CN103279746B - A kind of face identification method based on support vector machine and system - Google Patents
A kind of face identification method based on support vector machine and system Download PDFInfo
- Publication number
- CN103279746B CN103279746B CN201310210372.2A CN201310210372A CN103279746B CN 103279746 B CN103279746 B CN 103279746B CN 201310210372 A CN201310210372 A CN 201310210372A CN 103279746 B CN103279746 B CN 103279746B
- Authority
- CN
- China
- Prior art keywords
- sample
- face
- training
- difference
- identified
- 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
Abstract
The invention discloses a kind of face identification method based on support vector machine and system.Described method includes: obtain face sample training collection: for each face sample of face sample training concentration, randomly select k and belong to same class other face sample with this face sample as similar sample, randomly select the individual face sample belonged to a different category with this face sample of k as foreign peoples's sample；Difference sample is generated to set according to described similar sample and described foreign peoples's sample；Described difference sample is in set, for each face sample of described face sample training concentration, all has 2k similar difference sample pair, and 2k foreign peoples's difference sample pair；For difference sample to set, support vector machine training is used to obtain similarity judgment models；Obtain disaggregated model according to described similarity judgment models, use described disaggregated model to carry out recognition of face.Use method or the system of the present invention, the efficiency of recognition of face can be improved on the premise of ensureing quick sampling.
Description
Technical field
The present invention relates to field of face identification, particularly relate to a kind of face identification method based on support vector machine and be
System.
Background technology
Recognition of face refers to the computer technology utilizing com-parison and analysis face visual signature information to carry out identity discriminating.Mesh
Before, face identification method based on support vector machine (Support Vector Machine, SVM) development is the rapidest.
So-called support vector refers to those points of the training sample at edge, spacer.Here " machine (machine, machine) "
The actually implication of " algorithm ".In machine learning field, often some algorithms are regarded as a machine.Support vector machine is set up
(Vapnik-Chervonenkis Dimension) theory and Structural risk minization basis is tieed up at the VC of Statistical Learning Theory
On, according to limited sample information, in the complexity (i.e. the study precision to specific training sample) of model and learning capacity is (i.e.
Identify the ability of arbitrary sample error-free) between seek optimal compromise, in the hope of obtaining best Generalization Ability.Wherein, VC dimension
Reflecting the learning capacity of collection of functions, VC ties up the biggest then Learning machine the most complicated (capacity is the biggest).
P.Jonathon Phillips is at article " Support Vector Machines Applied to Face
Recognition " in propose support vector machine (SVM) is applied in recognition of face problem.In the method that this article proposes,
Support vector machine first has to learn a similarity function, constructs sample pair, is then come by the similarity between facial image
Carry out recognition of face.
But, the method structure sample to during have problems.On the one hand, the method can produce substantial amounts of training
Sample pair, may cause the operation overlong time even internal memory of the method to overflow and cannot perform.On the other hand the method can be produced
Raw unbalanced data problem.Unbalanced data problem refers to, due to the particularity of recognition of face problem, it may appear that similar sample pair
And dissimilar sample between very big uneven, this can affect the performance of support vector machine to a great extent.
Summary of the invention
It is an object of the invention to provide a kind of face identification method based on support vector machine and system, it is possible to reduce training
The number of sample pair, makes similar sample that the number with dissimilar sample pair keeps balance simultaneously, and then can ensure quickly
The efficiency of recognition of face is improved on the premise of sampling.
For achieving the above object, the invention provides following scheme:
A kind of face identification method based on support vector machine, including:
Obtain the step of face sample training collection: described face sample training concentrates the face sample including multiple classification
Set, comprises multiple face sample in the face sample set of each classification；
Choose similar sample and the step of foreign peoples's sample: for each face sample of described face sample training concentration
This, randomly select k and belong to same class other face sample with this face sample as similar sample, randomly select k and
The face sample that this face sample belongs to a different category is as foreign peoples's sample；
Generate the difference sample step to set: generate difference sample to collection according to described similar sample and described foreign peoples's sample
Close；Described difference sample is in set, for each face sample of described face sample training concentration, all has 2k similar difference
Sample pair, and 2k foreign peoples's difference sample pair；
Generate the step of disaggregated model: for described difference sample to set, use support vector machine training to obtain similarity
Judgment models, obtains disaggregated model according to described similarity judgment models；
The step of recognition of face: use described disaggregated model to carry out recognition of face.
Optionally, the training of described employing support vector machine obtains similarity judgment models, including:
The support vector machine training using kernel function to be gaussian radial basis function obtains described similarity judgment models.
Optionally, before described generation difference sample is to set, also include:
Each sample concentrating described face sample training carries out random dimensionality reduction, the dimension phase after each sample dimensionality reduction
With；
The face sample after dimensionality reduction is used to generate difference sample to set；
Accordingly, when using described disaggregated model to carry out recognition of face, including:
Face sample to be identified is carried out random dimensionality reduction, after the dimension of the face sample described to be identified after dimensionality reduction and dimensionality reduction
The dimension of face sample identical.
