Signature correction method, system, device and storage medium based on multi-algorithm fusion

Abstract

The present invention claims protection for an electronic signature correction technology based on eigenvalue decomposition, which relates to the field of electronic signature technology. The technology comprises the following steps: obtaining a signature stroke position sequence to establish a covariance matrix; performing eigenvalue decomposition on the covariance matrix, projecting the coordinates of all stroke points in the direction of the eigenvector to obtain the signature sequence coordinates, and obtaining the corrected coordinates of the stroke points based on the eigenvalue decomposition; obtaining a minimum bounding rectangle based on the convex hull of the stroke point set, calculating the stroke rotation angle based on the minimum bounding rectangle, and rotating the stroke points according to the angle to obtain the corrected coordinates of the stroke points based on the minimum bounding rectangle; and fusing the two corrected coordinates to obtain the corrected position coordinates of the signature strokes. This method can be widely applied in electronic signature verification scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic signatures, and in particular to a method for correcting electronic signatures. Background Art

[0002] Electronic signature verification involves numerous signature comparisons. However, due to the signing device or the signing posture, the signatory's signature may be skewed. This can lead to a loss of stroke feature information during electronic signature verification and comparison, as well as inaccurate feature information, resulting in incorrect verification results and reduced accuracy. Therefore, angle correction for electronic signatures is an essential part of pre-processing for signature comparison.

[0003] Publication number CN108256529A, Chinese invention patent application titled "Global point cloud description method based on point cloud projection contour signature and distribution matrix", discloses a global point cloud description method based on projection contour signature and distribution matrix, which uses the spatial coordinate information of the target point cloud to construct a local reference coordinate system with rotation and translation invariance, and then transforms the target point cloud to the reference coordinate system. Construct a spatial bounding box of the target point cloud after posture normalization, and project the target point cloud onto three adjacent planes of the bounding box. In order to enhance the descriptor's ability to represent the visible part of the target point cloud, the projection contour signature and distribution matrix of the point cloud projection on each projection plane are calculated respectively, and the relevant concatenation rules are determined based on the statistical characteristics of the point cloud projection. The sub-features of the target point cloud on the three projection planes are sequentially concatenated in the form of a histogram to obtain the final CSDM descriptor. This document publicly constructs a reference coordinate system with rotation, and then transforms the target point to the reference coordinate system. However, in the case where the original signature is skewed, projection transformation is performed without necessary angle correction, and the projection contour signature and distribution matrix are calculated, which is not suitable for signature verification and comparison scenarios with high accuracy requirements. Summary of the Invention

[0004] In order to solve the problem that the existing technology does not correct the skewed signature during the signature verification and comparison process, resulting in low recognition and verification accuracy, the present invention proposes an electronic signature correction method based on eigenvalue decomposition to reduce the inaccurate signature comparison results caused by the angle deviation of the electronic signature.

[0005] The technical solution of the present invention to solve the above technical problems is, on the one hand, a signature correction algorithm based on multi-algorithm fusion uses the minimum enclosing rectangle on the graph, integrates the signature correction algorithm based on eigenvalue decomposition, and provides an electronic signature correction method based on eigenvalue decomposition, including the following steps: a covariance calculation module obtains the signature stroke position sequence and establishes the covariance matrix of the sequence x-axis and y-axis; performs eigenvalue decomposition on the covariance matrix to obtain two linearly independent eigenvectors; a direction projection module projects the coordinates of all stroke points in the direction of the two eigenvectors to obtain the coordinates of the signature sequence, and obtains the corrected coordinates (x, y) of the stroke point (x, y) based on the eigenvalue decomposition. svd ,y svd ); calculate the convex hull of the stroke point set to obtain the minimum enclosing rectangle, calculate the stroke needle rotation angle based on the minimum enclosing rectangle, and obtain the corrected coordinates (x, y) of the stroke point (x) based on the minimum enclosing rectangle after rotating according to the direction angle. rect ,y rect ); fusion correction coordinates (x svd ,y svd ) and (x rect ,y rect ), get the fusion coordinates of the signature stroke points; the fusion coordinates (x fusion ,y fusion ) rotate θ clockwise final Degrees, get the corrected position coordinates of the signature strokes.

