CN106250913B - A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis - Google Patents
A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis Download PDFInfo
- Publication number
- CN106250913B CN106250913B CN201610581551.0A CN201610581551A CN106250913B CN 106250913 B CN106250913 B CN 106250913B CN 201610581551 A CN201610581551 A CN 201610581551A CN 106250913 B CN106250913 B CN 106250913B
- Authority
- CN
- China
- Prior art keywords
- sample
- classifier
- test sample
- feature
- test
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/625—License plates
Abstract
The invention discloses a kind of combining classifiers licence plate recognition methods based on local canonical correlation analysis, including 1, processing sample set, and sample set is divided into original training sample To, test sample Ts, second test sample Tp, and sample characteristics are extracted according to sample feature；2, using the sample characteristics of extraction, a variety of relatively independent single classifiers are trained with kernel function by parameter when changing training single classifier；3, using the method for class KNN, the sample set similar with test sample is found, and considers the relationship between fractional sample feature and whole classifier classification results using the method for canonical correlation analysis, obtains integrated model of classifiers so as to adjust each classifier weight；4, the final detection result of sample to be tested is codetermined by the classifier after step 3 is integrated, judges sample to be tested generic.The present invention can be self adaptive the different test sample of reply and change classifier weight, while can effectively improve classifier classification accuracy.
Description
Technical field
The present invention relates to pattern-recognition, integrated study, computer vision fields, are based especially on multimedia application technical side
To image processing techniques, specifically a kind of combining classifiers licence plate recognition method based on local canonical correlation analysis.
Background technique
License plate recognition technology can effectively reinforce supervision of the relevant department to road traffic in intelligent transportation field, drop significantly
The use of low human and material resources is the important component of modern smart city development.Main Car license recognition skill at this stage
Art, such as: being based on the licence plate recognition method of support vector machines (SVM), be based on the licence plate recognition method and base of rarefaction representation (SRC)
In the licence plate recognition method etc. of fuzzy classification.Above-mentioned licence plate recognition method can not be played sufficiently according to different sample characteristics
The maximum utility of classifier.Traditional licence plate recognition method is often because of the classification results excessively centralization, list of single classifier
One the characteristics of changing and classification results occur with contingency.The training of single classifier can depend on training sample set sheet unduly simultaneously
Body.In the case where working as lack of training samples or training sample excessively complexity, the sorter model that training obtains can not expire
The identification and detection of sufficient normal road traffic license plate.
To sum up, the concept that combining classifiers have been introduced on the basis of above-mentioned technology develops, by multiple relatively independent points
Class device codetermines the classification of sample to be tested, to improve the verification and measurement ratio of license plate.Now, the scheme of combining classifiers is for example, base
In the classification ensemble method of Adaboost (adaptive boosting), it is based on Bagging (Bootstrap
Aggregating categorizer integration method etc.).AdaBoost method is the classifier optimization method based on weak point of thought.It should
The principle that method can be learnt again the classifier that training obtains using its detection mistake classification, optimization traditional classifier classification effect
Fruit.Bagging is a kind of algorithm based on ballot selection thought.The advantage is that can carry out extensive melt to any classifier
It closes, not special requirement is defined to classifier, for the learning algorithm of similar neural network, Bagging can be instructed parallel
Practice manifold classification device, can greatly shorten the time overhead of algorithm, classification results are chosen in a vote by majority.
It can be seen that the above categorizer integration method does not take into account fractional sample feature and whole classifier classification results
Between connection, balance weight of the fractional sample feature in sample classification.To be depended on unduly because of whole classifying quality
Training sample set, the poor Weak Classifier of training classifying quality.For example, AdaBoost algorithm is easy to happen the feelings of over training
Condition can not balance contacting between whole training sample and classifier, and with the increase of training the number of iterations, its accuracy rate is anti-
And it can decline.And Bagging method is facing small sample problem since the classifier of its training depends on training sample unduly
When, especially in the case where the complicated numerous training samples of classification, classifying quality is barely satisfactory instead, can not take into account part
Sample and whole Weak Classifier play the role of in classification.