Optionally, the described disaggregated model of described employing carries out recognition of face, including:
Obtain face sample to be identified；
From the sample set of the described each classification of face training set, randomly select k face sample respectively, generate 2k
Difference sample pair to be identified, obtains difference sample to be identified to set；
Utilize described disaggregated model that set is analyzed by described difference sample to be identified, obtain described face sample to be identified
Basis and the similarity probabilities of each class in described face training set；
According to described similarity probabilities, determine the classification that described face sample to be identified belongs to.
Optionally, described method also includes:
Repeat and choose the step of similar sample and foreign peoples's sample and generate the difference sample step to set, obtain multiple
Similarity judgment models, obtains disaggregated model according to the plurality of similarity judgment models；
Accordingly, described utilize described disaggregated model that set is analyzed by described difference sample to be identified, including:
Utilize the plurality of similarity judgment models that set is analyzed by described difference sample to be identified, obtain multiple institute
State face sample to be identified and the similarity probabilities of each class in described face training set；
Multiple described similarity probabilities are averaged, obtains average similarity probability；
The described classification determining that described face sample to be identified belongs to, including:
By the classification of average similarity maximum probability, it is defined as the classification of described face sample to be identified ownership.
A kind of face identification system based on support vector machine, including:
Training set acquisition module, is used for obtaining face sample training collection；Described face sample training concentration includes multiple
The face sample set of classification, comprises multiple face sample in the face sample set of each classification；
Module chosen by sample, for performing to choose similar sample and the step of foreign peoples's sample: instruct for described face sample
Practice each the face sample concentrated, randomly select k and belong to same class other face sample as same with this face sample
Class sample, randomly selects k the face sample belonged to a different category with this face sample as foreign peoples's sample；
Difference sample is to set generation module, for performing the step generating difference sample to set: according to described similar sample
Difference sample is generated to set with described foreign peoples's sample；Described difference sample, in set, is concentrated for described face sample training
Each face sample, all has 2k similar difference sample pair, and 2k foreign peoples's difference sample pair；
Disaggregated model generation module, is used for for described difference sample set, uses support vector machine training to obtain similar
Property judgment models, obtains disaggregated model according to described similarity judgment models；
Face recognition module, is used for using described disaggregated model to carry out recognition of face.
Optionally, described disaggregated model generation module, including:
Similarity judgment models signal generating unit, for the support vector machine training using kernel function to be gaussian radial basis function
Obtain described similarity judgment models.
Optionally, also include:
Training set sample dimensionality reduction module, carries out random dimensionality reduction for each sample concentrating described face sample training,
Dimension after each sample dimensionality reduction is identical；So that the face sample after described difference sample uses dimensionality reduction to set generation module generates
Difference sample is to set；
Accordingly, described face recognition module, including:
Sample dimensionality reduction unit to be identified, for face sample to be identified is carried out random dimensionality reduction, waits described in after dimensionality reduction to know
The dimension of others' face sample is identical with the dimension of the face sample after dimensionality reduction.
Optionally, described face recognition module, including:
Face sample acquisition unit to be identified, is used for obtaining face sample to be identified；
Difference sample to be identified is to set signal generating unit, for respectively from the sample set of the described each classification of face training set
In randomly select k face sample, generate 2k difference sample pair to be identified, obtain difference sample to be identified to set；
Similarity probabilities computing unit, is used for utilizing described disaggregated model to carry out described difference sample to be identified to set point
Analysis, obtains described face sample to be identified and the similarity probabilities of each class in described face training set；
Classification determination unit, for according to described similarity probabilities, determines the classification that described face sample to be identified belongs to.
Optionally, also include:
Repeat module, be used for controlling described sample and choose module and described difference sample to set generation module repetition
Perform choose the step of similar sample and foreign peoples's sample and generate the difference sample step to set, obtain multiple similarity and judge mould
Type, obtains disaggregated model according to the plurality of similarity judgment models；
Accordingly, described similarity probabilities computing unit includes:
Similarity probabilities computation subunit, is used for utilizing the plurality of similarity judgment models to described difference sample to be identified
Set is analyzed, obtains multiple described face sample to be identified general with the similarity of each class in described face training set
Rate；
Average similarity probability calculation subelement, for averaging multiple described similarity probabilities, obtains average phase
Like property probability；
Described classification determination unit, including:
Classification determines subelement, for by the classification of average similarity maximum probability, being defined as described face sample to be identified
The classification of this ownership.
