CN110245615B - Signature authentication method, system and storage medium - Google Patents
Signature authentication method, system and storage medium Download PDFInfo
- Publication number
- CN110245615B CN110245615B CN201910521170.7A CN201910521170A CN110245615B CN 110245615 B CN110245615 B CN 110245615B CN 201910521170 A CN201910521170 A CN 201910521170A CN 110245615 B CN110245615 B CN 110245615B
- Authority
- CN
- China
- Prior art keywords
- signature
- points
- image
- cluster
- matching
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 239000013598 vector Substances 0.000 claims abstract description 44
- 238000004422 calculation algorithm Methods 0.000 claims description 40
- 239000011159 matrix material Substances 0.000 claims description 22
- 230000011218 segmentation Effects 0.000 claims description 14
- 238000012545 processing Methods 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 10
- 230000008859 change Effects 0.000 claims description 9
- 238000000605 extraction Methods 0.000 claims description 7
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 238000010586 diagram Methods 0.000 description 6
- 238000001514 detection method Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 3
- 238000012216 screening Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000005316 response function Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000003707 image sharpening Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/30—Writer recognition; Reading and verifying signatures
- G06V40/37—Writer recognition; Reading and verifying signatures based only on signature signals such as velocity or pressure, e.g. dynamic signature recognition
- G06V40/382—Preprocessing; Feature extraction
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Collating Specific Patterns (AREA)
Abstract
The invention discloses a signature authentication method, a system and a storage medium, wherein the method comprises the following steps: extracting image skeletons of an original signature image and a signature image to be identified, extracting characteristic points according to the image skeletons, and describing and matching the characteristic points; rejecting error matching points among each cluster, carefully and locally registering the signature image to be identified, judging the signature to be a false signature if the number of the screened feature points is less than a threshold value, otherwise, calculating the similarity among each matching cluster, roughly dividing the signature stroke, and finely dividing inflection points in the signature stroke; extracting the feature vectors of all strokes, carrying out stroke matching according to the feature vectors, and carrying out signature authentication according to the stroke matching result. The invention can improve the accuracy of signature authentication.
Description
Technical Field
The invention relates to a signature authentication method, a system and a storage medium, belonging to the technical field of image processing.
Background
In the departments of banks, insurance companies, public security and judicial law and the like, it is essential to verify the authenticity of the user identity, and most institutions confirm the user by using passwords and radio frequency identification cards, but the user identity cannot be substantially reflected. There are many cases in which a password-stealing card is stolen by others for illegal use. Therefore, signatures are widely used in various industries, and the legal status and importance of the signatures serving as a traditional identity identification mark cannot be replaced. However, signatures also exist in the situations of signature substitution, forged signatures and the like, so that the true and false authentication of signatures of some important files is very necessary.
The writing habit of people is a fixed writing activity mode formed by repeated practice and writing practice of people, has stability and specificity, and can be reflected in personal handwriting. The handwriting identification is a verification process for comparing and identifying two parts of handwriting on the basis of researching and recognizing personal writing habits to determine whether the two parts of handwriting are written by one person or not. Signature authentication is an important basis for verifying the identity of a signature person, and a writer of handwriting can be distinguished by authenticating the handwriting.
At present, many organizations still adopt a manual signature authentication mode, manual signature authentication needs special authentication personnel to authenticate signature fonts, has the defects of strong subjectivity, poor instantaneity, narrow application range, low efficiency and the like, and cannot timely solve the signature authentication problem in life. If the mutual matching mechanism is lacked in the counterfeit identifying mechanism, the qualification standard of the counterfeit identifying personnel is not standard, the counterfeit identifying level is uneven, and the like, the identification result is negatively influenced, and the authenticity judgment is influenced.
Disclosure of Invention
The invention aims to overcome the defects in the prior art and provides a signature authentication method, a system and a storage medium, which can improve the accuracy of signature authentication.
In order to achieve the purpose, the invention is realized by adopting the following technical scheme:
in a first aspect, the present invention provides a method for authenticating a signature, the method comprising the steps of:
respectively extracting image skeletons of an original signature image and a signature image to be identified;
extracting characteristic points according to the image skeleton, and describing and matching the characteristic points of the original signature image and the characteristic points of the signature image to be identified;
rejecting error matching points among each cluster and carrying out local registration on the signature image to be identified;
if the number of the screened feature points is smaller than the threshold value of the number of the set feature points, judging that the signature in the signature image to be identified is a pseudo signature; otherwise, continuing the identification;
calculating the similarity between each matched cluster, and if the distance mean value between the feature vectors of the clusters corresponding to the original signature image and the signature image to be identified is greater than a preset distance mean value threshold value, judging that the signature in the signature image to be identified is a pseudo signature; otherwise, continuing to identify;
respectively carrying out rough segmentation on the signature strokes in the original signature image and the signature image to be identified, detecting inflection points in the strokes, then carrying out fine segmentation on the strokes, and extracting the feature vectors of all the strokes;
and screening by adopting cosine similarity according to the feature vectors of each pair of matched strokes, if the number of the screened matched strokes is greater than a preset matched stroke number threshold value, judging that the signature in the signature image to be authenticated is a true signature, and otherwise, judging that the signature in the signature image to be authenticated is a false signature.