The above problem is coped with, the present invention proposes a kind of combining classifiers Car license recognition side based on local canonical correlation analysis
Method.The thought searching feature and classification results of present invention application canonical correlation analysis, the correlativity of this two groups of variables, and then really
The confidence level of fixed each single classifier, to distribute weight to each single classifier.And in order to balance fractional sample and entirety
The method that relationship between classification results utilizes class KNN, searches out K neighbour of sample to be tested, so as to close using this K
Neighbour's analysis fractional sample feature and classification results contact, at the same can be self adaptive dynamic adjustment classifier weight.Experiment
It proves that the present invention can effectively improve the accuracy rate of Car license recognition, and copes with different sample set robustness with higher.
Summary of the invention
It is an object of the invention to provide to play maximum utility of each single classifier when coping with different sample sets
A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis.Further research fractional sample feature and classification
Connection between device classification results, to effectively improve Car license recognition accuracy rate and Dynamic Programming.
In order to solve the above-mentioned technical problem, the specific technical solution that the present invention uses is as follows:
A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis, includes the following steps:
Step 1: data set is divided into training sample T by processing set of data sampleso, test sample Ts, second test sample
Tp, and the HOG feature according to sample feature extraction sample；
Step 2: on the basis of step 1, the training sample T of extraction is utilizedoFeature, pass through and change the single classification of training
Parameter when device trains a variety of relatively independent single classifiers with kernel function；
Step 3: it using the method for class KNN, finds and test sample TsSimilar sample, and utilize canonical correlation point
The method of analysis considers the relationship between fractional sample feature and whole classifier classification results, so as to adjust each classifier weight
Obtain integrated model of classifiers；
Step 4: the final detection result of sample to be tested is codetermined by the classifier after step 3 integrates, judgement
Sample to be tested generic.
Further, it includes three parts that the specific implementation of the step 1, which includes: the extraction sample characteristics: first part
It is training sample ToHOG feature extraction, for training a variety of different and relatively independent classifiers；Second part is to utilize
Class KNN method find with test sample TsSimilar K second test sample TpFeature extraction, for single classifying to each
Device detects its sample accuracy rate to realize that the dynamic to single classifier confidence level adjusts；Part III is test sample TsSpy
Sign is extracted, for detecting the final classification results of sample.
Further, the specific implementation of the step 2 includes: the classifier that uses for support vector machines, to obtain
Multiple and different classifiers punishes related coefficient C and gamma related coefficient G using adjustment when carrying out classifier training, and
And different kernels is taken in different parameter combinations, to form a variety of different classifiers.
Further, the specific implementation of the step 3 includes the following steps:
A. utilize the method for class KNN from original training sample TrIn to each test sample TsSelect several second tests
Sample, and then this several second test sample is tested to obtain classification results probability by the single classifier that training obtains
Value DTp(x)；And the probability value DT by obtainingp(x) each classifier is calculated for the test accuracy rate AT of sample Xp, calculation formula
It is as follows:
ATp i=DTp(XC=j) (i=1.2.3 ... L)
Wherein dLC(x) indicating that n-th test sample X (N≤K) is measured in l-th classifier is the probability value of classification C,
DTp(XC=j) indicate the probability that the classification C for measuring sample X is sample concrete class j, i.e., detection of each single classifier to sample X
Accuracy rate；
By obtained ATp iValue obtains the test result values Y as K second test sample in L classifierj i:
WhereinIndicate jth (j≤K) a sample in each correct probability value of single classifier classification results；
B. the feature vector, X of K sample is constructedK i, and fractional sample feature is calculated using the method for canonical correlation analysis
XnWith L classifier classification accuracy Yc iBetween whole correlation matrix Rxy, calculation formula is as follows:
XK i=(b1 … bq)
Wherein bqIndicate the feature in sample X q dimension, i=1.2.3 ... K；M11Indicate variable XKThe covariance matrix of itself
COV(X,X)；M12It indicates COV (X, Y)；M21It indicates COV (Y, X)；M22It indicates COV (Y, Y)；Cq+k,q+kIt indicates between each variable
Correlativity；
C. the correlativity matrix R of sample characteristics Yu classifier classification accuracy is obtained by step Bxy, then apply this step
3rd formula, which is asked, in rapid works as variable XKWith YcCanonical variable coefficient a when in maximal correlation relationshipi T, bi T, obtain XkWith YcIt
Between maximum linear correlation combiner Uq、Vp(p≤q, p, q=1.2.3 ...) is required canonical variable, and calculation formula is as follows:
Up=aΤXk q
Vp=bΤYc p
Subject to:
aTM11A=1
bTM22B=1
D. it is obtained between feature and classifier classification results for K neighbour of some test sample most by step C
The classifier linear combination V that big correlativity obtainsp, while according to the coefficient b of obtained canonical variablei TDistribute single classifier power
Weight, and then the approximation of classifier linear combination Yu sample characteristics maximum correlation can be calculated, thus obtain different classifications device
Combine confidence level ￡ (X, the V when predicting sample Xp), calculation formula is as follows:
Wherein DijIndicate the feature and classifier linear combination V of test sample XpCorrelativity.
