CN107392136B - Signature authenticity identification system based on feature self-adaptive oscillation attenuation - Google Patents
Signature authenticity identification system based on feature self-adaptive oscillation attenuation Download PDFInfo
- Publication number
- CN107392136B CN107392136B CN201710581714.XA CN201710581714A CN107392136B CN 107392136 B CN107392136 B CN 107392136B CN 201710581714 A CN201710581714 A CN 201710581714A CN 107392136 B CN107392136 B CN 107392136B
- Authority
- CN
- China
- Prior art keywords
- signature
- image
- characteristic
- module
- point
- 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
Images
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
- G06V40/37—Writer recognition; Reading and verifying signatures based only on signature signals such as velocity or pressure, e.g. dynamic signature recognition
- G06V40/376—Acquisition
-
- 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
- G06V40/388—Sampling; Contour coding; Stroke extraction
-
- 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
- G06V40/394—Matching; Classification
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Collating Specific Patterns (AREA)
Abstract
The invention discloses a signature authenticity identification system based on feature self-adaptive oscillation attenuation. The handwriting input device comprises a handwriting sample database module, an image acquisition and input module, a preprocessing module, a characteristic analysis and extraction module, a receiving module, a modulation module and a pulse generation and identification module. The signature and counterfeit identification process is that a preprocessing module preprocesses an image acquired by an image acquisition module in real time, a characteristic analysis and extraction module extracts pressure distribution characteristics, direction angle time distribution characteristics, word length proportion time sequence fitting characteristics and cross-scale inertia corner point characteristics, the characteristics are fused and then input into a system receiving module in an independent vector group mode to generate connecting signals, two connecting signals are respectively modulated into sequence signals in a modulation module to generate characteristic correlation coefficients among sequences, and a pulse generation and identification module judges authenticity according to the input characteristic correlation coefficients and outputs a final identification result. The invention can quickly identify the handwriting of the text independently, and the identification result is stable and objective.
Description
Technical Field
The invention relates to a signature authenticity identification system based on feature self-adaptive oscillation attenuation, and belongs to the technical field of digital image processing and handwriting identification.
Background
The rapid development of information technology brings great convenience to daily life of people, and meanwhile, accurate authentication of personal identity and information security protection become a key problem which needs to be solved urgently in the information era at present. Handwriting is a relatively stable behavior characteristic of a person, and the authentication of the personal identity by using the handwriting has the characteristics of non-invasiveness (or non-contact property), easiness in acquisition, easiness in acceptance by the person and the like, and is an important means for identifying the personal identity.
The traditional handwriting identification method mostly adopts manual handwriting identification, but practical application shows that the method has obvious defects and shortcomings and easily causes negative influence on the identification result. 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. Therefore, the computer technology is adopted to process the original non-engineering field of handwriting authentication in an automatic and standardized mode, and the development of the field is greatly promoted.
At present, handwriting authentication can be divided into an online mode and an offline mode according to different actual operation modes, both the online mode and the offline mode have wide application backgrounds, and the handwriting authentication can play an important role in the fields of criminal investigation and court judgement of finance, insurance and public security and judicial departments and the like. In practice, the on-line handwriting identification technology is mature and enters a practical stage, however, the off-line handwriting identification technology cannot acquire dynamic information such as stroke sequence, writing speed, pen transportation pressure and the like when a writer writes like the on-line handwriting identification technology, and the difficulty of the on-line handwriting identification technology is undoubtedly higher.
Most of the existing offline handwriting authentication methods are specific to specific characters, the essence of the existing offline handwriting authentication methods is that the authentication of personal handwriting is converted into the authentication of personal specific character handwriting, in the practical application process, specific characters need to be collected, and the excessive dependence on specific samples causes the authentication methods to lack universality and robustness. Therefore, the system for identifying the text independent off-line handwriting is not mature at present, but becomes a current research hotspot.
Disclosure of Invention
The invention discloses a signature authenticity identification system based on characteristic self-adaptive oscillation attenuation aiming at the technical vacancy in the handwritten signature authenticity identification of the current social popularity, so as to improve the current situation of the independent offline handwriting authenticity identification of a text.