With reference to the first aspect, further, the method further includes: before extracting the image skeleton, performing sharpening processing and binaryzation processing on the original signature image and the signature image to be identified respectively.
With reference to the first aspect, further, a Harris algorithm is used to extract feature points on the image skeleton, and the feature points are described and matched by an EOH feature descriptor.
In combination with the first aspect, further, a RANSAC algorithm of adaptive clustering is adopted to eliminate the error matching points between each cluster, and local registration is performed on the signature image to be identified.
With reference to the first aspect, further, the similarity between each matching cluster is calculated by extracting the Log-Gabor feature vector of each cluster based on a minimum circumscribed circle, where the minimum circumscribed circle represents a circle that can contain all feature points in the cluster and has a minimum area.
With reference to the first aspect, further, a path planning algorithm is used to perform coarse segmentation on the signature strokes.
With reference to the first aspect, further, the method for extracting feature vectors of all strokes includes: and adopting an LSD straight line extraction algorithm to extract straight line segments of the strokes.
In a second aspect, the present invention provides a signature authentication system, including a processor and a storage medium;
the storage medium is to store instructions;
the processor is configured to operate in accordance with the instructions to perform steps in accordance with the aforementioned method.
In a third aspect, the invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the aforementioned method.
Compared with the prior art, the signature authentication method, the signature authentication system and the storage medium provided by the embodiment of the invention have the beneficial effects that:
1) when the wrong matching points are removed, the RANSAC algorithm based on the self-adaptive clustering is used for removing the wrong matching points among each cluster: the overall structures of fonts written by the same person are similar, but the local structures are different, such as the upper and lower structures and the left and right structures of the fonts, so that the characteristic points are clustered to achieve the effect of signature block processing, then the corresponding homography matrix is found out according to the clustering, the RANSAC algorithm based on the self-adaptive clustering is adopted to eliminate the wrong matching points among each clustering, and the difference among the local structures of the fonts is avoided to screen out a large number of correct matching points;
2) the method comprises the following steps of roughly dividing a signature stroke by adopting a path planning algorithm, detecting an inflection point in the stroke, and finely dividing the stroke: when wrong feature points are screened, some correct and important feature points are easily screened out by mistake, the step not only can retrieve the important feature points screened by mistake, but also can detect some missed feature points again, and the precision of signature stroke segmentation is improved. Unmatched strokes can be screened out through the quantity relation of the characteristic points among the matched strokes, and excessive useless data generated in subsequent detection is avoided;
3) the method comprises the steps of extracting feature vectors of all strokes, measuring similarity of the feature vectors of each pair of matched strokes by using cosine similarity, and identifying the authenticity of a signature by judging the number of the strokes with the cosine similarity exceeding a threshold value;
4) the method improves the accuracy and the practicability of signature authentication, and is suitable for the authentication of signatures in different occasions such as banks, insurance companies and the like.
Drawings
Fig. 1 is a flowchart of a signature authentication method according to an embodiment of the present invention.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
As shown in fig. 1, which is a flowchart of a signature authentication method according to an embodiment of the present invention, a signature image is preprocessed, then a Harris algorithm is used to extract feature points on an image skeleton, and the feature points are described and matched by an EOH feature descriptor; then using a RANSAC algorithm based on self-adaptive clustering to eliminate error matching points among each cluster, and locally registering the image to be identified, if the number of the screened feature points is less than a threshold value, judging the signature as a pseudo signature, otherwise, calculating the similarity among each matching cluster by extracting the Log-Gabor feature vector of each cluster based on the minimum circumscribed circle, and identifying whether the signature is a pseudo signature according to the similarity; if the signature is not a pseudo signature, the path planning algorithm is adopted to carry out rough segmentation on the signature stroke, and inflection points in the stroke are detected and then fine segmentation is carried out on the stroke; then extracting feature vectors of all strokes by using an angle fluctuation feature calculation method based on straight lines; and finally, screening by adopting cosine similarity according to the feature vector of each pair of matched strokes, and realizing signature authentication according to the number of the screened matched strokes.
The signature authentication method provided by the embodiment of the invention specifically comprises the following steps:
step 1) preprocessing a signature image, including image sharpening, binarization, image skeleton extraction and the like:
11) and denoising the image by using a median filtering algorithm, filtering isolated points in the image and stains on the strokes of the signature, and realizing the clarity of the signature image.
12) And binarizing the clarified image according to an Otsu threshold segmentation algorithm. And (3) representing the original signature image after binarization by using leftotsu, and representing the signature image to be authenticated after binarization by using rightotsu.