[0006] Preferably, the covariance matrix can be established by: Construct the covariance matrix C: Among them, n is the sequence length, x i 、y i is the x and y coordinates of the i-th stroke point; perform eigenvalue decomposition according to the formula: Cv = λv, and remove In addition, two linearly independent orthogonal eigenvectors (v1, v2) are obtained, and their corresponding eigenvalues ​​are λ1 and λ2 respectively.

[0007] Preferably, the covariance matrix can be established by extracting the x and y sequences of the signature stroke sequence, and according to the formula: ij =Cov(X i , Y j )=E[X i -E(X)][Y j -E(Y)] calculate the covariance c of the i-th point of the signature stroke ij , establish the corresponding covariance matrix C as: in, E(x) represents the mean of the sequence X of the stroke points in the x direction, E(y) represents the mean of the sequence Y of the stroke points in the y direction, n is the number of stroke points, (xi ,y i ) represents the coordinate value of the i-th stroke point.

[0008] Preferably, according to the formula: (x * ,y * )=(x,y)·(v1,v2) Project the coordinates of each stroke point in the direction of the two eigenvectors to obtain the projected signature sequence (x * ,y * ), (x * ,y * ) According to the formula: x svd =x * -min(x * ),y svd =y * -min(y * ) moves, and the stroke coordinate value (x, y) is obtained. The stroke point correction coordinate (x svd ,y svd ). For the stroke point sequence matrix (X, Y) composed of all the coordinates of the stroke points, according to the formula: (X * ,Y * )=(X,Y)·[v1,v2] calculates the result of the projection of the stroke point sequence matrix (X,Y).

[0009] Preferably, the divide-and-conquer method can be used to calculate the convex hull of the point set, and the point corresponding to the minimum cosine value of the angle between the initial point of the stroke and any point in the stroke except the initial point is found. Then, with this corresponding point as the starting point, the point corresponding to the minimum cosine value of the angle between the starting point and any other points in the stroke is found until the initial point of the stroke is returned. The polygon connected by all points is the convex hull of the electronic signature; one endpoint of any side of the convex hull of the electronic signature is used as the origin to rotate so that the side is parallel to the x-axis, and the area of ​​the candidate circumscribed rectangle corresponding to the side is calculated. The rotation and area calculation of the candidate circumscribed rectangle are completed for all sides of the signature. The convex hull corresponding to the side with the smallest area constitutes the minimum circumscribed rectangle of the signature. According to the corresponding side of the minimum circumscribed rectangle, the signature stroke is rotated clockwise by an angle α to obtain the coordinates (x # ,y # ), (x # ,y # ) According to the formula: x rect =x # -min(x # ),y rect =y # -min(y # ), after moving, we get the corrected coordinates (x rect ,y rect ),in, Among them, the coordinates of the two end points corresponding to edge j are (x j ,y j ),(x j+1 ,y j+1 ).

[0010] Preferably, according to the formula Correct the coordinates (x svd ,y svd ) and the coordinates of the signature sequence after rotation (x # ,y # ) to obtain the corrected fusion coordinates of the signature stroke point coordinates (x, y). The weight coefficients of the numerator and denominator in the formula can be adjusted according to the situation (the optimal values ​​are 2 and 3).