The specific embodiment provided according to the present invention, the invention discloses techniques below effect:
The face identification method of the present invention and system, by each the face sample concentrated for described face sample training
This, randomly select k similar sample and k foreign peoples's sample, and according to described similar sample and described foreign peoples's sample, it is poor to generate
Sample is to set, for described difference sample to set, uses support vector machine training to obtain similarity judgment models；According to described
Multiple similarity judgment models obtain disaggregated model；Described disaggregated model is used to carry out recognition of face；On the one hand due to from described
In the face sample set of each classification that face sample training is concentrated, randomly select k similar sample and k foreign peoples's sample
In structure difference sample pair, with prior art, each face sample that described face sample training is concentrated, choose except this people
All face sample architecture difference samples outside face sample are to comparing, it is possible to reduce the number of training sample pair；On the other hand, due to
The similar sample chosen and the number of foreign peoples's sample are identical, it is possible to generate equal number of similar difference sample to and foreign peoples
Difference sample pair, makes similar sample that the number with dissimilar sample pair is kept balance.In conjunction with above-mentioned both sides beneficial effect, this
The method of invention or system can improve the efficiency of recognition of face on the premise of ensureing quick sampling.
Accompanying drawing explanation
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to institute in embodiment
The accompanying drawing used is needed to be briefly described, it should be apparent that, the accompanying drawing in describing below is only some enforcements of the present invention
Example, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to according to these accompanying drawings
Obtain other accompanying drawing.
Fig. 1 is the flow chart of the face identification method embodiment based on support vector machine of the present invention；
Fig. 2 is the structure chart of the face identification system embodiment based on support vector machine of the present invention.
Detailed description of the invention
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under not making creative work premise
Embodiment, broadly falls into the scope of protection of the invention.
Understandable for enabling the above-mentioned purpose of the present invention, feature and advantage to become apparent from, real with concrete below in conjunction with the accompanying drawings
The present invention is further detailed explanation to execute mode.
Fig. 1 is the flow chart of the face identification method embodiment based on support vector machine of the present invention.As it is shown in figure 1, institute
The method of stating includes:
Step 101: obtain face sample training collection；Described face sample training concentrates the face sample including multiple classification
This set, comprises multiple face sample in the face sample set of each classification；
Face sample in the face sample set of each classification, can be the face sample belonging to same person.With
Different face samples in one classification, sampling angle, or expression during sampling can be different.
Step 102: each the face sample concentrated for described face sample training, randomly selects k and this face
Sample belongs to same class other face sample as similar sample, randomly selects k and belongs to a different category with this face sample
Face sample as foreign peoples's sample；Wherein, k is positive integer.
Face number of samples under usual each classification is fewer, such as 10 to 20.K in step 102 takes
Value is less than the face number of samples under the category.
Step 103: generate difference sample to set according to described similar sample and described foreign peoples's sample；Described difference sample is to collection
In conjunction, each face sample that described face sample training is concentrated, all there is 2k similar difference sample pair, and k is individual different
Class difference sample pair；
Assume the face sample that described face sample training is concentrated be A, k similar sample be A1, A2, A3, A4,
A5, then the similar difference sample that can generate is to for A-A1, A-A2, A-A3, A-A4, A-A5；A1-A, A2-A, A3-A, A4-A,
A5-A.Adopt and can generate 2k foreign peoples's difference sample pair in the same way.
Step 104: for described difference sample to set, uses support vector machine training to obtain similarity judgment models, root
Disaggregated model is obtained according to described similarity judgment models；
The support vector machine training that kernel function can be used to be gaussian radial basis function obtains described similarity judgment models.
Disaggregated model can be obtained, it is also possible to obtain according to multiple similarity judgment models according to a similarity judgment models
To disaggregated model.When obtaining disaggregated model according to a similarity judgment models, described disaggregated model can be similar to described
Property judgment models is identical；When obtaining disaggregated model according to multiple similarity judgment models, described disaggregated model can be multiple
The set of similarity judgment models.
If x, z be ∈ X, X belongs to R(n) space, nonlinear function Φ realizes the mapping to feature space F of input space X, its
Middle F belongs to R(m), n < < m.Have according to kernel function technology:
K (x, z)=<Φ (x), Φ (z)>(1)
Wherein:<,>is inner product, (x z) is kernel function to K.From formula (1) it can be seen that m is tieed up in higher dimensional space by kernel function
Long-pending computing is converted into the kernel function of the n dimension low-dimensional input space and calculates, thus solves " the dimension calculated in high-dimensional feature space
Disaster " etc. problem.
Gaussian radial basis function in this step refers to the gaussian kernel function in RBF.
So-called RBF (Radial Basis Function is called for short RBF), is exactly certain radially symmetrical scalar
Function.Be normally defined in space the monotonic function of Euclidean distance between any point x to a certain center xc, can be denoted as k (| | x-xc
| |), its effect local often, i.e. when x is away from xc, function value is the least.
The most frequently used RBF is gaussian kernel function, and form is k (| | x-xc | |)=exp{-| | x-xc | | ^2/2* σ ^
2) } wherein xc is kernel function center, and σ is the width parameter of function, controls the radial effect scope of function.
Step 105: use described disaggregated model to carry out recognition of face.