13) Extracting image skeletons of the binarized images leftotsu and rightotsu by a K3M skeleton extraction algorithm, recording the preprocessed original signature image as leftImg, and recording the preprocessed signature image to be authenticated as rightImg.
Step 2) extracting feature points on the image skeleton by using a Harris algorithm, and describing and matching the feature points through an EOH (edge organized hierarchy) feature descriptor:
21) and respectively extracting feature points of the original signature image leftImg and the signature image rightImg to be authenticated by using a Harris algorithm. Let leftPoints represent the feature point set extracted from the original signature image leftImg, and rightPoints represent the feature point set extracted from the signature image rightImg to be authenticated. leftPoints [ i ] P ]Representing ith in characteristic point set leftPoints P Individual feature points, rightPoints [ j ] P ]Representing the jth of feature point sets rightPoints P And (4) a characteristic point.
22) Feature points in the feature point sets leftPoints and rightPoints are described through EOH feature descriptors.
Calculating characteristic point leftPoints [ i ] P ]And rightPoints [ j ] P ]EOH feature descriptor of (a): change i P And j P The feature descriptors of the feature points in the feature point set leftPoints and rightPoints are all calculated. The sets of EOH feature descriptors of all feature points in the feature point sets leftPoints and rightPoints are respectively expressed by Lehd and Rehd, Lehd [ i [ ] P ]Representing a characteristic point leftPoints [ i ] P ]EOH feature descriptor of (1), Rehd [ j) P ]Representing a point rightPoints j P ]EOH feature descriptor(s).
23) And matching the feature points in the feature point set leftPoints and rightPoints.
Calculating Lehd [ i ] P ]And Rehd [ j P ]Describing Euclidean distance between vectors, if the Euclidean distance is less than a constant dis1, considering Lehd [ i [ ] P ]And Rehd [ j P ]Match and note that they are a pair ofAnd (4) matching points. Change i P And j P Each feature descriptor in the EOH feature descriptor set Lehd is subjected to a match determination with each feature descriptor in the EOH feature descriptor set Rehd.
M to be found 1 Marking the matching points as matching point set match, match [ i m ]Indicating ith in match point set match m And matching points are matched. The match point set match represents the characteristic points existing in the original signature image leftImg, the match R represents the characteristic points existing in the signature image rightImg to be authenticated, and the match L [ i [ m ]Indicating the ith located in the feature point set matchL m A characteristic point, matchR [ i ] m ]Indicating the ith located in the feature point set matchR m A feature point, and match [ i m ]From matchL [ i ] m ]And matchR [ i ] m ]And (4) forming.
And 3) eliminating error matching points among each cluster by using a RANSAC (random Sample consensus) algorithm based on self-adaptive clustering, and locally registering the signature image to be identified.
31) And (3) aiming at the pixel positions of the characteristic points in the characteristic point set matchR, carrying out adaptive clustering on the characteristic points by using a mean-shift algorithm, and recording the obtained adaptive clustering number as r. The cluR is the cluster combination of the characteristic point set matchR, the cluL is the cluster set formed by the characteristic point set matchR, and the cluR [ i ] is the ith cluster in the cluster set cluR. And finding out the matched characteristic points of all the characteristic points in the cluster cluR [ i ] in the characteristic point set matchL, clustering the characteristic points into a class, and representing the class by the cluL [ i ]. Changing the value of i to make i get r from 1, so that r clusters can be obtained in the feature point set matchL, and storing the r clusters in the cluster set cluR. The matching clusters in the original signature image leftImg and the signature image rightImg to be authenticated are denoted by clu, and clu [ i ] denotes the ith pair of matching clusters in the set of matching clusters clu, wherein the matching cluster clu [ i ] consists of the cluster cluR [ i ] and the cluster cluL [ i ].
32) And eliminating the error matching points between each cluster by using a RANSAC algorithm based on a homography matrix.
4 pairs of matching points are randomly extracted from the ith pair of clusters clu [ i ], and the four pairs of matching points are guaranteed to be not collinear. And calculating a homography matrix by adopting a Direct Linear Transformation (DLT) algorithm, and then removing error matching points among each cluster by using a RANSAC algorithm based on the homography matrix. The method for eliminating the error matching points among each cluster by using the RANSAC algorithm based on the homography matrix comprises the following steps:
a. homography matrix
Is marked as a model
b. Compute clusters clu [ i ]]All data and models in
If the error is less than the threshold value Err 1 Then the point is determined to be the correct matching point, also called the local point. The number of all the local points measured by the model is recorded as
c. And (4) randomly extracting other 4 groups of non-collinear matching points, calculating a new homography matrix by adopting a direct linear transformation algorithm, and repeating the steps a and b. Finally, all homography matrixes which can be calculated by 4 pairs of matching points in the cluster can be found out and are respectively marked as
Can obtain a model consistent with the homography quantity
Comparing the number of local points contained in all models, marking the model containing the most local points as the optimal model, and using M i Represents; its corresponding homography matrix is denoted as H i 。
d. Change clustering clu [ i]The value of i is such that i is taken from 1 to r. Repeating the steps a, b and c, and calculating the optimal model M in all the clusters clu 1 ，M 2 …M r The models are respectively composed of homography matrix H 1 ，H 2 ，…H r And (4) obtaining the product. The set of local points included in the model is the set of feature points from which the mismatching points are removed. And removing the corresponding error characteristic points of the characteristic point set match according to the local interior points in the optimal model, and updating the data in the characteristic point set match.
33) Recording the number of the feature points in the feature point set match after the error matching points are removed as m 2 Setting a corresponding threshold Num 1 If m is 2 Is less than or equal to threshold Num 1 Then the signature is determined to be a false signature, otherwise step 34) is entered.
34) Local registration of the images is completed.
Finding matching clusters clu [1 ] in step 32)]、clu[2]…clu[r]R is respectively H 1 、H 2 …H r . And locally registering the rightImg of the signature image to be authenticated according to the homography matrix, wherein the relation between each pixel position of the image before registration and each pixel position in the image after registration is as follows:
wherein, (x, y) represents the pixel position of the image before registration, (x ', y') is the pixel position of the image after registration, and H represents the homography matrix. After the local registration of the image is completed, the locally registered picture is represented by RegImg.
And 4) calculating the similarity between each matched cluster by extracting the Log-Gabor characteristic vector of each cluster based on the minimum circumcircle, and identifying whether the signature is a pseudo signature according to the similarity.
41) The minimum circumscribed circles of the ith pair of matching clusters cluL [ i ] and cluR [ i ] in the cluster set clu are respectively obtained, wherein the minimum circumscribed circle represents a circle which can contain all the feature points in the clusters and has the smallest area. And changing the value of i to make i taken from 1 to r, and respectively solving the minimum circumcircle of r pairs of clusters. Taking an image area where each circumcircle is located, wherein cirL represents a set of r minimum circumcircle image areas in an original signature image leftImg; cirR represents the set of r smallest circumscribed circle image regions in the signature image to be authenticated, rightImg. And cirL [ i ] represents the minimum circumscribed circle image area of the ith cluster in the original signature image leftImg, and cirR [ i ] represents the minimum circumscribed circle image area of the ith cluster in the signature image rightImg to be authenticated.
42) And respectively convolving the cirL [ i ] and the cirR [ i ] with Log-Gabor filters containing S scales and O different directions to respectively form S x O Log-Gabor characteristic matrixes. Wherein
θ represents the O directions of the Log-Gabor filter.
And changing the value of i to obtain r from 1, and obtaining S x O characteristic matrixes formed by convolving each minimum circumcircle image area in the cirL and the cirR with a Log-Gabor filter.
43) Mixing cirL [ i ]]And cirR [ i ]]The method is divided into A fan-shaped areas respectively, S different scales can be counted out from each fan-shaped area, and the fan-shaped areas are provided with O direction histograms in different directions. Therefore, the sector area A can count S A direction histograms containing O directions, the S A direction histograms containing O directions form a characteristic vector of S A O dimensions, and the cirL [ i ] is obtained]The formed feature vector is represented as (x) Li1 ,x Li2 ,…，x LiS*A*O ) Introduction of cirR [ i ]]The formed feature vector is represented as (x) Ri1 ,x Ri2 ,…，x RiS*A*O ). Changing the value of i to make i get r from 1, and obtaining the characteristic vectors of r clusters in the cirL as (x) L11 ,x L12 ,…，x L1S*A*O )，(x L21 ,x L22 ,…，x L2S*A*O )，…，(x Lr1 ,x Lr2 ,…，x LrS*A*O ) (ii) a The feature vectors of r clusters in the cirR are obtained and are respectively (x) R11 ,x R12 ,…，x R1S*A*O )，(x R21 ,x R22 ,…，x R2S*A*O )，…，(x Rr1 ,x Rr2 ,…，x RrS*A*O )。
44) Calculating cirL[i]And cirR [ i ]]Euclidean distance d of formed characteristic vector of dimension S, A and O i Wherein
changing the value of i to make i get r from 1, and calculating the Euclidean distance between r vectors corresponding to cirL and cirR by d 1 ，d 2 ，…，d r And (4) showing. Then r distances d are obtained 1 ，d 2 ，…，d r Mean value of d m And then, obtaining the composite material,
setting a threshold value for judging the distance mean value as D, if the measured distance mean value D m >And D, judging that the signature is a pseudo signature, and otherwise, entering the next authentication.
And 5) carrying out coarse segmentation on the signature stroke by adopting a path planning algorithm, detecting an inflection point in the stroke, and carrying out fine segmentation on the stroke.