Further, the specific implementation of the step 4 includes:
Determine final test sample TsThe classification of middle sample to be tested X needs to detect its K related neighbour and tests the category
The related approximation Ψ of samplej(x), and sample to be tested X is handled, obtains its relationship D (x, y) with sample characteristics, Jin Erji
Calculate the approximation MT of sample to be tested Xs n(x), calculation formula is as follows:
Wherein MTs n(n=1.2.3 ...) indicates approximation when sample to be tested X is belonging respectively to the n-th class；
Finally by the difference calculated between test sample approximation and such sample class approximation, which is
It is the final classification of sample to be tested, calculation formula is as follows:
Beneficial effects of the present invention:
1, the present invention has been evaded tradition and has been asked based on testing result contingency present in single classifier licence plate recognition method
Topic.Using the thought of integrated study, multiple classifier Shared Decision Making results are analyzed, finally to the label of sample to be tested into
Row decision.This method can effectively improve classification accuracy.
2, the present invention has biggish difference with existing categorizer integration method.Traditional classifier Integrated Solution, not
Have and take into account contacting between part and entirety, so that will lead to the Weak Classifier that training obtains depends on training sample set sheet unduly
Body.When sample size is small or training sample set complexity, training to the effect to single classifier can also weaken.Simultaneously
In combining classifiers optimization, can also there are problems that optimization, so that Detection accuracy declines instead.And side used in the present invention
Method finds K neighbour of sample using the thought of class KNN, so as to balance between fractional sample feature and whole detection result
Connection, while to each test sample can dynamically adjust multi-categorizer classification when confidence level, to effectively solve
The certainly above problem.
3, the scheme of traditional Multi-classifers integrated is to consider that the optimization of multi-categorizer is integrated, there is no consider sample with
Connection between classifier.The thought of present invention application canonical correlation analysis (CCA), has studied sample characteristics and classifier result
Between canonical correlation relationship, multi-categorizer is understood to the sensibility of sample characteristics by the research to it, it is thus possible to find
Multi-categorizer weight adjusting parameter when certain class pattern detection can be more suitble to out.
Detailed description of the invention
Fig. 1 is the Multi-classifers integrated model based on CCA.
Fig. 2 is the Multi-classifers integrated strategic process figure based on CCA.
Fig. 3 is license plate data set experimental result picture.
Fig. 4 is data set Madelon experimental result picture.
Specific embodiment
With reference to the accompanying drawing, technical solution of the present invention is described in further detail.
Fig. 1 describes general thought of the invention.Method proposed by the present invention the following steps are included:
Step 1: processing set of data samples, by data set be divided into original training sample, test sample, second test sample,
Training sample, and sample characteristics are extracted according to sample feature.It is specific as follows:
In the embodiment of licence plate recognition method of the present invention, set of data samples S is existing license plate sample.Include in sample set
Original training sample TrWith test sample Ts.Wherein original training sample TrIn be divided into training sample T againoWith second test sample Tp
(i.e. by the method for class KNN from original training sample TrIn pick out with test sample TsMost similar sample), i.e. S=Tr+
Ts=To+Tp+Ts.It include in total 2100 sample objects, wherein original training sample has T in license plate data setr=1550,
Test sample has Ts=550, second test sample Tp=9.
It includes three parts that the sample characteristics, which extract: all data sets of this patent are all to extract histograms of oriented gradients
(Histogram of Oriented Gradient, HOG) feature.First part is training sample ToHOG feature extraction,
Purpose is to train a variety of different and relatively independent classifiers；Second part be using class KNN method find with test
Sample TsSimilar K second test sample TpFeature extraction, this part sample be mainly used for each single classifier detect
Its sample accuracy rate is to realize that the dynamic to single classifier confidence level adjusts；Part III is test sample TsFeature mention
It takes, for detecting the final classification results of sample.
Step 2: on the basis of step 1, the training sample T of extraction is utilizedoFeature, pass through and change the single classification of training
Parameter when device trains a variety of relatively independent single classifiers with kernel function.It is specific as follows:
The classifier used is support vector machines, to obtain multiple and different classifiers, is carrying out classifier training
When, related coefficient C and gamma related coefficient G are punished using adjustment, and different kernels is taken in different parameter combinations
Function forms a variety of different classifiers with this；The kernel function includes: RBF, Liner, Polynomial, Sigmoid.
It is specific as shown in table 1.
A sorter model of table more than 1
Step 3: it using the method for class KNN, finds and test sample TsSimilar sample, and utilize canonical correlation point
The method of analysis considers the relationship between fractional sample feature and whole classifier classification results, so as to adjust each classifier weight
Obtain integrated model of classifiers.As shown in Fig. 2, being implemented as follows:
A. utilize the method for class KNN from original training sample TrIn to each test sample TsSelect Tp=9 second tests
Sample.And then this 9 second test samples are tested to obtain classification results probability value by the single classifier that training obtains
DTp(x).And the probability value DT by obtainingp(x) each classifier is calculated for the test accuracy rate AT of sample Xp.Calculation formula is such as
Under:
ATp i=DTp(XC=j) (i=1.2.3 ... L) (2)
Wherein dLC(x) indicating that n-th test sample X (N≤K) is measured in l-th classifier is the probability value of classification C.
DTp(XC=j) indicate the probability that the classification C for measuring sample X is sample concrete class j, i.e., detection of each single classifier to sample X
Accuracy rate；The number of K expression second test sample.
By obtained ATp iValue obtains the test result values Y as K second test sample in L classifierj i。
WhereinIndicate jth (j≤K) a sample in each correct probability value of single classifier classification results.
B. the feature vector, X of K sample is constructedK i, and fractional sample is calculated using the method for canonical correlation analysis (CCA)
Feature XnWith L classifier classification accuracy Yc iBetween whole correlation matrix Rxy.Calculation formula is as follows:
XK i=(b1 … bq) (4)
Wherein bqIndicate the feature in sample X q dimension, i=1.2.3 ... K.M11Indicate variable XKThe covariance matrix of itself
COV(X,X)；M12It indicates COV (X, Y)；M21It indicates COV (Y, X)；M22It indicates COV (Y, Y)；Cq+k,q+kIt indicates between each variable
Correlativity.
C. the correlativity matrix R of sample characteristics Yu classifier classification accuracy is obtained by step Bxy, then apply formula
(8) it asks and works as variable XkWith YcCanonical variable coefficient a when in maximal correlation relationshipi T, bi T, obtain XkWith YcBetween maximum linear
Correlation combiner Uq、Vp(p≤q, p, q=1.2.3 ...) is required canonical variable, and calculation formula is as follows:
Up=aΤXk q (6)
Vp=bΤYc p (7)
Subject to:
aTM11A=1
bTM22B=1
D. it is obtained between feature and classifier classification results for K neighbour of some test sample most by step C
The classifier linear combination V that big correlativity obtainsp, while according to the coefficient b of obtained canonical variablei TDistribute single classifier power
Weight, and then the approximation of classifier linear combination Yu sample characteristics maximum correlation can be calculated, thus obtain different classifications device
Combine confidence level ￡ (X, the V when predicting sample Xp), calculation formula is as follows:
Wherein DijIndicate the feature and classifier linear combination V of test sample XpCorrelativity.
Step 4: the final detection result of sample to be tested is codetermined by the classifier after step 3 integrates, judgement
Sample to be tested generic.It is specific as follows:
Determine final test sample TsThe classification of middle sample to be tested X needs to detect its K related neighbour and tests the category
The related approximation Ψ of samplej(x), and sample to be tested X is handled, obtains its relationship D (x, y) with sample characteristics, Jin Erji
Calculate the approximation MT of sample to be tested Xs n(x), calculation formula is as follows:
Wherein MTs n(n=1.2.3 ...) indicates approximation when sample to be tested X is belonging respectively to the n-th class.
Finally by the difference calculated between test sample approximation and such sample class approximation, which is
It is the final classification of sample to be tested.Calculation formula is as follows:
Experiment conclusion
Below from several groups of experimental datas, to verify effectiveness of the invention.It is made simultaneously in order to evade the contingency of experiment
At error, provide the test result of one group of open source data set here, data set Madelon data set is from NIPS
It is obtained in 2003Feature Selection Challenge database resulting.It is the test for different data collection below
As a result.
Fig. 3 describes experimental result of this method in license plate data set.In license plate data set we have found that in phase
Under same kernel, different C (loss parameter), G (gamma parameter) parameter have a great impact to experiment.Single classifier is minimum
41.82% can be reached, 97.82% can be reached.And after Multiple Classifier Fusion, that low a part of classifier of classification accuracy
Assigned weight is close to 0.And the mentioned method of this experiment can reach 98.32%.
Fig. 4 describes the experimental result in Madelon data set, we extract the experiment knot under different parameters
Fruit research, when using multinomial kernel (Polynomial), between relatively independent single classifier, sample classification is accurate for discovery
Rate can reach 67.5%.And 75.28% can be reached using classification accuracy rate proposed by the invention.
To sum up, a kind of combining classifiers licence plate recognition method based on local canonical correlation analysis proposed by the present invention can
Effectively promote Car license recognition accuracy rate.Experimental data discovery is analyzed in the single preferable situation of classifier classifying quality, this reality
Test that mentioned classification effect promoting is limited, and maximum can improve 1%.It is found in the experimental result of similar Madelon data set
The mentioned method of the present invention between script single classifier classification results in the biggish situation of gap has sample classification accuracy rate
Obvious raising, the classification accuracy after Multiple Classifier Fusion improve nearly 8%.Classification of the present invention based on CCA
Device integrated approach can effectively improve the accuracy rate of picture classification in complex samples, and robustness with higher.
The series of detailed descriptions listed above only for feasible embodiment of the invention specifically
Protection scope bright, that they are not intended to limit the invention, it is all without departing from equivalent implementations made by technical spirit of the present invention
Or change should all be included in the protection scope of the present invention.
Claims (4)
1. a kind of combining classifiers licence plate recognition method based on local canonical correlation analysis, which is characterized in that including walking as follows
It is rapid:
Step 1: data set is divided into training sample T by processing set of data sampleso, test sample Ts, second test sample Tp, and
And the HOG feature of sample is extracted according to sample feature；
Step 2: on the basis of step 1, the training sample T of extraction is utilizedoFeature, by change training single classifier when
Parameter and kernel function train a variety of relatively independent single classifiers；
Step 3: it using the method for class KNN, finds and test sample TsSimilar sample, and utilize the side of canonical correlation analysis
Method considers the relationship between fractional sample feature and whole classifier classification results, is divided so as to adjust each classifier weight
Class device integrated model；
The specific implementation of the step 3 includes the following steps:
A. utilize the method for class KNN from original training sample TrIn to each test sample TsSeveral second test samples are selected,
And then this several second test sample is tested to obtain classification results probability value DT by the single classifier that training obtainsp
(x)；And the probability value DT by obtainingp(x) each classifier is calculated for the test accuracy rate AT of sample Xp, calculation formula is as follows:
ATp i=DTp(XC=j) (i=1.2.3 ... L)
Wherein dLC(x) indicating that n-th test sample X (N≤K) is measured in l-th classifier is the probability value of classification C, DTp
(XC=j) indicate that the probability that the classification C for measuring sample X is sample concrete class j, i.e., each single classifier are quasi- to the detection of sample X
True rate；
By obtained ATp iValue obtains the test result values Y as K second test sample in L classifierj i:
Wherein Yj i(i=1 ... L) indicates jth (j≤K) a sample in each correct probability value of single classifier classification results；
B. the feature vector, X of K sample is constructedK i, and fractional sample feature X is calculated using the method for canonical correlation analysisnWith L
A classifier classification accuracy Yc iBetween whole correlation matrix Rxy, calculation formula is as follows:
XK i=(b1 … bq)
Wherein bqIndicate the feature in sample X q dimension, i=1.2.3 ... K；M11Indicate variable XKThe covariance matrix COV of itself
(X,X)；M12It indicates COV (X, Y)；M21It indicates COV (Y, X)；M22It indicates COV (Y, Y)；Cq+k,q+kIt indicates between each variable
Correlativity；
C. the correlativity matrix R of sample characteristics Yu classifier classification accuracy is obtained by step Bxy, then using in this step
3rd formula is asked and works as variable XKWith YcCanonical variable coefficient a when in maximal correlation relationshipi T, bi T, obtain XkWith YcBetween most
Big linearly related combination Uq、Vp(p≤q, p, q=1.2.3 ...) is required canonical variable, and calculation formula is as follows:
Up=aTXk q
Vp=bTYc p
Subject to:
aTM11A=1
bTM22B=1
D. maximum phase between feature and classifier classification results for K neighbour of some test sample is obtained by step C
The classifier linear combination V that pass relationship obtainsp, while according to the coefficient b of obtained canonical variablei TSingle classifier weight is distributed,
And then the approximation of classifier linear combination Yu sample characteristics maximum correlation can be calculated, thus obtain the combination of different classifications device
Confidence level ￡ (X, V when to sample X predictionp), calculation formula is as follows:
Wherein DijIndicate the feature and classifier linear combination V of test sample XpCorrelativity；
Step 4: the final detection result of sample to be tested is codetermined by the classifier after step 3 integrates, and is judged to be measured
Sample generic.
2. a kind of combining classifiers licence plate recognition method based on local canonical correlation analysis according to claim 1,
It is characterized in that, the specific implementation of the step 1 includes: that the extraction sample characteristics include three parts: first part is trained sample
This ToHOG feature extraction, for training a variety of different and relatively independent classifiers；Second part is to utilize the class side KNN
Method find with test sample TsSimilar K second test sample TpFeature extraction, for detecting it to each single classifier
Sample accuracy rate is to realize that the dynamic to single classifier confidence level adjusts；Part III is test sample TsFeature extraction,
For detecting the final classification results of sample.
3. a kind of combining classifiers licence plate recognition method based on local canonical correlation analysis according to claim 1,
It is characterized in that, the specific implementation of the step 2 includes: the classifier that uses for support vector machines, multiple and different to obtain
Classifier related coefficient C and gamma related coefficient G are punished using adjustment, and in difference when carrying out classifier training
Parameter combination on take different kernels, to form a variety of different classifiers.
4. a kind of combining classifiers licence plate recognition method based on local canonical correlation analysis according to claim 1,
It is characterized in that, the specific implementation of the step 4 includes:
Determine final test sample TsThe classification of middle sample to be tested X needs to detect its K related neighbour to category test sample
Related approximation Ψj(x), and handle sample to be tested X, obtain its relationship D (x, y) with sample characteristics, so calculate to
The approximation MT of this X of test samples n(x), calculation formula is as follows:
Wherein MTs n(n=1.2.3 ...) indicates approximation when sample to be tested X is belonging respectively to the n-th class；
Finally by the difference calculated between test sample approximation and such sample class approximation, the difference minimum be to
The final classification of test sample sheet, calculation formula are as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610581551.0A CN106250913B (en) | 2016-07-21 | 2016-07-21 | A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610581551.0A CN106250913B (en) | 2016-07-21 | 2016-07-21 | A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106250913A CN106250913A (en) | 2016-12-21 |
| CN106250913B true CN106250913B (en) | 2019-08-02 |
Family
ID=57604416
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610581551.0A Active CN106250913B (en) | 2016-07-21 | 2016-07-21 | A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106250913B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106682985B (en) * | 2016-12-26 | 2020-03-27 | 深圳先进技术研究院 | Financial fraud identification method and system |
| CN109034201B (en) * | 2018-06-26 | 2022-02-01 | 创新先进技术有限公司 | Model training and rule mining method and system |
| CN111046926B (en) * | 2019-11-26 | 2023-09-19 | 山东浪潮科学研究院有限公司 | Computer vision image classification integrated learning method |
| CN112488098A (en) * | 2020-11-16 | 2021-03-12 | 浙江新再灵科技股份有限公司 | Training method of target detection model |
| CN114778374A (en) * | 2022-05-05 | 2022-07-22 | 江苏中烟工业有限责任公司 | Tobacco leaf classification method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103679191A (en) * | 2013-09-04 | 2014-03-26 | 西交利物浦大学 | An automatic fake-licensed vehicle detection method based on static state pictures |
| CN105740914A (en) * | 2016-02-26 | 2016-07-06 | 江苏科海智能系统有限公司 | Vehicle license plate identification method and system based on neighboring multi-classifier combination |
| CN105740886A (en) * | 2016-01-25 | 2016-07-06 | 宁波熵联信息技术有限公司 | Machine learning based vehicle logo identification method |
-
2016
- 2016-07-21 CN CN201610581551.0A patent/CN106250913B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103679191A (en) * | 2013-09-04 | 2014-03-26 | 西交利物浦大学 | An automatic fake-licensed vehicle detection method based on static state pictures |
| CN105740886A (en) * | 2016-01-25 | 2016-07-06 | 宁波熵联信息技术有限公司 | Machine learning based vehicle logo identification method |
| CN105740914A (en) * | 2016-02-26 | 2016-07-06 | 江苏科海智能系统有限公司 | Vehicle license plate identification method and system based on neighboring multi-classifier combination |
Non-Patent Citations (4)
| Title |
|---|
| Ensemble canonical correlation analysis;C. Okan Sakar 等;《AVEC "14 Proceedings of the 4th International Workshop on Audio/Visual Emotion Challenge》;20130813;第40卷(第2期);291-304 |
| Multi-Label Output Codes using Canonical Correlation Analysis;Yi Zhang 等;《Proceedings of the 14th International Conference on Artificial Intelligence and Statistics (AISTATS)》;20111231;873-882 |
| 基于距离学习的集成KNN分类器的研究;于飞;《中国优秀硕士学位论文全文数据库》;20100715(第7期);I140-62 |
| 车牌字符识别技术研究;黎涛;《中国优秀博硕士学位论文全文数据库 (硕士)》;20070315(第3期);I138-197 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106250913A (en) | 2016-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106250913B (en) | A kind of combining classifiers licence plate recognition method based on local canonical correlation analysis | |
| CN103632168B (en) | Classifier integration method for machine learning | |
| CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
| CN106897738B (en) | A kind of pedestrian detection method based on semi-supervised learning | |
| CN104156734B (en) | A kind of complete autonomous on-line study method based on random fern grader | |
| CN106295601A (en) | A kind of Safe belt detection method of improvement | |
| CN102163281B (en) | Real-time human body detection method based on AdaBoost frame and colour of head | |
| CN103440478B (en) | A kind of method for detecting human face based on HOG feature | |
| CN106372666B (en) | A kind of target identification method and device | |
| CN105389593A (en) | Image object recognition method based on SURF | |
| CN103871077B (en) | A kind of extraction method of key frame in road vehicles monitoring video | |
| CN105160317A (en) | Pedestrian gender identification method based on regional blocks | |
| CN109934269A (en) | A kind of opener recognition methods of electromagnetic signal and device | |
| CN108985360A (en) | Hyperspectral classification method based on expanding morphology and Active Learning | |
| CN105740914A (en) | Vehicle license plate identification method and system based on neighboring multi-classifier combination | |
| CN108197636A (en) | A kind of paddy detection and sorting technique based on depth multiple views feature | |
| CN103839078A (en) | Hyperspectral image classifying method based on active learning | |
| CN105975611A (en) | Self-adaptive combined downsampling reinforcing learning machine | |
| CN110717554A (en) | Image recognition method, electronic device, and storage medium | |
| CN103310235B (en) | A kind of steganalysis method based on parameter identification and estimation | |
| CN108877947A (en) | Depth sample learning method based on iteration mean cluster | |
| Karunasena et al. | Tea bud leaf identification by using machine learning and image processing techniques | |
| CN103631753A (en) | Progressively-decreased subspace ensemble learning algorithm | |
| CN104978569A (en) | Sparse representation based incremental face recognition method | |
| CN113191359B (en) | Small sample target detection method and system based on support and query samples |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210115 Address after: Room 3-53, room B2, No.15, Jinyang Road, Huaqiao Town, Kunshan City, Suzhou City, Jiangsu Province Patentee after: Jiangsu abid Information Technology Co.,Ltd. Address before: Zhenjiang City, Jiangsu Province, 212013 Jingkou District Road No. 301 Patentee before: JIANGSU University |