Concrete, the process using described disaggregated model to carry out recognition of face can comprise the steps:
Step A: obtain face sample to be identified；
Step B: randomly select k face sample from the sample set of the described each classification of face training set respectively, raw
Become 2k difference sample pair to be identified, obtain difference sample to be identified to set；
Step C: utilize described disaggregated model that set is analyzed by described difference sample to be identified, obtain described to be identified
Face sample and the similarity probabilities of each class in described face training set；
Step D: according to described similarity probabilities, determine the classification that described face sample to be identified belongs to.
Wherein, in step C, utilize described disaggregated model that set is analyzed by described difference sample to be identified, refer to
Using described difference sample to be identified to as input value, substitute into described disaggregated model and calculate.The result of calculation table of disaggregated model
Show described face sample to be identified and the similarity probabilities of each class in described face training set.
In step D, can be defined as classification corresponding for maximum similarity probabilities according to the size of similarity probabilities
The classification of described face sample to be identified ownership.
In sum, the face identification method of the present embodiment, by described face sample training is concentrated each
Face sample, randomly selects k similar sample and k foreign peoples's sample, according to described similar sample and described foreign peoples's sample,
Generation difference sample is to set, for described difference sample to set, uses support vector machine training to obtain similarity judgment models；Root
Disaggregated model is obtained according to described similarity judgment models；Described disaggregated model is used to carry out recognition of face；On the one hand due to from institute
State in the face sample set of each classification that face sample training is concentrated, randomly select k similar sample and k foreign peoples's sample
This structure difference sample pair, with prior art, for each face sample of described face sample training concentration, chooses except being somebody's turn to do
All face sample architecture difference samples outside face sample are to comparing, and the method for the present embodiment can reduce the number of training sample pair
Mesh；On the other hand, identical due to the number of the similar sample chosen and foreign peoples's sample, it is possible to generate equal number of same
Class difference sample to and foreign peoples's difference sample pair, make similar sample that the number with dissimilar sample pair is kept balance.In conjunction with above-mentioned two
The beneficial effect of aspect, the method for the present embodiment and then the effect of recognition of face can be improved on the premise of ensureing quick sampling
Rate.
In order to improve the efficiency of the face identification method of the present invention further, before step 103, it is also possible to include following
Step:
Each sample concentrating described face sample training carries out random dimensionality reduction, the dimension phase after each sample dimensionality reduction
With；
The face sample after dimensionality reduction is used to generate difference sample to set.
Wherein, each sample that face sample training is concentrated can be picture format.Described dimensionality reduction can be that reduction is described
The number of pixels of each sample.
Accordingly, after face sample is carried out dimension-reduction treatment, when using described disaggregated model to carry out recognition of face, permissible
Comprise the steps:
Face sample to be identified is carried out random dimensionality reduction, after the dimension of the face sample described to be identified after dimensionality reduction and dimensionality reduction
The dimension of face sample identical.
In order to improve the accuracy of the face identification method of the present invention further, the face identification method of the present invention is all right
Employing following steps:
Repeated execution of steps 102 and step 103, obtain multiple similarity judgment models, sentence according to the plurality of similarity
Disconnected model obtains disaggregated model；
For example, it is possible to repeated execution of steps 102 and each 50 times of step 103, it is possible to obtain 50 similarity judgment models.
Then can be according to 50 respective classification results of similarity judgment models, comprehensive analysis (such as averaging) obtains average phase
Like property probability.
The set of above-mentioned multiple similarity judgment models may be constructed disaggregated model.The present invention can be according to described classification mould
Type, carries out recognition of face.
Accordingly, when utilizing described disaggregated model that set is analyzed by described difference sample to be identified, under can including
State step:
Utilize the plurality of similarity judgment models that set is analyzed by described difference sample to be identified, obtain multiple institute
State face sample to be identified and the similarity probabilities of each class in described face training set；
Multiple described similarity probabilities are averaged, obtains average similarity probability；
When determining the classification that described face sample to be identified belongs to, can comprise the steps:
By the classification of average similarity maximum probability, it is defined as the classification of described face sample to be identified ownership.
The invention also discloses a kind of face identification system based on support vector machine.Fig. 2 be the present invention based on support
The structure chart of the face identification system embodiment of vector machine.As in figure 2 it is shown, described system may include that
Training set acquisition module 201, is used for obtaining face sample training collection；Described face sample training concentration includes many
The face sample set of individual classification, comprises multiple face sample in the face sample set of each classification；
Module 202 chosen by sample, chooses similar sample and the step of foreign peoples's sample for execution: for described face sample
Each face sample in training set, randomly selects k and belongs to same class other face sample conduct with this face sample
Similar sample, randomly selects k the face sample belonged to a different category with this face sample as foreign peoples's sample；
Difference sample is to set generation module 203, for performing the step generating difference sample to set: according to described similar sample
This and described foreign peoples's sample generate difference sample to set；Described difference sample, in set, is concentrated for described face sample training
Each face sample, all have 2k similar difference sample pair, and 2k foreign peoples's difference sample pair；
Disaggregated model generation module 204, for obtaining phase for described difference sample to set, employing support vector machine training
Like property judgment models, obtain disaggregated model according to described similarity judgment models；
Face recognition module 205, is used for using described disaggregated model to carry out recognition of face.
Wherein, described disaggregated model generation module 204, may include that
Similarity judgment models signal generating unit, for the support vector machine training using kernel function to be gaussian radial basis function
Obtain described similarity judgment models.
Described face recognition module 205, may include that
Face sample acquisition unit to be identified, is used for obtaining face sample to be identified；
Difference sample to be identified is to set signal generating unit, for respectively from the sample set of the described each classification of face training set
In randomly select k face sample, generate 2k difference sample pair to be identified, obtain difference sample to be identified to set；
Similarity probabilities computing unit, is used for utilizing described disaggregated model to carry out described difference sample to be identified to set point
Analysis, obtains described face sample to be identified and the similarity probabilities of each class in described face training set；
Classification determination unit, for according to described similarity probabilities, determines the classification that described face sample to be identified belongs to.
Described system can also include:
Training set sample dimensionality reduction module, carries out random dimensionality reduction for each sample concentrating described face sample training,
Dimension after each sample dimensionality reduction is identical；So that the face sample after described difference sample uses dimensionality reduction to set generation module generates
Difference sample is to set；
Accordingly, described face recognition module 205, may include that
Sample dimensionality reduction unit to be identified, for face sample to be identified is carried out random dimensionality reduction, waits described in after dimensionality reduction to know
The dimension of others' face sample is identical with the dimension of the face sample after dimensionality reduction.
Described system can also include:
Repeat module, be used for controlling described sample and choose module 202 and described difference sample to set generation module
203 repeat step and the generation difference sample step to set choosing similar sample and foreign peoples's sample, obtain multiple similar
Property judgment models, obtains disaggregated model according to the plurality of similarity judgment models；
Accordingly, described similarity probabilities computing unit may include that
Similarity probabilities computation subunit, is used for utilizing the plurality of similarity judgment models to described difference sample to be identified
Set is analyzed, obtains multiple described face sample to be identified general with the similarity of each class in described face training set
Rate；
Average similarity probability calculation subelement, for averaging multiple described similarity probabilities, obtains average phase
Like property probability；
Described classification determination unit, may include that
Classification determines subelement, for by the classification of average similarity maximum probability, being defined as described face sample to be identified
The classification of this ownership.
In order to be further appreciated by method and the beneficial effect of the present invention, say with a more specific example below
Bright.
In this example, face sample training collection uses UMIST face database.UMIST human face data collection comprises 564 figures
Picture, totally 20 people, cover different race, sexs and appearance.In this data set, the photo of each individual shooting has from side
To the continuous attitudes vibration of the different angles in front, it is a kind of conventional face recognition database.In this example from 20 class samples
Every class randomly selects half as training set, and second half is as test set, each 282 samples in training set, test set.
It is embodied as step as follows:
(1) step 1: process face training sample, constructs new training sample.If existing view data is
First the sample in training set is carried out random dimensionality reduction, sample is dropped to 50 dimensions, keeps the classification after dimensionality reduction simultaneously
Labelling is constant.Then the training sample set after dimensionality reduction is
After random dimensionality reduction, new difference sample to be produced to training set.For any two sample For, permissible
Produce two different difference samplesWithIfWithClassification identical, then remember the class label+1 of new sample, i.e.
Positive sample, is otherwise designated as-1, i.e. negative sample.
To each sample in training setRandomly choose k similar sample point and foreign peoples's sample point respectively, respectively
These sample points are respectively stored in setWithIn, k takes 5.The newest training set difference sample is represented by zq∈R50, q=1 ..., 5640,5640 is the number of difference sample pair.
Multiple similarity judgment models can be generated, construct difference sample respectively to set for each similarity judgment models.
Assuming to construct altogether enum similarity judgment models, enum takes 50, then repeat above-mentioned construction process 50 times.The difference sample pair obtained
In set, including the difference sample pair that each classification is corresponding, using described difference sample to gathering as new training sample set, respectively
It is designated as Trn1,Trn2..., Trn50。
(2) step 2: be trained support vector machine with new training sample set, obtains 50 similaritys and judges mould
Type.The set that these 50 similarity judgment models are constituted, it is simply that the disaggregated model in the present invention.Concrete, each is instructed
Practice sample set Trni, i=1,2 ..., 50, it is respectively adopted support vector machine training and obtains a similarity judgment models, totally 50
Individual, independent each other, wherein pth similarity judgment models is to obtain according to the methods below:
The most widely used gaussian radial basis function of Selection of kernel function of support vector machineWherein σ
It it is nuclear parameter.After specifying the parameter that support vector machine needs, input training sample set, with gaussian radial basis function for core letter
Number, training one similarity judgment models of generation:
Wherein And bpIt it is pth the similarity judgment models produced by support vector machine training
Coefficient, sgn () represents sign function.
