CN106778586A - Offline handwriting signature verification method and system - Google Patents
Offline handwriting signature verification method and system Download PDFInfo
- Publication number
- CN106778586A CN106778586A CN201611122474.9A CN201611122474A CN106778586A CN 106778586 A CN106778586 A CN 106778586A CN 201611122474 A CN201611122474 A CN 201611122474A CN 106778586 A CN106778586 A CN 106778586A
- Authority
- CN
- China
- Prior art keywords
- signature
- sample
- measured
- image
- static
- 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.)
- Granted
Links
Classifications
-
- 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
- G06V40/33—Writer recognition; Reading and verifying signatures based only on signature image, e.g. static signature recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Abstract
The invention discloses a kind of offline handwriting signature verification method and system, comprise the following steps：Static signature sample in S1, static signature Sample Storehouse is pre-processed；S2, the extraction that multiple features are carried out to pretreated signature image, including moment characteristics, local binary patterns feature, gray level co-occurrence matrixes feature and Pulse Coupled Neural Network feature；S3, the multiple features to the static signature sample of extraction are trained, the standard sample database after being trained；S4, acquisition signature to be measured, and signature to be measured is pre-processed, obtain multiple features of signature to be measured；S5, multiple features of signature to be measured are matched with the character pair of static signature sample in standard sample database, identified that signature to be measured is actual signature or forges a signature.The present invention effectively can carry out accurate discriminating to offline signature to be measured.
Description
Technical field
The present invention relates to CRT technology field, more particularly to a kind of offline handwriting signature verification method and system.
Background technology
With the fast development of information technology, people's living standard increasingly improve during safety problem be subject to it is preceding not
Some challenges, accurately carry out personal identification and differentiate extremely important in real time.Traditional identity differentiates based on the side such as password, IC-card
Formula, there is a various shortcomings, such as password may be stolen forgetting, and IC-card may be lost stolen etc..People are badly in need of a kind of reliable and side
It is easy with personal identification authentication technique come in overcoming the shortcomings of conventional method.And the identity identification method for being based on handwritten signature can
Fundamentally to solve disadvantages mentioned above.
Signature authentication technique compared with other biological determination techniques, with enough multidate informations, it is difficult imitate, distinction compared with
It is high, respect the advantage such as the right of privacy and acquisition of information high efficiency, the property collected, human injury's acceptable degree in signature character and
Robustness aspect is all protruded very much, is had broad application prospects and application value.Therefore, to signature carry out it is effective, reliable,
Quickly differentiate that there is important social value and Practical significance.
Signing, it is online and offline to differentiate and be divided into, and on-line signature provides more multidate informations, and this information is difficult
Imitate, so being identified than easy offline.The intersection error rate of current on-line signature system has dropped down to less than 1%, foreign countries
Also practical product is had to emerge.Static signature checking is then writer after plain paper is submitted a written statement to a higher authority and writes signature, recycles and shines
The optical imaging apparatus such as camera, scanner extract the signature of writer.Off-line approximate computation for facility environment requirement than
Online mode is loosely many, if largely improving its checking accuracy, it will have bigger application preceding than online mode
Scape.
The content of the invention
The technical problem to be solved in the present invention is to be directed to the unstable defect of offline handwriting signature verification in the prior art,
There is provided a kind of stabilization effective offline handwriting signature verification method and system.
The technical solution adopted for the present invention to solve the technical problems is：
A kind of offline handwriting signature verification method is provided, is comprised the following steps：
S1, the static signature sample in static signature Sample Storehouse is pre-processed；
S2, the extraction that multiple features are carried out to pretreated signature image, including moment characteristics, local binary patterns are special
Levy, gray level co-occurrence matrixes feature and Pulse Coupled Neural Network feature；
S3, the multiple features to the static signature sample of extraction are trained, the standard sample database after being trained；
S4, acquisition signature to be measured, and signature to be measured is pre-processed, obtain multiple features of signature to be measured；
S5, multiple features of signature to be measured are matched with the character pair of static signature sample in standard sample database,
Identify that signature to be measured is actual signature or forges a signature.
In method of the present invention, step S3 is specially：
, used as test sample, other are used as training sample for any one static signature sample in selection static signature Sample Storehouse
This, and calculates the average of training sample characteristic vector, and training sample the distance between characteristic vector, obtain the equal of distance
Value variance；
The mean distance of the test sample and the characteristic vector of training sample is calculated, the average side of training sample distance is compared
Difference, calculates the similarity degree of test sample and training sample, is actual signature if similarity is more than predetermined threshold value, is otherwise puppet
Signature is made, the standard sample database after being trained；
Count the false rejection rate and false acceptance rate in the standard sample database.
In method of the present invention, step S3 is specially：Set up the classification of the different characteristic of correspondence static signature sample
Device, and be trained, obtain meeting the grader of resolution；
Step S5 is specially：Multiple features of signature to be measured are identified by corresponding grader, and according to difference
The output result of grader judges whether signature to be measured is actual signature.
In method of the present invention, step S3 is specially：Set up the classification of the different characteristic of correspondence static signature sample
Device, and corresponding characteristic vector is labeled by true and false sample, randomly select the spy of the part sample of static signature Sample Storehouse
Levy vector to be trained, remainder sample counts predicting the outcome for test sample as test sample, is reflected
Other accuracy.
In method of the present invention, pretreatment in step S1 includes binaryzation, clipping Boundaries, size normalization, inclines
Tiltedly correction and contract away from；
Binaryzation is specially：The grey level histogram of image is divided into by two parts with optimal threshold, is made between two parts
Variance takes maximum；
Clipping Boundaries are specially：Horizontal and vertical projection is carried out to signature image, border is carried out according to projection histogram
Shearing, obtains removing the signature image on border；
Size normalization is specially：Signature is located at picture middle part by filling up-and-down boundary, then picture is contracted in proportion
Put；
Slant Rectify is specially：Using the pixel of signature image as characteristic point, using characteristic point in figure and the pass of baseline
System, characteristic point least square fitting is gone out the direction of baseline, the incline direction as signed；
Contracting is away from specially：Make the projection of vertical direction to signature image, obtain projection histogram；In statistics projection histogram
The number of lowest part, and distance therebetween, Distance Judgment according to lowermost extent its whether be signature image white space, if
It is to shear this section of white space, the signature map for obtaining contracting after.
In method of the present invention, different features are extracted to different preprocessing process, specially：To clipping Boundaries
Gray level image extract Pulse Coupled Neural Network feature, to normalized gray level image texture feature extraction include local binary
Pattern feature and gray level co-occurrence matrixes feature, low order moment characteristics are extracted to normalized bianry image.
