A method for fingerprint matching based on feature point matching

Abstract

The application relates to the technical field of fingerprint matching, and discloses a fingerprint matching method based on feature point matching, which realizes fingerprint matching based on the matching of feature points in a sample fingerprint feature point set and feature points in a template fingerprint feature point set. The application solves the problems of large power consumption, long time, and non-universality existing in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fingerprint matching, and particularly relates to a fingerprint matching method based on feature point matching. BACKGROUND

[0002] With the rapid development of science and technology and digital life, people have higher requirements for the security of identity information. Decades ago, user identity was usually verified by token (such as password or smart card). They are easy to be forgotten or lost. However, in recent years, biometric recognition technology has developed rapidly and has entered all aspects of people's life, and has a trend of replacing traditional token recognition and authentication. Among all biometric recognition technologies, fingerprint recognition is the most widely used method in personal identity recognition systems due to its relative persistence and uniqueness. Fingerprint recognition algorithms are mainly divided into three categories: matching based on image correlation, matching based on image pattern and matching based on feature points. Among them, the feature point based fingerprint matching is the most widely used algorithm in artificial matching and automatic identification system (AFIS). Compared with the other two algorithms, the feature point algorithm is more memory-saving and less computationally intensive, and is suitable for low-power devices.

[0003] Matching two sets of feature points one by one is a very difficult task, because the direction, position, angle and force of the finger during the acquisition process are not exactly the same, resulting in almost always some rotation, translation or deformation between the two sets of points. In order to overcome this problem, existing algorithms need to align the feature points in advance. The two sets of feature points are aligned before feature point matching, and finally the similarity between the two is calculated according to the number and degree of feature point matching.

[0004] The implementation of the present application mainly solves the following technical problems:

[0005] 1) The existing fingerprint matching algorithm commonly uses matching based on image correlation and matching based on image pattern, which is computationally intensive, requires high hardware devices and is not suitable for low-power devices.

[0006] 2) The existing feature point based fingerprint matching algorithm needs to align the feature points in advance, which will result in a large amount of calculation and a long matching time.

[0007] 3) The fingerprint feature definition used in the existing feature point based fingerprint matching algorithm does not conform to the general specification and does not have portability.

[0008] In summary, the existing technology has the problems of high power consumption, long time and non-universal. SUMMARY

[0009] In order to overcome the shortcomings of the prior art, the application provides a fingerprint matching method based on feature point matching, which solves the problems of large power consumption, long time, and non-universality in the prior art.

[0010] The application solves the above problems by using the technical scheme of:

[0011] The application provides a fingerprint matching method based on feature point matching, which realizes fingerprint matching based on matching of feature points in a sample fingerprint feature point set and feature points in a template fingerprint feature point set.

[0012] As a preferred technical scheme, the method comprises the following steps:

[0013] S1, reference point selection: calculating a plurality of sample reference points for sample feature point sorting according to the center coordinates of the sample fingerprint effective area and the upper, lower, left and right boundaries of the sample fingerprint effective area, and calculating a plurality of template reference points for template feature point sorting according to the center coordinates of the template fingerprint effective area and the upper, lower, left and right boundaries of the template fingerprint effective area;

[0014] S2, feature point sorting: respectively sorting the sample feature point set before sorting and the template feature point set before sorting according to the coordinates of the template reference points selected in step S1, to obtain the sample feature point set after sorting and the template feature point set after sorting;

[0015] S3, searching for effective matching pairs: establishing a plurality of sample feature point matching pairs of a sample center point in the sample feature point set, establishing a plurality of template feature point matching pairs of a template center point in the template feature point set, and separately pairing each sample feature point matching pair with each template feature point matching pair, and the successfully matched sample feature point matching pair and template feature point matching pair are collectively referred to as effective matching pairs;

[0016] S4, center point iteration: if there is a feature point in the sample feature point set that meets the set condition and has not been used as a sample center point to perform S3 operation, the feature point is used as a new sample center point of S3, and S3 is repeated; if there is a template feature point in the template feature point set that meets the set condition and has not been used as a template center point to perform S3 operation, the template feature point is used as a new template center point of S3, and S3 is repeated; if none of the above conditions is met, the method jumps to S5;

[0017] S5, global matching constraint: calculate the transformation parameters of all valid matching pairs, obtain the optimal transformation parameters according to the transformation parameters of the valid matching pairs, and then eliminate the valid matching pairs located outside the set tolerance range according to the optimal transformation parameters; wherein the optimal transformation parameters refer to the global optimal solution taking the transformation parameters of the valid matching pairs as the decision variables and the global matching result of the sample fingerprint feature point set and the template fingerprint feature point set as the target;

[0018] S6, matching result prediction: constructing an ANN prediction model, taking the matching index corresponding to the global optimal solution as the ANN prediction model, and predicting the similarity between the sample fingerprint feature point set and the template fingerprint feature point set.

[0019] As a preferred technical solution, in step S1, the sample fingerprint effective area represents the area surrounded by the maximum convex hull containing all fingerprint feature points, the center coordinates of the sample fingerprint effective area are obtained by averaging the X coordinates and Y coordinates of all pixel points in the sample fingerprint effective area, the upper boundary of the sample fingerprint effective area is the minimum Y coordinate of all points constituting the convex hull, the lower boundary of the sample fingerprint effective area is the maximum Y coordinate of all points constituting the convex hull, the left boundary of the sample fingerprint effective area is the minimum X coordinate of all points constituting the convex hull, and the right boundary of the sample fingerprint effective area is the maximum Y coordinate of all points constituting the convex hull; the template fingerprint effective area represents the area surrounded by the maximum convex hull containing all fingerprint feature points, the center coordinates of the template fingerprint effective area are obtained by averaging the X coordinates and Y coordinates of all pixel points in the template fingerprint effective area, the upper boundary of the template fingerprint effective area is the minimum Y coordinate of all points constituting the convex hull, the lower boundary of the template fingerprint effective area is the maximum Y coordinate of all points constituting the convex hull, the left boundary of the template fingerprint effective area is the minimum X coordinate of all points constituting the convex hull, and the right boundary of the template fingerprint effective area is the maximum Y coordinate of all points constituting the convex hull; wherein the coordinate axes for measuring X coordinates and Y coordinates use the definition in section 6.4.2 of ISO / IEC 19794-2:2011.

[0020] As a preferred technical solution, in step S1, the template reference point calculation method is as follows:

[0021]

[0022]

[0023] C center =(x c ,y c ),

[0024] C up =(x c ,y c -Δyoffset ),

[0025] C down =(x c ,y c +Δy offset ),

[0026] C left =(x c -Δx offset ,y c ),

[0027] C rigth =(x c +Δx offset ,y c ),

[0028] Among them, (x c ,y c ) represents the center coordinates of the valid fingerprint region, y min It is the upper boundary of the valid fingerprint area, y max It is the lower boundary of the valid fingerprint area, x min It is the left boundary of the valid fingerprint area, x max It is the right boundary of the valid fingerprint area, C center It is the fingerprint center reference point, C up It is the reference point on the fingerprint, C down It is the fingerprint reference point, C left It is the left reference point of the fingerprint, C right It is the right reference point of the fingerprint, Δx offset It is the x-coordinate offset of the left and right reference points of the fingerprint relative to the central reference point, Δy offset It is the y-coordinate offset of the upper and lower reference points of the fingerprint relative to the center reference point; the fingerprint refers to the sample fingerprint or the template fingerprint. The reference point of the template fingerprint is calculated using the center coordinates, upper, lower, left and right boundaries of the valid area of ​​the template fingerprint, and the reference point of the sample fingerprint is calculated using the center coordinates, upper, lower, left and right boundaries of the valid area of ​​the sample fingerprint.

[0029] As a preferred technical solution, the feature point sorting method in step S2 includes the following steps:

[0030] S21. Establish an empty set of fingerprint feature points, called the sorted set of fingerprint feature points. Sort all feature points in the original fingerprint feature point set according to their distance from the fingerprint center reference point from near to far. Select the two feature points closest to the fingerprint center reference point and add them to the sorted set of fingerprint feature points in order from near to far.

[0031] S22, Select the feature point in the fingerprint feature point set before sorting that is closest to the upper / lower / left / right reference point of the fingerprint and has not yet been added to the fingerprint feature point set after sorting, and add this point as the third / fourth / fifth / sixth feature point to the fingerprint feature point set after sorting;

[0032] S23, all feature points in the fingerprint feature point set that have not yet been added to the sorted fingerprint feature point set before sorting are sorted from near to far according to the distance of the feature points to the fingerprint center reference point, and added to the sorted fingerprint feature point set in this order.

[0033] S24. Through the above steps, sort the sample fingerprint feature point set and the template fingerprint feature point set before sorting, respectively, to obtain the sorted sample fingerprint feature point set and the sorted template fingerprint feature point set.

[0034] As a preferred technical solution, in step S3, the method for constructing sample feature point matching pairs is as follows: Select a sample feature point other than the sample center point from the sample feature point set, and use it as the sample surrounding point of the sample center point. The distance between the sample surrounding point and the sample center point should be within a certain range; establish a connection between the sample center point and the sample surrounding point of the sample center point, which is a sample feature point matching pair; the sample center point needs to establish a connection with all its sample surrounding points that meet the conditions.

[0035] The construction method of template feature point matching pair is as follows: Select a template feature point other than the template center point from the template feature point set, and use it as the template surrounding point of the template center point. The distance between the template surrounding point and the template center point should be within a certain range; establish a connection between the template center point and the template surrounding point of the template center point, which is a template feature point matching pair; the template center point needs to establish a connection with all its template surrounding points that meet the conditions.

[0036] As a preferred technical solution, in step S3, a successful match is considered to be achieved if the sample feature point matching pair and the template feature point matching pair meet the following conditions:

[0037]

[0038] Where, d p It is the distance between template feature point matching pairs, d q It is the distance between matching pairs of sample feature points, γ p It is the number of ridges traversed by the template feature point matching pair, γ q This is the number of ridges traversed by the sample feature point matching pair, and the number of ridges follows the definition in section 8.5.2 of the international standard ISO / IEC 19794-2:2011; α p It is the orientation angle of the center point of the template feature point matching pair, α qβ is the orientation angle of the center point of the sample feature point matching pair. p It is the orientation angle of the template feature point matching pair around the point, β q It is the orientation angle of the sample feature point matching pair around the point; θ p The angle θ is the angle between the line connecting the center point of the template and the surrounding point and the positive direction of the X-axis. q It is the angle between the line connecting the center point of the sample pair and the surrounding point and the positive direction of the X-axis, with the center point of the sample pair as the center; T γ It is the tolerance for the number of ridge lines, T d It is the distance tolerance, T αβ It is the feature point angle tolerance, T βθ It is the tolerance for the angle difference of the lines connecting the feature points.

[0039] As a preferred technical solution, in step S4, the transformation parameters of the effective matching pair are calculated using the following formula:

[0040]

[0041]

[0042] Δθ=θ p -θ q ,

[0043] Δx=x pc -(x qc cosΔθ-y qc sinΔθ),

[0044] Δy=y pc -(x qc sinΔθ+y qc cosΔθ),

[0045] Where, x pc This indicates the effective match for the X and Y coordinates of the template center point. pc Indicates the Y-coordinate of the center point of the template for a valid match; x pr Indicates the effective match to the template surrounding point's X and Y coordinates. pr Indicates the Y coordinate of the valid match around the template point; x qc This indicates the X and Y coordinates of the center point of the sample for a valid match. qc This indicates the Y-coordinate of the center point of the sample for a valid match, x qr Indicates the effective match to the X coordinates and Y coordinates of the points surrounding the sample. qrThe center point and surrounding point are essentially feature points. Δθ represents the rotation angle of the sample matching pair relative to the template matching pair in the effective matching pair, where Δθ∈(-π,π]; Δx represents the translation of the sample matching pair relative to the template matching pair in the x-axis direction in the effective matching pair, and Δy represents the translation of the sample matching pair relative to the template matching pair in the y-axis direction in the effective matching pair.

[0046] As a preferred technical solution, in step S5, if the transformation parameters of a valid matching pair are within the tolerance range of the fingerprint transformation parameters, the pair is determined to be a correct matching pair; otherwise, it is an incorrect matching pair. The tolerance range of the transformation parameters is as follows:

[0047]

[0048] Where (Δθ, Δx, Δy) are the transformation parameters of the effective matching pair, θ b It is the optimal transformation angle between the sample fingerprint feature point set and the template fingerprint feature point set, x b It is the optimal x-coordinate translation, y b It is the optimal y-coordinate translation amount, T θ It is the angle tolerance, T x It is the tolerance for x-axis translation, T y It is the tolerance for y-axis translation. The optimal transformation angle, optimal x-coordinate translation, and optimal y-coordinate translation are derived from the optimal transformation parameters described in step S4.

[0049] As a preferred technical solution, in step S6, the ANN model includes a first fully connected layer, a first BN layer, a first ReLU layer, a second fully connected layer, a second BN layer, a second ReLU layer, a Dropout layer, a third fully connected layer, and a Sigmoid classifier connected in sequence; wherein, the first fully connected layer is used to input parameters representing the global similarity of fingerprints, and the Sigmoid classifier is used to output the similarity of fingerprints.

[0050] Compared with the prior art, the present invention has the following advantages:

[0051] (1) The fingerprint matching algorithm proposed in this invention has good portability and can be used for low-power devices. It is suitable for personnel fingerprint authentication in daily environments. At the same time, the data source complies with the ISO / IEC19794-2 standard and has wide substitutability.

[0052] (2) The surrounding point matching algorithm designed in this invention avoids the problem that common feature point-based fingerprint recognition algorithms need to perform prior alignment of feature point sets, simplifies the calculation process, reduces the amount of calculation, and can effectively improve the algorithm time of fingerprint matching while ensuring recognition accuracy, thus improving the user experience.

[0053] (3) The present invention uses the ANN algorithm, which greatly reduces the data dimension without losing important features and improves the fingerprint matching speed. Attached Figure Description

[0054] Figure 1 This is a schematic flowchart of a fingerprint matching method based on feature point matching according to the present invention.

[0055] Figure 2 A diagram illustrating the relevant parameters for matching;

[0056] Figure 3 Flowchart for matching surrounding points;

[0057] Figure 4 This is a schematic diagram of the center point iteration;

[0058] Figure 5 This is a schematic diagram of an ANN network model. Detailed Implementation

[0059] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0060] Example 1

[0061] like Figures 1 to 5 As shown, in order to address the above problems, this invention proposes a fingerprint matching method based on feature point matching.

[0062] The fingerprint matching method includes calculating the coordinate offset of each fingerprint; sorting fingerprint feature points in a specific order; establishing connections between sample and template feature points to find effective matching pairs; calculating the globally optimal transformation parameters of the feature point set based on the effective matching pairs, applying global constraints to the effective matching pairs, and obtaining the optimal feature point matching pairs; calculating ten matching indicators for the two fingerprints based on the optimal matching pairs, and using an ANN to predict the similarity between the template data and the sample data. This scheme avoids the inefficiency caused by prior alignment of feature points, while applying global constraints to the matching pairs to ensure the accuracy of the matching pair structure, and can adapt to fingerprint images with large deformations. In addition, the scheme mostly uses simple integer and logical operations, so the performance requirements of hardware devices are not high, and it has good portability.

[0063] The present invention specifically adopts the following overall process:

[0064] For the sample fingerprint and template fingerprint to be matched, a subset of representative feature points (minutia) are selected to represent the fingerprint. Each feature point contains four parameters (x, y, θ, t), all conforming to the definitions in the international standard ISO / IEC 19794-2:2011. The definitions of the x and y coordinates and their axes are derived from Section 6.4.2, the angle θ from Section 6.4.3, and the type t includes endpoints and bifurcation points, defined from Section 6.3.1. The corresponding fingerprint matching algorithm comprises the following steps; see the flowchart for details. Figure 1 :

[0065] S1, Reference point selection: Based on the center coordinates of the effective region of the sample fingerprint and the upper, lower, left, and right boundaries of the effective region of the sample fingerprint, calculate a number of sample reference points for sorting sample feature points; and, based on the center coordinates of the effective region of the template fingerprint and the upper, lower, left, and right boundaries of the template fingerprint region, calculate a number of template reference points for sorting template feature points.

[0066] S2, Feature point sorting: Based on the coordinates of the reference point selected in step S1, sort the feature points of the sample fingerprint feature point set and the template fingerprint feature point set before sorting, respectively, to obtain the sample fingerprint feature point set and the template fingerprint feature point set after sorting.

[0067] S3, Finding valid matching pairs: According to a certain iteration order, several feature points in the sample fingerprint feature point set are respectively used as sample center points, and several sample feature point matching pairs are established for each sample center point; according to a certain iteration order, several feature points in the template fingerprint feature point set are respectively used as template center points, and template feature point matching pairs are established for each template center point; then each sample feature point matching pair is paired with each template feature point matching pair individually, and the successfully matched sample feature point matching pairs and template feature point matching pairs are collectively called valid matching pairs;

[0068] S4, Center Point Iteration: If there are feature points in the sample fingerprint feature point set that meet specific conditions but are not used as sample center points for operation S3, then use these feature points as the new sample center points for S3 and repeat S3; if there are template feature points in the template fingerprint feature point set that meet specific conditions but are not used as template center points for operation S3, then use these template feature points as the new template center points for S3 and repeat S3; if none of the above conditions are met, then jump to S5.

[0069] S5, Global Matching Constraint: Calculate the transformation parameters of all valid matching pairs, obtain the optimal transformation parameters based on the transformation parameters of the valid matching pairs, and then remove valid matching pairs that are outside a certain tolerance range based on the optimal transformation parameters; where the optimal transformation parameters refer to the globally optimal solution with the transformation parameters of the valid matching pairs as the decision variable and the global pairing result of the sample fingerprint feature point set and the template fingerprint feature point set as the objective.

[0070] S6, Matching Result Prediction: Construct an ANN prediction model, using the matching index corresponding to the global optimal solution as the ANN prediction model to predict the similarity between the sample fingerprint feature point set and the template fingerprint feature point set.

[0071] The more specific technical solution is as follows:

[0072] 1. Benchmark point calculation:

[0073] The benchmark point calculation method designed in this invention is as follows:

[0074] (1) Calculate the maximum convex hull containing all fingerprint feature points as the effective fingerprint region. The minimum Y coordinate of all points forming the convex hull is the upper boundary of the effective fingerprint region, the maximum Y coordinate of all points forming the convex hull is the lower boundary of the effective fingerprint region, the minimum X coordinate of all points forming the convex hull is the left boundary of the effective fingerprint region, and the maximum Y coordinate of all points forming the convex hull is the right boundary of the effective fingerprint region.

[0075] (2) Calculate the fingerprint center and upper, lower, left, and right boundaries based on the valid fingerprint area, and calculate the five reference points used for fingerprint sorting, as shown in the following formula:

[0076]

[0077]

[0078] C center =(x c ,y c ),

[0079] C up =(x c ,y c -Δy offset ),

[0080] C down =(x c ,y c +Δy offset ),

[0081] C left =(x c -Δx offset ,yc ),

[0082] C rigth =(x c +Δx offset ,y c ),

[0083] Among them, (x c ,y c ) represents the center coordinates of the valid fingerprint region, y min It is the upper boundary of the valid fingerprint area, y max It is the lower boundary of the valid fingerprint area, x min It is the left boundary of the valid fingerprint area, x max It is the right boundary of the valid fingerprint area, C center It is the central reference point, C up It is the upper reference point, C down It is the lower reference point, C left It is the left reference point of the fingerprint, C right It is the right reference point of the fingerprint; Δx offset It is the x-coordinate offset of the left and right reference points of the fingerprint relative to the central reference point, Δy offset It is the y-coordinate offset of the upper and lower reference points of the fingerprint relative to the center reference point.

[0084] Perform the above operations on the template fingerprint feature point set and the sample fingerprint feature point set to obtain the template fingerprint reference point and the sample fingerprint reference point, respectively.

[0085] 2. Feature point sorting:

[0086] The feature point sorting method designed in this invention is as follows:

[0087] (1) Establish an empty set of fingerprint feature points, called the sorted fingerprint feature point set, or simply the feature point set. Sort all feature points in the unsorted fingerprint feature point set according to their distance from the fingerprint center reference point, from closest to farthest. Select the two feature points closest to the fingerprint center reference point and add them to the feature point set in the order from closest to farthest.

[0088] (2) Select the feature point in the fingerprint feature point set that is closest to the fingerprint upper / lower / left / right reference point and has not yet been added to the feature point set, and add this point as the third / fourth / fifth / sixth feature point to the feature point set;

[0089] (3) Before sorting, all feature points in the fingerprint feature point set that have not yet been added to the feature point set are sorted from near to far according to the distance of the feature points to the fingerprint center reference point, and added to the feature point set in this order.

[0090] Through the above steps, the fingerprint feature point set of the sample before sorting and the fingerprint feature point set of the template before sorting are sorted respectively to obtain the fingerprint feature point set of the sample after sorting and the fingerprint feature point set of the template after sorting.

[0091] 3. Find valid matching pairs:

[0092] For the template feature point set {P1, P2, ..., P...} n} and the sample feature point set {Q1,Q2,…,Q} n First, assume that one feature point is selected from the sample feature point set and one feature point from the template feature point set according to a specific order, and these are respectively used as the sample center point P. s and template center point Q s .

[0093] Select sample feature points other than the sample center point from the sample feature point set as the sample surrounding points P of the sample center point. r Similarly, template feature points other than the template center point are selected from the template feature point set as the template surround points Q of the template center point. r .

[0094] Connecting the center point of the template to the surrounding points is called the template feature point pair {P}. s ,P r}, connecting the center point of the sample with the surrounding points, is called the sample feature point pair {Q}. s Q r}

[0095] Assuming that the center point pairs formed by the sample center point and the template center point can match each other, the following matching parameters can be derived, such as... Figure 2 As shown. The left side represents template feature point pairs, and the right side represents sample feature point pairs. The coordinates of the template center point are (x...). pc y pc The coordinates of the center point of the template are (x, y). pr y pr The coordinates of the center point of the template are (x, y). qc y qc The coordinates of the center point of the template are (x, y). qr y qr ). d p It is the distance between template feature point matching pairs, d q It is the distance between matching pairs of sample feature points; t pc It is point P c Feature point type, t pr It is point P r Feature point type, t qc It's point Q. c Feature point type, t qr It's point Q.r Feature point types; γ p It is the number of ridges traversed by the template feature point matching pair, γ q This refers to the number of ridges traversed by the sample feature point matching pairs. The number of ridges follows the definition in Section 8.5.2 of the international standard ISO / IEC 19794-2:2011, and the same applies below; α p It is the orientation angle of the center point of the template feature point matching pair, α q β is the orientation angle of the center point of the sample feature point matching pair. p It is the orientation angle of the template feature point matching pair around the point, β q It is the orientation angle of the sample feature point matching pair around the point; θ p It is the angle between the line connecting the center point of the template and the surrounding point and the positive direction of the X-axis. θ q It is the angle between the line connecting the center point of the sample pair and the surrounding points and the positive direction of the X-axis, with the center point of the sample pair as the center. These features are only related to the relative positional relationship between the two feature points, and are not related to the translation or rotation angle of the fingerprint.

[0096] To accommodate issues such as quantization errors, deformation of fingers when applied to the sensor, and inherent sensor defects, a tolerance mechanism needs to be designed for the matching pairs. Successfully matched pairs must meet the following conditions:

[0097]

[0098] In the above formula, T γ It is the tolerance for the number of ridge lines, T d It is the distance tolerance, T αβ It is the feature point angle tolerance, T βθ This refers to the tolerance of the angle difference between the lines connecting the feature points. All of the above parameters can be discretized to fully utilize the integer arithmetic units of low-power devices, reducing computational load and improving operational efficiency.

[0099] This invention will use template feature point pairs {P} that satisfy the above conditions s ,P r} and sample feature point pairs {Q s Q r} collectively referred to as a valid matching pair A = {P} s Q s ,P r Q r Record valid matching pairs and reselect surrounding feature points P. r and Q r Repeat the above steps until all template surrounding feature points and sample surrounding feature points of the center point pair have been matched with each other at least once.

[0100] When creating a matching pair, only points whose distance falls within a specific interval are selected, i.e., d. s ,d t ∈[d down ,d up ], where the upper bound is d up The lower bound is d down Because fingerprint deformation typically causes significant angular quantization errors at close range and significant distance quantization errors at long range, the selection of upper and lower bounds should comprehensively consider parameters such as sensor size, resolution, and sensor characteristics to obtain the best matching effect. This invention selects an upper bound of 120 pixels and a lower bound of 40 pixels.

[0101] Figure 3 This demonstrates the matching results of surrounding points centered on feature points P1 and Q1. The left side of the figure shows the template feature point set, and the right side shows the sample feature point set. The circular dashed lines in the figure represent the upper and lower bounds of the matching for the central feature points. Successfully matched feature point pairs are connected by dashed lines. Some surrounding feature points in the figure exceed the matching boundaries or fail to match.

[0102] 4. Center point iteration:

[0103] (1) Select the first six feature points from the sorted template dataset as the template center points for the first six iterations. All feature points in the sorted sample dataset are used as sample center points. Each template center point establishes a center point pair with each sample center point. Find valid matching pairs according to the method in Section 2.

[0104] (2) Select the first six feature points from the sorted sample dataset as the sample center points for the 7th to 12th iterations. All feature points in the sorted template dataset are used as template center points. Each sample center point is paired with a template center point (the center point pairs established in the previous step do not need to be repeated). Find valid matching pairs according to the method in Section 2.

[0105] (3) Obtain a set of several valid matching pairs {A1, A2, ..., A...} from the first 12 iterations through the above steps. n After that, surround points are extracted from each valid matching pair to form surround point pairs (P). rn Q rn The extracted surrounding point pairs that have not been matched as center point pairs are used as new center point pairs, and their valid matching pairs are found according to the method in Section 2. This step is iterated repeatedly until all surrounding point pairs of valid matching pairs have been matched as center point pairs. If a sufficient number of valid matching pairs cannot be obtained after multiple iterations, it indicates that the fingerprint is likely not a match, and the result can be output directly.

[0106] Figure 4This demonstrates the feature point matching order of the template dataset and the sample dataset. The row coordinates represent the 12 feature points in the template dataset, numbered {A, B, C, ..., L}; the columns represent the 13 feature points in the sample dataset, numbered {a, b, c, ..., m}. The number in the cell indicates in which iteration the template feature point and the sample feature point were matched as center points; blank cells indicate that no matching was performed.

[0107] The feature points of the template and sample datasets have already been sorted, so the matching structure in steps 1 and 2 can roughly assess the fingerprint similarity. If the number of valid matches is too small after 12 matches in steps 1 and 2, it means that the template fingerprint and the sample fingerprint are likely not a match, so subsequent steps can be skipped and the result returned directly.

[0108] 5. Global matching degree constraint:

[0109] The effective matching pair model only performs local structural matching of fingerprints and cannot effectively evaluate the global similarity of fingerprints. Imposing global matching constraints on the effective matching pairs by transforming their parameters can yield better matching results. The calculation method for the effective matching pair transformation parameters is as follows: Δθ = θ p -θ q ,

[0110] Δx=x pc -(x qc cosΔθ-y qc sinΔθ),

[0111] Δy=y pc -(x qc sinΔθ+y qc cosΔθ),

[0112] Where, x pc This indicates the effective match for the X and Y coordinates of the template center point. pc Indicates the Y-coordinate of the center point of the template for a valid match; x pr Indicates the effective match to the template surrounding point's X and Y coordinates. pr Indicates the Y coordinate of the valid match around the template point; x qc This indicates the X and Y coordinates of the center point of the sample for a valid match. qc This indicates the Y-coordinate of the center point of the sample for a valid match, x qr Indicates the effective match to the X coordinates and Y coordinates of the points surrounding the sample. qr θ represents the Y-coordinate of the valid match around the sample points; p The angle θ is the angle between the line connecting the center point of the template and the surrounding point and the positive direction of the X-axis. qThe angle between the center point of the sample pair and the line connecting the center point and the surrounding point and the positive direction of the X-axis is taken as the center. Δθ represents the rotation angle of the sample matching pair relative to the template matching pair in the effective matching pair, Δθ∈(-π,π]; Δx represents the translation of the sample matching pair relative to the template matching pair in the x-axis direction in the effective matching pair, and Δy represents the translation of the sample matching pair relative to the template matching pair in the y-axis direction in the effective matching pair. To speed up the comparison, the above parameters can be restricted. In this invention, to accommodate arbitrary angle rotation of the fingerprint image, Δθ∈(-π,π) is selected. Δx and Δy are generally not constrained, but the transformed feature point coordinates should not exceed the effective area of ​​another fingerprint.

[0113] By establishing a data distribution window and statistically analyzing the data distribution of transformation parameters, the optimal transformation parameters for two fingerprints can be calculated. This invention divides the rotation angle interval into 360 equal parts, each representing 1 degree in radians. The rotation angle data of all transformation parameters are accumulated into the corresponding angle data distribution window. The angle data distribution window will then display a maximum value, which is the optimal rotation angle θ for fingerprint transformation. b Similarly, for the remaining matching pairs, the x-coordinate transformation and y-coordinate transformation need to be calculated separately. The maximum value is the optimal x-coordinate transformation of the sample fingerprint relative to the template fingerprint. b With y coordinate b .

[0114] Under ideal conditions, after performing an affine transformation on the corresponding sample matching pair using the optimal transformation parameters of the effective matching pair, the sample matching pair should be able to completely coincide with the template matching pair (i.e., the sample center point and the template center point are in the same position and have the same orientation angle, and the sample surrounding point and the template surrounding point are in the same position and have the same orientation angle). However, in reality, due to fingerprint deformation and other reasons, they cannot be completely coincident. Therefore, this invention sets a certain tolerance, as shown below.

[0115]

[0116] Where (Δθ, Δx, Δy) are the transformation parameters of the effective matching pair, θ b It is the rotation angle of the optimal transformation parameters, x b It is the x-coordinate translation of the optimal transformation parameters, y b T is the y-coordinate translation of the optimal transformation parameters. θ It is the angle tolerance, T x It is the tolerance for x-coordinate translation, T yThis is the tolerance for the y-coordinate translation. For each valid match pair, if its transformation parameters are within this tolerance range, then the valid match pair can be determined as a correct match pair; otherwise, it is an incorrect match pair. If the number of global match pairs that meet the requirements is insufficient, then the sample fingerprint and the template fingerprint can be considered mismatched. The minimum number of global match pairs selected in this invention is 10.

[0117] 6. Matching Result Prediction

[0118] This invention extracts the 10 parameters that best represent the global similarity of fingerprints from the above steps, and establishes an ANN model based on these parameters, as shown below:

[0119] Valid matching pair count score: The number of valid matching pairs is mapped to the 0-1 interval using the sigmoid function;

[0120] Correct match count score: The number of correct match pairs after global match filtering, mapped to the 0-1 interval by the sigmoid function;

[0121] Ridge Count Correct Match Percentage: The number of ridges γ between template matching pairs in a correct match pair. p The number of ridges γ between the sample matching pairs q If they are equal, it is marked as correct; if they are not equal, it is marked as incorrect. The percentage of correct ridge number matches is the ratio of the number of correct ridge number matches to the total number of correct matches.

[0122] Point type correct match percentage: Among the correct matches, the template surrounds point type t pr and sample surrounding point type t qr If they are the same, it is recorded as a correct match; if they are not equal, it is recorded as a wrong match. The percentage of correct matches for point types is the ratio of the number of correct point type matches to the total number of correct match pairs.

[0123] Fingerprint effective region area difference rate: The ratio of the difference between the effective region areas of the template fingerprint and the sample fingerprint to the smaller of the effective region areas of the template fingerprint and the sample fingerprint; where the area of ​​the effective region is the area of ​​the convex hull.

[0124] Template fingerprint actual matching area ratio: The ratio of the effective area of ​​the template fingerprint to the actual acquisition area of ​​the template fingerprint image (i.e., the sum of the effective area and the background area).

[0125] Sample fingerprint actual matching area ratio: The ratio of the effective area of ​​the sample fingerprint to the actual acquisition area of ​​the sample fingerprint image (i.e., the sum of the effective area and the background area).

[0126] Template fingerprint matching interval feature point matching rate: The ratio of the number of feature points in the template fingerprint that successfully establish correct matching pairs (hereinafter referred to as the number of successfully matched feature points in the template fingerprint) to the total number of feature points in the template fingerprint;

[0127] Sample fingerprint matching interval feature point matching rate: The ratio of the number of feature points in the sample fingerprint that successfully establish correct matching pairs (hereinafter referred to as the number of successfully matched feature points in the sample fingerprint) to the total number of feature points in the sample fingerprint;

[0128] Effective matching interval feature point actual matching rate: the ratio of the smaller value of the number of successfully matched feature points between the template fingerprint and the sample fingerprint to the larger value of the total number of feature points between the template fingerprint and the sample fingerprint;

[0129] ANN model structure:

[0130] The ANN structure for fingerprint classification designed in this invention consists of three fully connected layers. The output of the last fully connected layer is fed into a Sigmoid classifier, which outputs the similarity between fingerprints. The network structure is as follows: Figure 5 As shown.

[0131] The first and second fully connected layers each have 32 neurons, while the third fully connected layer has only 1 neuron. The first two fully connected layers are followed by a batch normalization (BN) layer and a ReLU layer. The second ReLU layer is followed by a dropout layer to overcome overfitting. The dropout layer disables neurons with a given probability; disabled neurons neither participate in forward nor backward propagation. Through this strategy, the network learns more robust features during training. The dropout probability chosen in this invention is 0.5.

[0132] Model training:

[0133] Model training is performed according to the following steps:

[0134] 1. Extract fingerprint matching parameters from the global fingerprint matching results using the formula above;

[0135] 2. Split the dataset so that the ratio of training set to test set is 7:3;

[0136] 3. Initialize the weights: use Kaiming for the first and second fully connected layers, and use Xavier for the third layer.

[0137] 4. Set the loss function. This invention uses the binary cross-entropy loss function (BCE focal loss) to solve the problem of mismatch between the amount of positive and negative sample data.

[0138] 5. Train the ANN model.

[0139] After training, the ANN algorithm is tested using a test set.

[0140] The fingerprint matching algorithm proposed in this invention has good portability and can be used in low-power devices, adapting to personnel fingerprint authentication in daily environments. At the same time, the data source complies with the ISO / IEC 19794-2 standard and has wide substitutability.

[0141] The surrounding point matching algorithm designed in this invention avoids the problem of prior alignment of feature point sets required by common feature point-based fingerprint recognition algorithms. It simplifies the calculation process, reduces the amount of computation, and can effectively improve the algorithm time of fingerprint matching while ensuring recognition accuracy, thus improving the user experience.

[0142] This invention uses the ANN algorithm, which greatly reduces the data dimensionality without losing important features, thereby improving the fingerprint matching speed.

[0143] As described above, the present invention can be implemented well.

[0144] All features disclosed in all embodiments of this specification, or steps in all methods or processes implied in the disclosure, may be combined and / or extended or replaced in any way, except for mutually exclusive features and / or steps.

[0145] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Based on the technical essence of the present invention, any simple modifications, equivalent substitutions, and improvements made to the above embodiments within the spirit and principles of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A fingerprint matching method based on feature point matching, characterized in that, Fingerprint matching is achieved by matching feature points in the sample fingerprint feature point set and feature points in the template fingerprint feature point set. Includes the following steps: S1, Reference point selection: Based on the center coordinates of the effective region of the sample fingerprint and the upper, lower, left, and right boundaries of the effective region of the sample fingerprint, calculate a number of sample reference points for sorting sample feature points; and, based on the center coordinates of the effective region of the template fingerprint and the upper, lower, left, and right boundaries of the effective region of the template fingerprint, calculate a number of template reference points for sorting template feature points. S2, Feature point sorting: Based on the coordinates of the template reference point selected in step S1, sort the sample fingerprint feature point set and the template fingerprint feature point set before sorting respectively to obtain the sorted sample fingerprint feature point set and the template fingerprint feature point set. S3, Finding valid matching pairs: Take several feature points from the sample fingerprint feature point set as sample center points and establish several sample feature point matching pairs for each sample center point; take several feature points from the template fingerprint feature point set as template center points and establish template feature point matching pairs for each template center point; then pair each sample feature point matching pair with each template feature point matching pair individually. The successfully matched sample feature point matching pairs and template feature point matching pairs are collectively called valid matching pairs. S4, Center Point Iteration: If there are feature points in the sample fingerprint feature point set that meet the set conditions but are not used as sample center points for operation S3, then use these feature points as the new sample center points for S3 and repeat S3; if there are template feature points in the template fingerprint feature point set that meet the set conditions but are not used as template center points for operation S3, then use these template feature points as the new template center points for S3 and repeat S3; if none of the above conditions are met, then jump to S5. S5, Global Matching Constraint: Calculate the transformation parameters of all valid matching pairs, obtain the optimal transformation parameters based on the transformation parameters of the valid matching pairs, and then remove valid matching pairs that are outside the set tolerance range based on the optimal transformation parameters; where the optimal transformation parameters refer to the globally optimal solution with the transformation parameters of the valid matching pairs as the decision variable and the global pairing result of the sample fingerprint feature point set and the template fingerprint feature point set as the objective. S6, Matching Result Prediction: Construct an ANN prediction model, using the matching index corresponding to the global optimal solution as the ANN prediction model to predict the similarity between the sample fingerprint feature point set and the template fingerprint feature point set.

2. The fingerprint matching method based on feature point matching according to claim 1, characterized in that, In step S1, the effective region of the sample fingerprint represents the region enclosed by the largest convex hull containing all fingerprint feature points. The center coordinates of the effective region are obtained by averaging the X and Y coordinates of all pixels within the effective region. The minimum Y coordinate of all points forming the convex hull is the upper boundary of the effective region, the maximum Y coordinate is the lower boundary, the minimum X coordinate is the left boundary, and the maximum Y coordinate is the right boundary. The effective region of the template fingerprint represents... The effective region of the template fingerprint is the area enclosed by the largest convex hull containing all fingerprint feature points. The center coordinates of the effective region of the template fingerprint are obtained by averaging the X-coordinates and Y-coordinates of all pixels in the effective region of the template fingerprint. The minimum Y-coordinate of all points forming the convex hull is the upper boundary of the effective region of the template fingerprint, and the maximum Y-coordinate of all points forming the convex hull is the lower boundary of the effective region of the template fingerprint. The minimum X-coordinate of all points forming the convex hull is the left boundary of the effective region of the template fingerprint, and the maximum Y-coordinate of all points forming the convex hull is the right boundary of the effective region of the template fingerprint. The coordinate axes used to measure the X and Y coordinates are defined according to Section 6.4.2 of ISO / IEC 19794-2:2011.

3. The fingerprint matching method based on feature point matching according to claim 2, characterized in that, In step S1, the template reference point is calculated as follows: in, These are the center coordinates of the valid fingerprint area. It is the upper boundary of the valid fingerprint area. It is the lower boundary of the valid fingerprint area. It is the left boundary of the valid fingerprint area. It is the right boundary of the valid fingerprint area. It is the fingerprint center reference point. It is the reference point on the fingerprint. It is the fingerprint reference point. It is the left reference point of the fingerprint. It is the right reference point of the fingerprint. It is the x-coordinate offset of the left and right reference points of the fingerprint relative to the central reference point. It is the y-coordinate offset of the upper and lower reference points of the fingerprint relative to the center reference point; the fingerprint refers to the sample fingerprint or the template fingerprint. The reference point of the template fingerprint is calculated using the center coordinates, upper, lower, left and right boundaries of the valid area of ​​the template fingerprint, and the reference point of the sample fingerprint is calculated using the center coordinates, upper, lower, left and right boundaries of the valid area of ​​the sample fingerprint.

4. The fingerprint matching method based on feature point matching according to claim 3, characterized in that, Step S2, the feature point sorting method, includes the following steps: S21. Establish an empty set of fingerprint feature points, called the sorted set of fingerprint feature points. Sort all feature points in the original fingerprint feature point set according to their distance from the fingerprint center reference point from near to far. Select the two feature points closest to the fingerprint center reference point and add them to the sorted set of fingerprint feature points in order from near to far. S22, Select the feature point in the fingerprint feature point set before sorting that is closest to the upper / lower / left / right reference point of the fingerprint and has not yet been added to the fingerprint feature point set after sorting, and add this point as the third / fourth / fifth / sixth feature point to the fingerprint feature point set after sorting; S23, all feature points in the fingerprint feature point set that have not yet been added to the sorted fingerprint feature point set before sorting are sorted from near to far according to the distance of the feature points to the fingerprint center reference point, and added to the sorted fingerprint feature point set in this order. S24. Through the above steps, sort the sample fingerprint feature point set and the template fingerprint feature point set before sorting, respectively, to obtain the sorted sample fingerprint feature point set and the sorted template fingerprint feature point set.

5. The fingerprint matching method based on feature point matching according to claim 4, characterized in that, In step S3, the method for constructing sample feature point matching pairs is as follows: Select a sample feature point other than the sample center point from the sample feature point set, and use it as the sample surrounding point of the sample center point. The distance between the sample surrounding point and the sample center point should be within a certain range; establish a connection between the sample center point and the sample surrounding point of the sample center point, which is a sample feature point matching pair; the sample center point needs to establish a connection with all its sample surrounding points that meet the conditions. The construction method of template feature point matching pair is as follows: Select a template feature point other than the template center point from the template feature point set, and use it as the template surrounding point of the template center point. The distance between the template surrounding point and the template center point should be within a certain range; establish a connection between the template center point and the template surrounding point of the template center point, which is a template feature point matching pair; the template center point needs to establish a connection with all its template surrounding points that meet the conditions.

6. The fingerprint matching method based on feature point matching according to claim 5, characterized in that, In step S3, a successful match is considered to be a pair of sample feature points and a pair of template feature points that satisfy the following conditions: ； in, It is the distance between template feature point matching pairs. It is the distance between matching pairs of sample feature points. It is the number of ridges traversed by the template feature point matching pair. It is the number of ridges traversed by the sample feature point matching pair, and the number of ridges follows the definition in section 8.5.2 of the international standard ISO / IEC 19794-2:2011; It is the orientation angle of the center point of the template feature point matching pair. It is the orientation angle of the center point of the sample feature point matching pair. It is the orientation angle of the template feature point matching pair around the point. It is the orientation angle of the sample feature point matching pair around the point; It is the angle between the line connecting the center point of the template and the surrounding point and the positive direction of the X-axis. It is the angle between the center point of the sample pair and the line connecting the center point of the sample pair and the surrounding point and the positive direction of the X-axis; It is the tolerance for the number of ridge lines. It is distance tolerance. It is the feature point angle tolerance. It is the tolerance for the angle difference of the lines connecting the feature points.

7. The fingerprint matching method based on feature point matching according to claim 6, characterized in that, In step S4, the transformation parameters of the effective matching pair are calculated using the following formula: ； in, This indicates that the X coordinate of the center point of the template is a valid match. This indicates the Y-coordinate of the valid match relative to the center point of the template; This indicates the X coordinate of the valid match around the template points. Indicates the Y coordinate of the valid match around the template point; This indicates the X-coordinate of the center point of a valid match sample. This indicates the Y-coordinate of the center point of the sample that represents a valid match. This indicates the X coordinates of the points surrounding the sample that are in a valid match. This indicates the Y-coordinate of the valid match for the sample's surrounding points; the aforementioned center points and surrounding points are essentially all feature points. This indicates the rotation angle of the sample match pair relative to the template match pair in the valid match pair. ; This represents the amount of translation of the sample match pair relative to the template match pair in the x-axis direction within the valid match pair. This indicates the amount of translation of the sample match pair relative to the template match pair in the y-axis direction within the valid match pair.

8. The fingerprint matching method based on feature point matching according to claim 7, characterized in that, In step S5, if the transformation parameters of a valid matching pair are within the tolerance range of the fingerprint transformation parameters, it is determined to be a correct matching pair; otherwise, it is an incorrect matching pair. The tolerance range of the transformation parameters is as follows: ； in, These are the transformation parameters of the valid matching pairs. It is the optimal transformation angle between the sample fingerprint feature point set and the template fingerprint feature point set. It is the optimal x-coordinate translation amount. It is the optimal y-coordinate translation amount. It's the angle tolerance. It is the tolerance for x-axis translation. It is the tolerance for y-axis translation. The optimal transformation angle, optimal x-coordinate translation, and optimal y-coordinate translation are derived from the optimal transformation parameters described in step S4.

9. A fingerprint matching method based on feature point matching according to any one of claims 1-8, characterized in that, In step S6, the ANN model includes a first fully connected layer, a first batch normalization (BN) layer, a first ReLU layer, a second fully connected layer, a second batch normalization (BN) layer, a second ReLU layer, a Dropout layer, a third fully connected layer, and a Sigmoid classifier, which are connected in sequence. The first fully connected layer is used to input parameters representing the global similarity of fingerprints, and the Sigmoid classifier is used to output the similarity of fingerprints.

Images