(3) step 3: process the test sample of test set, constructs new test sample.If any one is to be tested
Sample x, x ∈ R10304。
Same, test sample being carried out random dimensionality reduction, the dimension of test sample is dropped to 50 dimensions, sample x to be tested becomes
ForThe process of the random dimensionality reduction of test sample and the random dimensionality reduction of corresponding training sample are duplicate.
After random dimensionality reduction, the new test set comprising difference sample pair to be produced.For reducing the number of new test sample,
Num sample of random choose in each class sample of former training set, num takes 5, for sample to be testedEvery with pick out
Individual training sampleAll generate two difference samples pairWithThe newest test set is represented byq
=1 ..., 200,200 is the number of difference sample pair.
(4) step 4: the difference sample pair that will obtain in new test setIt is sequentially inputted to each similarity sentence
In disconnected model, the z in former similarity judgment models, each similarity judgment models is replaced to can get 200 judged results.Often
The result of individual similarity judgment models is represented sequentially asAccording to classification results, can count
X and the similarity probabilities size of each class in former training sample in each similarity judgment models
(5) step 5: carry out integrated by the result of each similarity judgment models, draws final classification results.By each
Similarity judgment models can draw sample to be tested x and similarity probabilities size P of each class in former training sampleij,i=1,
2,...,50,j=1,2,...,20.Then sample to be tested and class average similarity probability P are calculatedj', calculation is as follows:
Finally sample to be tested x is classified, i.e. according to average likelihood probability maximal criterion
The effect of the present invention can be by following experimental verification:
During experiment, the value of regular factor takes 10, and the parameter of Gauss radial kernel function is then new in generation every time
All pass through after the training sample of difference sample pair to calculate, calculate process specific as follows: first determine the average of training sample set,
Then each training sample Euclidean distance to average point is obtained, the intermediate value of the meansigma methods of all distances then taking for nuclear parameter
Value.According to the model drawn in training set, test set is estimated performance obtains discrimination.Choosing of training sample and test sample
Choosing is random, is repeated 10 times test, last experimental result such as table 1.
Table 1
The control methods of the present invention is four kinds of typical Downsapling methods, the most random down-sampling, NearMiss-1,
NearMiss-2 and NearMiss-3, wherein the sample rate of Downsapling method reaches balance be as the criterion with new training set after sampling
Determine.
Pass through experimental result, it can be seen that the recognition of face effect of the present invention is substantially better than other four kinds of Downsapling methods,
And show stronger stability, there is certain advantage.
| The present invention | Random down-sampling | NearMiss-1 | NearMiss-2 | NearMiss-3 | |
| Time (second) | 1321.5(50 it is secondary) | 604(1 time) | 2291.5(1 it is secondary) | 2467.5(1 it is secondary) | 1109.3(1 it is secondary) |
Table 2
Table 2 gives the time that these four method is spent in certain is once trained.Note present invention comprises 50 secondary
Become sample to generating and the time of 50 SVM training, say, that have only to every time 26.43 seconds the most permissible.Other four kinds of methods
The execution time, be all that 1 sample is to generation time and the time of 1 SVM training.It is obvious that the present invention's adopts the sampling time
It is the fastest comparatively speaking.
In this specification, each embodiment uses the mode gone forward one by one to describe, and what each embodiment stressed is and other
The difference of embodiment, between each embodiment, identical similar portion sees mutually.For system disclosed in embodiment
For, owing to it corresponds to the method disclosed in Example, so describe is fairly simple, relevant part sees method part and says
Bright.
Principle and the embodiment of the present invention are set forth by specific case used herein, saying of above example
Bright method and the core concept thereof being only intended to help to understand the present invention；Simultaneously for one of ordinary skill in the art, foundation
The thought of the present invention, the most all will change.In sum, this specification content is not
It is interpreted as limitation of the present invention.