51) And finding the shortest path from each feature point in the feature point set matchL in the original signature image leftImg to the next feature point by adopting a Dijkstra shortest path algorithm, wherein each path represents a stroke section in the image leftImg. Denote the collection of signature strokes in the original signature image leftImg by StrL, StrL [ k ] 1 ]Representing the kth in the set of strokes StrL 1 A segment stroke. Then, according to the matching relation between the characteristic points in the matchL and matchR, finding out the stroke StrL [ k ] 1 ]Two matching points of two characteristic points in the characteristic point set matchR are found, the shortest path between the two points is found by using StrR [ k ] 1 ]And (4) showing. Changing k 1 The value of (1) is to enable all signature strokes in a signature stroke set StrL in an original signature image leftImg to find matched signature strokes in a signature image rightImg to be authenticated, wherein the signature stroke set in the signature image rightImg to be authenticated is represented by StrR.
52) Set a size of w 1 *w 1 Window omega along the stroke StrL k 1 ]Moves this window omega through the intra-window mapThe gray level of the image area is changed to determine whether an inflection point in the stroke is encountered. The degree of gray scale change is reflected by the response function value of the corner detection operator. Wherein the corner detection operator is:
wherein (x) 1 ,y 1 ) Indicating the location of the pixel in the stroke.
And
are respectively x 1 And y 1 Partial derivative of direction, ω (x) 1 ,y 1 ) Is a gaussian window function.
By M 1 Represents the autocorrelation matrix in the corner detection operator:
let λ 1 ，λ 2 Representing an autocorrelation matrix M 1 Defining a corner response function of
R(x 1 ,y 1 )＝det(M 1 )-htr 2 (M 1 )
Wherein det (M) 1 )＝λ 1 λ 2 ，tr(M 1 )＝λ 1 +λ 2 And h is a constant threshold.
If angular point discriminant R (x) 1 ,y 1 ) Greater than a set threshold R a Then, the corner point is determined as an inflection point in the stroke, and a new feature point is added at the inflection point.
Following the same method on stroke StrR [ k ] 1 ]Finding and marking the inflection point of the stroke if the stroke StrL [ k ] is matched 1 ]And StrR [ k ] 1 ]If the number of the newly added feature points is not consistent, the pair of strokes is filtered, otherwise, the matched strokes with the feature points are foundThe strokes are divided from the feature points.
Changing k 1 Until all strokes in StrL and StrR are subjected to stroke inflection point search once, and the number of inflection points in each pair of matched strokes is calculated. And filtering out matched strokes with unmatched feature point numbers, and dividing strokes containing inflection points again. The stroke quantity finally obtained by the original signature image leftImg and the signature image rightImg to be authenticated is k a By StrL a Representing the set of all split strokes, StrR, in the original signature image leftImg a Representing the collection of all split strokes in the authentication image, rightImg. StrL a [k 2 ]Representing a set of strokes StrL a K of (1) 2 Segment strokes, StrR a [k 2 ]Representing a set of strokes StrR a K of (1) 2 Segment strokes, and StrL a [k 2 ]And StrR a [k 2 ]Is a pair of matching strokes.
And step 6) extracting the feature vectors of all strokes by using a linear-based angle fluctuation feature calculation method, then screening by adopting cosine similarity according to the feature vectors of each pair of matched strokes, and finally realizing signature authentication according to the number of the screened matched strokes.
61) Stroke StrL pair by LSD (line Segment detector) linear extraction algorithm a [k 2 ]Extracting straight line segment to obtain StrL a [k 2 ]The number of straight line segments extracted is recorded as
Representing by strokes StrL a [k 2 ]The first step of extraction 1 Straight line segments; calculate the StrL of strokes a [k 2 ]The number of pixels in the image is denoted as pi L From l 1 Starting with 1, calculate the straight line segment
The number of occupied pixels is recorded as pi a1 (ii) a Stroke StrR a [k 2 ]The number of pixels in the image is denoted as pi R The number of pixels is
As stroke StrR a [k 2 ]The 1 st straight line segment extracted from the step (1) is marked as
Changing l 1 Is taken to be value of l 1 Take l in sequence 1 ＝2，l 1 3, … according to a straight line segment
The number of pixels and the stroke StrL a [k 2 ]The ratio relation and the position relation of the total pixel number are extracted in sequence to extract the stroke StrR a [k 2 ]All of the straight line segments of (a),
representing a stroke StrR a [k 2 ]L of (b) extraction 1 Straight line segments.
62) Stroke StrL a [k 2 ]And stroke StrR a [k 2 ]All the straight line segments comprise two end points, and the direction vector of the straight line in which the straight line segment is positioned is obtained according to the two end points of the straight line segment, namely a stroke StrL a [k 2 ]Middle straight line segment
Are respectively expressed as
Representing straight line segments
The direction vector of (2). And (4) solving the change angle between the adjacent vectors, namely the included angle between the adjacent vectors.
And
is varied by an angle of
Wherein,
stroke StrL a [k 2 ]In (1)
Linear segment formation
Each direction vector, the change angle between every two adjacent direction vectors is calculated as
Will this
Each variable angle forms one
Feature vector of dimension is
By the same token, StrR in stroke a [k 2 ]Middle straight line segment
Are respectively expressed as
Representing straight line segments
The direction vector of (2). Stroke StrR a [k 2 ]Can also form one
Feature vector of dimension
Wherein
To represent
And
the angle of change of (c).
63) Changing k 2 A value of (a) k 2 Get k from 1 a And step 61) is repeated. Finding the feature vector formed by all strokes in StrL and StrR, StrL a StrL of Chinese StrL a [1]，StrL a [2]，…，StrL a [k a ]The feature vectors formed are respectively
StrR a Stroke StrR in (1) a [1]，StrR a [2]，…，StrR a [k a ]The formed feature vectors are respectively
64) Computing corresponding feature vectors
And
cosine similarity therebetween, wherein the cosine similarity is expressed as:
wherein,
the angle between the two vectors is shown,
representing cosine similarity.
Changing k 2 A value of (a) k 2 Get k from 1 a . Find out
And
and
and
respectively has cosine similarity of
Let
Respectively with a threshold value K 1 Comparing and calculating to obtain a value greater than a threshold K 1 The number of cosine similarities of (c), which is denoted as K. If the final value of K is less than the threshold value K 2 If not, the signature can be identified as signed by the original signer.
The signature authentication method provided by the embodiment of the invention comprises the following steps: 1) when the wrong matching points are removed, the RANSAC algorithm based on the self-adaptive clustering is used for removing the wrong matching points among each cluster. The overall structure of the fonts written by the same person is similar, but the local structures of the fonts are different, such as the upper and lower structures and the left and right structures of the fonts. Therefore, the characteristic points are clustered to achieve the effect of signature blocking processing, then corresponding homography matrixes are found out according to the clustering, and the RANSAC algorithm based on self-adaptive clustering is adopted to eliminate the wrong matching points among each cluster, so that the difference existing among local structures of fonts is avoided, and a large number of correct matching points are screened out. 2) And carrying out coarse segmentation on the signature stroke by adopting a path planning algorithm, detecting an inflection point in the stroke, and then carrying out fine segmentation on the stroke. When wrong feature points are screened, some correct and important feature points are easily screened out by mistake, the step not only can retrieve the important feature points screened by mistake, but also can detect some missed feature points again, and the precision of signature stroke segmentation is improved. Unmatched strokes can be screened out through the quantity relation of the characteristic points among the matched strokes, and excessive useless data generated in subsequent detection are avoided. 3) The method comprises the steps of extracting feature vectors of all strokes, and then carrying out similarity measurement on each pair of feature vectors of matched strokes by using cosine similarity. And identifying the authenticity of the signature by judging the stroke number of which the cosine similarity exceeds a threshold value. The method compares the stroke similarity by fitting the strokes with straight line segments and describing the variation trend of the strokes with the variation angles between the line segments. This method reflects the most character of the font, not only accommodates the acceptable differences between fonts, but also holds up the key stroke change characteristics.
The signature authentication method provided by the embodiment of the invention improves the accuracy and the practicability of signature authentication, and is suitable for the authentication of signatures in different occasions such as banks, insurance companies and the like.
The embodiment of the invention also provides a signature authentication system which can be used for realizing the signature authentication method and comprises a processor and a storage medium;
the storage medium is used for storing instructions;
the processor is configured to operate in accordance with the instructions to perform the steps according to the aforementioned method.
Embodiments of the present invention also provide a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the steps of the aforementioned method.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.
Claims (8)
1. A method for authenticating a signature, the method comprising the steps of:
respectively extracting image skeletons of an original signature image and a signature image to be identified;
extracting characteristic points according to the image skeleton, and carrying out characteristic point description and matching on the original signature image and the signature image to be identified;
rejecting error matching points among each cluster and carrying out local registration on the signature image to be identified;
if the number of the screened feature points is smaller than the threshold value of the number of the set feature points, judging that the signature in the signature image to be identified is a pseudo signature; otherwise, continuing the identification;
calculating the similarity between each matched cluster, and if the distance mean value between the feature vectors of the clusters corresponding to the original signature image and the signature image to be identified is greater than a preset distance mean value threshold value, judging that the signature in the signature image to be identified is a pseudo signature; otherwise, continuing the identification;
respectively carrying out rough segmentation on the signature strokes in the original signature image and the signature image to be identified, detecting inflection points in the strokes, then carrying out fine segmentation on the strokes, and extracting the feature vectors of all the strokes;
selecting by adopting cosine similarity according to the feature vectors of each pair of matched strokes, if the number of the selected matched strokes is larger than a preset matched stroke number threshold value, judging that the signature in the signature image to be authenticated is a true signature, otherwise, judging that the signature in the signature image to be authenticated is a false signature;
the method comprises the following steps of adopting a RANSAC algorithm of self-adaptive clustering to eliminate error matching points among clusters, and locally registering a signature image to be identified, wherein the RANSAC algorithm specifically comprises the following steps:
31) aiming at the pixel positions of the feature points in the feature point set matchR, carrying out adaptive clustering on the pixel positions by using a mean-shift algorithm, and recording the obtained adaptive clustering number as r; the cluR is the cluster combination of the characteristic point set matchR, the cluL is the cluster set formed by the characteristic point set matchR, and the cluR [ i ] is the ith cluster in the cluster set cluR; finding out the matched characteristic points of all the characteristic points in the clustered cluR [ i ] in a characteristic point set matchL, clustering the characteristic points into a class, and expressing the class by the cluL [ i ]; changing the value of i to enable i to be taken from 1 to r, enabling r clusters to be obtained in the feature point set matchL, and storing the r clusters into a cluster set cluR; clu represents matching clusters in the original signature image leftImg and the signature image rightImg to be authenticated, clu [ i ] represents the ith pair of matching clusters in the matching cluster set clu, wherein the matching cluster clu [ i ] consists of a cluster cluR [ i ] and a cluster cluL [ i ];
32) eliminating error matching points among each cluster by using a RANSAC algorithm based on a homography matrix;
randomly extracting 4 pairs of matching points from the ith pair of clusters clu [ i ], and ensuring that the four pairs of matching points are not collinear; calculating a homography matrix by adopting a direct linear transformation algorithm, and then eliminating error matching points among each cluster by using a RANSAC algorithm based on the homography matrix; the method for eliminating the error matching points among each cluster by using the RANSAC algorithm based on the homography matrix comprises the following steps:
a. homography matrix
Is marked as a model
b. Compute clusters clu [ i]All data and models in
If the error is less than the threshold value Err 1 If so, judging the matching point to be a correct matching point, also called an in-office point; the number of all the local interior points measured by the model is recorded as
c. Randomly extracting other 4 groups of non-collinear matching points, calculating a new homography matrix by adopting a direct linear transformation algorithm, and repeating the steps a and b; finally, all homography matrixes which can be calculated by 4 pairs of matching points in the cluster can be found out and are respectively marked as
Can obtain a model consistent with the homography quantity
Comparing the number of the local points contained in all the models, marking the model containing the most local points as the optimal model, and using M i Represents; its corresponding homography matrix is denoted as H i ；
d. Change clustering clu [ i]Taking i from 1 to r; repeating the steps a, b and c, and calculating the optimal model M in all the clusters clu 1 ，M 2 …M r The models are respectively composed of homography matrix H 1 ，H 2 ，…H r Obtaining; the collection of local points contained in the model is the collection of the characteristic points after the error matching points are removed; removing error characteristic points corresponding to the characteristic point set match according to the local interior points in the optimal model, and updating data in the characteristic point set match;
33) recording the number of the feature points in the feature point set match after the error matching points are removed as m 2 Setting a corresponding threshold Num 1 If m is 2 Is less than or equal to threshold Num 1 If yes, judging the signature to be a pseudo signature, otherwise, entering a step 34);
34) completing local registration of the images;
step 32) of finding matching clusters clu [1 ]]、clu[2]…clu[r]R is respectively H 1 、H 2 …H r (ii) a According to a homography matrixAnd locally registering the rightImg of the signature image to be authenticated, wherein the relation between each pixel position of the image before registration and each pixel position in the image after registration is as follows:
wherein, (x, y) represents the pixel position of the image before registration, (x ', y') is the pixel position of the image after registration, and H represents a homography matrix; after the local registration of the images is completed, the locally registered pictures are represented by RegImg.
2. The method of claim 1, further comprising: before extracting an image skeleton, performing sharpening processing and binarization processing on an original signature image and a signature image to be identified respectively, wherein the sharpening processing comprises denoising processing on the image by adopting a median filtering algorithm.
3. The signature authentication method according to claim 1, wherein Harris algorithm is used to extract feature points on the image skeleton, and the feature points are described and matched by EOH feature descriptors.
4. The signature authentication method as claimed in claim 1, wherein the similarity between each matching cluster is calculated by extracting the Log-Gabor feature vector of each cluster based on the smallest circumscribed circle representing the circle which can contain all the feature points in the cluster and has the smallest area.
5. The method of claim 1, wherein a path planning algorithm is used to roughly segment the signature strokes.
6. The signature authentication method as claimed in claim 1, wherein the method of extracting feature vectors of all strokes comprises: and adopting an LSD straight line extraction algorithm to extract straight line segments of the strokes.