Present invention also offers a kind of offline handwriting signature verification system, including：
Static signature Sample Storehouse acquisition module, for gathering static signature sample；
Sample Storehouse pretreatment module for pre-processing and right to the static signature sample in static signature Sample Storehouse
Pretreated signature image carries out the extraction of multiple features, including moment characteristics, local binary patterns feature, gray level co-occurrence matrixes
Feature and Pulse Coupled Neural Network feature；
Sample Storehouse training module, the multiple features for the static signature sample to extracting are trained, after being trained
Standard sample database；
Signature processing module to be measured, for obtaining signature to be measured, and pre-processes to signature to be measured, obtains signature to be measured
Multiple features；
Signature recognition module, for by multiple features of signature to be measured with standard sample database static signature sample it is corresponding
Feature is matched, and identifies that signature to be measured is actual signature or forges a signature.
In system of the present invention, the Sample Storehouse training module specifically for selection static signature Sample Storehouse in appoint
Used as test sample, other calculate the average of training sample characteristic vector to one static signature sample of meaning as training sample,
The distance between and the characteristic vector of training sample, obtain the mean variance of distance；
The Sample Storehouse training module is additionally operable to calculate the distance of the test sample and the average of the characteristic vector of training sample,
The mean variance of training sample distance is compared, the similarity degree of test sample and training sample is calculated, if similarity is more than default
Threshold value is then actual signature, otherwise to forge a signature, the standard sample database after being trained；Count the mistake in the standard sample database
False rejection rate and false acceptance rate.
In system of the present invention, the Sample Storehouse training module is specifically for setting up correspondence static signature sample not
With the grader of feature, and it is trained, obtains meeting the grader of resolution；
The signature recognition module specifically for multiple features of signature to be measured are identified by corresponding grader,
And judge whether signature to be measured is actual signature according to the output result of different classifications device.
In system of the present invention, the Sample Storehouse pretreatment module to sample in Sample Storehouse specifically for carrying out two-value
Change, clipping Boundaries, size normalization, Slant Rectify and contracting are away from treatment；
Binaryzation is specially：The grey level histogram of image is divided into by two parts with optimal threshold, is made between two parts
Variance takes maximum；
Clipping Boundaries are specially：Horizontal and vertical projection is carried out to signature image, border is carried out according to projection histogram
Shearing, obtains removing the signature image on border；
Size normalization is specially：Signature is located at picture middle part by filling up-and-down boundary, then picture is contracted in proportion
Put；
Slant Rectify is specially：Using the pixel of signature image as characteristic point, using characteristic point in figure and the pass of baseline
System, characteristic point least square fitting is gone out the direction of baseline, the incline direction as signed；
Contracting is away from specially：Make the projection of vertical direction to signature image, obtain projection histogram；In statistics projection histogram
The number of lowest part, and distance therebetween, Distance Judgment according to lowermost extent its whether be signature image white space, if
It is to shear this section of white space, the signature map for obtaining contracting after.
The beneficial effect comprise that：The present invention is pre-processed by signature, eliminates some unnecessary
The influence of extraneous factor, and extract multiple features, including moment characteristics, local binary patterns spy according to different pre-processed results
Levy, gray level co-occurrence matrixes feature and Pulse Coupled Neural Network feature, the extraction of feature take into account shape, texture and puppet dynamic
Feature；Accurate discriminating just effectively can be carried out to offline signature to be measured by the matching of these features.
The present invention has further used simple statistical sorter to count the identification result of various discrimination methods,
It is also possible to use grader carries out the statistics of identification result to signature image.It is found through experiments that, these methods can realize stabilization
Effective offline handwriting signature verification.
Brief description of the drawings
Below in conjunction with drawings and Examples, the invention will be further described, in accompanying drawing：
Fig. 1 is the flow chart of the offline handwriting signature verification method of the embodiment of the present invention；
The offline handwriting signature verification FB(flow block)s of Fig. 2；
Fig. 3 a are the signature map after the shearing of border；
Fig. 3 b are the signature map after size normalization；
Fig. 4 a-4c are the signature map that sciagraphy cutting is contracted during；
Fig. 5 is signature identification system structured flowchart；
Fig. 6 is local binary patterns feature extraction schematic diagram；
Fig. 7 is gray level co-occurrence matrixes feature extraction schematic diagram.
Specific embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the present invention, not
For limiting the present invention.
The discrimination method for offline handwritten signature of the embodiment of the present invention, as shown in figure 1, comprising the following steps：
S1, the static signature sample in static signature Sample Storehouse is pre-processed；The main purpose of pretreatment is removal
Interference information, and it is ready to be characterized extraction；Mainly usable smoothing denoising, size place normalization, binaryzation and
Clipping Boundaries, there are segmentation, sciagraphy contracting to be pre-processed away from methods such as, slant corrections.Static signature sample is all to use pen
Write on the signature on printing paper, then by scanner carry out gray scale scanning be input into computer, but currently without it is relatively good from
Line handwritten signature storehouse, oneself gathered data workload is too big.What the embodiment of the present invention was selected is the static signature contest of 2011
Signature storehouse, 10 signers, 200 actual signatures and 200 forge a signature, and have obtained corresponding differentiating accurate by assessment
Rate.
S2, feature extraction is carried out to pretreated signature image, because handwritten signature is random larger and easily receives environment
Etc. the influence of factor, one-side feature can not accurately represent the writing style of signer, need to consider from many aspects
Ensure that the feature ga s safety degree for extracting is strong, reliability is high, well simultaneously intrinsic dimensionality can not be excessive for independence, the present invention is in order to accurately have
The expression signer feature of effect, is extracted shape, texture and the pseudo- behavioral characteristics of signature, using different characteristic comprehensive description respectively
Method.Shape facility includes moment characteristics, and moment characteristics can describe the overall structures such as signature profile, word bit inclination, centre-of gravity shift
Feature；Textural characteristics include local binary patterns and gray level co-occurrence matrixes, and texture can intuitively reflect signature image in simple terms
Visual signature, signature image is described by the intensity profile of pixel and its surrounding space neighborhood；Based on coupled neural network
Pseudo- behavioral characteristics extract the multidate informations such as pressure when can then write signature by greyscale transformation secondary indication signer
Change, and the model of coupled neural network is similar to optic nerve imaging, can preferably represent some pseudo- dynamic letters of signature
Breath.The present invention from these different feature extracting methods can more complete and accurate expression signer writing style.S3, to extract
Multiple features of static signature sample be trained, the standard sample database after being trained；
S4, acquisition signature to be measured, and signature to be measured is pre-processed, the pre- place mentioned in preprocessing process and step S1
Reason is identical, then carries out feature extraction, and the feature of extraction is identical with the feature extraction mentioned in step S2, obtains signature to be measured
Multiple features；
S5, multiple features of signature to be measured are matched with the character pair of static signature sample in standard sample database,
Identify that signature to be measured is actual signature or forges a signature.
In step S1, the static signature sample in static signature Sample Storehouse is pre-processed, including binaryzation, size are returned
One change and slant correction, sciagraphy contracting away from etc., extract image correlated characteristic, including square, city block distance conversion be also a kind of mirror
Other method, calculates stain in signature map after a binaryzation it to the distance between nearest stain in another signature map, with most
Closely sum, as the distance between two signatures, is a kind of conventional identification algorithm.Local two innings of pattern (Local Binary
Pattern, LBP), gray level co-occurrence matrixes (Gray Level Co-occurrence Matrix, GLCM) and pulse coupled neural
Network (PCNN-Pulse Coupled Neural Network, PCNN) feature etc., can be used weighted euclidean distance as feature
Distance metric.
Specifically：
1) what binaryzation was used is OTSU algorithms, and its general principle is to be split the grey level histogram of image with optimal threshold
Into two parts, the variance between two parts is set to take maximum, i.e. separation property maximum.
2) what clipping Boundaries were used is sciagraphy, horizontal and vertical projection is carried out to signature image, according to projection histogram
Border is sheared, obtains removing the signature image on border；
3) what size normalization was used is that filling up-and-down boundary makes signature be located at picture middle part, then picture is contracted in proportion
Put；
4) what Slant Rectify was used is least square method, using the pixel of signature image as characteristic point, using special in figure
The relation a little with baseline is levied, characteristic point least square fitting is gone out the direction of baseline, the incline direction as signed；
5) sciagraphy contracting makees the throwing of vertical direction to signature image away from the signature for being mainly used in having obvious distance between word and word
Shadow, judges whether this is sheared in projection histogram intermediate value minimum region.
In a preferred embodiment of the invention, size normalization is concretely comprised the following steps：
First four borders of signature image are determined, by border beyond all blank spaces cut off；
The length-width ratio of image after judging to shear, if fruit width is more than length, phase is filled in the up-and-down boundary of signature image
With the blank of width, signature is set to be located at the middle part of square picture after filling；
Assuming that square size is N*N after filling, the size to be reached after specification is m*m, then square is pressed into m/64's
Scale factor is zoomed in and out, the signature image after being normalized, as shown in Figure 3 b.
Sciagraphy contracting is away from concretely comprising the following steps：
Make the projection of vertical direction to signature image, obtain projection histogram, as shown in fig. 4 a；
The number of lowest part in statistics projection histogram, and distance therebetween；
Whether it is the white space of signature image for Distance Judgment according to lowermost extent, if then shearing this section of clear area
Domain, the signature map for obtaining contracting after, as shown in figure 4 b and 4 c.
The extraction of multiple features is carried out to pretreated signature image, specially：To the binary picture of size normalization
As entering line-spacing feature extraction, the binary image to clipping Boundaries carries out coupled pulse neural network characteristics extraction, and size is returned
One gray level image changed carries out gray level co-occurrence matrixes and local binary patterns feature extraction；
Wherein, moment characteristics are mainly description character shape and physical significance, extract length-width ratio, word bit direction, elongation degree, stretch
Eight features such as the latitude of emulsion, the horizontal degree of bias, the vertical degree of bias, horizontal stretching equilibrium degree and vertical stretching equilibrium degree composition moment characteristics to
Amount.
Local binary patterns feature mainly describes image local textural characteristics, can be using the statistic histogram of LBP characteristic spectrums
As characteristic vector；The collection of illustrative plates of the LBP conversion of equivalent formulations, the number of times that each pixel value occurs in statistics LBP collection of illustrative plates can be calculated
Histogram, as characteristic vector.
Gray level co-occurrence matrixes feature describes two Joint Distributions of pixel with certain spatial relation, can see
Into two joint histograms of pixel grey scale pair, the gray level image to size normalization carries out GLCM feature extractions, by image elder generation
Carry out four direction GLCM conversion, then respectively extract four energy of matrix, correlation, entropy, contrasts, as GLCM features to
Amount；16 features composition characteristic vectors such as the energy of the four direction of extraction, contrast, correlation and entropy.
Pulse Coupled Neural Network feature, the texture information of main reflection image.PCNN models are anti-to the details of dark pixel
Should be more sensitive, the result of successive ignition just reflects the texture information of image, and the pseudo- multidate information of some for being included, right
The gray level image of signature carries out PCNN interative computations, the entropy of output matrix after each iteration is calculated, by the defeated of its each iteration
Go out the entropy of matrix as characteristic value, iteration 30 times obtains the entropy time series of 30 dimensions as PCNN characteristic vectors.
In a preferred embodiment of the invention, feature extraction is mapped to low-dimensional especially by change high-dimensional feature space of changing commanders
Feature space, then remove again redundancy and incoherent feature further reduce dimension.Obtain the relatively low spy of intrinsic dimensionality
The distance between after levying vector, calculate, its similarity is estimated, then judge its true and false with threshold method, or directly with grader pair
It is trained and predicts, then statistical experiment result, including Moment Feature Extraction submodule, local binary patterns feature extraction
Module, gray level co-occurrence matrixes feature extraction submodule and Pulse Coupled Neural Network feature extraction submodule
In a preferred embodiment of the invention, feature extraction is carried out in the following manner：
Moment Feature Extraction：It is the digital picture of M*N for a size, (p+q) rank geometric moment is defined such as formula：
The center of gravity of signature image can be obtained by zeroth order and single order geometric moment, and center is just obtained by the origin of coordinates of center of gravity
Square, central moment is to reflect that gray scale is the measurement how to be distributed relative to grey scale centre of gravity in corresponding region.Second moment can obtain figure
The variance of stain coordinate as in, obtain in horizontal direction and vertical direction on range of extension, accordingly can just calculate length-width ratio,
Word bit direction, elongation degree, range of extension, the horizontal degree of bias, the vertical degree of bias, horizontal stretching equilibrium degree and vertical stretching equilibrium degree etc. eight
Feature, as Character eigenvector；
Local binary patterns feature extraction is comprised the following steps that：
Normalized gray level image is divided into the zonule of 3*3 first；
For the pixel in each zonule, 8 adjacent gray values of pixel are compared with it, if surrounding pixel
Value is more than center pixel value, then the position of the pixel is marked as 1, is otherwise 0.So, 8 points in 3*3 neighborhoods through than
Compared with 8 bits can be produced, that is, obtain the LBP values of the window center pixel as shown in Figure 6；
Excessive binary pattern is all unfavorable for extraction, classification and the access of information of texture, it is contemplated that of equal value
The LBP of pattern, when the circulation binary number corresponding to certain LBP is from 0 to 1 or from 1 to 0 be up to saltus step twice, the LBP institutes
Corresponding binary system is known as an equivalent formulations class, and the LBP of general modfel can be transformed to the LBP of equivalent formulations；
Then the histogram of each zonule, i.e., the frequency that each numeral occurs are calculated；Then the histogram is returned
One change is processed.
The statistic histogram in each region that will finally obtain is attached as a characteristic vector, that is, view picture figure
LBP texture feature vectors；
Gray level co-occurrence matrixes feature extraction：Gray level co-occurrence matrixes can reflect the gray scale of image on direction, adjacent spaces,
The integrated information of amplitude of variation, is the basis of the local mode and queueing discipline for analyzing image, if f (x, y) is a width two-dimemsional number
Word image, its size is M × N, and grey level is Ng, then the gray level co-occurrence matrixes for meeting certain space relation are P (i, j)=#
(x1, y1), (x2, y2) ∈ M × N | f (x1, y1)=i, f (x2, y2)=j }
Wherein # (x) represents the element number in set x, it is clear that P is the matrix of Ng*Ng, if between (x1, y1) and (x2, y2)
Distance is d, and both angles with abscissa line are θ, then can obtain various spacing and angle gray level co-occurrence matrixes P (i, j,
D, θ).
Image gray levels are first quantified as 8 grades in the present invention, 1 is taken apart from d, angle distinguishes 0,45,90,135 degree, to signing
Name image carries out 4 GLCM conversion in direction, and the schematic diagram of GLCM conversion is as shown in fig. 7, in order to be able to more intuitively with symbiosis square
Battle array description texture situation, from co-occurrence matrix derive some reflection matrix situations parameters, such as energy, correlation, contrast, entropy,
The co-occurrence matrix of four direction calculates these four parameters respectively, used as GLCM characteristic vectors；
Pulse Coupled Neural Network feature extraction：Feature extraction is carried out to signature image using the entropy time series of PCNN.
Specific model parameter is set, with the pixel value of signature gray level image as the external drive of neuron, is entered by PCNN models
Row iteration computing, output matrix is obtained after each iteration, then according to the calculating of shannon formula entropy, obtains entropy as the figure
The characteristic vector of picture.
Characteristic matching is carried out in step S5, following several method can be used, first method is weighted euclidean distance, is led to
The distance between calculating characteristic vector is crossed, signature identification result is estimated with threshold method, second method is to use support
Vector machine classifier, is directly trained to the characteristic vector signed and then predicts last statistical forecast result, assesses corresponding
Resolution, the third method is to use BP neural network grader, directly the characteristic vector signed is trained and then prediction
Last statistical forecast result, assesses corresponding resolution.
Weighted euclidean distance：The sample for needing training enough, obtains corresponding substantially threshold value.System first using sign into
Row training, obtains a stencil value.When signature needs to differentiate, the weighting Europe between its characteristic vector and stencil value is directly calculated
Family name's distance value.It is trained and determines a threshold values using a large amount of signatures, when waiting survey sample distance value to be more than threshold values, is then judged to
It is false；Otherwise it is true.It is as follows apart from computing formula：
Wherein dist is the distance value of sample to be tested, FiIt is the characteristic value of sample to be tested, μiIt is equal for training sample feature
Value, σiIt is standard deviation, n is characterized number.
If not having great amount of samples in experiment, threshold method can only be made substantially to the identification result of every kind of signature discrimination method
Assessment.
Support vector machine classifier：The linear classifier of interval maximum spatially is characterized, learning strategy is to make interval
Maximize, solve two classification problems that signature differentiates.System is pre-processed and feature extraction to all samples first, then to carrying
The feature for taking carries out the true and false and marks and store, and selected part sample characteristics are used as training in the characteristic vector space of all samples
Collection, the accuracy of the result that remainder predict as test set, last statistical test sample by grader is realized signing
Differentiate.The topmost design of certain supporting vector is the parameter designing of kernel function, and the ginseng of kernel function can be adjusted by testing
Number, reaches optimal identification result.
BP neural network grader：The characteristic vector that the signature image of the true and false will have been marked first is normalized, and then chooses one
Partial data as training, remainder data as test data, then statistical identification accuracy.BP neural network grader makes
With similar with SVMs, the standard of the true and false will be carried out to characteristic vector, selected part sample is right as training sample
All sampling feature vectors are normalized, then test prediction, statistical identification accuracy.BP neural network grader is important
It is the design of neutral net, design of input layer hidden layer and output layer etc, this mainly uses for reference associated specialist many experiments
Statistical value set.
Test result indicate that signature discrimination method of the invention, can carry out off-line approximate computation and identification result visitor
See, stablize.To improve identification result, one embodiment of the present of invention is proposed signature image piecemeal, and selection importance is higher
Part merged and differentiated again.
Signature image piecemeal uses simple uniform Method of Partitioning, and signature image after normalization is divided into 4*4 parts, unites
The width of meter signature is high, cutting in proportion.
Offline handwriting signature verification system construction drawing is substantially wrapped as shown in figure 5, pre-processed to existing signature sample
Binaryzation, clipping Boundaries and size normalization are included, normalized bianry image, normalized gray level image and clipping Boundaries are obtained
Gray level image afterwards, different features are extracted for different preprocessing process, and the gray level image of clipping Boundaries extracts PCNN spies
Levy, normalized gray level image texture feature extraction includes LBP and GLCM, normalized bianry image extracts low order away from feature.
Feature extraction is obtained carrying out taxonomic history after characteristic vector, and support vector machine classifier, BP neural network point have been used respectively
Class device, simple statistical sorter.
The offline handwriting signature verification system of the embodiment of the present invention, including：
Static signature Sample Storehouse acquisition module, for gathering static signature sample；
Sample Storehouse pretreatment module for pre-processing and right to the static signature sample in static signature Sample Storehouse
Pretreated signature image carries out the extraction of multiple features, including moment characteristics, local binary patterns feature, gray level co-occurrence matrixes
Feature and Pulse Coupled Neural Network feature；
Sample Storehouse training module, the multiple features for the static signature sample to extracting are trained, after being trained
Standard sample database；
Signature processing module to be measured, for obtaining signature to be measured, and pre-processes to signature to be measured, obtains signature to be measured
Multiple features；
Signature recognition module, for by multiple features of signature to be measured with standard sample database static signature sample it is corresponding
Feature is matched, and identifies that signature to be measured is actual signature or forges a signature.
Sample Storehouse training module is specifically for any one the static signature sample conduct in selection static signature Sample Storehouse
Test sample, other calculate the average of training sample characteristic vector as training sample, and training sample characteristic vector
The distance between, obtain the mean variance of distance；
The Sample Storehouse training module is additionally operable to calculate the distance of the test sample and the average of the characteristic vector of training sample,
The mean variance of training sample distance is compared, the similarity degree of test sample and training sample is calculated, if similarity is more than default
Threshold value is then actual signature, otherwise to forge a signature, the standard sample database after being trained；Count the mistake in the standard sample database
False rejection rate and false acceptance rate.
Sample Storehouse training module is instructed specifically for setting up the grader of the different characteristic of correspondence static signature sample
Practice, obtain meeting the grader of resolution；
Signature recognition module is specifically for multiple features of signature to be measured are identified by corresponding grader, and root
Judge whether signature to be measured is actual signature according to the output result of different classifications device.
Sample Storehouse pretreatment module specifically for sample in Sample Storehouse is carried out binaryzation, clipping Boundaries, size normalization,
Slant Rectify and contracting are away from treatment；
Binaryzation is specially：The grey level histogram of image is divided into by two parts with optimal threshold, is made between two parts
Variance takes maximum；
Clipping Boundaries are specially：Horizontal and vertical projection is carried out to signature image, border is carried out according to projection histogram
Shearing, obtains removing the signature image on border；
Size normalization is specially：Signature is located at picture middle part by filling up-and-down boundary, then picture is contracted in proportion
Put；
Slant Rectify is specially：Using the pixel of signature image as characteristic point, using characteristic point in figure and the pass of baseline
System, characteristic point least square fitting is gone out the direction of baseline, the incline direction as signed；
Contracting is away from specially：Make the projection of vertical direction to signature image, obtain projection histogram；In statistics projection histogram
The number of lowest part, and distance therebetween, Distance Judgment according to lowermost extent its whether be signature image white space, if
It is to shear this section of white space, the signature map for obtaining contracting after.
To sum up, the present invention has carried out clipping Boundaries and size normalization in pretreatment to signature, and eliminating some need not
The influence of the extraneous factor wanted；During feature extraction, low order moment characteristics, that is, shape are extracted for normalized bianry image
Feature, the gray level co-occurrence matrixes features to Normalized Grey Level image zooming-out, the gray level image to clipping Boundaries is extracted local two
Value pattern feature and Pulse Coupled Neural Network feature, feature extraction take into account shape, texture and pseudo- behavioral characteristics；Matching
When, use simple statistical sorter to count the identification result of various discrimination methods, also use comparing classical
SVMs and BP neural network grader the statistics of identification result is carried out to signature image.By experiment it can be found that
These methods can realize the effective offline handwriting signature verification of stabilization.
It should be appreciated that for those of ordinary skills, can according to the above description be improved or converted,
And all these modifications and variations should all belong to the protection domain of appended claims of the present invention.
Claims (10)
1. a kind of offline handwriting signature verification method, it is characterised in that comprise the following steps：
S1, the static signature sample in static signature Sample Storehouse is pre-processed；
S2, the extraction that multiple features are carried out to pretreated signature image, including moment characteristics, local binary patterns feature, ash
Degree co-occurrence matrix feature and Pulse Coupled Neural Network feature；
S3, the multiple features to the static signature sample of extraction are trained, the standard sample database after being trained；
S4, acquisition signature to be measured, and signature to be measured is pre-processed, obtain multiple features of signature to be measured；
S5, multiple features of signature to be measured are matched with the character pair of static signature sample in standard sample database, recognized
Going out signature to be measured is actual signature or forges a signature.
2. method according to claim 1, it is characterised in that step S3 is specially：
Selection static signature Sample Storehouse in any one static signature sample as test sample, other as training sample,
And calculate the average of training sample characteristic vector, and training sample the distance between characteristic vector, obtain the average of distance
Variance；
The mean distance of the test sample and the characteristic vector of training sample is calculated, the mean variance of training sample distance is compared,
The similarity degree of test sample and training sample is calculated, is actual signature if similarity is more than predetermined threshold value, be otherwise forgery
Signature, the standard sample database after being trained；
Count the false rejection rate and false acceptance rate in the standard sample database.
3. method according to claim 1, it is characterised in that step S3 is specially：Set up correspondence static signature sample
The grader of different characteristic, and be trained, obtain meeting the grader of resolution；
Step S5 is specially：Multiple features of signature to be measured are identified by corresponding grader, and according to different classifications
The output result of device judges whether signature to be measured is actual signature.
4. method according to claim 3, it is characterised in that step S3 is specially：Set up correspondence static signature sample
The grader of different characteristic, and corresponding characteristic vector is labeled by true and false sample, randomly select static signature Sample Storehouse
The characteristic vector of part sample be trained, remainder sample predicts the outcome as test sample to test sample
Counted, the accuracy for being differentiated.
5. method according to claim 1, it is characterised in that the pretreatment in step S1 include binaryzation, clipping Boundaries,
Size normalization, Slant Rectify and contracting away from；
Binaryzation is specially：The grey level histogram of image is divided into by two parts with optimal threshold, makes the variance between two parts
Take maximum；
Clipping Boundaries are specially：Horizontal and vertical projection is carried out to signature image, border is sheared according to projection histogram,
Obtain removing the signature image on border；
Size normalization is specially：Signature is set to be located at picture middle part by filling up-and-down boundary, then by picture bi-directional scaling；
Slant Rectify is specially：Using the pixel of signature image as characteristic point, using characteristic point in figure and the relation of baseline, will
Characteristic point least square fitting goes out the direction of baseline, the incline direction as signed；
Contracting is away from specially：Make the projection of vertical direction to signature image, obtain projection histogram；It is minimum in statistics projection histogram
The number at place, and distance therebetween, Distance Judgment according to lowermost extent its whether be signature image white space, if then
This section of white space is sheared, the signature map for obtaining contracting after.
6. method according to claim 5, it is characterised in that different features, tool are extracted to different preprocessing process
Body is：Gray level image to clipping Boundaries extracts Pulse Coupled Neural Network feature, and texture is extracted to normalized gray level image
Feature includes local binary patterns feature and gray level co-occurrence matrixes feature, and low order moment characteristics are extracted to normalized bianry image.
7. a kind of offline handwriting signature verification system, it is characterised in that including：
Static signature Sample Storehouse acquisition module, for gathering static signature sample；
Sample Storehouse pretreatment module, for being pre-processed to the static signature sample in static signature Sample Storehouse, and to pre- place
Signature image after reason carries out the extraction of multiple features, including away from feature, local binary patterns feature, gray level co-occurrence matrixes feature
With Pulse Coupled Neural Network feature；
Sample Storehouse training module, the multiple features for the static signature sample to extracting are trained, the mark after being trained
Quasi- Sample Storehouse；
Signature processing module to be measured, for obtaining signature to be measured, and pre-processes to signature to be measured, obtains many of signature to be measured
Individual feature；
Signature recognition module, for by the character pair of static signature sample in multiple features of signature to be measured and standard sample database
Matched, identified that signature to be measured is actual signature or forges a signature.
8. system according to claim 7, it is characterised in that the Sample Storehouse training module is signed specifically for selection is offline
Used as test sample, other calculate training sample to any one static signature sample in name Sample Storehouse as training sample
The average of characteristic vector, and training sample the distance between characteristic vector, obtain the mean variance of distance；
The Sample Storehouse training module is additionally operable to calculate the distance of the test sample and the average of the characteristic vector of training sample, compares
The mean variance of training sample distance, calculates the similarity degree of test sample and training sample, if similarity is more than predetermined threshold value
It is then actual signature, otherwise to forge a signature, the standard sample database after being trained；The mistake counted in the standard sample database is refused
Exhausted rate and false acceptance rate.
9. system according to claim 7, the Sample Storehouse training module is specifically for setting up correspondence static signature sample
Different characteristic grader, and be trained, obtain meeting the grader of resolution；
The signature recognition module is specifically for multiple features of signature to be measured are identified by corresponding grader, and root
Judge whether signature to be measured is actual signature according to the output result of different classifications device.
10. system according to claim 6, it is characterised in that the Sample Storehouse pretreatment module is specifically for sample
Sample carries out binaryzation, clipping Boundaries, size normalization, Slant Rectify and contracting away from treatment in storehouse；
Binaryzation is specially：The grey level histogram of image is divided into by two parts with optimal threshold, makes the variance between two parts
Take maximum；
Clipping Boundaries are specially：Horizontal and vertical projection is carried out to signature image, border is sheared according to projection histogram,
Obtain removing the signature image on border；
Size normalization is specially：Signature is set to be located at picture middle part by filling up-and-down boundary, then by picture bi-directional scaling；
Slant Rectify is specially：Using the pixel of signature image as characteristic point, using characteristic point in figure and the relation of baseline, will
Characteristic point least square fitting goes out the direction of baseline, the incline direction as signed；
Contracting is away from specially：Make the projection of vertical direction to signature image, obtain projection histogram；It is minimum in statistics projection histogram
The number at place, and distance therebetween, Distance Judgment according to lowermost extent its whether be signature image white space, if then
This section of white space is sheared, the signature map for obtaining contracting after.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611122474.9A CN106778586B (en) | 2016-12-08 | 2016-12-08 | Off-line handwritten signature identification method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611122474.9A CN106778586B (en) | 2016-12-08 | 2016-12-08 | Off-line handwritten signature identification method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106778586A true CN106778586A (en) | 2017-05-31 |
| CN106778586B CN106778586B (en) | 2020-11-17 |
Family
ID=58881715
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611122474.9A Expired - Fee Related CN106778586B (en) | 2016-12-08 | 2016-12-08 | Off-line handwritten signature identification method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106778586B (en) |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944336A (en) * | 2017-10-11 | 2018-04-20 | 中国科学院自动化研究所 | Handwriting signature verification system based on cloud computing |
| CN108921006A (en) * | 2018-05-03 | 2018-11-30 | 西北大学 | The handwritten signature image true and false identifies method for establishing model and distinguishing method between true and false |
| CN108921209A (en) * | 2018-06-21 | 2018-11-30 | 杭州骑轻尘信息技术有限公司 | Image identification method, device and electronic equipment |
| CN108960055A (en) * | 2018-05-30 | 2018-12-07 | 广西大学 | A kind of method for detecting lane lines based on local line's stage mode feature |
| CN108960210A (en) * | 2018-08-10 | 2018-12-07 | 武汉优品楚鼎科技有限公司 | It is a kind of to grind the method, system and device for reporting board-like identification and segmentation |
| CN109344856A (en) * | 2018-08-10 | 2019-02-15 | 华南理工大学 | A kind of off-line signature verification method based on multilayer discriminate feature learning |
| CN109409254A (en) * | 2018-10-10 | 2019-03-01 | 成都优易数据有限公司 | A kind of electronic contract handwritten signature identification method based on twin neural network |
| CN109657498A (en) * | 2018-12-28 | 2019-04-19 | 广西师范大学 | The difference method for secret protection that top-k Symbiotic Model excavates in a plurality of stream |
| CN109817348A (en) * | 2019-01-18 | 2019-05-28 | 郑静 | Portable medical signature management method and Related product |
| CN110096977A (en) * | 2019-04-18 | 2019-08-06 | 中金金融认证中心有限公司 | The training method and handwriting verification method, equipment and medium of handwriting verification model |
| CN110222666A (en) * | 2019-06-14 | 2019-09-10 | 河海大学常州校区 | A kind of signature false distinguishing method and system |
| CN110245615A (en) * | 2019-06-17 | 2019-09-17 | 河海大学常州校区 | A kind of signature false distinguishing method, system and storage medium |
| CN110263740A (en) * | 2019-06-26 | 2019-09-20 | 四川新网银行股份有限公司 | Different type block letter document dubbing method based on OCR technique |
| CN111178290A (en) * | 2019-12-31 | 2020-05-19 | 上海眼控科技股份有限公司 | Signature verification method and device |
| CN111178203A (en) * | 2019-12-20 | 2020-05-19 | 江苏常熟农村商业银行股份有限公司 | Signature verification method and device, computer equipment and storage medium |
| CN111209567A (en) * | 2019-12-30 | 2020-05-29 | 北京邮电大学 | Method and device for judging perceptibility of improving robustness of detection model |
| CN111275070A (en) * | 2019-12-26 | 2020-06-12 | 厦门商集网络科技有限责任公司 | Signature verification method and device based on local feature matching |
| CN111753809A (en) * | 2020-07-10 | 2020-10-09 | 上海眼控科技股份有限公司 | Method and equipment for correcting handwritten signature |
| CN112036323A (en) * | 2020-09-01 | 2020-12-04 | 中国银行股份有限公司 | Signature handwriting identification method, client and server |
| WO2021027336A1 (en) * | 2019-08-14 | 2021-02-18 | 深圳壹账通智能科技有限公司 | Authentication method and apparatus based on seal and signature, and computer device |
| CN112395919A (en) * | 2019-08-16 | 2021-02-23 | 河海大学常州校区 | Off-line signature authentication method and system based on multi-feature metric learning |
| CN112836636A (en) * | 2021-02-02 | 2021-05-25 | 北京惠朗时代科技有限公司 | Method and device for identifying authenticity of signature image |
| CN113095156A (en) * | 2021-03-23 | 2021-07-09 | 西安深信科创信息技术有限公司 | Double-current network signature identification method and device based on inverse gray scale mode |
| WO2021148692A1 (en) * | 2020-01-24 | 2021-07-29 | Universidad Autónoma de Madrid | Method and system for verifying dynamic handwriting and signatures by means of deep learning |
| CN113269136A (en) * | 2021-06-17 | 2021-08-17 | 南京信息工程大学 | Offline signature verification method based on triplet loss |
| CN113313092A (en) * | 2021-07-29 | 2021-08-27 | 太平金融科技服务(上海)有限公司深圳分公司 | Handwritten signature recognition method, and claims settlement automation processing method, device and equipment |
| CN113449543A (en) * | 2020-03-24 | 2021-09-28 | 百度在线网络技术（北京）有限公司 | Video detection method, device, equipment and storage medium |
| CN113920589A (en) * | 2021-10-28 | 2022-01-11 | 平安银行股份有限公司 | Signature identification method, device, equipment and medium based on artificial intelligence |
| CN114155613A (en) * | 2021-10-20 | 2022-03-08 | 杭州电子科技大学 | Offline signature comparison method based on convenient sample acquisition |
| TWI777188B (en) * | 2020-07-07 | 2022-09-11 | 新光人壽保險股份有限公司 | Contract signature authentication method and device |
| CN115880304A (en) * | 2023-03-08 | 2023-03-31 | 曲阜市巨力铁路轨道工程股份有限公司 | Method for identifying defects of sleeper based on complex scene |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1477580A (en) * | 2003-06-12 | 2004-02-25 | 上海交通大学 | Off-line Chinese signature identification method |
-
2016
- 2016-12-08 CN CN201611122474.9A patent/CN106778586B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1477580A (en) * | 2003-06-12 | 2004-02-25 | 上海交通大学 | Off-line Chinese signature identification method |
Non-Patent Citations (3)
| Title |
|---|
| 刘蕾: "加权DTW方法及其在手写签名鉴别中的应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 李亚婷等: "基于结构特征的离线手写签名鉴别", 《中国科技论文在线》 * |
| 陈文颉等: "一种基于特征层数据融合的目标识别方法", 《第四届全球智能控制与自动化大会论文集》 * |
Cited By (46)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944336A (en) * | 2017-10-11 | 2018-04-20 | 中国科学院自动化研究所 | Handwriting signature verification system based on cloud computing |
| CN108921006A (en) * | 2018-05-03 | 2018-11-30 | 西北大学 | The handwritten signature image true and false identifies method for establishing model and distinguishing method between true and false |
| CN108921006B (en) * | 2018-05-03 | 2020-08-04 | 西北大学 | Method for establishing handwritten signature image authenticity identification model and authenticity identification method |
| CN108960055A (en) * | 2018-05-30 | 2018-12-07 | 广西大学 | A kind of method for detecting lane lines based on local line's stage mode feature |
| CN108960055B (en) * | 2018-05-30 | 2021-06-08 | 广西大学 | Lane line detection method based on local line segment mode characteristics |
| CN108921209A (en) * | 2018-06-21 | 2018-11-30 | 杭州骑轻尘信息技术有限公司 | Image identification method, device and electronic equipment |
| CN108960210A (en) * | 2018-08-10 | 2018-12-07 | 武汉优品楚鼎科技有限公司 | It is a kind of to grind the method, system and device for reporting board-like identification and segmentation |
| CN109344856A (en) * | 2018-08-10 | 2019-02-15 | 华南理工大学 | A kind of off-line signature verification method based on multilayer discriminate feature learning |
| CN109344856B (en) * | 2018-08-10 | 2022-04-08 | 华南理工大学 | Offline signature identification method based on multilayer discriminant feature learning |
| CN109409254A (en) * | 2018-10-10 | 2019-03-01 | 成都优易数据有限公司 | A kind of electronic contract handwritten signature identification method based on twin neural network |
| CN109657498A (en) * | 2018-12-28 | 2019-04-19 | 广西师范大学 | The difference method for secret protection that top-k Symbiotic Model excavates in a plurality of stream |
| CN109657498B (en) * | 2018-12-28 | 2021-09-24 | 广西师范大学 | Differential privacy protection method for top-k symbiotic mode mining in multiple streams |
| CN109817348A (en) * | 2019-01-18 | 2019-05-28 | 郑静 | Portable medical signature management method and Related product |
| CN109817348B (en) * | 2019-01-18 | 2020-04-14 | 郑静 | Mobile medical signature management method and related product |
| CN110096977A (en) * | 2019-04-18 | 2019-08-06 | 中金金融认证中心有限公司 | The training method and handwriting verification method, equipment and medium of handwriting verification model |
| CN110096977B (en) * | 2019-04-18 | 2021-05-11 | 中金金融认证中心有限公司 | Training method of handwriting authentication model, handwriting authentication method, device and medium |
| CN110222666A (en) * | 2019-06-14 | 2019-09-10 | 河海大学常州校区 | A kind of signature false distinguishing method and system |
| CN110222666B (en) * | 2019-06-14 | 2022-09-13 | 河海大学常州校区 | Signature authentication method and system |
| CN110245615B (en) * | 2019-06-17 | 2022-09-13 | 河海大学常州校区 | Signature authentication method, system and storage medium |
| CN110245615A (en) * | 2019-06-17 | 2019-09-17 | 河海大学常州校区 | A kind of signature false distinguishing method, system and storage medium |
| CN110263740A (en) * | 2019-06-26 | 2019-09-20 | 四川新网银行股份有限公司 | Different type block letter document dubbing method based on OCR technique |
| WO2021027336A1 (en) * | 2019-08-14 | 2021-02-18 | 深圳壹账通智能科技有限公司 | Authentication method and apparatus based on seal and signature, and computer device |
| CN112395919A (en) * | 2019-08-16 | 2021-02-23 | 河海大学常州校区 | Off-line signature authentication method and system based on multi-feature metric learning |
| CN111178203A (en) * | 2019-12-20 | 2020-05-19 | 江苏常熟农村商业银行股份有限公司 | Signature verification method and device, computer equipment and storage medium |
| CN111275070A (en) * | 2019-12-26 | 2020-06-12 | 厦门商集网络科技有限责任公司 | Signature verification method and device based on local feature matching |
| CN111275070B (en) * | 2019-12-26 | 2023-11-14 | 厦门商集网络科技有限责任公司 | Signature verification method and device based on local feature matching |
| CN111209567B (en) * | 2019-12-30 | 2022-05-03 | 北京邮电大学 | Method and device for judging perceptibility of improving robustness of detection model |
| CN111209567A (en) * | 2019-12-30 | 2020-05-29 | 北京邮电大学 | Method and device for judging perceptibility of improving robustness of detection model |
| CN111178290A (en) * | 2019-12-31 | 2020-05-19 | 上海眼控科技股份有限公司 | Signature verification method and device |
| WO2021148692A1 (en) * | 2020-01-24 | 2021-07-29 | Universidad Autónoma de Madrid | Method and system for verifying dynamic handwriting and signatures by means of deep learning |
| CN113449543A (en) * | 2020-03-24 | 2021-09-28 | 百度在线网络技术（北京）有限公司 | Video detection method, device, equipment and storage medium |
| CN113449543B (en) * | 2020-03-24 | 2022-09-27 | 百度在线网络技术（北京）有限公司 | Video detection method, device, equipment and storage medium |
| TWI777188B (en) * | 2020-07-07 | 2022-09-11 | 新光人壽保險股份有限公司 | Contract signature authentication method and device |
| CN111753809A (en) * | 2020-07-10 | 2020-10-09 | 上海眼控科技股份有限公司 | Method and equipment for correcting handwritten signature |
| CN112036323B (en) * | 2020-09-01 | 2024-02-27 | 中国银行股份有限公司 | Signature handwriting authentication method, client and server |
| CN112036323A (en) * | 2020-09-01 | 2020-12-04 | 中国银行股份有限公司 | Signature handwriting identification method, client and server |
| CN112836636A (en) * | 2021-02-02 | 2021-05-25 | 北京惠朗时代科技有限公司 | Method and device for identifying authenticity of signature image |
| CN113095156A (en) * | 2021-03-23 | 2021-07-09 | 西安深信科创信息技术有限公司 | Double-current network signature identification method and device based on inverse gray scale mode |
| CN113269136A (en) * | 2021-06-17 | 2021-08-17 | 南京信息工程大学 | Offline signature verification method based on triplet loss |
| CN113269136B (en) * | 2021-06-17 | 2023-11-21 | 南京信息工程大学 | Off-line signature verification method based on triplet loss |
| CN113313092B (en) * | 2021-07-29 | 2021-10-29 | 太平金融科技服务(上海)有限公司深圳分公司 | Handwritten signature recognition method, and claims settlement automation processing method, device and equipment |
| CN113313092A (en) * | 2021-07-29 | 2021-08-27 | 太平金融科技服务(上海)有限公司深圳分公司 | Handwritten signature recognition method, and claims settlement automation processing method, device and equipment |
| CN114155613A (en) * | 2021-10-20 | 2022-03-08 | 杭州电子科技大学 | Offline signature comparison method based on convenient sample acquisition |
| CN114155613B (en) * | 2021-10-20 | 2023-09-15 | 杭州电子科技大学 | Offline signature comparison method based on convenient sample acquisition |
| CN113920589A (en) * | 2021-10-28 | 2022-01-11 | 平安银行股份有限公司 | Signature identification method, device, equipment and medium based on artificial intelligence |
| CN115880304A (en) * | 2023-03-08 | 2023-03-31 | 曲阜市巨力铁路轨道工程股份有限公司 | Method for identifying defects of sleeper based on complex scene |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106778586B (en) | 2020-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106778586A (en) | Offline handwriting signature verification method and system | |
| CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
| AU2004271639B2 (en) | Systems and methods for biometric identification using handwriting recognition | |
| Sigari et al. | Offline handwritten signature identification and verification using multi-resolution gabor wavelet | |
| CN107657241B (en) | Signature pen-oriented signature authenticity identification system | |
| CN104464079A (en) | Multi-currency-type and face value recognition method based on template feature points and topological structures of template feature points | |
| Zois et al. | Offline handwritten signature modeling and verification based on archetypal analysis | |
| CN110826408B (en) | Face recognition method by regional feature extraction | |
| Buse et al. | A structural and relational approach to handwritten word recognition | |
| Srihari et al. | Biometric and forensic aspects of digital document processing | |
| Verma et al. | A novel approach for structural feature extraction: contour vs. direction | |
| CN106778491A (en) | The acquisition methods and equipment of face 3D characteristic informations | |
| Roy et al. | Offline signature verification using grid based and centroid based approach | |
| Prashanth et al. | DWT based offline signature verification using angular features | |
| Sudharshan et al. | Handwritten signature verification system using deep learning | |
| CN110222660B (en) | Signature authentication method and system based on dynamic and static feature fusion | |
| Houtinezhad et al. | Off-line signature verification system using features linear mapping in the candidate points | |
| Shah et al. | Signature recognition and verification: The most acceptable biometrics for security | |
| Aravinda et al. | South Indian character recognition using statistical feature extraction and distance classifier | |
| Lee et al. | Combining macro and micro features for writer identification | |
| Yahyatabar et al. | Online signature verification: A Persian-language specific approach | |
| Blessy et al. | Deep Learning Approach to Offline Signature Forgery Prevention | |
| Rahman et al. | Writer-independent Offline Handwritten Signature Verification using Novel Feature Extraction Techniques | |
| Pervouchine et al. | Extraction and Analysis of Document Examiner Features from Vector Skeletons of Grapheme ‘th’ | |
| CN112395919B (en) | Off-line signature authentication method and system based on multi-feature metric learning |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20201117 Termination date: 20211208 |