[0011] Preferably, based on the fusion result (x fusion ,y fusion ), use a sliding window to slide on the signature sequence, and for the w points in the window, calculate the minimum x coordinate x in its convex hull min , the largest x-coordinate x max ,exist In the interval, from x min At the beginning, the step size is 1 to iterate, calculate the height difference between the upper and lower bounds of the corresponding convex hull, and obtain the height difference feature under the sliding window Among them, h i The height difference between the upper and lower bounds of the convex hull corresponding to the i-th iteration value. According to the stroke corresponding to the maximum value of the average height difference in all sliding windows, the vertical stroke with the best verticality is determined, and the final rotation angle θ is determined based on the vertical stroke. final .

[0012] Preferably, the eigenvector with the largest eigenvalue is obtained, and the index value I corresponding to the larger λ is obtained. max , The corresponding eigenvector is The two values ​​of and The vectors it consists of are Determine whether the angle formed by the vector is greater than the angle θ final According to the angle θ final The corresponding unit vector (a,b) T , and the vector Composition matrix A:

[0013] If det(A)<0, then according to the formula: Determine the final coordinate point of the stroke

[0014] In the second aspect, the present invention also proposes an electronic signature correction system based on eigenvalue decomposition, which includes: a covariance calculation module, an eigenvalue decomposition module, a directional projection module, a signature correction module, and a fusion module. The covariance calculation module obtains the signature stroke position sequence and establishes the covariance matrix of the sequence x-axis and y-axis; performs eigenvalue decomposition on the covariance matrix C to obtain two linearly independent eigenvectors; the directional projection module projects the coordinates of all stroke points in the direction of the two eigenvectors to obtain the x-coordinate and y-coordinate of the signature sequence, and moves the stroke point (x, y) to obtain the correction coordinate (x) based on the eigenvalue decomposition. svd ,y svd ); The signature correction module calculates the convex hull of the stroke point set to obtain the minimum enclosing rectangle, calculates the rotation angle of the stroke needle based on the minimum enclosing rectangle, and obtains the corrected coordinates (x, y) of the stroke point based on the minimum enclosing rectangle after rotation according to the direction angle. rect ,y rect ); fusion module fusion correction coordinates (x svd ,y svd ) and (x rect ,y rect ), get the fusion coordinates of the signature stroke points, and convert the fusion coordinates (x fusion ,y fusion ) rotate θ clockwise final Degrees, get the corrected position coordinates of the signature strokes.

[0015] In a third aspect, the present invention further proposes a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the electronic signature correction method described in the present invention.

[0016] In a fourth aspect, the present invention also proposes an electronic device comprising: one or more processors; a memory; and one or more applications stored in the memory and configured to be loaded and run by the one or more processors in order to execute the above-mentioned electronic signature correction method of the present invention.

[0017] The present invention is based on a signature correction algorithm that integrates multiple algorithms. It uses a minimum enclosing rectangle method, integrating signature correction algorithms based on eigenvalue decomposition and minimum enclosing rectangle. The two complement each other, ultimately making the overall algorithm more robust. The method proposes a convex hull feature based on a sliding window, so that the final correction result can be correctly evaluated. The method uses the search results as the final result, making the correction result more accurate, with an average error of no more than 2°. It can quickly and accurately determine the correct direction of the electronic signature, with simple, efficient, and robust operation, and can quickly and effectively correct electronic signatures in various acquisition environments. It can quickly and accurately detect and correct signature skew that occurs during the signing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Attachment Figure 1 Schematic diagram of the electronic signature correction system based on multi-algorithm fusion of the present invention;

[0019] like Figure 2 Shown is a schematic diagram of the signature correction effect of the present invention. DETAILED DESCRIPTION

[0020] In order to facilitate a clear understanding of the present invention, the technical solution of the present invention is described below in conjunction with the accompanying drawings and specific examples. Obviously, the accompanying drawings in the following description are some embodiments of the present invention.

[0021] The present invention provides a simple and efficient electronic signature correction system based on multi-algorithm fusion, such as Figure 1 The figure shows an electronic signature correction system based on eigenvalue decomposition, which includes: covariance calculation module, eigenvalue decomposition module, direction projection module, signature correction module, and fusion module. The covariance calculation module obtains the signature stroke position sequence and establishes the covariance matrix of the sequence x-axis and y-axis; the covariance matrix is ​​subjected to eigenvalue decomposition to obtain two linearly independent eigenvectors; the direction projection module projects the coordinates of all stroke points in the direction of the two eigenvectors to obtain the coordinates of the signature sequence, and moves the stroke points to obtain the corrected coordinates (x svd ,y svd ); The signature correction module calculates the convex hull of the stroke point set to obtain the minimum enclosing rectangle, calculates the stroke needle rotation angle based on the minimum enclosing rectangle, and obtains the corrected coordinates (x, y) of the stroke point based on the minimum enclosing rectangle after rotation according to the direction angle. rect ,y rect ); fusion module fusion correction coordinates (x svd ,y svd ) and (x rect ,y rect ), get the fusion coordinates of the signature stroke points, and convert the fusion coordinates (x fusion ,y fusion ) Rotate clockwise to obtain the corrected position coordinates of the signature strokes.

[0022] Among them, the electronic signature correction module is constructed based on the minimum enclosing rectangle, including: calculating the convex hull of the point set, calculating the minimum enclosing rectangle, and angle rotation; fine-tuning based on the sliding window, result fusion, and convex hull feature calculation based on the sliding window to determine the final rotation angle.

[0023] Calculate the covariance matrix of the signature position sequence, take out the position sequence, and calculate the covariance matrix of the x-axis and y-axis.

[0024] Specifically, we can obtain the coordinates (x, y) of the signature stroke points, extract the x and y sequences of the signature stroke sequence, the sequence length is n, and calculate the covariance cov(x, y) based on the formula Construct the covariance matrix C: Among them, the coordinate values ​​of the i-th stroke point are x i and y i (The horizontal bars on x and y indicate the average value, and the cov function is used to calculate the covariance.)

[0025] C can also be expressed as:

[0026] The method of establishing the covariance matrix can also be: calculating the covariance c of the i-th point of the signature stroke ij , the specific formula can be: c ij =Cov(X i , Y j )=E[X i -E(X)][Y j -E(Y)], establish the corresponding covariance matrix in, E(x) represents the mean of the stroke point sequence X, E(y) represents the mean of the stroke point sequence Y, n is the number of stroke points, (x i ,y i ) represents the coordinate value of the i-th point.

[0027] The covariance matrix can also be established in other ways known to those skilled in the art.

[0028] Eigenvalue decomposition. Perform eigenvalue decomposition on the covariance matrix C to obtain two linearly independent eigenvectors (v1, v2). Perform eigenvalue decomposition according to the formula: Cv = λv, where all λ that meet the above formula are the eigenvalues ​​of the eigenvalue decomposition, and v that meets the above formula is the eigenvector of the eigenvalue decomposition. According to C, if and only if When the covariance matrix C is not full rank, for natural electronic signatures, since the data distribution is random, the covariance matrix C must be full rank, that is, the two v must be linearly independent, and since C is a real symmetric matrix, two orthogonal eigenvectors are obtained. Two linearly independent eigenvectors (v1, v2) are obtained, and their corresponding eigenvalues ​​are λ1 and λ2 respectively.

[0029] Directional projection. Project the coordinates of each stroke point in the direction of the feature vector. According to the formula: (x * ,y * )=(x,y)·(v1,v2) The coordinates of each stroke point (x i ,y i) is projected in the direction of the two eigenvectors, where (x, y) is the x-coordinate and y-coordinate of the signature sequence, with a size of n×2, [v1, v2] represents a matrix composed of two orthogonal eigenvectors, with a size of 2×2, (x * ,y * ) represents the x-coordinate and y-coordinate of the signature sequence obtained after projection.

[0030] In order to obtain more accurate projection position coordinates, the coordinates of the signature sequence obtained after projection (x * ,y * ) According to the following formula: svd =x * -min(x * ),y svd =y * -min(y * ) is moved, the stroke coordinate value (x, y) is obtained based on the stroke point correction coordinate (x svd ,y svd The above moving process moves all points to the first quadrant, and the min function finds the minimum value of the set.

[0031] Furthermore, a minimum bounding rectangle-based electronic signature correction model is constructed. First, the convex hull of the stroke point set is calculated. This can be done using a divide-and-conquer approach. The convex hull calculation method finds the point with the minimum cosine value of the angle between the stroke's initial point and any point in the stroke (except the initial point) in the y-coordinate direction. Using this corresponding point as the starting point, the corresponding point with the minimum cosine value of the angle between the y-coordinate direction and any other point in the stroke is then found. A line is drawn from the origin to this corresponding point, traversing all points. This process is repeated until the initial point is reached, forming the convex hull of the electronic signature.

[0032] The specific steps are as follows: (x0, y0) is recorded as the initial point of the stroke, where y0 is the y coordinate point of the signature data corresponding to the smallest y value of the collected stroke. If the smallest y value corresponds to multiple x values, the smallest x value is x0. The positive direction of y=y0 is an edge, and the intersection of (x0, y0) and any point (x i ,y i ) is another edge, with (x0,y0) as the vertex, and the constructed angle is θ i , let cos(θ i )The smallest (x i ,y i ) is (x1,y1), after obtaining (x1,y1), record the straight line formed by (x0,y0)->(x1,y1) as an edge, and then use (x1,y1) as the vertex and repeat the above steps to obtain (x2,y2)...(x k ,y k ), until (xk ,y k )=(x0,y0), then (x0,y0),...,(x k ,y k ) is the convex polygon formed by the electronic signature.

[0033] Next, calculate the minimum enclosing rectangle. For example, the convex hull of an electronic signature (x0, y0),..., (x k ,y k ) has k-1 edges. For each edge, take one of the endpoints as the origin, construct a coordinate system, rotate it so that the edge coincides with the x-axis, and record the new convex hull coordinates as Where i represents the number of the edge. The minimum and maximum coordinates are marked as: Then the candidate minimum circumscribed rectangle of the convex hull is The area of ​​the rectangle is Then the area of ​​all the minimum circumscribed rectangle candidates of the convex hull is S 1 ,...,S k , let S j =min(S 1 ,...,S k ), then the candidate minimum circumscribed rectangle corresponding to edge j is the minimum circumscribed rectangle of the convex hull.

[0034] Finally, the signature stroke is rotated clockwise according to the corresponding side of the minimum circumscribed rectangle. The rotation angle can be determined according to the slope of the side. The endpoint corresponding to side j is (x j ,y j ),(x j+1 ,y j+1 ), then its slope is The clockwise rotation angle α should be Note (x # ,y # ) represents the x-coordinate and y-coordinate of the signature sequence obtained after rotation.

[0035] In order to be more accurate, fine-tune the coordinate points and move all the points to the first quadrant. # ,y # ) According to the following formula: rect =x # -min(x # ),y rect =y # -min(y # ) after moving, we get (x rect ,y rect ) are the final corrected coordinates based on the minimum bounding rectangle.

[0036] Fine-tuning based on sliding window. First, fuse the above two correction results, and use the formula according to the predetermined weights. The two correction coordinates are fused to obtain the signature correction fusion result.

[0037] Next, based on the fusion result (x fusion ,y fusion ), use a sliding window of length w to slide on the signature sequence to verify the signature result. For example, when the window starts from subscript 1, the sliding window includes (x1, y1),..., (x w ,y w ) There are w points in total. For the w points in the window starting from the index i, it is recorded as (x window ,y window ), calculate its convex hull, and record the smallest x coordinate in the convex hull as x min , the maximum x coordinate is x max , then the x-value range of the convex hull is [x min ,x max ], and then we can get In this interval, from x min At the beginning, the step size is 1 to iterate, and each iteration is for the current x value, recorded as x i , the height difference between the upper and lower bounds of the corresponding convex hull is calculated, that is, x = x i The difference between the y values ​​of the two intersection points with the convex hull is recorded as h i , for the x coordinate outside the convex hull, if min(x fusion )≤x <min(x window ) or max(x window ) <x≤max(x fusion ), set the corresponding h to 0, so that the feature under the sliding window is obtained Finally, after all windows are calculated, the average of h corresponding to different windows is taken to obtain remember The largest value in Generally speaking, The size of represents whether the vertical strokes in the electronic signature are vertical. The bigger the better. Determine the vertical rotation angle θ final .

[0038] by Figure 2 For example, from left to right and from top to bottom, the original data, the correction based on eigenvalue decomposition, and the correction based on convex hull features are shown respectively. Set h max =0,θ final=0, the search domain is [-5°, 5°], the step size is 0.25°, and the fusion result is rotated clockwise by the step size angle in each step to calculate the like Greater than the existence of h max ,but At the same time, the rotation angle of this step is set as the final rotation angle θ final Assume that in all the search results, when -5° is found, h=2.04 is the maximum value, and -5° is selected as the final rotation angle. After the search is completed, θ final This is the final rotation angle. At this time, the fusion result (x fusion ,y fusion ) rotate θ clockwise final degrees, and we get (x final ,y final ), which are the final x- and y-coordinates of the electronic signature.

[0039] The embodiment described above is only one embodiment of the present invention. Any common changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included in the protection scope of the present invention.

Claims

1. An electronic signature correction method based on eigenvalue decomposition, characterized in that: The method comprises the following steps: a covariance calculation module obtains a signature stroke point sequence and establishes a covariance matrix of the sequence on the x-axis and the y-axis; an eigenvalue decomposition is performed on the covariance matrix to obtain two linearly independent eigenvectors; a direction projection module projects the coordinates of the stroke point position sequence in the direction of the two eigenvectors to obtain the signature sequence coordinates, and obtains the corrected coordinates (x, y) of the stroke point (x, y) based on the eigenvalue decomposition. svd ,y svd ); calculate the convex hull of the stroke point set to obtain the minimum enclosing rectangle, calculate the stroke rotation angle based on the minimum enclosing rectangle, and obtain the corrected coordinates (x) of the stroke point based on the minimum enclosing rectangle after rotating according to the direction angle rect ,y rect ); fusion correction coordinates (x svd ,y svd ) and (x rect ,y rect ) to obtain the fusion coordinates of the signature stroke points; rotate the fusion coordinates clockwise to obtain the corrected position coordinates of the signature stroke points; The specific steps of establishing the covariance matrix are: Based on covariance Construct the covariance matrix C: Among them, n is the number of stroke points in the sequence, (x i ,y i ) represents the coordinate value of point i; Or, according to the formula: c ij =Cov(X i ,Y j )=E[X i -E(X)][Y j -E(Y)] calculate the covariance c of the i-th point of the signature stroke ij , establish the corresponding covariance matrix in, E(x) represents the mean of the stroke point sequence X, E(y) represents the mean of the stroke point sequence Y, n is the number of stroke points, (x i ,y i ) represents the coordinate value of point i; According to the formula: Cv=λv, perform eigenvalue decomposition, except In addition, two linearly independent orthogonal eigenvectors (v1, v2) are obtained; According to the formula (x * ,y * )=(x,y)·(v1,v2) Project the coordinates of each stroke point in the direction of the two eigenvectors to obtain the projected signature sequence (x * ,y * ), (x * ,y * ) According to the formula: x svd =x * -min(x * ),y svd =y * -min(y * ) moves, and the stroke coordinate value (x, y) is obtained. The stroke point correction coordinate (x svd ,y svd ); Based on the corrected fusion coordinates (x fusion ,y fusion ), use a sliding window of length w to slide on the signature sequence, and for the w points in the window, calculate the minimum x coordinate x in its convex hull min , the largest x-coordinate x max ,exist In the interval, from x min At the beginning, the step size is 1 to iterate, calculate the height difference between the upper and lower bounds of the corresponding convex hull, and obtain the height difference feature under the sliding window Among them, h i is the height difference between the upper and lower bounds of the convex hull corresponding to the i-th iteration value; according to the stroke corresponding to the maximum value of the average value of the height difference in all sliding windows, the vertical stroke with the best verticality is determined, thereby determining the rotation angle θ final , the search domain of the rotation angle is [-5°, 5°].

2. The method according to claim 1, characterized in that The divide-and-conquer method is used to calculate the convex hull of the stroke point set. The point corresponding to the minimum cosine value of the angle between the y-coordinate direction of the initial point of the stroke and any point in the stroke except the initial point is found. Then, with this corresponding point as the starting point, the point corresponding to the minimum cosine value of the angle between the y-coordinate direction of the starting point and any other points in the stroke is found until the initial point of the stroke is returned. The polygon connected by all points is the convex hull of the signature. One endpoint of any side of the signature convex hull is used as the origin to rotate the side so that the side is parallel to the x-axis. The area of ​​the candidate circumscribed rectangle corresponding to the side is calculated. The rotation and area calculation of the candidate circumscribed rectangle are completed for all sides of the signature. The convex hull corresponding to the side with the smallest area constitutes the minimum circumscribed rectangle of the signature. According to the corresponding side of the minimum circumscribed rectangle, the signature stroke is rotated clockwise by an angle α to obtain the signature stroke coordinates (x # ,y # ), (x # ,y # ) According to the formula: x rect =x # -min(x # ),y rect =y # -min(y # ), after moving, we get the corrected coordinates (x rect ,y rect ),in, Among them, the coordinates of the two end points corresponding to edge j are (x j ,y j ),(x j+1 ,y j+1 ).

3. The method according to claim 1, characterized in that According to the formula Correct the coordinates (x svd ,y svd ) and the signature stroke coordinates (x # ,y # ) to obtain the corrected fusion coordinates of the signature stroke points.

4. An electronic signature correction system based on eigenvalue decomposition that implements the method according to any one of claims 1 to 3, characterized in that: include: Covariance calculation module, eigenvalue decomposition module, direction projection module, signature correction module, fusion module, the covariance calculation module obtains the signature stroke point sequence, establishes the covariance matrix of the sequence x-axis and y-axis; performs eigenvalue decomposition on the covariance matrix to obtain two linearly independent eigenvectors; the direction projection module projects the coordinates of all stroke points in the direction of the two eigenvectors to obtain the signature sequence coordinates, and moves the stroke point (x, y) based on the eigenvalue decomposition to obtain the corrected coordinates (x svd ,y svd ); The signature correction module calculates the convex hull of the stroke point set to obtain the minimum enclosing rectangle, calculates the rotation direction angle based on the minimum enclosing rectangle, and obtains the corrected coordinates (x, y) of the stroke point based on the minimum enclosing rectangle after rotation according to the direction angle. rect ,y rect ); fusion module fusion correction coordinates (x svd ,y svd ) and (x rect ,y rect ), get the fusion coordinates of the signature stroke points, and convert the fusion coordinates (x fusion ,y fusion ) Rotate clockwise by the angle θ determined based on the best vertical stroke final , get the corrected position coordinates of the signature strokes.

5. A computer-readable storage medium, characterized in that A computer program is stored thereon, and the program is executed by a processor to implement the electronic signature correction method described in any one of claims 1 to 3.

6. An electronic device, characterized in that: include: one or more processors; Memory; One or more application programs are stored in the memory and configured to be loaded and executed by the one or more processors so as to execute the electronic signature correction method according to any one of claims 1 to 3.

Images