7. A signature authentication system is characterized by comprising a processor and a storage medium;
the storage medium is used for storing instructions;
the processor is configured to operate in accordance with the instructions to perform the steps of the method according to any one of claims 1 to 6.
8. Computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910521170.7A CN110245615B (en) | 2019-06-17 | 2019-06-17 | Signature authentication method, system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910521170.7A CN110245615B (en) | 2019-06-17 | 2019-06-17 | Signature authentication method, system and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110245615A CN110245615A (en) | 2019-09-17 |
| CN110245615B true CN110245615B (en) | 2022-09-13 |
Family
ID=67887447
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910521170.7A Active CN110245615B (en) | 2019-06-17 | 2019-06-17 | Signature authentication method, system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110245615B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110809284B (en) * | 2019-09-25 | 2022-09-16 | 福建新大陆软件工程有限公司 | Positioning method, system, device and readable storage medium based on MR data |
| CN111461015B (en) * | 2020-05-07 | 2023-10-13 | 中金金融认证中心有限公司 | RNN (RNN) -model-based user-independent online signature identification method and device |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110235920A1 (en) * | 2009-03-13 | 2011-09-29 | Nec Corporation | Image signature matching device |
| CN104200239A (en) * | 2014-09-09 | 2014-12-10 | 河海大学常州校区 | Image feature fusion identification based signature authentic identification system and method |
| CN104751465A (en) * | 2015-03-31 | 2015-07-01 | 中国科学技术大学 | ORB (oriented brief) image feature registration method based on LK (Lucas-Kanade) optical flow constraint |
| CN105160654A (en) * | 2015-07-09 | 2015-12-16 | 浙江工商大学 | Towel label defect detecting method based on feature point extraction |
| CN106778586A (en) * | 2016-12-08 | 2017-05-31 | 武汉理工大学 | Offline handwriting signature verification method and system |
-
2019
- 2019-06-17 CN CN201910521170.7A patent/CN110245615B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110235920A1 (en) * | 2009-03-13 | 2011-09-29 | Nec Corporation | Image signature matching device |
| CN104200239A (en) * | 2014-09-09 | 2014-12-10 | 河海大学常州校区 | Image feature fusion identification based signature authentic identification system and method |
| CN104751465A (en) * | 2015-03-31 | 2015-07-01 | 中国科学技术大学 | ORB (oriented brief) image feature registration method based on LK (Lucas-Kanade) optical flow constraint |
| CN105160654A (en) * | 2015-07-09 | 2015-12-16 | 浙江工商大学 | Towel label defect detecting method based on feature point extraction |
| CN106778586A (en) * | 2016-12-08 | 2017-05-31 | 武汉理工大学 | Offline handwriting signature verification method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110245615A (en) | 2019-09-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Bayar et al. | Towards open set camera model identification using a deep learning framework | |
| Warif et al. | SIFT-symmetry: a robust detection method for copy-move forgery with reflection attack | |
| Lee | Copy-move image forgery detection based on Gabor magnitude | |
| CN110443128B (en) | Finger vein identification method based on SURF feature point accurate matching | |
| Prakash et al. | Detection of copy-move forgery using AKAZE and SIFT keypoint extraction | |
| Ortega et al. | Retinal verification using a feature points-based biometric pattern | |
| Tao et al. | Unsupervised detection of built-up areas from multiple high-resolution remote sensing images | |
| CN110298376B (en) | Bank bill image classification method based on improved B-CNN | |
| EP2294531A1 (en) | Scale robust feature-based identifiers for image identification | |
| CN106056122B (en) | A kind of image zone duplicating stickup altering detecting method based on KAZE characteristic point | |
| CN107622489A (en) | A kind of distorted image detection method and device | |
| CN110245615B (en) | Signature authentication method, system and storage medium | |
| Al-Qershi et al. | Enhanced block-based copy-move forgery detection using k-means clustering | |
| CN108140107B (en) | Quickly, high-precision large-scale fingerprint verification system | |
| CN111310628A (en) | Paper currency forming mode inspection and identification method based on paper currency printing pattern characteristics | |
| Iwasokun et al. | Fingerprint singular point detection based on modified poincare index method | |
| CN113392856B (en) | Image forgery detection device and method | |
| Akbar et al. | Palm vein biometric identification system using local derivative pattern | |
| Zaharieva et al. | Image based recognition of ancient coins | |
| CN113011426A (en) | Method and device for identifying certificate | |
| CN115311746A (en) | Off-line signature authenticity detection method based on multi-feature fusion | |
| Dixit et al. | Region duplication detection in digital images based on Centroid Linkage Clustering of key–points and graph similarity matching | |
| CN107977600A (en) | A kind of signature false distinguishing method based on side palm feature | |
| Zebari et al. | Analysis of dense descriptors in 3D face recognition | |
| Fu et al. | A circle detection algorithm based on mathematical morphology and chain code |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |