CN107330429A - A kind of localization method and device of certificate entry - Google Patents
A kind of localization method and device of certificate entry Download PDFInfo
- Publication number
- CN107330429A CN107330429A CN201710349083.9A CN201710349083A CN107330429A CN 107330429 A CN107330429 A CN 107330429A CN 201710349083 A CN201710349083 A CN 201710349083A CN 107330429 A CN107330429 A CN 107330429A
- Authority
- CN
- China
- Prior art keywords
- certificate
- sample
- entry
- feature point
- region
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
- G06F18/2193—Validation; Performance evaluation; Active pattern learning techniques based on specific statistical tests
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a kind of localization method and device of certificate entry, it is related to certificate identification technology field.The localization method and device for the certificate entry that the present invention is provided, is the initial position based on entry in certificate when carrying out entry positioning using certificate regression model, returns the physical location for obtaining each entry.Because entry initial position is fixed in certificate and the standardized degree of certificate image does not interfere with the initial position of entry in certificate, therefore entry in certificate is positioned by certificate regression model, the accuracy rate for enabling to entry to position is improved.
Description
Technical field
The present invention relates to certificate identification technology field, more particularly to a kind of localization method and device of certificate entry.
Background technology
At present, the application of certificate identification is more and more extensive.Certificate identification is exactly to gather various certificate figures by mobile terminal
Picture, for example, China second-generation identity card, passport, driving license, driving license image etc., are then scanned to certificate image, automatic identification card
The information of each entry on part.In order to realize the identification to certificate, it is necessary to which each entry is positioned to additional clause.
In the prior art, when being positioned to additional clause each entry, it is typically, binaryzation first is carried out to certificate
Processing, obtains bianry image, then carries out printed page analysis to the bianry image, to position the entry for needing to recognize.
It is lack of standardization because printed page analysis requires higher to the standard degree of image when carrying out entry positioning using prior art
Image can make it that the result of printed page analysis is impacted.And the certificate image collected in practical application is typically nonstandard
Image, for example, the certificate image collected often has that irregular illumination, surface reflection, background are present.Cause
This, in the prior art, when positioning the entry for needing to recognize by printed page analysis, often occurs the problem of positioning is inaccurate.
The content of the invention
In view of the above problems, it is proposed that the present invention overcomes above mentioned problem or solved at least in part to provide one kind
A kind of localization method and device of certificate entry of above mentioned problem.
According to the first aspect of the present invention there is provided a kind of localization method of certificate entry, this method includes:
Obtain certificate region;
According to the certificate region, determine that each needing in the certificate region is known using default certificate regression model
Other bar destination locations;The certificate regression model is obtained according to fisrt feature point and the training of second feature point;Described first
Characteristic point is entry initial position characteristic point, and the second feature point is entry physical location characteristic point.
Optionally, methods described also includes:
The certificate regression model is obtained according to fisrt feature point and the training of second feature point.
Optionally, the step that the certificate regression model is obtained according to fisrt feature point and the training of second feature point
Suddenly, including:
First eigenvector is determined according to the fisrt feature point, according to the second feature point determine second feature to
Amount;
Using the first eigenvector and the default characteristics algorithm of second feature vector training, obtain certificate and return mould
Type.
Optionally, the step of acquisition certificate region, including:
Obtain target image;
According to the target image, certificate region is determined using default certificate detection model;The certificate detection model
It is to be obtained according to sample training;Wherein, the sample includes positive sample and negative sample, and the positive sample is to include certificate area
The image in domain, the negative sample is the image not comprising certificate region.
Optionally, the step in certificate region is determined using default certificate detection model according to the target image described
Before rapid, methods described also includes:
Feature based algorithm, calculates the characteristic set of each sample;Wherein, comprising each in the characteristic set
The corresponding at least one characteristic value of sample;
Using the characteristic set of each sample, at least one strong classifier is built;
Cascade classifier is built according at least one described strong classifier, the cascade classifier is defined as the card
Part detection model.
According to the second aspect of the present invention there is provided a kind of positioner of certificate entry, the device includes:
Acquisition module, for obtaining certificate region;
Determining module, for according to the certificate region, the certificate region to be determined using default certificate regression model
In need to each recognize bar destination locations;The certificate regression model is according to fisrt feature point and the training of second feature point
Obtain;The fisrt feature point is entry initial position characteristic point, and the second feature point is entry physical location characteristic point.
Optionally, described device also includes:
Training module, for obtaining the certificate regression model according to fisrt feature point and the training of second feature point.
Optionally, the training module, including:
First determination sub-module, it is special according to described second for determining first eigenvector according to the fisrt feature point
Levy a determination second feature vector;
First training submodule, for utilizing the first eigenvector and the default feature of second feature vector training
Algorithm, obtains certificate regression model.
Optionally, the acquisition module, including:
Acquisition submodule, for obtaining target image;
Second determination sub-module, for according to the target image, certificate area to be determined using default certificate detection model
Domain;The certificate detection model is obtained according to sample training;Wherein, the sample includes positive sample and negative sample, described
Positive sample is the image for including certificate region, and the negative sample is the image not comprising certificate region.
Optionally, described device also includes:
Calculating sub module, for feature based algorithm, calculates the characteristic set of each sample;Wherein, the feature
The corresponding at least one characteristic value of each sample is included in set;
Submodule is built, for the characteristic set using each sample, at least one strong classifier is built;
3rd determination sub-module, for building cascade classifier according at least one described strong classifier, by the cascade
Grader is defined as the certificate detection model.
A kind of localization method and device of certificate entry provided in an embodiment of the present invention, is entered using certificate regression model
It is the initial position based on entry in certificate when row entry is positioned, returns the physical location for obtaining each entry.Due to certificate
Middle entry initial position is fixed and the standardized degree of certificate image does not interfere with the initial position of entry in certificate, therefore is passed through
Certificate regression model is positioned to entry in certificate, and the accuracy rate for enabling to entry to position is improved.
Described above is only the general introduction of technical solution of the present invention, in order to better understand the technological means of the present invention,
And can be practiced according to the content of specification, and in order to allow above and other objects of the present invention, feature and advantage can
Become apparent, below especially exemplified by the embodiment of the present invention.
Brief description of the drawings
By reading the detailed description of hereafter preferred embodiment, various other advantages and benefit is general for this area
Logical technical staff will be clear understanding.Accompanying drawing is only used for showing the purpose of preferred embodiment, and is not considered as to this hair
Bright limitation.And in whole accompanying drawing, identical part is denoted by the same reference numerals.In the accompanying drawings:
Fig. 1 is a kind of step flow chart of the localization method of certificate entry provided in an embodiment of the present invention;
Fig. 2-1 is the step flow chart of the localization method of another certificate entry provided in an embodiment of the present invention;
Fig. 2-2 is a kind of certificate area obtaining method flow chart of steps provided in an embodiment of the present invention;
Fig. 2-3 is a kind of method and step flow chart for obtaining certificate regression model provided in an embodiment of the present invention;
Fig. 2-4 is a kind of fisrt feature point schematic diagram provided in an embodiment of the present invention;
Fig. 2-5 is a kind of second feature point schematic diagram provided in an embodiment of the present invention;
Fig. 3 is a kind of block diagram of the positioner of certificate entry provided in an embodiment of the present invention;
Fig. 4-1 is the block diagram of the positioner of another certificate entry provided in an embodiment of the present invention;
Fig. 4-2 is the block diagram of the positioner of another certificate entry provided in an embodiment of the present invention.
Embodiment
The exemplary embodiment of the present invention is more fully described below with reference to accompanying drawings.Although showing this hair in accompanying drawing
Bright exemplary embodiment, it being understood, however, that may be realized in various forms the implementation of the invention without that should be illustrated here
Example is limited.It is opposite to be able to be best understood from the present invention there is provided these embodiments, and can be by the present invention's
Scope completely convey to those skilled in the art.
Fig. 1 is a kind of step flow chart of the localization method of certificate entry provided in an embodiment of the present invention, as shown in figure 1,
This method can include:
Step 101, acquisition certificate region.
Certificate region in the embodiment of the present invention is the actual area of certificate.The certificate can for identity card, driver's license,
Student's identity card etc., the embodiment of the present invention is not construed as limiting for the specific species of certificate.Typically, when carrying out certificate identification, figure can be passed through
As harvester carries out IMAQ to certificate.In the certificate image collected, it will usually include certificate region and background area
Domain.For example, certificate is placed on desk, when carrying out IMAQ to the certificate by imaging sensor, the certificate figure collected
The certificate region and the subregion of desk can be included as in.
Additional clause entry is positioned for convenience, while improving the accuracy rate of positioning.Entry is determined in certificate
, it is necessary to determine certificate region in the certificate image collected when position.
Step 102, according to the certificate region, determined using default certificate regression model in the certificate region
Bar destination locations need to each be recognized.
In embodiments of the present invention, certificate can be obtained by fisrt feature point and the training of second feature point and returns mould
Type, then the certificate region obtained in above-mentioned steps 101 is inputted in certificate regression model, obtain each needing to know in certificate region
Other bar destination locations.
Wherein, the certificate regression model is obtained according to fisrt feature point and the training of second feature point, the fisrt feature
Point can be entry initial position characteristic point, and the second feature point can be entry physical location characteristic point.Due to the version of certificate
Formula is fixed, i.e. therefore entry initial position, it is known that can utilize the certificate regression model, according to entry initial position, constantly
Iteration, you can obtain the physical location of entry in certificate region, is completed to needing the positioning of identification entry in certificate.And certificate figure
The standardization degree of picture will not produce influence to the initial position of entry in certificate, therefore carry out bar by certificate regression model
When mesh is positioned, it is possible to increase the accuracy rate of positioning.
In summary, the localization method of certificate entry provided in an embodiment of the present invention, is carried out using certificate regression model
It is the initial position based on entry in certificate when entry is positioned, returns the physical location for obtaining each entry.Due in certificate
Entry initial position is fixed and the standardized degree of certificate image does not interfere with the initial position of entry in certificate, therefore passes through card
Part regression model is positioned to entry in certificate, and the accuracy rate for enabling to entry to position is improved.
Fig. 2-1 is the step flow chart of the localization method of another certificate entry provided in an embodiment of the present invention, such as Fig. 2-
Shown in 1, this method can include:
Step 201, acquisition certificate region.
Specifically, Fig. 2-2 is a kind of certificate area obtaining method flow chart of steps provided in an embodiment of the present invention, such as scheme
Shown in 2-2, step 201 can include:
Step 2011, feature based algorithm, calculate the characteristic set of each sample.
Sample in the embodiment of the present invention includes positive sample and negative sample.Wherein, the positive sample can be certificate region
Image, the negative sample can be the image not comprising certificate region, or include the image in part certificate region.Obtaining
Can be that the certificate image for including certificate region is gathered from actual scene, by manually marking certificate when sample data
Region, obtains certificate region and is used as positive sample.By way of artificial screening, a number of non-certificate region conduct is obtained
Negative sample.Certificate image is gathered from actual scene, the sample got is truer, the card obtained using the sample training
Part detection model, the accuracy rate when obtaining certificate region is higher.The embodiment of the present invention is in sample, positive sample and negative sample
This quantitative proportion is not limited, and both quantitative proportions can be set according to actual needs.It is preferred that, both ratio of number
Can be 1: 5.
Wherein, can be the direct acquisition non-certificate administrative division map consistent with positive sample size when negative sample is obtained
Size as not considering as negative sample or negative sample, after all negative samples are got, passes through normalization
Processing, the size of all negative samples and positive sample is sized to consistent.
The certificate detection model obtained using sample training, can be in certificate image, soon by the certificate detection model
Speed accurately determines certificate region.It is determined that when certificate region, any in certificate image can choose and certificate region
Whether image-region of the same size, then judge the image-region as certificate region, through excessive using the certificate detection model
It is secondary to search for obtain certificate region.
Accordingly, it is determined that certificate region can the abstract image-region classification problem for feature based, that is, be exactly, it is any to choose
The image-region of one certificate area size, classifies to the graphics field, and it is to belong to certificate area to judge the image-region
Domain, is also non-certificate region.
In the embodiment of the present invention, manifold characteristic value of the sample can be included in the characteristic set of each sample,
Various features value can be included in namely one characteristic set.Example, the feature of the sample can be brightness, gray scale or
Gray scale difference etc., it is corresponding can using brightness value, gray value or gray scale difference value as sample various features value.
Example, haar features are based in the embodiment of the present invention, by taking the gray scale difference feature of sample as an example, each sample are calculated
This characteristic set, to be schematically illustrated.Wherein, gray scale difference feature can be divided into above and below gray scale difference, left and right gray scale difference with
And 45 degree of angle gray scale differences, three kinds of features, calculate respectively the characteristic value of gray scale difference up and down of each sample, left and right gray scale difference characteristic value with
And 45 degree of angle gray scale difference characteristic values, obtain the characteristic set of each sample.Example, it can be calculated using integration nomography
To the characteristic value of each sample.
So, a sample has corresponding three kinds of characteristic values, and these three characteristic values may be constructed the feature of a sample
Set.In practical application, the corresponding characteristic value of other features of sample can also be included in characteristic set, for example, it is also possible to wrap
Local binary patterns (Local Binary Pattern, LBP) corresponding characteristic value of feature, the embodiment of the present invention containing sample
This is not construed as limiting.Simultaneously for the characteristic value number included in a characteristic set, it can be set according to actual needs
Put, the embodiment of the present invention is not also limited this.
By taking 10 samples as an example, the process for building certificate detection model is illustrated.Wherein, positive sample is X1, X2,
X3, X7, X8, X9, negative sample are X4, X5, X6, X10.
Calculate respectively in 10 samples, the characteristic value of gray scale difference up and down of each sample, left and right gray scale difference characteristic value and
45 degree of angle gray scale difference characteristic values, obtain the characteristic set of each sample.The characteristic set of each sample includes gray scale difference up and down
Characteristic value, left and right gray scale difference characteristic value and 45 degree of angle gray scale difference characteristic values.It is specific as follows:X1 characteristic set can be
(X1a, X1b, X1c), X2 characteristic set can be (X2a, X2b, X2c), X3 characteristic set can for (X3a, X3b,
X3c), X4 characteristic set can be (X4a, X4b, X4c), and X5 characteristic set can be (X5a, X5b, X5c), X6 spy
It can be (X6a, X6b, X6c) to collect conjunction, and X7 characteristic set can be (X7a, X7b, X7c), and X8 characteristic set can be
(X8a, X8b, X8c), X9 characteristic set can be (X9a, X9b, X9c), X10 characteristic set can for (X10a,
X10b, X10c).
Step 2012, the characteristic set using each sample, build at least one strong classifier.
Example, when building strong classifier, can using adaptive enhancing (Adaptive Boosting,
Adaboost) algorithm.The Adaboost algorithm is a kind of iterative algorithm, and its core concept is for the training of same training set
Go out multiple different Weak Classifiers, then these weak classifier sets are got up, constitute a strong classifier.Adaboost is calculated
Method is during each round repetitive exercise, and the sample of previous Weak Classifier misclassification can be strengthened, all samples after weighting
Next Weak Classifier can be used to train.Meanwhile, a new Weak Classifier is added in each round, until reaching certain
Individual predetermined sufficiently small error rate reaches preassigned maximum iteration.
Due in the characteristic set of a sample, at least one characteristic value can be included, therefore utilize the feature of each sample
Set, can obtain at least one strong classifier.For example, using the characteristic value of gray scale difference up and down of each sample as training set, can
To build a strong classifier, the left and right gray scale difference characteristic value using each sample can also build one strong point as training set
Class device, 45 degree of angle gray scale difference characteristic values using each sample can also build a strong classifier, either as training set
On the basis of using the characteristic value of gray scale difference up and down of each sample as training set, increase new characteristic value, be used as new training
Collection, rebuilds a strong classifier.The number of specific strong classifier, the characteristic value that can be included in characteristic set
Count to determine.Exemplified by having three kinds of extraordinary characteristic values in characteristic set in above-mentioned steps, using in each sample feature set
Three kinds of feature description values, can build at least three strong classifiers.
Example, with X1, X2, X3, X4, X5, X6, X7, X8, X9, the X10 characteristic value of gray scale difference up and down is used as training set
Exemplified by, the process for building a strong classifier is illustrated.Assuming that the corresponding gray scale difference characteristic value point up and down of 10 samples
It is not:X1a=1, X2a=1, X3a=1, X3a=1, X3a=4, X3a=5, X3a=6, X3a=7, X3a=8, X10a=9,
10 characteristic values can be divided into two classes, and a class is that " 1 " represents positive sample, i.e. certificate region, another kind of to represent negative sample for " -1 "
This, i.e., non-certificate region.
Specifically, a strong classifier can be built by following steps:
Step A, each gray scale difference characteristic value up and down of initialization weights distribution.
If N number of sample, then identical weights are set to each training sample:1/N.For above-mentioned 10 spies
When value indicative initialization weights distribution, identical weights can be set to each characteristic value:1/10, make each sample
Weights Wmi=W1i=1/N=0.1, wherein, N represents number of samples, N=10, i=1,2 ..., 10, then respectively for m
Represent iterations.
Step B, utilize it is each up and down gray scale difference characteristic value training obtain Weak Classifier.
In the embodiment of the present invention, by being iterated training to sample, multiple Weak Classifiers can be obtained, are then utilized
The plurality of Weak Classifier builds strong classifier, specifically, first to 10 samples in training set, first time repetitive exercise is carried out,
First Weak Classifier is obtained, when training is iterated, multiple threshold values can be determined according to characteristic value, the threshold value can
As class node, to classify to characteristic value, then according to the accuracy of each threshold classification, Weak Classifier is determined.Most
The error rate of the Weak Classifier is calculated afterwards, and using the error rate as the coefficient of the Weak Classifier, the coefficient represents the Weak Classifier
The shared weight in strong classifier.After a wheel training is completed, the weights of each sample in training set can be carried out
Update, when weight is updated, can improve not by the weights of the sample of Accurate classification in epicycle training, it is accurate to reduce
The weights of the sample of classification.Then the repetitive exercise of a new round is carried out using the training set after updating, new weak typing is obtained
Device., then can be according to the plurality of weak typing so after many wheel repetitive exercises of progress, it is possible to obtain multiple Weak Classifiers
Device constitutes a strong classifier.
Specifically, training process is as follows:
D is distributed using with weights1Training dataset, carry out first round repetitive exercise, process is as follows:
Carried out for above-mentioned sample characteristics in first round repetitive exercise, epicycle iteration, the weights of each sample characteristics
It is distributed as D1, the weights of each sample characteristics are 0.1.In training, first.According to above-mentioned sample characteristics, threshold is determined
Value;Then, using threshold value as characteristic value class node, above-mentioned sample characteristics is classified;Finally, according to classification results,
Determine Weak Classifier.
Specifically, when threshold value v takes 2.5, classifying using v=2.5 as class node to sample, will be greater than 2.5
Sample point enters " -1 " class, and the sample less than 2.5 point is entered into " 1 " class.Due to 012 respectively less than 2.5, so being in " 1 " class
In, and " 0,1,2 " corresponding class itself are " 1 " to sample.Thus may determine that " 0,1,2 " are correctly classified sample.Due to sample
" 3,4,5,6,7,8,9 " are all higher than 2.5, so be in " -1 " class, and sample " 3,4,5,9 " corresponding class itself for " -
1 ", " 6,7,8 " corresponding class itself are " 1 " to sample.Thus may determine that sample " 3,4,5,9 " are correctly classified, sample " 6,
7,8 " by mistake classification.
Therefore, when threshold value takes 2.5, error rate is that 3*0.1=0.3 (takes -1, then sample when 1, x > 2.5 are taken during x < 2.5
" 6,7,8 " misclassifications, 0.3) error rate is.
When threshold value v takes 5.5, sample is classified using v=5.5 as class node, 5.5 sample point is will be greater than
Enter " -1 " class, the sample less than 5.5 point is entered into " 1 " class.Due to sample, " 0,1,2,3,4,5 " are respectively less than 5.5, so being in
In " 1 " class, and " 0,1,2 " corresponding class itself are " 1 " to sample, and " 3,4,5 " corresponding class itself are " -1 " to sample.Therefore can
To determine sample, " 0,1,2 " are correctly classified, and " 3,4,5 " by mistake classification for sample.Due to sample, " 6,7,8,9 " are all higher than
5.5, so be in " -1 " class, and " 6,7,8 " corresponding class itself are " 1 " to sample, and the corresponding class of sample " 9 " itself is
“-1”.Thus may determine that sample " 9 " is correctly classified, " 6,7,8 " by mistake classification for sample.
Therefore, when threshold value v takes 5.5, the sample point that will be greater than 5.5 enters " -1 " class, and the sample less than 5.5 point is entered into " 1 " class
When, error rate is 6*0.1=0.6.If threshold value v takes 5.5, the sample less than 5.5 point is entered into " -1 " class, 5.5 sample is will be greater than
When one's duty enters " 1 " class, then " 0,1,2,9 " be 0.4 by mistake classification, now error rate to sample.Therefore, threshold value 5.5 is corresponding most
Small error rate is 0.4.
When threshold value v takes 8.5, sample is classified using v=8.5 as class node, 8.5 sample point is will be greater than
Enter " -1 " class, the sample less than 8.5 point is entered into " 1 " class.Then " 3,4,5 " by mistake classification, the corresponding error of threshold value 8.5 for sample
Rate is 0.3
No matter threshold value v takes 2.5, or 8.5, must have 3 sample characteristics by misclassification, therefore can be in threshold value 2.5
With optional one in threshold value 8.5, determine Weak Classifier.It is 2.5 that threshold value is taken in the embodiment of the present invention, constitutes first weak typing
Device G1(x):
Then calculate by G1(x) the weights sum of misclassification sample, G is defined as by the weights sum1(x) on training set
Error rate e1, e1=P (G1(xi) ≠ yi)=3*0.1=0.3.
It is determined that calculating G1(x) error in classification rate e1Afterwards, according to error in classification rate e1Calculate G1(x) coefficient, should
Coefficient represents G1(x) weight shared in strong classifier.Following is a in coefficient formulas, the coefficient formulasmTable
Give the impression of weakness the coefficient of grader, emThe error in classification rate of Weak Classifier is represented, m represents the number of times of current iteration, log (*) expressions pair
Number function.G can be calculated by the coefficient formulas1(x) coefficient.
By e1=0.3 substitutes into above-mentioned formula, can obtain:a1=0.4236
Then, the weights distribution to training data is updated, for next round iteration., can when weight is updated
To be updated by following formula:
Dm+l=(WM+1,1, WM+l, 2... WM+l, i…WM+l, N)
Wherein, Dm+1Represent the weights distribution of m+1 wheels, WM+1, NRepresent the weights of sample N in m+1 wheels, WmiRepresent the
Sample i weights, Z in m wheelsmStandardizing factor is represented, D can be causedm+1As a probability distribution, exp (*) represents index
Function, yiRepresent m wheel in the corresponding classifications of sample i, example, the category can be 1 can also be -1.
Weights distribution D after renewal2=(0.0715,0.0715,0.0715,0.0715,0.0715,0.0715,
0.1666,0.1666,0.1666,0.0715).
By first round iteration, the classification function f of Weak Classifier is obtained1(x)=a1*G1(x)=0.4236G1(x)。
The Weak Classifier is sign (f1(x)), wherein sign (*) presentation class function, it is realized the symbol of classification function
Number isolate out.
D is distributed using with weights2Training dataset, carry out second and take turns repetitive exercise, process is as follows:
In epicycle iteration, the weights of each sample characteristics are distributed as D2, the weights of each sample characteristics are
0.0715,0.0715,0.0715,0.0715,0.0715,0.0715,0.1666,0.1666,0.1666,0.0715.Work as threshold value
When v takes 2.5, sample is classified using v=2.5 as class node, the sample point that will be greater than 2.5 enters " -1 " class, will be less than
2.5 sample point enters " 1 " class.Then " 6,7,8 " are sample by mistake classification, error in classification rate now:3*0.1666=
0.4998。
When threshold value v takes 5.5, sample is classified using v=5.5 as class node, 5.5 sample point is will be greater than
Enter " 1 " class, the sample less than 5.5 point is entered into " -1 " class.Then " 0,1,2,9 " by mistake classification, error in classification rate now for sample
For:4*0.0715=0.286.
When threshold value v takes 8.5, sample is classified using v=8.5 as class node, 8.5 sample point is will be greater than
Enter " -1 " class, the sample less than 8.5 point is entered into " 1 " class.Then " 3,4,5 " by mistake classification, error in classification rate now for sample
For:3*0.0715=0.2143.
As can be seen that classification error rate is minimum when threshold value v takes 8.5, therefore obtains second basic classification device and be:
According to error in classification rate calculation formula, G can be obtained2(x) the error rate e on training dataset2=P (G2
(xi) ≠ yi)=0.0715*3=0.2143.
By e2Above-mentioned coefficient formulas is substituted into, can be obtained, a2=0.6496
Then, the weights distribution to training data is updated, and obtains D3=0.0455,0.0455,0.0455,
0.1667,0.1667,0.01667,0.1060,0.1060,0.1060,0.0455.
By the second wheel iteration, the classification function f of Weak Classifier is obtained2(x)=a1*G1(x)+a2*G2(x)=
0.4236G1(x)+0.6496G2(x) Weak Classifier, is obtained for sign (f2(x))。
D is distributed using with weights3Training dataset study, carry out third round repetitive exercise, process is as follows:
In epicycle iteration, the weights of each sample characteristics are distributed as D3, the weights of each sample characteristics are
0.0455,0.0455,0.0455,0.1667,0.1667,0.1667,0.1060,0.1060,0.1060,0.0455)
When threshold value v takes 2.5, sample is classified using v=2.5 as class node, 2.5 sample point is will be greater than
Enter " -1 " class, the sample less than 2.5 point is entered into " 1 " class.Then " 6,7,8 " by mistake classification, error in classification rate now for sample
For:3*0.1060=XXX.
When threshold value v takes 5.5, sample is classified using v=5.5 as class node, 5.5 sample point is will be greater than
Enter " 1 " class, the sample less than 5.5 point is entered into " -1 " class.Then " 0,1,2,9 " by mistake classification, error in classification rate now for sample
For:3*0.0455+0.0715=XXX.
When threshold value v takes 8.5, sample is classified using v=8.5 as class node, 8.5 sample point is will be greater than
Enter " -1 " class, the sample less than 8.5 point is entered into " 1 " class.Then " 3,4,5 " by mistake classification, error in classification rate now for sample
For:3*0.1667=XXX.
As can be seen that classification error rate is minimum when threshold value v takes 5.5, therefore obtains the 3rd basic classification device and be:
According to error in classification rate calculation formula, G can be obtained3(x) the error rate e on training dataset3=P (G3
(xi) ≠ yi)=0.0455*4=0.1820.
By e3Above-mentioned coefficient formulas is substituted into, can be obtained, a3=0.7514.
By third round iteration, the classification function f of Weak Classifier is obtained3(x)=a1*G1(x)+a2*G2(x) +a3*G3
(x), Weak Classifier is sign (f3(x)).Due to sign (f3(x)) there is 0 misclassified gene on training set, therefore can terminate
Repetitive exercise process.
According to three Weak Classifiers obtained in above-mentioned three-wheel iteration, a strong classifier G (X) can be built.
G (x)=sign [f3(x)]
=sign [a1*G1(x)+a2*G2(x)+a3*G3(x)]
=sign [0.4236G1(x)+0.6496G2(x)+0.7514G3(x)]
Further, it is possible to use other characteristic values of each sample repeat the above steps, then build as training set
One strong classifier.By that analogy, at least one strong point is built using at least one of each sample feature set characteristic value
Class device.
Step 2013, at least one strong classifier according to build cascade classifier, and the cascade classifier is determined
For the certificate detection model.
Example, cascade classifier can be made up of multiple strong classifiers., can be according to above-mentioned in the embodiment of the present invention
The multiple strong classifiers built in step, build a cascade classifier.The cascade classifier is defined as certificate detection mould
Type.
Typically, the false drop rate for the strong classifier that training is obtained can be relatively low, but verification and measurement ratio is not very high, if it is desired that
Strong classifier has higher verification and measurement ratio, then its false drop rate can accordingly increase.Wherein, verification and measurement ratio refers to accurately be detected
To certificate number of regions and certificate image in contained certificate number of regions ratio.False drop rate be by flase drop be certificate region
Non- certificate number of regions and the ratio of all detected window numbers in certificate image.Therefore, if directly using strong classifier conduct
Certificate detection model, it is possible that the situation that certificate region is not detected accurately, certificate detection efficiency is relatively low.Therefore, may be used
To build a cascade classifier using multiple strong classifiers, certificate detection model is used as.By setting cascade classifier
Parameter, can make cascade classifier false drop rate while with compared with high detection rate relatively low.So utilizing certificate detection model
When determining certificate region, detection efficiency is higher.
Step 2014, acquisition target image.
The target image can be, when detection to certificate, the certificate image got.Example, if will
The identity card of user " Zhang San " is identified, now, the image for including " Zhang San " identity card that image capture device is collected
As target image.
Step 2015, according to the target image, determine certificate region using default certificate detection model.
After getting target image, it is possible to use the certificate detection model built in above-mentioned steps 2013, the mesh is determined
Certificate region in logo image.Also include other regions comprising certificate region in the image collected due to image capture device,
For example, " Zhang San " identity card region had both been contained in the target image comprising " Zhang San " identity card, while can also include background area
Domain, it is therefore desirable to determined identity card region using certificate detection model, is easy to be identified in subsequent step, improves
The accuracy rate of identification.
Specifically, target image can be inputted in certificate detection model, it is determined that when certificate region, Ke Yitong
Detection window is crossed, it is any in the target image to choose and the image-region of certificate area size always, by the certificate detection model
Whether be certificate region, travel through whole target image, determine certificate region if judging the image-region.Passing through detection window time
When going through target image, it can be exactly with random ergodic, in the target image, randomly select the area consistent with certificate area size
Domain, is judged, untill determining certificate region.Top-down traversal mode can also be taken, that is, is exactly, from mesh
The top of logo image starts, according to from top to bottom, and order from left to right is traveled through to target image, when getting
Behind certificate region, stop traversal, can so reduce follow-up unnecessary traversing operation, improve the efficiency for determining certificate region.
In the embodiment of the present invention, by building certificate detection model, then determined using the certificate detection model from target image
Certificate detection model, fast can accurately determine certificate region.
It should be noted that in the step of above-mentioned acquisition certificate region, the order of each step is not unique, for example, step
Rapid 2014 can be located at before step 2011, and the embodiment of the present invention is not limited this.At the same time it can also by other means
Obtain and this is not construed as limiting in certificate region, the embodiment of the present invention.
Step 202, the certificate regression model obtained according to fisrt feature point and the training of second feature point.
The certificate regression model is obtained according to fisrt feature point and the training of second feature point, and the fisrt feature point can
Think entry initial position characteristic point, the second feature point can be entry physical location characteristic point.Certificate regression model, can
To be positioned to each entry in certificate region, the task of entry positioning is exactly for a given certificate region, really
Make the physical location of entry in the certificate region.In daily life, the distribution of the entry of certificate be nearly all it is fixed, i.e., just
It is that the initial position of entry is fixed in each certificate, due to deviation present in manufacturing process, it is each in each certificate
The physical location of individual entry can be equipped with certain deviation with the initial bit of default entry.Therefore, entry positioning can be changed
For the process returned by entry initial position, the physical location to entry.
Optionally, Fig. 2-3 is a kind of method and step flow for obtaining certificate regression model provided in an embodiment of the present invention
Figure, as Figure 2-3, step 202 can include:
Step 2021, first eigenvector determined according to the fisrt feature point, is determined according to the second feature point
Two characteristic vectors.
Example, the first of each sample by way of manually being marked to certificate area sample, can be determined
Characteristic point and second feature point.When mark, the initial position that can mark entry in each sample is special as first
Levy a little, the physical location of entry is used as second feature point.Do not limited for the specific notation methods embodiment of the present invention, example
Such as, the upper left position of entry can be marked, the lower-left Angle Position of entry can also be marked, or by the upper left corner of entry and
The lower left corner is all marked.Fig. 2-4 is a kind of fisrt feature point schematic diagram provided in an embodiment of the present invention, as illustrated in figs. 2-4,
Entry " name ", " sex ", " nationality ", " address ", the initial position of " date of birth " and " neck card date " are marked
Note, it is the fisrt feature point of " the XX cards " to show the black color dots in the fisrt feature point in the certificate region, Fig. 2-4.Fig. 2-
5 be a kind of second feature point schematic diagram provided in an embodiment of the present invention, as seen in figs. 2-5, entry " name ", " sex ",
The physical location in " nationality ", " address ", " date of birth " and " neck card date " is marked, and shows the certificate region
Second feature point, the Grey Point in Fig. 2-5 is the second feature point of " the XX cards ".
It is then possible to determined according to the Feature Descriptor of setting the fisrt feature of the fisrt feature point of each sample to
Amount and the second feature of second feature point vector.This feature, which describes son, to change (Scale- for scale invariant feature
Invariant feature transform, SIFT) feature or histograms of oriented gradients (Histogram of Oriented
Gradient, HOG) feature, the embodiment of the present invention is not construed as limiting to this.
Specifically, the embodiment of the present invention is by taking SIFT feature as an example, schematically illustrated.SIFT feature describes son can be with
Gaussian image gradient in characteristic point neighbors around is counted, the characteristic vector of characteristic point is obtained.It is a three-dimensional battle array
Row, can obtain a vector, the vector is characteristic vector by carrying out arrangement according to certain rules to cubical array.Due to
Feature Descriptor is relevant with the yardstick where characteristic point, therefore, it is determined that characteristic point characteristic vector when, can be in feature
Carried out in the corresponding Gaussian image of point., can be first by the attached of characteristic point when determining characteristic vector according to SIFT feature description
Neighbour domain is divided into 16 sub-regions.
Centered on characteristic point, the position and direction of image gradient in characteristic point neighbors around are rotated into a deflection
θ.In characteristic point neighbors around after the position and direction rotation of image gradient, the gradient in 8 directions in per sub-regions is calculated
Direction histogram, draws the accumulated value of each gradient direction, forms a seed point.Now, the gradient direction per sub-regions
Histogram is divided into 8 direction scopes by 0 degree~360 degree, and each scope is 45 degree.So, 8 sides are had per sub-regions
To gradient intensity information.Due to there are 16 sub-regions, so, 16*8=128 data are had, 128 dimensions are ultimately formed
SIFT feature vector.
After characteristic vector is determined by SIFT feature, Gauss weighting processing can also be carried out to characteristic vector.Show
Example, weighting processing can be carried out using standard gaussian function.Processing is weighted to characteristic vector can prevent characteristic point
The minor variations of position bring very big change to characteristic vector, and assign less weight to the point away from characteristic point, protect
The accuracy of characteristic vector calculating is demonstrate,proved.
Step 2022, using the first eigenvector and the default characteristics algorithm of second feature vector training, demonstrate,proved
Part regression model.
Example, it is possible to use supervision descent method (ervised Descent Method, SDM) algorithm, according to each sample
This first eigenvector and second feature vector, iteration training, computation model parameter obtains certificate regression model.
By taking sample of 10 certificate region pictures to build certificate regression model as an example, illustrate.
With { diCertificate region pictures are represented, wherein i represents the picture number that certificate region picture is concentrated.The pictures
10 certificate region pictures, i.e. i=1,2,3...10 can be included in conjunction.With x*Represent certificate region picture i second feature
Point.With x0Represent certificate region picture i fisrt feature point.
WithP feature point coordinates in the picture of certificate region is represented, h is the non-thread at each characteristic point
Property feature extraction function, by taking SIFT feature as an example, that is, each characteristic point will extract the SIFT feature of 128 dimensions, i.e.It is x by the second feature point manually demarcated*, in x*The SIFT feature that place is extracted can be designated as Φ*=h
(d(x*)).Thus, entry positioning can be regarded as, by characteristic point from initial value x0Converge to minimum value x*, ask so that descending array function most
The difference Δ x of small characteristic point:
f(x0+ Δ x)=| | h (d (x0+Δx))-Φ*||2
Example, can be by determining the coefficient of repetitive exercise, by characteristic point from initial value x0Converge to minimum value x*。
With x in the picture of certificate region0The SIFT feature Φ at place0As input feature vector, pass through object function:
The coefficients R of kth time repetitive exercise can be determinedkAnd bk, wherein, RkAnd bkThe path of each iteration is represented,
Independent variable value when argmin (*) representative functions value is minimum, according to the parameter of each repetitive exercise, determines that certificate is returned
Return model.
Step 203, according to the certificate region, determined using default certificate regression model in the certificate region
Bar destination locations need to each be recognized.
When actually identification, certificate region can be input in the certificate regression model, due to the format of certificate
It is fixed, i.e. known to entry initial position.Therefore the certificate regression model can be utilized, according to entry initial position, constantly repeatedly
In generation, you can obtain the physical location of entry in certificate region, complete to needing the positioning of identification entry in certificate.Due to certificate figure
The standardization degree of picture will not produce influence to the initial position of entry in certificate, therefore carry out bar by certificate regression model
When mesh is positioned, it is possible to increase the accuracy rate of positioning.
Meanwhile, the localization method of certificate entry provided in an embodiment of the present invention needs identification entry to enter in certificate region
During row positioning, it can be completed by certificate regression model, compared to the positioning for needing multiple steps to complete in the prior art
Method, simplifies the operation of entry positioning, improves the efficiency of entry positioning.
In summary, the localization method of certificate entry provided in an embodiment of the present invention, by building certificate detection model,
Certificate detection model is determined from target image using the certificate detection model, certificate region is fast accurately defined,
Then, entry positioning is carried out using certificate regression model, bar destination locations need to each be recognized by determining in certificate region.Profit
It is the initial position based on entry in certificate when carrying out entry positioning with certificate regression model, returns the reality for obtaining each entry
Border position.Due in certificate entry initial position fix and certificate image standardized degree do not interfere with entry in certificate just
Beginning position, therefore entry in certificate is positioned by certificate regression model, the accuracy rate for enabling to entry to position is carried
It is high.
Fig. 3 is a kind of block diagram of the positioner of certificate entry provided in an embodiment of the present invention, as shown in figure 3, the device
30 can include:
Acquisition module 301, for obtaining certificate region.
Determining module 302, for according to the certificate region, the certificate to be determined using default certificate regression model
Bar destination locations need to be each recognized in region;The certificate regression model is according to fisrt feature point and second feature point
Training is obtained;The fisrt feature point is entry initial position characteristic point, and the second feature point is entry physical location feature
Point.
In summary, the positioner of certificate entry provided in an embodiment of the present invention, is carried out using certificate regression model
It is the initial position based on entry in certificate when entry is positioned, returns the physical location for obtaining each entry.Due in certificate
Entry initial position is fixed and the standardized degree of certificate image does not interfere with the initial position of entry in certificate, therefore passes through card
Part regression model is positioned to entry in certificate, and the accuracy rate for enabling to entry to position is improved.
Fig. 4-1 is the block diagram of the positioner of another certificate entry provided in an embodiment of the present invention, as shown in figure 4, should
Device 40 can include:
Acquisition module 401, for obtaining certificate region.
Determining module 402, for according to the certificate region, the certificate to be determined using default certificate regression model
Bar destination locations need to be each recognized in region;The certificate regression model is according to fisrt feature point and second feature point
Training is obtained;The fisrt feature point is entry initial position characteristic point, and the second feature point is entry physical location feature
Point.
Fig. 4-2 is the block diagram of the positioner of another certificate entry provided in an embodiment of the present invention, as shown in the Fig. 4-2,
The device 40 can include:Acquisition module 401, determining module 402, training module 403,
Acquisition module 401, for obtaining certificate region.
Determining module 402, for according to the certificate region, the certificate to be determined using default certificate regression model
Bar destination locations need to be each recognized in region;The certificate regression model is according to fisrt feature point and second feature point
Training is obtained;The fisrt feature point is entry initial position characteristic point, and the second feature point is entry physical location feature
Point.
Training module 403, mould is returned for obtaining the certificate according to fisrt feature point and the training of second feature point
Type.
Optionally, above-mentioned training module 403, can include:
First determination sub-module, it is special according to described second for determining first eigenvector according to the fisrt feature point
Levy a determination second feature vector;
First training submodule, for utilizing the first eigenvector and the default feature of second feature vector training
Algorithm, obtains certificate regression model.
Optionally, above-mentioned acquisition module 401, can include:
Acquisition submodule, for obtaining target image;
Second determination sub-module, for according to the target image, certificate area to be determined using default certificate detection model
Domain;The certificate detection model is obtained according to sample training;Wherein, the sample includes positive sample and negative sample, described
Positive sample is the image for including certificate region, and the negative sample is the image not comprising certificate region.
Optionally, above-mentioned acquisition module 401 can also include:
Calculating sub module, for feature based algorithm, calculates the characteristic set of each sample;Wherein, the feature
The corresponding at least one characteristic value of each sample is included in set;
Submodule is built, for the characteristic set using each sample, at least one strong classifier is built;
3rd determination sub-module, for building cascade classifier according at least one described strong classifier, by the cascade
Grader is defined as the certificate detection model.
In summary, the positioner of certificate entry provided in an embodiment of the present invention, is carried out using certificate regression model
It is the initial position based on entry in certificate when entry is positioned, returns the physical location for obtaining each entry.Due in certificate
Entry initial position is fixed and the standardized degree of certificate image does not interfere with the initial position of entry in certificate, therefore passes through card
Part regression model is positioned to entry in certificate, and the accuracy rate for enabling to entry to position is improved.
Each embodiment in this specification is described by the way of progressive, and what each embodiment was stressed is
With the difference of other embodiment, between each embodiment identical similar part mutually referring to.
It would have readily occurred to a person skilled in the art that be:Any combination application of each above-mentioned embodiment is all feasible, therefore
Any combination between each above-mentioned embodiment is all embodiment of the present invention, but this specification exists as space is limited,
This is not just detailed one by one.
In the specification that this place is provided, numerous specific details are set forth.It is to be appreciated, however, that the implementation of the present invention
Example can be put into practice in the case of these no details.In some instances, known method, knot is not been shown in detail
Structure and technology, so as not to obscure the understanding of this description.
Similarly, it will be appreciated that in order to simplify the present invention and help to understand one or more of each inventive aspect,
Above in the description of the exemplary embodiment of the present invention, each feature of the invention is grouped together into single reality sometimes
Apply in example, figure or descriptions thereof.However, the method for the disclosure should be construed to reflect following intention:Want
Seek the application claims features more more than the feature being expressly recited in each claim of protection.More precisely, such as
As claims reflect, inventive aspect is all features less than single embodiment disclosed above.Therefore,
Thus the claims for following embodiment are expressly incorporated in the embodiment, wherein each claim sheet
Body is all as the separate embodiments of the present invention.
Those skilled in the art, which are appreciated that, to carry out adaptivity to the module in the equipment in embodiment
Ground changes and they is arranged in one or more equipment different from the embodiment.Can be the module in embodiment
Or unit or component are combined into a module or unit or component, and multiple submodule or son can be divided into addition
Unit or sub-component., can be with addition at least some in such feature and/or process or unit exclude each other
Using any combinations to all features disclosed in this specification (including adjoint claim, summary and accompanying drawing) and such as
Any method of the displosure or all processes or unit of equipment are combined.Unless expressly stated otherwise, this specification
Each feature disclosed in (including adjoint claim, summary and accompanying drawing) can or similar purpose identical, equivalent by offer
Alternative features replace.
Although in addition, it will be appreciated by those of skill in the art that some embodiments described herein include other embodiments
In included some features rather than further feature, but the combination of the feature of be the same as Example does not mean in the present invention
Within the scope of and form different embodiments.For example, in detail in the claims, embodiment claimed it is any
One of mode can use in any combination.
It should be noted that the present invention will be described rather than limits the invention for above-described embodiment, and this
Art personnel can design alternative embodiment without departing from the scope of the appended claims.In claim
In, any reference symbol between bracket should not be configured to limitations on claims.Word "comprising" is not excluded for depositing
In element or step not listed in the claims.Word "a" or "an" before element does not exclude the presence of multiple
Such element.The present invention can be by means of including the hardware of some different elements and by means of properly programmed calculating
Machine is realized.In if the unit claim of equipment for drying is listed, several in these devices can be by same
Hardware branch is embodied.The use of word first, second, and third does not indicate that any order.Can be by these word solutions
It is interpreted as title.
Claims (10)
1. a kind of localization method of certificate entry, it is characterised in that methods described includes:
Obtain certificate region;
According to the certificate region, determine each recognize entry in the certificate region using default certificate regression model
Position;The certificate regression model is obtained according to fisrt feature point and the training of second feature point;The fisrt feature point
For entry initial position characteristic point, the second feature point is entry physical location characteristic point.
2. according to the method described in claim 1, it is characterised in that methods described also includes:
The certificate regression model is obtained according to fisrt feature point and the training of second feature point.
3. method according to claim 2, it is characterised in that described to be trained according to fisrt feature point and second feature point
The step of obtaining the certificate regression model, including:
First eigenvector is determined according to the fisrt feature point, second feature vector is determined according to the second feature point;
Using the first eigenvector and the default characteristics algorithm of second feature vector training, certificate regression model is obtained.
4. according to the method described in claim 1, it is characterised in that the step of the acquisition certificate region, including:
Obtain target image;
According to the target image, certificate region is determined using default certificate detection model;The certificate detection model is root
Obtained according to sample training;Wherein, the sample includes positive sample and negative sample, and the positive sample is the figure for including certificate region
Picture, the negative sample is the image not comprising certificate region.
5. method according to claim 4, it is characterised in that described according to the target image, utilize default card
Before the step of part detection model determines certificate region, methods described also includes:
Feature based algorithm, calculates the characteristic set of each sample;Wherein, each sample pair is included in the characteristic set
At least one characteristic value answered;
Using the characteristic set of each sample, at least one strong classifier is built;
Cascade classifier is built according at least one described strong classifier, the cascade classifier is defined as into the certificate detects
Model.
6. a kind of positioner of certificate entry, it is characterised in that described device includes:
Acquisition module, for obtaining certificate region;
Determining module, for according to the certificate region, being determined using default certificate regression model in the certificate region
Bar destination locations need to each be recognized;The certificate regression model is obtained according to fisrt feature point and the training of second feature point;
The fisrt feature point is entry initial position characteristic point, and the second feature point is entry physical location characteristic point.
7. device according to claim 6, it is characterised in that described device also includes:
Training module, for obtaining the certificate regression model according to fisrt feature point and the training of second feature point.
8. device according to claim 7, it is characterised in that the training module, including:
First determination sub-module, for determining first eigenvector according to the fisrt feature point, according to the second feature point
Determine second feature vector;
First training submodule, for presetting characteristics algorithm using the first eigenvector and second feature vector training,
Obtain certificate regression model.
9. device according to claim 6, it is characterised in that the acquisition module, including:
Acquisition submodule, for obtaining target image;
Second determination sub-module, for according to the target image, certificate region to be determined using default certificate detection model;Institute
Stating certificate detection model is obtained according to sample training;Wherein, the sample includes positive sample and negative sample, the positive sample
To include the image in certificate region, the negative sample is the image not comprising certificate region.
10. device according to claim 9, it is characterised in that the acquisition module, in addition to:
Calculating sub module, for feature based algorithm, calculates the characteristic set of each sample;Wherein, the characteristic set
In include the corresponding at least one characteristic value of each sample;
Submodule is built, for the characteristic set using each sample, at least one strong classifier is built;
3rd determination sub-module, for building cascade classifier according at least one described strong classifier, by the cascade sort
Device is defined as the certificate detection model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710349083.9A CN107330429B (en) | 2017-05-17 | 2017-05-17 | Certificate item positioning method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710349083.9A CN107330429B (en) | 2017-05-17 | 2017-05-17 | Certificate item positioning method and device |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107330429A true CN107330429A (en) | 2017-11-07 |
CN107330429B CN107330429B (en) | 2021-03-09 |
Family
ID=60192861
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710349083.9A Active CN107330429B (en) | 2017-05-17 | 2017-05-17 | Certificate item positioning method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107330429B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109284757A (en) * | 2018-08-31 | 2019-01-29 | 湖南星汉数智科技有限公司 | A kind of licence plate recognition method, device, computer installation and computer readable storage medium |
WO2019218334A1 (en) * | 2018-05-18 | 2019-11-21 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for user analysis |
US11250291B2 (en) | 2018-09-04 | 2022-02-15 | Advanced New Technologies, Co., Ltd. | Information detection method, apparatus, and device |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6281928B1 (en) * | 1998-05-13 | 2001-08-28 | Chuo Hatsujo Kabushiki Kaisha | Positional detector device for a vehicular license plate |
CN101352061A (en) * | 2006-03-31 | 2009-01-21 | 西门子企业通讯有限责任两合公司 | Method and apparatus for determining the location of a mobile object |
CN102194256A (en) * | 2011-02-23 | 2011-09-21 | 上海博路信息技术有限公司 | Ticketing system based on bar code and identity card detection |
CN105447490A (en) * | 2015-11-19 | 2016-03-30 | 浙江宇视科技有限公司 | Vehicle key point detection method based on gradient regression tree and apparatus thereof |
CN105528601A (en) * | 2016-02-25 | 2016-04-27 | 华中科技大学 | Identity card image acquisition and recognition system as well as acquisition and recognition method based on contact type sensor |
CN105701488A (en) * | 2016-01-01 | 2016-06-22 | 广州恒巨信息科技有限公司 | Identity card identification method |
CN105825243A (en) * | 2015-01-07 | 2016-08-03 | 阿里巴巴集团控股有限公司 | Method and device for certificate image detection |
CN106056080A (en) * | 2016-05-30 | 2016-10-26 | 东莞市中控电子技术有限公司 | Visualized biometric information acquisition device and acquisition method |
CN106203454A (en) * | 2016-07-25 | 2016-12-07 | 重庆中科云丛科技有限公司 | The method and device that certificate format is analyzed |
-
2017
- 2017-05-17 CN CN201710349083.9A patent/CN107330429B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6281928B1 (en) * | 1998-05-13 | 2001-08-28 | Chuo Hatsujo Kabushiki Kaisha | Positional detector device for a vehicular license plate |
CN101352061A (en) * | 2006-03-31 | 2009-01-21 | 西门子企业通讯有限责任两合公司 | Method and apparatus for determining the location of a mobile object |
CN102194256A (en) * | 2011-02-23 | 2011-09-21 | 上海博路信息技术有限公司 | Ticketing system based on bar code and identity card detection |
CN105825243A (en) * | 2015-01-07 | 2016-08-03 | 阿里巴巴集团控股有限公司 | Method and device for certificate image detection |
CN105447490A (en) * | 2015-11-19 | 2016-03-30 | 浙江宇视科技有限公司 | Vehicle key point detection method based on gradient regression tree and apparatus thereof |
CN105701488A (en) * | 2016-01-01 | 2016-06-22 | 广州恒巨信息科技有限公司 | Identity card identification method |
CN105528601A (en) * | 2016-02-25 | 2016-04-27 | 华中科技大学 | Identity card image acquisition and recognition system as well as acquisition and recognition method based on contact type sensor |
CN106056080A (en) * | 2016-05-30 | 2016-10-26 | 东莞市中控电子技术有限公司 | Visualized biometric information acquisition device and acquisition method |
CN106203454A (en) * | 2016-07-25 | 2016-12-07 | 重庆中科云丛科技有限公司 | The method and device that certificate format is analyzed |
Non-Patent Citations (2)
Title |
---|
SHEN LU 等: "ID Numbers Recognition by Local Similarity Voting", 《INTERNATIONAL JOURNAL OF ADVANCED COMPUTER SCIENCE AND APPLICATIONS,SPECIAL ISSUE ON IMAGE PROCESSING AND ANALYSIS》 * |
伍凯 等: "级联回归的多姿态人脸配准", 《中国图象图形学报》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2019218334A1 (en) * | 2018-05-18 | 2019-11-21 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for user analysis |
US10990500B2 (en) | 2018-05-18 | 2021-04-27 | Beijing Didi Infinity Technology And Development Co., Ltd. | Systems and methods for user analysis |
CN109284757A (en) * | 2018-08-31 | 2019-01-29 | 湖南星汉数智科技有限公司 | A kind of licence plate recognition method, device, computer installation and computer readable storage medium |
US11250291B2 (en) | 2018-09-04 | 2022-02-15 | Advanced New Technologies, Co., Ltd. | Information detection method, apparatus, and device |
Also Published As
Publication number | Publication date |
---|---|
CN107330429B (en) | 2021-03-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Gupta et al. | Breast cancer histopathological image classification: is magnification important? | |
WO2017190574A1 (en) | Fast pedestrian detection method based on aggregation channel features | |
CN105488536B (en) | A kind of agricultural pests image-recognizing method based on multiple features depth learning technology | |
CN104484681B (en) | Hyperspectral Remote Sensing Imagery Classification method based on spatial information and integrated study | |
CN105512684B (en) | Logo automatic identifying method based on principal component analysis convolutional neural networks | |
CN107563428A (en) | Classification of Polarimetric SAR Image method based on generation confrontation network | |
CN104239902B (en) | Hyperspectral image classification method based on non local similitude and sparse coding | |
CN109271991A (en) | A kind of detection method of license plate based on deep learning | |
CN109117885B (en) | Stamp identification method based on deep learning | |
Seff et al. | Leveraging mid-level semantic boundary cues for automated lymph node detection | |
CN105512657B (en) | Character identifying method and equipment | |
CN104850822B (en) | Leaf identification method under simple background based on multi-feature fusion | |
CN109815979A (en) | A kind of weak label semantic segmentation nominal data generation method and system | |
CN104680545B (en) | There is the detection method of well-marked target in optical imagery | |
CN104573699B (en) | Trypetid recognition methods based on middle equifield intensity magnetic resonance anatomy imaging | |
CN107330429A (en) | A kind of localization method and device of certificate entry | |
CN104268552B (en) | One kind is based on the polygonal fine classification sorting technique of part | |
CN108764302A (en) | A kind of bill images sorting technique based on color characteristic and bag of words feature | |
CN108197636A (en) | A kind of paddy detection and sorting technique based on depth multiple views feature | |
CN111310628A (en) | Paper currency forming mode inspection and identification method based on paper currency printing pattern characteristics | |
CN107169496A (en) | A kind of character recognition method and device | |
CN104699781B (en) | SAR image search method based on double-deck anchor figure hash | |
CN109002463A (en) | A kind of Method for text detection based on depth measure model | |
CN107292225A (en) | A kind of face identification method | |
CN108268865A (en) | Licence plate recognition method and system under a kind of natural scene based on concatenated convolutional network |
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 |