A signature authenticity identification system based on feature self-adaptive oscillation attenuation comprises:
image acquisition type-in module: recording the handwriting to be detected in real time, performing analog-to-digital conversion on the generated image of the handwriting to be detected, and transmitting and storing the image to the image preprocessing module for processing of the subsequent module;
the image database module: pre-storing a personal signature sample image and signature sample image information of a registered user, wherein the signature sample image information comprises signature time and signature gravity;
an image preprocessing module: sequentially carrying out graying processing, denoising processing, binarization processing, size standardization processing and signature stroke skeleton image extraction processing on the handwriting image to be detected and the signature sample image;
a feature vector extraction module: extracting each feature vector from the stroke skeleton image processed by the image preprocessing module; the characteristic vector comprises a pressure distribution characteristic, a direction angle time distribution characteristic, a word length proportion time sequence fitting characteristic and a cross-scale inertia corner point characteristic;
pressure distribution characteristics: carrying out feature extraction by using pen carrying pressure during signature, taking the average value of the maximum value and the minimum value of the pen carrying pressure as a pressure threshold, and selecting a part of the signature handwriting, of which the pen carrying pressure is greater than the threshold, as a research object; sampling the time used by a signer for signing, uniformly selecting a certain number of sampling points, calculating the proportion of the part of the signature handwriting corresponding to each time sampling point, which is larger than the threshold value, in the signed length, taking the proportion as a longitudinal axis, taking the time sampling points as a transverse axis, obtaining a proportional relation graph of the time and the part of the signature handwriting, which corresponds to each sampling point, in which the pen-moving pressure is larger than the threshold value, in the signature handwriting, in which the time and the pen-moving pressure are larger than the threshold value, in the signed length, by using a least square method to perform curve fitting on two-dimensional discrete points in the relation graph, and obtaining an associated fitting curve of the time and the part of the signature handwriting;
directional angular time distribution characteristics: the speeds of the pen point in the X, Y shaft direction in the rectangular coordinate system are respectively V during the writing processx(t)、Vy(t) according to the velocity V in the X-axis directionx(t) and speed V in Y-axis directiony(t) calculating the writing direction angle theta (t) of the movement of the pen point in the signature process
θ(t)＝tan-1(Vy(t)/Vx(t)), t is the writing time;
sampling the time used by a signer for signing, uniformly selecting a certain number of sampling points, calculating a pen point direction angle corresponding to each time sampling point, taking the time sampling points as a transverse axis and the pen point direction angle as a longitudinal axis to obtain a two-dimensional time and direction angle relation graph, and performing curve fitting on two-dimensional discrete points in the time and direction angle relation graph by using a least square method to obtain an associated fitting curve of the time and direction angle as a direction angle time distribution characteristic;
word length ratio time sequence fitting characteristics: extracting the sum of character pixel values in a signature skeleton image and recording the sum as word length, sampling the time used by a signer for signing, uniformly selecting a certain number of sampling points, counting the ratio of the word length written by the signer to the word length of the whole name on each sampling point as the word length proportion, taking the time sampling point as a transverse axis and the word length proportion as a longitudinal axis to obtain a two-dimensional time and word length proportional relation graph, and performing curve fitting on two-dimensional discrete points in the time and word length proportional relation graph by using a least square method to obtain an associated fitting curve of the time and word length proportion as a word length proportional time sequence fitting characteristic;
cross-scale inertial corner feature: introducing a Gaussian pyramid model, carrying out corner fast detection on a signature skeleton image by using Harris algorithm by using included angles and position information such as signature character connection points, turning points, corners and the like of pictures with different scales under a Gaussian pyramid in each scale plane, recording proportional positions of all V-shaped corners in a signature image, recording stroke feature values of the V-shaped corners, and finally, recording proportional position features under each scaleThe value and the stroke edge characteristic value are sequentially put into the corresponding characteristic vector v under each scaleiFinally, the feature vector v under each scale is utilizediGenerating a feature vector group V, taking the feature vector group V as a cross-scale inertial corner feature, and i is a label of a picture with different scales;
V＝{v0,v1,v2...vi,}
the V-shaped angular point refers to an angular point of a stroke included angle between 0 and 90 degrees;
the proportional position refers to the position distribution of the V-shaped angular points in the signature image relative to the font, namely on a horizontal axis and a vertical axis, the projection point of the V-shaped angular point of each character is divided into the proportional relation that the projection line segment of the stroke skeleton image on two number axes accounts for the total length of character projection on each number axis;
a receiving module: after extracting each characteristic value, the characteristic vector extraction module performs characteristic fusion on each characteristic value of the handwriting image to be detected according to a set fusion rule; meanwhile, feature fusion is carried out on each feature value of the signature sample image according to a set fusion rule, the receiving module outputs two connecting signals as an input part of the modulation module, and the two connecting signals comprise a feature connecting signal to be detected and a sample feature connecting signal;
a modulation module: and modulating the connection signal into a sequence signal, taking a correlation coefficient corresponding to the characteristic sequence to be detected and the sample characteristic sequence as a pulse excitation signal of the characteristic sequence element, and sending the pulse excitation signal to the pulse generation and identification module.
Modulating the connection signal into a sequence signal, adding a positive offset which is summarized as 1 to the characteristic sequence to be detected after the connection signal is modulated into the sequence signal, and then performing link modulation with the sample characteristic sequence in a nonlinear multiplication mode to generate a correlation coefficient as a pulse excitation signal of the characteristic sequence element and sending the pulse excitation signal to a pulse generation and identification module;
the pulse generation and identification module: receiving the pulse excitation signal to output a pulse signal, superposing the pulse signals for multiple times, comparing the superposed pulse with a set threshold value, outputting a signature authenticity identification result, and visually feeding back corresponding identification information to an operator.
And (3) identifying the signature image by combining a biological neuron multi-signal transmission pulse oscillation model and a multi-feature cooperative identification method by taking a signature multi-feature cooperative identification pseudo starting point. Comprises a pulse generating part and a pulse identifying part;
the pulse generating part comprises a comparator with variable threshold value and a pulse generator, when the pulse generator is switched on, the frequency of the pulse is constant, when the neuron outputs a pulse, the threshold value of the neuron is rapidly increased through feedback, when the characteristic correlation coefficient is smaller than the threshold value of the neuron, the pulse generator is switched off, the pulse is stopped from being emitted, the threshold value starts to exponentially decrease, when the characteristic correlation coefficient is larger than the threshold value of the neuron, the pulse generator is switched on, and the neuron is ignited to output a pulse or a pulse sequence;
the pulse identification part superposes the pulses output for many times, judges the authenticity of the image to be identified, outputs feedback information of the identification result on a system interface, and visually feeds back corresponding identification information to an operator.
The invention achieves the following beneficial effects: the invention discloses a signature authenticity identification system based on feature self-adaptive oscillation attenuation, which aims to improve the current situation of text independent off-line handwriting authenticity identification. Compared with the traditional authenticity identification method, the signature authenticity identification model takes signature multi-feature collaborative identification as a starting point, combines the biological neuron multi-signal transmission pulse oscillation model with the multi-feature collaborative identification method, can eliminate the influence caused by the feature data fluctuation of different signatures through self-adaptive control of the signal attenuation amplitude and the oscillation frequency, and can avoid the error phenomenon caused by mutual interference among multiple features.
Drawings
FIG. 1 is a block diagram of a system module;
FIG. 2 is a system process flow diagram;
FIG. 3 is a diagram of a main module model;
FIG. 4 is a diagram of a pulse coupled neuron model;
FIG. 5 is a flow chart of adaptive oscillation attenuation;
FIG. 6 is a schematic diagram of a time sampling model;
fig. 7 is a schematic view of a pressure distribution characteristic.
Detailed Description
The system module structure of the invention is shown in figure 1, the main processing steps are shown in figure 2, firstly, the signature board is controlled by the software operating system to collect the handwriting sample to be identified in real time, the person who needs to sign the handwriting sample writes 5 times own name on the signature board, the collected sample is preprocessed to extract the handwriting characteristic, the characteristic of the image to be detected is fused with the sample characteristic, the similarity between the characteristic to be detected and the sample characteristic is judged by adopting characteristic self-adaptive oscillation attenuation, the relevant pulse is output, and the identification information is fed back on the software operating interface. The whole working process of the handwriting identification system is that a hardware system is matched with a software system to complete the identification work of a handwriting sample in real time, and the authenticity information of the handwriting sample to be identified can be fed back to an operator quickly.
The main module model of the system of the invention is shown in fig. 3, and the specific processing method of each module is as follows:
(1) when the authenticity of the handwriting needs to be judged, firstly, a handwriting image to be detected is recorded in real time through the signature board, the user writes 5 own signatures on the signature board, the time and the gravity distribution used by the user signature are recorded, and the generated handwriting image to be detected is transmitted and stored to a preprocessing module of the system for the cyclic processing of a subsequent characteristic vector analyzing and extracting module.
(2) The image preprocessing module works: setting the handwriting image to be detected stored in the image acquisition data storage area as I1Copying one copy to be stored in data storage area and making image gray scale to obtain gray scale image I2And for the gray image I2Carrying out threshold segmentation by using a maximum inter-class variance method (otsu algorithm) to obtain a target binary image I3Finally, the target binary image I is processed3Standardized image I with size of 600 × 300 is obtained by size standardization4For standardized images I4Selecting a corrosion factor ofLinear type, with 8 pixel points in length, and selecting four directions of 0 degree, 90 degree, 45 degree and 135 degree (corresponding to four strokes of Chinese character horizontal, vertical, left falling and right falling) respectively for standardization image I4Performing closed operation, storing the stroke image as a stroke image, and rapidly refining the one-way stroke image to obtain a stroke skeleton image I with 0 degree, 90 degrees, 45 degrees and 135 degrees5 horizontal、I5 vertical、I5 left-falling off、I5 right falling downAre temporarily saved in the data storage area for subsequent processing.
The fast thinning algorithm is used for peeling off a target layer by layer based on mathematical morphology and fast extracting a stroke skeleton image, the process is to scan the sequence and the reverse sequence of the stroke image twice, determine the number of layers of points in each stroke image, judge whether the points belong to boundary points or skeletons according to the number of layers, (x, y) is a current point, P (x, y) is a current point pixel value, the following three steps are mainly carried out, a black value is specified to be 1, and a white value is 0:
a. sequentially scanning each point from top to bottom from left to right, and setting the coordinates of the current point as (x, y), so that the coordinates of four surrounding points thereon are (x-1, y +1), (x-1, y-1), (x, y-1), and the coordinates of four surrounding points below are (x +1, y-1), (x +1, y +1), and (x, y + 1); if the current point is a black point, the number of layers surrounding the four points is determined, and an upper surrounding layer value F (x, y) is defined:
when the current point pixel value p (x, y) is 1;
F(x,y)＝Min[F(x-1,y+1),F(x-1,y),F(x-1,y-1),F(x,y-1)]+1
when p (x, y) is 0;
F(x,y)＝0
b. from bottom to top, scanning each point from right to left in sequence, if the current point is a black point, judging the number of layers surrounding the four points, and defining the value G (x, y) of the lower surrounding layer:
when the current point pixel value p (x, y) is 1;
G(x,y)＝Min[G(x+1,y-1),G(x+1,y),G(x+1,y+1),G(x,y+1)]+1
when p (x, y) is 0;
G(x,y)＝0
c. let the actual number of layers M (x, y), the actual number of layers per point is the minimum of the two numbers of layers above, i.e.:
M(x,y)＝Min[F(x,y),G(x,y)]
and scanning each point from top to bottom from left to right, judging the layer number condition of 8 points around the point, if the current point is the maximum point of the 9 points, keeping the current point, and if the current point is not the maximum point, deleting the current point. The rapid refining work can be completed through the steps.
(3) An image I containing only a single-direction stroke skeleton (here, the horizontal direction is taken as a representative) is taken5 horizontalRandomly selecting trisection points on the horizontal strokes in the image, reconstructing the circle, extracting and calculating the radius of the horizontal circle, storing the reciprocal curvature of the radius length as the horizontal stroke characteristic value reflecting the horizontal stable characteristic of the personal handwriting as the stroke characteristic value of the direction, and aligning the vertical direction image I5 verticalLeft-falling direction image I5 left-falling offImage of falling direction I5 right falling downAnd sequentially carrying out the same processing, respectively storing the characteristic values as unidirectional stroke characteristic values, and sequentially carrying out the above cyclic processing of image preprocessing and characteristic vector extraction on all the handwriting images to be detected and the signature sample images to obtain the stroke characteristic values of each handwriting image to be detected and the signature sample image. 5 images can be respectively taken from the handwriting image to be detected and the signature sample image in the database.
The specific process of the size standardization treatment comprises the following steps:
1) calculating a target binary image I3Detecting the position of the central pixel point of each pixel column, and adjusting the target binary image I according to the position information of the central pixel point3The inclination angle is such that a set number of central pixel points are in a target binary image I of the signature3On the horizontal axis of the signature image, the horizontal axis is referenced in pixel rows;
2) detecting the position of edge pixel points of the signature image adjusted in the step 1) from the upper direction, the lower direction, the left direction and the right direction, cutting the image according to the position information of the edge pixel points, and standardizing the image size in equal proportion.
(4) And the characteristic vector extraction module is used for extracting different stroke characteristic values in the handwriting image to be detected and the signature sample image after the image preprocessing module finishes operation. The feature vector extraction module comprises a plurality of different sub-modules, and the sub-modules respectively comprise a pen holding posture feature extraction module, a pressure distribution feature extraction module, a direction angle distribution feature extraction module, a word length proportion time sequence fitting feature module and a cross-scale inertia corner feature module.
a. A pressure distribution feature extraction module: the pressure of the pen point on the screen is changed when different persons or the same person signs, but due to the writing habit of each person, the change of the pressure of the sign pen on the screen is usually the same when the person writes the same characters, and the characteristic is not easy to imitate, so that the characteristic has higher identification for the identity of a signer. As shown in FIG. 7, the pressure sensors detect that the signature samples are at different parts of the word, and the pen-moving pressure of the signature samples is distributed between 0.01mN and 0.9 mN. And performing feature extraction by using pen-moving pressure during signature, sampling the average value of the maximum value and the minimum value of the pen-moving pressure in the signature as a pressure threshold, and selecting a part of the signature handwriting, of which the pen-moving pressure is greater than the threshold, as a research object. Uniformly selecting 64 sampling points in the period from the beginning of pen-down signature to the end of signature, calculating the proportion of the part, larger than the threshold, of the signature handwriting corresponding to each time sampling point to the length of the signed handwriting, taking the time sampling points as the horizontal axis and the proportion of the part, larger than the threshold, of the signature handwriting as the vertical axis, fitting a curve by using a least square method, and taking a pressure distribution curve of the signature as a pressure distribution characteristic.
b. The direction angle time distribution characteristic extraction module: the speeds of the pen point in the X, Y shaft direction in the rectangular coordinate system are respectively V during the writing processx(t)、Vy(t) according to the velocity V in the X-axis directionx(t) and speed V in Y-axis directiony(t) calculating the writing direction angle theta (t) of the movement of the pen point in the signature process
θ(t)＝tan-1(Vy(t)/Vx(t))
From the beginning of the signature falling to the end of the signature, as shown in fig. 6, 64 sampling points are uniformly selected in the period of time, the pen point direction angle corresponding to each time sampling point is calculated, the time sampling point is taken as the horizontal axis, the proportion of the part larger than the threshold value is taken as the vertical axis, the curve is fitted by using the least square method, and the direction angle fitting curve of the signature is taken as the direction angle time distribution characteristic.
c. The character length proportion time sequence fitting characteristic extraction module is used for extracting the sum of character pixel values in the stroke skeleton image to be recorded as the character length L, sampling the time for signing by a signer, uniformly selecting 64 sampling points as shown in figure 6, and counting the character length L of the signer writing on each sampling point1Ratio L to the overall name length L1Word length ratio of L with time sample point t as horizontal axis,/L as word length ratio1and/L, taking the vertical axis as the vertical axis, obtaining a two-dimensional time and word length proportional relation graph, and performing curve fitting on two-dimensional discrete points in the time and word length proportional relation graph by using a least square method to obtain a correlation fitting curve of the time and word length proportion, namely the word length proportion time sequence fitting characteristic.
d. A cross-scale inertia corner feature extraction module: introducing a Gaussian pyramid model, carrying out corner rapid detection on a signature skeleton image by using Harris algorithm by using included angles and position information such as signature character connection points, turning points, corners and the like of pictures with different scales under a Gaussian pyramid in each scale plane, recording proportional positions of all V-shaped corners in the signature image, recording stroke feature values of the V-shaped corners, and sequentially putting the proportional position feature values and the stroke feature values under each scale to corresponding feature vectors V under each scaleiFinally, the feature vector v under each scale is utilizediGenerating a feature vector group V, taking the feature vector group V as a cross-scale inertial corner feature, and i is a label of a picture with different scales;
V＝{v0,v1,v2...vi,}
the V-shaped angular point refers to an angular point of a stroke included angle between 0 and 90 degrees;
the proportional position refers to the position distribution of the V-shaped angular points in the signature image relative to the font, namely on the horizontal axis and the longitudinal axis, the projection line segments of the projection points of the V-shaped angular points of each character on the two number axes of the whole stroke skeleton image of the signature account for the total length of the character projection on each number axis;
(5) the receiving module, the modulating module and the pulse generating and identifying module construct the pulse coupled neural network by using the pulse coupled neuron model of fig. 4, and the processing flow of the three modules is shown in fig. 5. Wherein the parameters of the parameter part of the initialization model comprise AL、AF、VL、VFM, W, β and S, wherein A isLAnd AFThe decay time constants, V, of the sample input and the input to be measured, respectivelyLAnd VFThe model parameters of the calculation model include F, L, U, Y, E, where F is the input channel to be measured, L is the sample input channel, U is the multiplied modulated output signal, Y is the output signal of the pulse generator, and E is the dynamic threshold of the pulse generator.
After the characteristic vector extraction module extracts each characteristic value, the receiving module performs characteristic fusion on each characteristic value of the handwriting image to be detected according to the formula (1); meanwhile, the characteristic values of the specimen signature sample image are subjected to characteristic fusion according to the formula (2),
the input part of the receiving module comprises a sample characteristic input Fij[n]And a feature to be tested input Lij[n]Respectively by the connection weighting coefficient M of the sample input and the connection weighting coefficient W of the input to be measuredIts neighboring neurons are connected; sample feature input Fij[n]And a feature to be tested input Lij[n]The two functional units carry out iterative operation, and the iterative process is attenuated according to an exponential law; and sample feature input Fij[n]Adding an external stimulus Sij；FijIs the feedback of the (i, j) th neuron, LijIs a coupling connection, VFAnd VLAre respectively FijAnd LijThe intrinsic potential of (a); m and W are respectively a matrix which represents the size of the peripheral neurons of the central neuron and reflects the strength of information transmitted by the central neuron by the adjacent neurons; y isijIs the output of the neuron at the time of the sub-iteration; a. theFAnd only ALRespectively as sample feature inputs Fij[n]And a feature to be tested input Lij[n]Decay time constant of, WijklTo connect the weighting coefficients one, MijklIs a connection weight coefficient two.
(6) In the modulation module, the connection signal from the receiving module is modulated into a sequence signal under the action of a connection strength coefficient β, and a correlation coefficient corresponding to the characteristic sequence to be detected and the sample characteristic sequence is taken as a pulse excitation signal of a characteristic sequence element, the partial link coefficient is generated in a link modulation mode, the link modulation refers to an internal activity item of a neuron, and the internal activity item of the neuron, namely a connection domain, is input L by the characteristic to be detectedij[n]And sample feature input Fij[n]The two functional units are combined together in a nonlinear multiplication mode, the mathematical expression is shown as formula (3), and Uij[n]＝Fij[n](1+βLij[n]) Formula (3)
In equation (3), β is the connection strength factor between synapses, and the connection strength factor β regulates the extent to which peripheral neurons affect the firing cycle of the central neuronijThe effect of scaling, a larger link factor can result in a larger range of pulse synchronization.
(7) Relating the characteristic to a coefficient, i.e. the connection field UijThe pulse generator is input to a pulse generation and identification module, and a pulse generation part consists of a comparator with variable threshold and a pulse generator; pulse generation and modulation moduleThe formula is as follows:
Eij＝e-αEEij[n-1]+VEYij[n]
wherein, YijIs the output of the neuron at the time of the sub-iteration, EijIs the dynamic threshold, V, of the pulse generatorEIs a threshold amplitude coefficient, which is a closed-valued amplitude adjustment constant. When the pulse generator is turned on, the frequency of the firing pulses is constant. When a neuron outputs a pulse, the dynamic threshold E of the neuronijIs rapidly improved by feedback. When characteristic correlation coefficient UijLess than neuron dynamic threshold EijThe pulse generator is turned off, stopping the pulse from being delivered. Followed by a dynamic threshold EijBegin to exponentially decrease as the characteristic correlation coefficient UijGreater than neuron dynamic threshold EijThe pulse generator is turned on and the neuron is fired, outputting a pulse or pulse train, while returning the change in the adaptive decay parameter a to the receiving module. The pulse identification part finally superposes the pulses output for multiple times by setting certain iteration times, compares the superposed pulses with a set threshold value, judges the authenticity of the image to be identified and outputs feedback information of the authenticity identification result on a system interface.
The present invention has been disclosed in terms of the preferred embodiment, but it is not intended to be limited to the embodiment, and all technical solutions obtained by substituting or converting the equivalent embodiments fall within the scope of the present invention.
Claims (7)
1. A signature authenticity identification system based on feature self-adaptive oscillation attenuation is characterized by comprising the following components:
image acquisition type-in module: recording the handwriting to be detected in real time, performing analog-to-digital conversion on the generated image of the handwriting to be detected, and transmitting and storing the image to the image preprocessing module;
the image database module: pre-storing a personal signature sample image and signature sample image information of a registered user, wherein the signature sample image information comprises signature time and signature gravity;
an image preprocessing module: sequentially carrying out graying processing, denoising processing, binarization processing, size standardization processing and signature stroke skeleton image extraction processing on the handwriting image to be detected and the signature sample image;
a feature vector extraction module: extracting each feature vector from the stroke skeleton image processed by the image preprocessing module;
a receiving module: after extracting each characteristic value, the characteristic vector extraction module performs characteristic fusion on each characteristic value of the handwriting image to be detected according to a set fusion rule; meanwhile, feature fusion is carried out on each feature value of the signature sample image according to a set fusion rule, the receiving module outputs two connecting signals as an input part of the modulation module, and the two connecting signals comprise a feature connecting signal to be detected and a sample feature connecting signal;
a modulation module: modulating the connecting signal into a sequence signal, taking a correlation coefficient corresponding to the characteristic sequence to be detected and the sample characteristic sequence as a pulse excitation signal of a characteristic sequence element, and sending the pulse excitation signal to a pulse generation and identification module;
the pulse generation and identification module: receiving a pulse excitation signal to output a pulse signal, superposing the pulse signal for multiple times, comparing the superposed pulse with a set threshold value, and outputting a signature authenticity identification result;
the characteristic vector comprises a pressure distribution characteristic, a direction angle time distribution characteristic, a word length proportion time sequence fitting characteristic and a cross-scale inertia corner point characteristic;
pressure distribution characteristics: carrying out feature extraction by using pen carrying pressure during signature, taking the average value of the maximum value and the minimum value of the pen carrying pressure as a pressure threshold, and selecting a part of the signature handwriting, of which the pen carrying pressure is greater than the threshold, as a research object; sampling the time used by a signer for signing, uniformly selecting a certain number of sampling points, calculating the proportion of the part of the signature handwriting corresponding to each time sampling point, which is larger than the threshold value, in the signed length, taking the proportion as a longitudinal axis, taking the time sampling points as a transverse axis, obtaining a proportional relation graph of the time and the part of the signature handwriting, which corresponds to each sampling point, in which the pen-moving pressure is larger than the threshold value, in the signature handwriting, in which the time and the pen-moving pressure are larger than the threshold value, in the signed length, by using a least square method to perform curve fitting on two-dimensional discrete points in the relation graph, and obtaining an associated fitting curve of the time and the part of the signature handwriting;
directional angular time distribution characteristics: the speeds of the pen point in the X, Y shaft direction in the rectangular coordinate system are respectively V during the writing processx(t)、Vy(t) according to the velocity V in the X-axis directionx(t) and speed V in Y-axis directiony(t) calculating the writing direction angle theta (t) of the movement of the pen point in the signature process
θ(t)＝tan-1(Vy(t)/Vx(t)), t is the writing time;
sampling the time used by a signer for signing, uniformly selecting a certain number of sampling points, calculating a pen point direction angle corresponding to each time sampling point, taking the time sampling points as a transverse axis and the pen point direction angle as a longitudinal axis to obtain a two-dimensional time and direction angle relation graph, and performing curve fitting on two-dimensional discrete points in the time and direction angle relation graph by using a least square method to obtain an associated fitting curve of the time and direction angle as a direction angle time distribution characteristic;
word length ratio time sequence fitting characteristics: extracting the sum of character pixel values in a signature skeleton image and recording the sum as word length, sampling the time used by a signer for signing, uniformly selecting a certain number of sampling points, counting the ratio of the word length written by the signer to the word length of the whole name on each sampling point as the word length proportion, taking the time sampling point as a transverse axis and the word length proportion as a longitudinal axis to obtain a two-dimensional time and word length proportional relation graph, and performing curve fitting on two-dimensional discrete points in the time and word length proportional relation graph by using a least square method to obtain an associated fitting curve of the time and word length proportion as a word length proportional time sequence fitting characteristic;
cross-scale inertial corner feature: introducing a Gaussian pyramid model, wherein pictures with different scales exist under the Gaussian pyramid, and firstly, in each scale plane, by using included angles and position information such as signature character connection points, turning points, angular points and the likeCarrying out angular point fast detection on the name skeleton image by using a Harris algorithm, recording the proportional positions of all V-shaped angular points in the signature image, recording the stroke characteristic values of the V-shaped angular points, and finally sequentially putting the proportional position characteristic values and the stroke characteristic values under each scale into the corresponding characteristic vector V under each scaleiFinally, the feature vector v under each scale is utilizediGenerating a feature vector group V, taking the feature vector group V as the cross-scale inertial corner feature, i is the label of the picture with different scales,
V＝{v0,v1,v2...vi,}
the V-shaped angular point refers to an angular point of a stroke included angle between 0 and 90 degrees;
the proportional position refers to the position distribution of the V-shaped angular points in the signature image relative to the font, namely on the horizontal axis and the longitudinal axis, the projection point of the V-shaped angular point of each character is divided into the proportional relation that the projection line segment of the stroke skeleton image on the two number axes accounts for the total length of the character projection on each number axis.
2. The signature authenticity identification system based on feature adaptive shock attenuation according to claim 1, wherein: in the image pre-processing module, the image is pre-processed,
respectively selecting four directions of 0 degree, 90 degrees, 45 degrees and 135 degrees to respectively standardize the size I image4Performing closed operation, storing the stroke image as a stroke image, and rapidly refining the one-way stroke image to obtain a stroke skeleton image I with 0 degree, 90 degrees, 45 degrees and 135 degrees5 horizontal、I5 vertical、I5 left-falling off、I5 right falling down。
3. The signature authenticity identification system based on feature adaptive shock attenuation according to claim 2, wherein: the rapid refining processing comprises the following steps:
scanning the sequence and the reverse sequence of the stroke images twice, determining the layer number of points in each stroke image, judging whether the points belong to boundary points or skeletons according to the layer number, (x, y) is a current point, P (x, y) is a pixel value of the current point, and performing the following three steps, wherein a black value is 1 and a white value is 0:
a. sequentially scanning each point from top to bottom from left to right, and setting the coordinates of the current point as (x, y), so that the coordinates of four surrounding points thereon are (x-1, y +1), (x-1, y-1), (x, y-1), and the coordinates of four surrounding points below are (x +1, y-1), (x +1, y +1), and (x, y + 1); if the current point is a black point, the number of layers surrounding the four points is determined, and an upper surrounding layer value F (x, y) is defined:
when the current point pixel value p (x, y) is 1;
F(x,y)＝Min[F(x-1,y+1),F(x-1,y),F(x-1,y-1),F(x,y-1)]+1
when p (x, y) is 0;
F(x,y)＝0
b. from bottom to top, scanning each point from right to left in sequence, if the current point is a black point, judging the number of layers surrounding the four points, and defining the value G (x, y) of the lower surrounding layer:
when the current point pixel value p (x, y) is 1;
G(x,y)＝Min[G(x+1,y-1),G(x+1,y),G(x+1,y+1),G(x,y+1)]+1
when p (x, y) is 0;
G(x,y)＝0
c. let the actual number of layers M (x, y), the actual number of layers per point is the minimum of the two numbers of layers above, i.e.:
M(x,y)＝Min[F(x,y),G(x,y)]
and scanning each point from top to bottom from left to right, judging the layer number condition of 8 points around the point, if the current point is the maximum point of the 9 points, keeping the current point, and if the current point is not the maximum point, deleting the current point.
4. The signature authenticity identification system based on feature adaptive shock attenuation according to claim 1, wherein: in the receiving module, the receiving module is provided with a receiving module,
after extracting each characteristic value, the characteristic vector extraction module performs characteristic fusion on each characteristic value of the handwriting image to be detected according to the formula (1); meanwhile, the characteristic values of the specimen signature sample image are subjected to characteristic fusion according to the formula (2),
the input part of the receiving module comprises a sample characteristic input Fij[n]And a feature to be tested input Lij[n]The connecting weighting coefficient M input by the sample and the connecting weighting coefficient W input to be detected are respectively connected with the adjacent neurons; sijIs an external excitation signal; fijIs the feedback of the (i, j) th neuron, LijIs a coupling connection, VFAnd VLAre respectively FijAnd LijThe intrinsic potential of (a); a. theFAnd only ALRespectively as sample feature inputs Fij[n]And a feature to be tested input Lij[n]Decay time constant of, WijklTo connect the weighting coefficients one, MijklIs a connection weight coefficient two.
5. The signature authenticity identification system based on feature adaptive shock attenuation according to claim 1, wherein: in the modulation module, the connection signal is modulated into a sequence signal, after the connection signal is modulated into the sequence signal, a positive offset which is reduced to 1 is added to the characteristic sequence to be detected, then the positive offset and the sample characteristic sequence are subjected to link modulation in a nonlinear multiplication mode, and a correlation coefficient is generated and serves as a pulse excitation signal of the characteristic sequence element to be sent to the pulse generation and identification module.
6. The signature authenticity identification system based on feature adaptive shock attenuation according to claim 5, wherein: internal activity item of neuron, namely connecting domain Uij[n]From the feature under test Lij[n]And sample feature input Fij[n]The two functional units are combined together in a nonlinear multiplication mode, the mathematical expression is shown as formula (3), and Uij[n]＝Fij[n](1+βLij[n]) Formula (3)
β is the coefficient of connection strength between synapses.
7. The signature authenticity identification system based on feature adaptive shock attenuation according to claim 1, wherein: the pulse generation and identification module comprises a pulse generation part and a pulse identification part;
the pulse generating part comprises a comparator with variable threshold value and a pulse generator, when the pulse generator is switched on, the frequency of the pulse is constant, when the neuron outputs a pulse, the threshold value of the neuron is rapidly increased through feedback, when the characteristic correlation coefficient is smaller than the threshold value of the neuron, the pulse generator is switched off, the pulse is stopped from being emitted, the threshold value starts to exponentially decrease, when the characteristic correlation coefficient is larger than the threshold value of the neuron, the pulse generator is switched on, and the neuron is ignited to output a pulse or a pulse sequence;
the pulse identification part superposes the pulses output for many times, compares the superposed pulses with a set threshold value and outputs a signature authenticity identification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710581714.XA CN107392136B (en) | 2017-07-17 | 2017-07-17 | Signature authenticity identification system based on feature self-adaptive oscillation attenuation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710581714.XA CN107392136B (en) | 2017-07-17 | 2017-07-17 | Signature authenticity identification system based on feature self-adaptive oscillation attenuation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107392136A CN107392136A (en) | 2017-11-24 |
| CN107392136B true CN107392136B (en) | 2020-07-17 |
Family
ID=60339823
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710581714.XA Active CN107392136B (en) | 2017-07-17 | 2017-07-17 | Signature authenticity identification system based on feature self-adaptive oscillation attenuation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107392136B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086771B (en) * | 2018-08-16 | 2021-06-08 | 电子科技大学 | Optical character recognition method |
| CN109190351A (en) * | 2018-09-19 | 2019-01-11 | 宁辛 | On-line signature person identity authorization system based on mobile terminal, device and method |
| CN109829491B (en) * | 2019-01-22 | 2021-09-28 | 开易（北京）科技有限公司 | Information processing method, apparatus and storage medium for image detection |
| CN111046751B (en) * | 2019-11-22 | 2024-02-13 | 华中师范大学 | Formula identification method and device |
| CN111340810B (en) * | 2020-05-21 | 2020-08-25 | 深圳市儿童医院 | Intelligent evaluation method for Chinese character writing quality |
| CN113139435A (en) * | 2021-03-30 | 2021-07-20 | 北京思特奇信息技术股份有限公司 | Self-learning signature handwriting deepening identification method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN104809451A (en) * | 2015-05-15 | 2015-07-29 | 河海大学常州校区 | Handwriting authentication system based on stroke curvature detection |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI569176B (en) * | 2015-01-16 | 2017-02-01 | 新普科技股份有限公司 | Method and system for identifying handwriting track |
-
2017
- 2017-07-17 CN CN201710581714.XA patent/CN107392136B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN104809451A (en) * | 2015-05-15 | 2015-07-29 | 河海大学常州校区 | Handwriting authentication system based on stroke curvature detection |
Non-Patent Citations (3)
| Title |
|---|
| PCNN的原理及应用;顾晓东，余道衡;《电路与系统学报》;20010930;第6卷(第3期);全文 * |
| 基于优化DTW算法的在线手写签名认证系统研究与设计;罗勇军;《中国优秀硕士学位论文全文数据库 信息科技辑》;20141015(第10期);正文第9-10页 * |
| 基于神经网络的离线签名自动鉴别系统;钟建;《中国优秀硕士学位论文全文数据库 信息科技辑》;20120315(第3期);正文第9-58页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107392136A (en) | 2017-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107392136B (en) | Signature authenticity identification system based on feature self-adaptive oscillation attenuation | |
| CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
| CN106778586B (en) | Off-line handwritten signature identification method and system | |
| Yuan et al. | Deep residual network with adaptive learning framework for fingerprint liveness detection | |
| CN107657241B (en) | Signature pen-oriented signature authenticity identification system | |
| CN104809451B (en) | A kind of person's handwriting identification system based on stroke curvature measuring | |
| CN110543822A (en) | finger vein identification method based on convolutional neural network and supervised discrete hash algorithm | |
| Bawane et al. | Object and character recognition using spiking neural network | |
| Avola et al. | R-SigNet: Reduced space writer-independent feature learning for offline writer-dependent signature verification | |
| Engin et al. | Offline signature verification on real-world documents | |
| Ali et al. | Recognition system for Pakistani paper currency | |
| Sudharshan et al. | Handwritten signature verification system using deep learning | |
| Goud et al. | Smart attendance notification system using SMTP with face recognition | |
| Kumar | Signature verification using neural network | |
| Jain et al. | Online signature verification using energy, angle and directional gradient feature with neural network | |
| CN116229528A (en) | Living body palm vein detection method, device, equipment and storage medium | |
| Fariza et al. | Mobile Based Offline Handwritten Signature Forgery Identification using Convolutional Neural Network | |
| CN112395919B (en) | Off-line signature authentication method and system based on multi-feature metric learning | |
| Dewan et al. | Offline Signature Verification Using Neural Network | |
| Khan et al. | An offline signature recognition and verification system based on neural network | |
| Verma et al. | Static Signature Recognition System for User Authentication Based Two Level Cog, Hough Tranform and Neural Network | |
| Deore et al. | Offline signature recognition: Artificial neural network approach | |
| Sharma et al. | An Offline Signature Verification System Using Neural Network Based on Angle Feature and Energy Density | |
| Sharma et al. | Offline signature verification using supervised and unsupervised neural networks | |
| Shekar et al. | Offline Signature verification using CNN and SVM classifier |
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 |