distance difference and the angle difference are lower than the threshold T d ,T α When

[0044] |d1-d2|≤T d ,

[0045] If the two pairs of feature points pass the first level of verification, each will be awarded one point, otherwise no points will be awarded;

[0046] (44) When two pairs of feature points pass the first verification and are each awarded one point, traverse each set of point pairs in the remaining matching pairs and use the point pairs as the third pair of matching point pairs. (r≠p≠q) is added to form a triangular geometric relationship with the first two sets of matching point pairs, and a second verification is performed to calculate and The angle between and The angle between Calculate separately and The distance between and The distance between

[0047] (45) Set the threshold T dr ,T αr , when the distance and angle meet the following conditions, that is

[0048]

[0049] If the three pairs of feature points pass the second verification, one point will be added to each pair; otherwise, no points will be added.

[0050] (46) After performing the above operations on all feature point pairs, the feature point pairs with the top three scores are selected and retained as the correctly matched feature point pairs, and the remaining incorrectly matched feature point pairs are deleted to obtain the final matched set.

[0051] Furthermore, the step 5) includes the following steps:

[0052] (51) According to the coordinates of the matched two-dimensional point pairs in the set matched, a matched three-dimensional point pair is obtained on the corresponding three-dimensional fingerprint;

[0053] (52) According to the three-dimensional matching point pairs, the rigid body transformation parameters R0 and T0 are calculated, where R0 is a 3×3 rotation matrix and T0 = [t x ,t y ,t z ] T is the translation vector, R0 and T0 are the initial transformation relationship of the two 3D fingerprints for coarse registration, and the coarse registration of the source 3D fingerprint and the target 3D fingerprint is completed.

[0054] The working principle of the present invention is: combining the image information corresponding to the fingerprint point cloud data, extracting two-dimensional feature points, performing double matching verification on the two-dimensional feature points by constructing feature point descriptors to obtain matching point pairs, and calculating the initial transformation matrix of the three-dimensional fingerprint based on the three-dimensional point cloud feature point pairs corresponding to the two-dimensional matching point pairs to complete the coarse alignment of the three-dimensional fingerprint.

[0055] The advantages of the present invention are: fingerprint images and point cloud data are collected simultaneously and synchronously, which can achieve high-precision fingerprint coarse registration; descriptors are constructed based on fingerprint core points and ridge and valley line information, and double matching verification is performed simultaneously, which has strong robustness; coarse registration is performed by combining fingerprint images with point cloud information, which has high computational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 This is a flow chart of the three-dimensional fingerprint coarse registration steps in the present invention.

[0057] Figure 2a-2b is a comparison diagram of a two-dimensional image before and after enhancement provided by an embodiment of the present invention, wherein Figure 2a This is the two-dimensional fingerprint image before enhancement. Figure 2b It is the binary image after two-dimensional fingerprint enhancement.

[0058] Figure 3 This is the result of extracting feature points from a target fingerprint image in one embodiment of the present invention.

[0059] Figure 4 This is the result of extracting feature points from a source fingerprint image in one embodiment of the present invention.

[0060] Figure 5 It is a schematic diagram of feature point description of an embodiment of the present invention.

[0061] Figure 6 This is an effect diagram after matching the feature points of the source image and the target image, provided by an embodiment of the present invention.

[0062] Figure 7This is an effect diagram after fuzzy matching of matching feature point pairs in one embodiment of the present invention.

[0063] Figure 8 It is a schematic diagram of the geometric position relationship of double matching verification.

[0064] Figure 9 This is an effect diagram after performing matching verification on matching feature point pairs and deleting incorrect matching point pairs in an embodiment of the present invention.

[0065] FIG10 is a diagram showing the effect of the source point cloud and the target point cloud before and after registration according to an embodiment of the present invention, wherein Figure 10a This is the effect picture before registration. Figure 10b This is the effect picture after alignment. DETAILED DESCRIPTION

[0066] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0067] Reference Figure 1 ,The multi-view 3D fingerprint coarse registration method includes the following steps:

[0068] 1) Read two sets of 3D fingerprint data from different perspectives, one as the source fingerprint and the other as the target fingerprint. Each set of data contains a frame of point cloud data and a corresponding 2D fingerprint image taken simultaneously. Preprocess the two 2D fingerprint images as follows:

[0069] (11) Reading two two-dimensional fingerprint images and enhancing them to obtain a binary fingerprint image with distinct ridges and valleys and a corresponding fingerprint foreground mask image;

[0070] (12) Due to the influence of shooting, the edge area of the enhanced fingerprint image has poor quality. Therefore, the edge area of the foreground mask image is corroded by corrosion operation, and finally the mask image is used to filter and enhance the fingerprint edge area. The images before and after enhancement are as follows: Figure 2a and Figure 2b shown.

[0071] (13) Calculating the directional maps of the two two-dimensional fingerprint enhanced images and refining them to obtain a source fingerprint refinement map and a target fingerprint refinement map;

[0072] 2) Extracting feature points from the enhanced fingerprint image generated in step 1) and performing feature description, including the following steps:

[0073] (21) The endpoints, bifurcation points, and core points of the source fingerprint refinement map and the target fingerprint refinement map are extracted as feature points to obtain the two-dimensional coordinates of the feature points. The direction field of the feature points is calculated based on the direction map of the fingerprint image and the two-dimensional coordinates of the feature points. The initial feature point information obtained is (x, y, T, θ), (x, y) is the coordinate of the feature point, T is the feature point type, including endpoints, bifurcation points, and core points, and θ is the direction of the feature point. The feature point extraction results of the target fingerprint image are as follows: Figure 3 As shown, the result of feature point extraction of source fingerprint image Figure 4 As shown, the yellow triangle represents the core point, the green circle represents the endpoint, and the red circle represents the bifurcation point;

[0074] (22) Assume that the feature point set extracted from the source two-dimensional fingerprint M is F M ={f m1 ,f m2 ,…,f mm}, the feature point set extracted from the target two-dimensional fingerprint N is F N ={f n1 ,f n2 ,…,f nn};

[0075] (23) For the feature point set F M and F N For each detail point except the core point, calculate the number of ridges spanning from the detail point to the core point in the feature set, such as Figure 5 As shown, the red triangle represents the core point position, and the green circle represents the detail point position. The specific calculation steps are as follows:

[0076] (231) The coordinates of the core point C are (x C ,y C ), the coordinates of feature point P are (x P ,y P ), calculate the slope k and intercept b of the straight line passing through points C and P according to the following equations:

[0077]

[0078] (232) Calculate the horizontal distance d between the core point C and the feature point P respectively x and vertical distance d y , when d x >d y When the line from the core point C to the feature point P passes through the pixel point, the x value is recorded, and the corresponding y value is calculated according to the slope k and the intercept b; conversely, the y value of the pixel point is recorded, and the corresponding x value is calculated according to the slope k and the intercept b, and finally the pixel point set through which the line from the core point C to the feature point P passes is determined;

[0079] (233) Combined with the two-dimensional fingerprint image information, the grayscale value of each pixel in the pixel set is recorded. The grayscale value of the pixel located on the ridge line in the image is 1, and the grayscale value of the pixel located on the valley line is 0. Traverse each pixel point. If the grayscale value of the pixel point is 1 and the grayscale value of the next pixel point is 0, then the point is considered to be a mutation point from the ridge line to the valley line. In order to increase the accuracy of calculating the number of ridge lines, the threshold value of the width between ridge lines is set to 4. If the distance between the mutation point and the previous mutation point is greater than the threshold value of the width between ridge lines, the number of ridge lines crossed is increased by one.

[0080] (234) The calculated number of ridge crossings is recorded in the feature point information, and the feature point information is expanded to (x, y, T, θ, N), where N represents the number of ridge crossings between the detail point and the core point, as shown in Figure 5 The middle red dots indicate the number of ridges spanned;

[0081] (24) For the feature point set F M and F N For each detail point except the core point, calculate the direction difference between the detail point and the eight neighboring blocks of the block where it is located. Take the neighboring block pointed by the detail point as the starting block, and calculate the direction difference between each neighboring block and the block where the detail point is located in a clockwise direction. Finally, eight direction differences θ1-θ8 are obtained and recorded in the feature point information. The feature point information is expanded to (x, y, T, θ, N, θ1, θ2, θ3, θ4, θ5, θ6, θ7, θ8);

[0082] 3) Based on the feature point information of the two fingerprints obtained in step 2), a rough feature point matching is performed to obtain an initial fuzzy matching point pair. The matching process includes the following steps:

[0083] (31) For the feature point set F M Every detail point except the core point Traverse the feature point set F N The detail point f n ,Find the feature points that match the detail points. The matching process is as follows:

[0084] (311) Calculate f m 、f n The difference S1 between the number of ridges spanned by the corresponding core point is calculated as follows:

[0085]

[0086] (312) Calculate f m 、f n The Euclidean distance S2 of the direction difference of the eight neighborhood blocks is calculated as follows:

[0087]

[0088] (313) Set the threshold threshold_ridge of the difference in the number of ridge lines to 2 based on empirical values, and set the threshold threshold_ori of the Euclidean distance of the direction difference to 0.3. For the feature point f in the source fingerprint image m , select all feature points from F N that can satisfy S1 < threshold_ridge to form a set M1. Calculate the feature point in M1 with the smallest Euclidean distance S2 from f m . If S2 < threshold_ori is satisfied at this time, it is considered that this feature point matches f m , and save the record of the two matching feature points f m , f n coordinates and the corresponding S1, S2 in the matching set matched;

[0089] (32) The core point of the source fingerprint and the core point of the target fingerprint are defaulted to be matching feature points. Therefore, add the information of the two fingerprint core points to the set matched to obtain the initial set of feature points F M and the set of matching feature point pairs in F N :

[0090] matched = {m1, m2, m3…m i} (4)

[0091] Among them:

[0092]

[0093] In the example, the matching effect is as Figure 6 shown.

[0094] (33) Conduct a preliminary screening of the feature point pairs. For each matching point pair in the set matched, check whether there is a matching point pair with the same coordinates in the entire set; if it exists, it proves that there are feature points in F N that match multiple feature points in F M . Take the three matching point pairs with the smallest Euclidean distance S i2 of the direction difference among all the duplicate matching point pairs as the initially correctly matched fuzzy matching point pairs, and delete the remaining feature point pairs. The matching effect after screening is as Figure 7 shown.

[0095] (34) In summary, obtain the initial set of fuzzy matching point pairs matched for the source two-dimensional fingerprint and the target two-dimensional fingerprint.

[0096] 4) Based on the initial fuzzy matching point pairs obtained in step 3), double matching verification is performed using the geometric position relationship between the feature point pairs to filter out incorrect matching point pairs, including the following steps:

[0097] (41) Further matching verification is performed using the mutual position relationship between the initial matching point pairs. A score is defined for each pair of feature points and initialized to 0;

[0098] (42) Feature point pairs and (p≠q) is two pairs of matching feature points. The first level of verification is performed on the matching point pairs. Calculate and The distance between them is denoted as d1. and The distance between them is recorded as d2; calculate the feature points Direction and point and The angle between the lines is recorded as Feature Points Direction and point and The angle between the lines is recorded as Calculate feature points Direction and point and The angle between the lines is recorded as Feature Points Direction and point and The angle between the lines is recorded as like Figure 8 As shown in the black part;

[0099] (43) Set the threshold T d ,T α , when both the distance difference and the angle difference are lower than the threshold T d ,T α When

[0100] |d1-d2|≤T d ,

[0101] If the two pairs of feature points pass the first level of verification, each will be awarded one point, otherwise no points will be awarded;

[0102] (44) When two pairs of feature points pass the first verification and are each awarded one point, each pair of point pairs in the remaining matching point pairs is traversed and the point pairs are used as the third pair of matching point pairs. (r≠p≠q) is added to form a triangular geometric relationship with the first two sets of matching point pairs, and a second verification is performed to calculate and The angle between and The angle between Calculate separately and The distance between and The distance between like Figure 8 As shown in the yellow part; (45) set the threshold T dr ,T αr , when the distance and angle meet the following conditions, that is

[0103]

[0104] If the three pairs of feature points pass the second verification, one point will be added to each pair; otherwise, no points will be added.

[0105] (46) After performing the above operations on all feature point pairs, the feature point pairs with the top three scores are selected and retained as the correctly matched feature point pairs, and the remaining incorrectly matched feature point pairs are deleted to obtain the final matched set. The matching effect after double matching verification is as follows: Figure 9 shown.

[0106] 5) Find the corresponding 3D matching feature point pairs based on the 2D matching feature point pairs obtained in step 4) and perform coarse registration, which includes the following steps:

[0107] (51) According to the coordinates of the matched two-dimensional point pairs in the set matched, a matched three-dimensional point pair is obtained on the corresponding three-dimensional fingerprint;

[0108] (52) According to the three-dimensional matching point pairs, the rigid body transformation parameters R0 and T0 are calculated, where R0 is a 3×3 rotation matrix and T0 = [t x ,t y ,t z ] T is the translation vector, R0 and T0 are the initial transformation relationship of the two three-dimensional fingerprints for coarse registration, and the coarse registration of the source three-dimensional fingerprint and the target three-dimensional fingerprint is completed. The registration effect of the example is as follows Figure 10a and Figure 10b shown.

[0109] The contents described in the embodiments of this specification are merely an enumeration of the implementation forms of the inventive concept. The scope of protection of the present invention should not be regarded as limited to the specific forms described in the embodiments. The scope of protection of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A multi-view 3D fingerprint coarse registration method, characterized in that: The steps include: 1) Read two sets of 3D fingerprint data from different perspectives, one as the source fingerprint and the other as the target fingerprint; each set of data contains a frame of point cloud data and a corresponding 2D fingerprint image taken simultaneously; preprocess the two 2D fingerprint images; 2) Extracting feature points from the two pre-processed two-dimensional fingerprint images and performing feature description; 3) Perform fuzzy matching on the two sets of feature point information to obtain an initial fuzzy matching pair; 4) Using the geometric position relationship between different feature point pairs to perform double matching verification on the fuzzy matching pairs and filter out incorrect matching point pairs; including the following steps: (41) Further matching verification is performed using the mutual position relationship between the initial matching point pairs; a score is defined for each pair of feature points and initialized to 0; (42) Feature point pairs and (p≠q) is two pairs of matching feature points. The first level of verification is performed on the matching point pairs. Calculate and The distance between them is denoted as d1. and The distance between them is recorded as d2; calculate the feature points Direction and point and The angle between the lines is recorded as Feature Points Direction and point and The angle between the lines is recorded as Calculate feature points Direction and point and The angle between the lines is recorded as Feature Points Direction and point and The angle between the lines is recorded as (43) Set the threshold T d ,T α , when both the distance difference and the angle difference are lower than the threshold T α ,T α When If the two pairs of feature points pass the first level of verification, each will be awarded one point, otherwise no points will be awarded; (44) When two pairs of feature points pass the first verification and are each awarded one point, traverse each set of point pairs in the remaining matching pairs and use the point pairs as the third pair of matching point pairs. (r≠p≠q) is added to form a triangular geometric relationship with the first two sets of matching point pairs, and a second verification is performed to calculate and The angle between and The angle between Calculate separately and The distance between and The distance between (45) Set the threshold T dr , T αr When the distance and angle meet the following conditions, that is, If the three pairs of feature points pass the second verification, each pair will be awarded one point, otherwise no points will be awarded; (46) After performing the above operations on all feature point pairs, the feature point pairs with the top three scores are selected and retained as the correctly matched feature point pairs, and the remaining incorrectly matched feature point pairs are deleted to obtain the final matched set; 5) According to the matched two-dimensional feature point pairs, the corresponding three-dimensional point cloud feature point pairs of the source three-dimensional fingerprint point cloud data and the target three-dimensional fingerprint point cloud data are found, and a rough registration is performed to obtain the initial transformation matrix.

2. The multi-view 3D fingerprint coarse registration method according to claim 1, characterized in that: The step 1) comprises the following steps: (11) Reading two two-dimensional fingerprint images and enhancing them to obtain a binary fingerprint image with distinct ridges and valleys and a corresponding fingerprint foreground mask image; (12) Due to the influence of shooting, the edge area of the enhanced fingerprint image has poor quality, so the edge area of the foreground mask image is corroded by corrosion operation, and finally the mask image is used to filter and enhance the fingerprint edge area; (13) Calculating the directional maps of the two binary fingerprint enhanced images and refining them to obtain the source fingerprint refinement map and the target fingerprint refinement map; 3. The multi-view 3D fingerprint coarse registration method according to claim 2, characterized in that: The step 2) comprises the following steps: (21) Extracting endpoints, bifurcation points, and core points from the source fingerprint refinement map and the target fingerprint refinement map as feature points, obtaining the two-dimensional coordinates of the feature points, and calculating the direction field of the feature points based on the direction map of the fingerprint image and the two-dimensional coordinates of the feature points. The obtained initial feature point information is (x, y, T, θ), where (x, y) is the coordinate of the feature point, T is the feature point type, including endpoints, bifurcation points, and core points, and θ is the direction of the feature point; (22) The feature point set extracted from the source two-dimensional fingerprint M is F M ={f m1 ,f m2 ,…,f mm }, the feature point set extracted from the target two-dimensional fingerprint N is F N ={f n1 ,f n2 ,…,f nn }; (23) For the feature point set F M and F N For each detail point except the core point, calculate the number of ridges between the detail point and the core point in the feature set. The specific calculation steps are as follows: (231) The coordinates of the core point C are (x C ,y C ), the coordinates of feature point P are (x P ,y P ), calculate the slope k and intercept b of the straight line passing through points C and P according to the following equations: (232) Calculate the horizontal distance d between the core point C and the feature point P respectively x and vertical distance d y , when d x >d y When the line from the core point C to the feature point P passes through the pixel point, the x value is recorded, and the corresponding y value is calculated according to the slope k and the intercept b; conversely, the y value of the pixel point is recorded, and the corresponding x value is calculated according to the slope k and the intercept b, and finally the pixel point set through which the line from the core point C to the feature point P passes is determined; (233) Combined with the two-dimensional fingerprint image information, the grayscale value of each pixel in the pixel set is recorded. The grayscale value of the pixel located on the ridge line in the image is 1, and the grayscale value of the pixel located on the valley line is 0. Each pixel is traversed. If the grayscale value of the pixel is 1 and the grayscale value of the next pixel is 0, the point is considered to be a mutation point from the ridge line to the valley line. In order to increase the accuracy of calculating the number of ridges, the threshold of the width between ridges is set to 4. If the distance between the mutation point and the previous mutation point is greater than the threshold of the width between ridges, the number of ridges crossed is increased by one. (234) The calculated number of ridge crossings is recorded in the feature point information, and the feature point information is expanded to (x, y, T, θ, N), where N represents the number of ridge crossings between the detail point and the core point; (24) For the feature point set F M and F N For each detail point except the core point, the direction difference between the detail point and the eight neighboring blocks of the block where it is located is calculated respectively. The neighboring block pointed by the detail point is taken as the starting block, and the direction difference between each neighboring block and the block where the detail point is located is calculated in a clockwise direction. Finally, eight direction differences θ1-θ8 are obtained and recorded in the feature point information. The feature point information is expanded to (x, y, T, θ, N, θ1, θ2, θ3, θ4, θ5, θ6, θ7, θ8).

4. The multi-view 3D fingerprint coarse registration method according to claim 3, characterized in that: The step 3) comprises the following steps: (31) For the feature point set F M Every detail point except the core point Traverse the feature point set F N The detail point f n ,Find the feature points that match the detail points. The matching process is as follows: (311) Calculate f m 、f n The difference S1 between the number of ridges spanned by the corresponding core point is calculated as follows: (312) Calculate f m 、f n The Euclidean distance S2 of the direction difference of the eight neighborhood blocks is calculated as follows: (313) Set the threshold threshold_ridge of the difference in the number of ridge lines to 2 based on empirical values, and the threshold threshold_ori of the Euclidean distance of the direction difference to 0.

3. For the feature point f in the source fingerprint image m , select all feature points that can satisfy S1 < threshold_ridge from F N to form a set M1. Calculate the feature point with the smallest Euclidean distance S2 from f m in M1. If S2 < threshold_ori is satisfied at this time, it is considered that this feature point matches f m . Save in the matching set matched the records of the two matched feature points f m , f n coordinates and the corresponding S1, S2; (32) The source fingerprint core point and the target fingerprint core point are by default the feature points that match each other. Therefore, the information of the two fingerprint core points is added to the set matched to obtain the initial feature point set F. M and F N The set of matching feature point pairs in : in: (33) Perform a preliminary screening of feature point pairs, and for each matching point pair in the set matched, find out whether there is the same point pair in the entire set. A matching pair of coordinates; if it exists, prove that F N There are characteristic points and F M Match multiple feature points in the , and calculate the Euclidean distance S of the direction difference between all matching points in the repeated matching i2 The three smallest matching point pairs are used as the fuzzy matching point pairs for preliminary correct matching, and the remaining feature point pairs are deleted; (34) In summary, the initial fuzzy matching point pair set matched of the source two-dimensional fingerprint and the target two-dimensional fingerprint is obtained.

5. The multi-view 3D fingerprint coarse registration method according to claim 4, characterized in that: The step 5) comprises the following steps: (51) According to the coordinates of the matched two-dimensional point pairs in the set matched, a matched three-dimensional point pair is obtained on the corresponding three-dimensional fingerprint; (52) According to the three-dimensional matching point pairs, the rigid body transformation parameters R0 and T0 are calculated, where R0 is a 3×3 rotation matrix and T0 = [t x ,t y ,t z ] T is the translation vector, R0 and T0 are the initial transformation relationship of the two 3D fingerprints for coarse registration, and the coarse registration of the source 3D fingerprint and the target 3D fingerprint is completed.

Images