Claims (6)
1. a face identification method based on support vector machine, it is characterised in that including:
Obtain the step of face sample training collection: described face sample training concentrates the face sample set including multiple classification
Close, the face sample set of each classification comprises multiple face sample；
Choose similar sample and the step of foreign peoples's sample: each face sample that described face sample training is concentrated, with
Machine is chosen k and is belonged to same class other face sample with this face sample as similar sample, randomly selects k and this face
The face sample that sample belongs to a different category is as foreign peoples's sample；
Generate the difference sample step to set: generate difference sample to set according to described similar sample and described foreign peoples's sample；Institute
State difference sample in set, each face sample that described face sample training is concentrated, all there is 2k similar difference sample
Right, and 2k foreign peoples's difference sample pair；
Generate the step of disaggregated model: for described difference sample to set, use support vector machine training to obtain similarity and judge
Model, obtains disaggregated model according to described similarity judgment models；
The step of recognition of face: use described disaggregated model to carry out recognition of face；
The described disaggregated model of described employing carries out recognition of face, including:
Obtain face sample to be identified；
From the sample set of the described each classification of face training set, randomly select k face sample respectively, generate 2k and wait to know
Other difference sample pair, obtains difference sample to be identified to set；
Utilize described disaggregated model that set is analyzed by described difference sample to be identified, obtain described face sample to be identified with
The similarity probabilities of each class in described face training set；
According to described similarity probabilities, determine the classification that described face sample to be identified belongs to；
Also include: repeat and choose the step of similar sample and foreign peoples's sample and generate the difference sample step to set, obtain
Multiple similarity judgment models, obtain disaggregated model according to the plurality of similarity judgment models；
Accordingly, described utilize described disaggregated model that set is analyzed by described difference sample to be identified, including:
Utilize the plurality of similarity judgment models that set is analyzed by described difference sample to be identified, obtain multiple described in treat
Identify face sample and the similarity probabilities of each class in described face training set；
Multiple described similarity probabilities are averaged, obtains average similarity probability；
The described classification determining that described face sample to be identified belongs to, including:
By the classification of average similarity maximum probability, it is defined as the classification of described face sample to be identified ownership.
Method the most according to claim 1, it is characterised in that the training of described employing support vector machine obtains similarity and judges
Model, including:
The support vector machine training using kernel function to be gaussian radial basis function obtains described similarity judgment models.
Method the most according to claim 1, it is characterised in that before described generation difference sample is to set, also include:
Each sample concentrating described face sample training carries out random dimensionality reduction, and the dimension after each sample dimensionality reduction is identical；
The face sample after dimensionality reduction is used to generate difference sample to set；
Accordingly, when using described disaggregated model to carry out recognition of face, including:
Face sample to be identified is carried out random dimensionality reduction, the people after the dimension of the face sample described to be identified after dimensionality reduction and dimensionality reduction
The dimension of face sample is identical.
4. a face identification system based on support vector machine, it is characterised in that including:
Training set acquisition module, is used for obtaining face sample training collection；Described face sample training is concentrated and is included multiple classification
Face sample set, the face sample set of each classification comprises multiple face sample；
Module chosen by sample, chooses similar sample and the step of foreign peoples's sample for execution: for described face sample training collection
In each face sample, randomly select k and belong to same class other face sample as similar sample with this face sample
This, randomly select k the face sample belonged to a different category with this face sample as foreign peoples's sample；
Difference sample is to set generation module, for performing the step generating difference sample to set: according to described similar sample and institute
State foreign peoples's sample and generate difference sample to set；Described difference sample to set in, for described face sample training concentrate each
Individual face sample, all has 2k similar difference sample pair, and 2k foreign peoples's difference sample pair；
Disaggregated model generation module, for obtaining similarity for described difference sample sentence set, employing support vector machine training
Disconnected model, obtains disaggregated model according to described similarity judgment models；
Face recognition module, is used for using described disaggregated model to carry out recognition of face；
Face sample acquisition unit to be identified, is used for obtaining face sample to be identified；
Difference sample to be identified to set signal generating unit, for respectively from the sample set of the described each classification of face training set with
K face sample chosen by machine, generates 2k difference sample pair to be identified, obtains difference sample to be identified to set；
Similarity probabilities computing unit, is used for utilizing described disaggregated model to be analyzed described difference sample to be identified to set,
Obtain described face sample to be identified and the similarity probabilities of each class in described face training set；
Classification determination unit, for according to described similarity probabilities, determines the classification that described face sample to be identified belongs to, also wraps
Include:
Repeat module, be used for controlling described sample and choose module and described difference sample set generation module is repeated
Choose the step of similar sample and foreign peoples's sample and generate the difference sample step to set, obtaining multiple similarity judgment models,
Disaggregated model is obtained according to the plurality of similarity judgment models；
Accordingly, described similarity probabilities computing unit includes:
Similarity probabilities computation subunit, is used for utilizing the plurality of similarity judgment models to described difference sample to be identified to collection
Conjunction is analyzed, and obtains multiple described face sample to be identified and the similarity probabilities of each class in described face training set；
Average similarity probability calculation subelement, for averaging multiple described similarity probabilities, obtains average similarity
Probability；
Described classification determination unit, including:
Classification determines subelement, for by the classification of average similarity maximum probability, being defined as described face sample to be identified and return
The classification belonged to.
System the most according to claim 4, it is characterised in that described disaggregated model generation module, including:
Similarity judgment models signal generating unit, obtains for the support vector machine training using kernel function to be gaussian radial basis function
Described similarity judgment models.
System the most according to claim 4, it is characterised in that also include:
Training set sample dimensionality reduction module, carries out random dimensionality reduction for each sample concentrating described face sample training, each
Dimension after sample dimensionality reduction is identical；So that the face sample after described difference sample uses dimensionality reduction to set generation module generates difference sample
This is to set；
Accordingly, described face recognition module, including:
Sample dimensionality reduction unit to be identified, for face sample to be identified is carried out random dimensionality reduction, the people described to be identified after dimensionality reduction
The dimension of face sample is identical with the dimension of the face sample after dimensionality reduction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310210372.2A CN103279746B (en) | 2013-05-30 | 2013-05-30 | A kind of face identification method based on support vector machine and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310210372.2A CN103279746B (en) | 2013-05-30 | 2013-05-30 | A kind of face identification method based on support vector machine and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103279746A CN103279746A (en) | 2013-09-04 |
| CN103279746B true CN103279746B (en) | 2016-12-07 |
Family
ID=49062261
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310210372.2A Active CN103279746B (en) | 2013-05-30 | 2013-05-30 | A kind of face identification method based on support vector machine and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103279746B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104008364B (en) * | 2013-12-31 | 2018-09-25 | 广西科技大学 | Face identification method |
| CN103679160B (en) * | 2014-01-03 | 2017-03-22 | 苏州大学 | Human-face identifying method and device |
| CN103793699A (en) * | 2014-02-24 | 2014-05-14 | 苏州大学 | Face recognition method and system |
| CN105139033B (en) * | 2015-08-24 | 2018-11-06 | 小米科技有限责任公司 | Grader construction method and device and image processing method and device |
| CN107688829A (en) * | 2017-08-29 | 2018-02-13 | 湖南财政经济学院 | A kind of identifying system and recognition methods based on SVMs |
| CN108446659A (en) * | 2018-03-28 | 2018-08-24 | 百度在线网络技术（北京）有限公司 | Method and apparatus for detecting facial image |
| CN108564067B (en) * | 2018-04-28 | 2021-07-20 | 深圳卡安兴科技有限公司 | Method and system for determining threshold value of face comparison |
| CN109145991B (en) * | 2018-08-24 | 2020-07-31 | 北京地平线机器人技术研发有限公司 | Image group generation method, image group generation device and electronic equipment |
| CN112819018A (en) * | 2019-10-31 | 2021-05-18 | 北京沃东天骏信息技术有限公司 | Method and device for generating sample, electronic equipment and storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663370A (en) * | 2012-04-23 | 2012-09-12 | 苏州大学 | Face identification method and system |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7050607B2 (en) * | 2001-12-08 | 2006-05-23 | Microsoft Corp. | System and method for multi-view face detection |
-
2013
- 2013-05-30 CN CN201310210372.2A patent/CN103279746B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663370A (en) * | 2012-04-23 | 2012-09-12 | 苏州大学 | Face identification method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103279746A (en) | 2013-09-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103279746B (en) | A kind of face identification method based on support vector machine and system | |
| CN106096538B (en) | Face identification method and device based on sequencing neural network model | |
| CN108564129B (en) | Trajectory data classification method based on generation countermeasure network | |
| CN107273845B (en) | Facial expression recognition method based on confidence region and multi-feature weighted fusion | |
| CN102663370B (en) | Face identification method and system | |
| Mori et al. | Recognizing objects in adversarial clutter: Breaking a visual CAPTCHA | |
| US20190205620A1 (en) | High-quality training data preparation for high-performance face recognition systems | |
| CN110852168A (en) | Pedestrian re-recognition model construction method and device based on neural framework search | |
| CN106599941A (en) | Method for identifying handwritten numbers based on convolutional neural network and support vector machine | |
| CN107103281A (en) | Face identification method based on aggregation Damage degree metric learning | |
| US10747961B2 (en) | Method and device for identifying a sentence | |
| CN106934012A (en) | A kind of question answering in natural language method and system of knowledge based collection of illustrative plates | |
| Singh et al. | A study of moment based features on handwritten digit recognition | |
| Liu et al. | Facial landmark machines: A backbone-branches architecture with progressive representation learning | |
| CN104318242A (en) | High-efficiency SVM active half-supervision learning algorithm | |
| Zeng et al. | Fine-grained image retrieval via piecewise cross entropy loss | |
| CN104392231A (en) | Block and sparse principal feature extraction-based rapid collaborative saliency detection method | |
| CN103745200A (en) | Facial image identification method based on word bag model | |
| Wu et al. | Some analysis and research of the AdaBoost algorithm | |
| CN109948735A (en) | A kind of multi-tag classification method, system, device and storage medium | |
| CN103065158A (en) | Action identification method of independent subspace analysis (ISA) model based on relative gradient | |
| CN103679160A (en) | Human-face identifying method and device | |
| Zeng et al. | Optimization of bagging classifiers based on SBCB algorithm | |
| US20200320387A1 (en) | Independent and dependent reading using recurrent networks for natural language inference | |
| CN105320764A (en) | 3D model retrieval method and 3D model retrieval apparatus based on slow increment features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |