CN106407915A - SVM (support vector machine)-based face recognition method and device - Google Patents
SVM (support vector machine)-based face recognition method and device Download PDFInfo
- Publication number
- CN106407915A CN106407915A CN201610798856.7A CN201610798856A CN106407915A CN 106407915 A CN106407915 A CN 106407915A CN 201610798856 A CN201610798856 A CN 201610798856A CN 106407915 A CN106407915 A CN 106407915A
- Authority
- CN
- China
- Prior art keywords
- face
- characteristic
- parameter
- characteristic parameter
- module
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Probability & Statistics with Applications (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention discloses an SVM (support vector machine)-based face recognition method. The method includes the following steps that: (a) a face photograph containing abnormal skin is acquired from a network; (b) image preprocessing is carried out; (c) the abnormal skin of a face is separated through the K-means algorithm; (d) abnormal skin feature parameters are extracted, so that a simple feature space can be obtained; (e) training is performed, so that an SVM classifier can be obtained; (f) an optimal feature space is formed; and (g) a face recognition device is adopted to carry out recognition and classification. Thus, the SVM algorithm can be used to transform a problem into a quadratic optimization problem so as to obtain a global optimal value, and an unavoidable local extreme value problem of a neural network algorithm can be solved; the generalization ability of the SVM algorithm is higher than that of the neural network; and the recognition rate of each feature parameter in the SVM classifier for a sample classification result is analyzed, feature parameters of which the contribution rates are positive are selected as the input items of the SVM classifier, and therefore, classification accuracy can be improved.
Description
Technical field
The invention belongs to characteristics of image technology of identification field is and in particular to a kind of face characteristic recognition methodss based on SVM
And device.
Background technology
In recent years, with communication technology, the Internet, cloud computing fast development, image procossing and characteristics of image are identified as
For the hot spot technology of domestic and international research, face characteristic identification is an important topic of characteristics of image identification, widely should have
With field, how quickly to identify that face becomes the actual and urgent demand of people.
In view of drawbacks described above, creator of the present invention passes through long research and practice obtains the present invention finally.
Content of the invention
For solving above-mentioned technological deficiency, the technical solution used in the present invention is, provide a kind of based on support vector machine
Face characteristic recognition methodss, it includes：
Step a, obtains the various human face photos comprising improper skin from network；
Step b, Image semantic classification, face part in photo is obtained by Face datection algorithm, intercepts the improper skin of face
Skin region simultaneously arranges picture size and resolution；
The improper skin of face is carried out separating by step c by K-means algorithm；
Step d, extracts color, shape and the textural characteristics parameter of improper skin speckle and melanotic nevus, to characteristic parameter
It is normalized operation, and reduces parameter dimension, obtain brief feature space；
Step e, carries out off-line training using SVM to the sample photo of input, and improper skin is divided into speckle and black
Nevuss, training obtains SVM classifier；
Step f, according to the contribution rate to recognition result for each characteristic parameter, selects the higher set of characteristic parameters of contribution rate
Composition optimal characteristics space；
Step g, is obtained test sample photo, execution step b～step d, using face identification device, it is identified
Classification, to check the speed of service and the accuracy in detection of this device.
Preferably, described step c includes
Step c1, given sample packages contain N number of sample space data set, if iterationses are R it is intended that cluster numbers are K, at random
Generate K pixel as initial cluster center；
Step c2, calculates the similarity distance of each data object and current cluster centre Y (k, r) in sample aperture；
Step c3, calculates K new cluster centre, and formula is as follows：
Whether rationally step c4, judge cluster, whether rationally to judge cluster with equation below,
| E (r+1)-E (r) | ＜ ε.
Preferably, described step d includes：
Step d1：Extract parameters for shape characteristic；
Step d2：Extract Color characteristics parameters；
Step d3：Texture feature extraction parameter；
Step d4：Characteristic parameter vector data is normalized with operation, unified dimension and parameter distribution are interval；
Step d5：Dimension-reduction treatment is carried out to characteristic parameter vector space；
Step d6：Obtain brief effective characteristic parameter space.
Preferably, SVM classifier is Radial basis kernel function in described step e, its function expression is as follows：
Preferably, in described step f, giving up the characteristic parameter item of negative contribution rate and zero contribution rate, filter out best features
Parameter combination, as the input item of SVM classifier.
A kind of and above-described corresponding device of face characteristic recognition methodss based on support vector machine, it includes：
Obtain human face photo module, from the various human face photo comprising improper skin of Network Capture；
Image pre-processing module, improper region picture attribute is set on the human face photo that intercept network obtains；
Separate face improper skin module, using K-means algorithm according to similarity size by the improper skin of face
Skin carries out separating：
Extract improper skin characteristic parameter module, extract the characteristic parameter of speckle and melanotic nevus, generate brief effective spy
Levy parameter space：
Training sample photo module, carries out off-line training using SVM to the sample photo of input, improper skin is divided into
Speckle and melanotic nevus, training obtains SVM classifier；
Optimal characteristics space generation module, according to the contribution rate to recognition result for each characteristic parameter, selects contribution rate relatively
High set of characteristic parameters composition optimal characteristics space；
Face recognition module, obtains test sample photo, execution image pre-processing module, separation face improper skin mould
Block and improper skin characteristic parameter extraction module, are identified to it classifying using face identification device, to check this device
The speed of service and accuracy in detection.
Preferably, the described face's improper skin module that separates includes：
Initial cluster center generates submodule, and given sample packages contain N number of sample space data set, if iterationses are R, refers to
Determining cluster numbers is K, generates K pixel at random as initial cluster center；
Calculate similarity apart from submodule, calculate each data object and current cluster centre Y (k, r) in sample aperture
Similarity distance；
New cluster centre calculating sub module, calculates K new cluster centre with following computing formula：
Whether rationally cluster judging submodule, judge cluster with equation below,
| E (r+1)-E (r) | ＜ ε
Preferably, the improper skin characteristic parameter module of described extraction includes：
Extract parameters for shape characteristic submodule, extract the characteristic parameter of speckle and melanotic nevus：Area, girth, circular similarity；
Extract Color characteristics parameters submodule, extract RGB color system in tri- component gray values of R, G, B low-order moment and
Color component ratio, also in hsv color system tri- component gray values of H, S, V low-order moment；
Texture feature extraction parameter sub-module, the characteristic parameter of speckle and melanotic nevus in texture feature extraction：RGB color system
The contrast of gray level co-occurrence matrixes, dependency and energy；
Characteristic parameter vector data is normalized operation submodule, carries out unifying dimension and parameter distribution is interval；
Characteristic parameter vector space carries out dimension-reduction treatment submodule, carries out dimension-reduction treatment to characteristic parameter vector space；
Characteristic parameter space generates submodule, generates brief effective characteristic parameter space.
Preferably, the kernel function that in described training sample pictures module, SVM classifier uses is Radial basis kernel function,
Its function expression is as follows：
Preferably, in the generation module of described optimal characteristics space, giving up the characteristic parameter item of negative contribution rate and zero contribution rate,
Filter out best features parameter combination, as the input item of SVM classifier.
Compared with the prior art the beneficial effects of the present invention is：Based on the face characteristic recognition methodss of SVM, using SVM
Problem is converted into quadratic form optimization problem by algorithm, obtains global optimum, solves the office that neural network algorithm is difficult to avoid that
Portion's extreme-value problem；SVM algorithm has strict theory and Fundamentals of Mathematics, and based on structural risk minimization, generalization ability is higher than
Neutral net；By analyzing the discrimination to sample classification result for each characteristic parameter in SVM classifier, screening wherein contribution rate
For positive characteristic parameter as the input item of SVM classifier, improve classification accuracy.
Brief description
For the technical scheme being illustrated more clearly that in various embodiments of the present invention, below will be to required in embodiment description
The accompanying drawing using is briefly described.
Fig. 1 is a kind of flow chart of face characteristic recognition methodss based on SVM of the embodiment of the present invention one；
Fig. 2 is a kind of flow chart of step b of face characteristic recognition methodss based on SVM of the embodiment of the present invention two；
Fig. 3 is a kind of flow chart of step c of face characteristic recognition methodss based on SVM of the embodiment of the present invention three；
Fig. 4 is a kind of flow chart of step d of face characteristic recognition methodss based on SVM of the embodiment of the present invention four；
Fig. 5 is a kind of functional schematic of face characteristic identifying device based on SVM of the embodiment of the present invention seven.
Specific embodiment
Below in conjunction with accompanying drawing, the above-mentioned He other technical characteristic of the present invention and advantage are described in more detail.
Embodiment one
A kind of face characteristic recognition methodss Ji Yu support vector machine (SVM) of the present invention, using " off-line training, online inspection
Survey " principle, the face sample photo obtaining various spottiness and melanotic nevus first is it is ensured that photo just comprises face, using K-
Means clustering algorithm is partitioned into the speckle of face and the region of melanotic nevus；Then the shape of extraction spotting out and melanotic nevus, face
Multiple characteristic parameters of normal complexion texture, and characteristic parameter is normalized with operation and dimension-reduction treatment；Finally using SVM to sample
It is identified classification based training, according to the contribution rate to recognition result for the sample characteristics parameter, select best features parameter space, from
And obtain the higher face characteristic recognition methodss of accuracy rate.
As shown in figure 1, being a kind of flow chart of the face characteristic recognition methodss Ji Yu support vector machine (SVM), wherein, institute
State and included based on the face characteristic recognition methodss of support vector machine (SVM)：
Step a：The various human face photos comprising improper skin (speckle and melanotic nevus) are obtained from network；
Step b：Image semantic classification, obtains face part in photo by Face datection algorithm, intercepts the improper skin of face
Skin region simultaneously arranges picture size and resolution；
Step c：By K-means algorithm, the improper skin speckle of face and melanotic nevus are carried out separating；
Mainly due to similarity between the color of the normal skin of face very close to, normal skin and speckle, melanotic nevus it
Between similarity gap remote it is possible to two taxonomic clusterings be carried out by K-means algorithm with the method for similarity measurement, realization
Face speckle and the separation of melanotic nevus；
Step d：Extract color, shape and the textural characteristics parameter of improper skin speckle and melanotic nevus, to characteristic parameter
It is normalized operation, and reduces parameter dimension, obtain brief feature space；
Step e：Using SVM, off-line training is carried out to the sample photo of input, improper skin is divided into speckle and melanotic nevus,
Training obtains SVM classifier；
Step f：According to the contribution rate to recognition result for each characteristic parameter, select the higher set of characteristic parameters of contribution rate
Composition optimal characteristics space；
Step g：Obtain test sample photo, execution step b～step d, using face identification device, it is identified
Classification, to check the speed of service and the accuracy in detection of this device.
Based on the face characteristic recognition methodss of SVM, using SVM algorithm, problem is converted into quadratic form optimization problem, obtains
Global optimum, solves the local extremum problem that neural network algorithm is difficult to avoid that；SVM algorithm has strict theoretical sum
Learn basis, based on structural risk minimization, generalization ability is higher than neutral net；By analyzing each spy in SVM classifier
Levy the discrimination to sample classification result for the parameter, screening wherein contribution rate is the input as SVM classifier for the positive characteristic parameter
, improve classification accuracy.
Embodiment two
Face characteristic recognition methodss based on SVM as above, the present embodiment is different from part and is, such as Fig. 2 is originally
Shown in the flow chart based on face characteristic recognition methodss step b of SVM for the invention, wherein, step b includes：
Step b1：The human face photo that input obtains；
Step b2：Face local is obtained by Face datection algorithm, intercepts face part；
Step b3：Set size and the resolution of picture.
Embodiment three
Face characteristic recognition methodss Ji Yu support vector machine (SVM) as above, the present embodiment is different from part
It is, as shown in the flow chart of face characteristic recognition methodss step c Ji Yu support vector machine (SVM) for Fig. 3 present invention, wherein,
Step c includes：
Step c1：Given sample packages contain N number of sample space data set, if iterationses are R it is intended that cluster numbers are K, at random
Generate K pixel as initial cluster center Y (k, r), wherein k=1,2 ..., K；R=1,2 ..., R；
Step c2：The similarity calculating each data object and current cluster centre Y (k, r) in sample aperture is apart from D
(xn,Y(k,r))；Wherein, n=1,2 ..., N, form cluster Wk, meet following formula,
Then xn∈Wk, xnIt is designated as ω, wherein ε is any positive number setting；
In order to reduce amount of calculation, using squared euclidean distance as similarity distance：
D2(X, Y)=(x-x0)2+(y-y0)2
Wherein X (x, y) is the coordinate of arbitrary data object, Y (x0,y0) it is its place cluster centre coordinate；
Step c3：Calculate K new cluster centre, computing formula is as follows：
Wherein k represents place cluster, and n represents the data object number that this cluster comprises；
Step c4：Whether rationally to judge cluster, discrimination formula is as follows：
| E (r+1)-E (r) | ＜ ε
| E (r+1)-E (r) | ＜ ε
The computing formula of clustering criteria functional value is as follows：
If cluster rationally, stops iteration, otherwise continues executing with step c2, c3 and c4.
Example IV
Face characteristic recognition methodss based on SVM as above, the present embodiment is different from part and is, such as Fig. 4 is originally
Shown in the flow chart of face characteristic recognition methodss step d Ji Yu support vector machine (SVM) for the invention, wherein, step d includes：
Step d1：Extract parameters for shape characteristic；
The characteristic parameter of the speckle in shape facility and melanotic nevus comprises area, girth, circular similarity.
The sum of area=this area pixel point
Girth=this regional edge is along the pixel number of contour line
Step d2：Extract Color characteristics parameters；
The characteristic parameter of the speckle in color characteristic and melanotic nevus includes tri- component gray values of R, G, B in RGB color system
Low-order moment and color component ratio, also in hsv color system tri- component gray values of H, S, V low-order moment.
The first moment of RGB color system and hsv color system reflects the color average in this region, and computing formula is as follows：
The second moment of RGB color system and hsv color system reflects that the color standard in this region is poor, and computing formula is as follows：
In RGB color system, R, G, B color component is respectively r, g, b, and shared ratio calculation formula is as follows：
R=R/ (R+G+B)
G=G/ (R+G+B)
B=B/ (R+G+B)
On expression Color Range, Lab color space is better than RGB color, therefore turns image from RGB color
Change on Lab color space, conversion formula is as follows：
Step d3：Texture feature extraction parameter；
In textural characteristics, the characteristic parameter of speckle and melanotic nevus has the contrast of RGB color system gray level co-occurrence matrixes, correlation
Property and energy；
The contrast reflection clean mark degree of gray level co-occurrence matrixes, numerical value is bigger, and texture is deeper, and computing formula is as follows：
The energy reflection gradation of image of gray level co-occurrence matrixes is evenly distributed degree and texture fineness, and computing formula is such as
Under：
The dependency of gray level co-occurrence matrixes weigh be expert at or column direction on similarity degree, reflection local gray level is related
Property, computing formula is as follows：
The characteristic parameter extracting is needed to be shown in Table 1 in step d1, d2 and d3：
The characteristic parameter that table 1 extracts
Step d4：Characteristic parameter vector data is normalized with operation, unified dimension and parameter distribution are interval；
For the statistical distribution of unified samples, eliminate dimension difference, accelerate the convergence of data training, need to feature
The normalization operation of parameter space, characteristic parameter is distributed between 0 to 1.Normalization computing formula is as follows：
Wherein xmaxRepresent greatest measure in this vector, xminRepresent minimum value in this vector, xinBefore representing normalization
Numerical value, xoutRepresent the numerical value after normalization.
Step d5：Dimension-reduction treatment is carried out to characteristic parameter vector space, it is to avoid " dimension disaster "；
Step d6：Terminate, obtain brief effective characteristic parameter space.
Embodiment five
Face characteristic recognition methodss based on SVM as above, the present embodiment is different from part and is,
In step e, SVM classifier commonly use the linear kernel function of kernel function, Polynomial kernel function, Radial basis kernel function and
It is contemplated that Radial basis kernel function classification performance is higher and more stable, the present invention uses radial direction base core letter to sigmoid kernel function
Number, its function expression is as follows：
Embodiment six
Face characteristic recognition methodss based on SVM as above, the present embodiment is different from part and is,
In step f, each characteristic parameter item has corresponding discrimination as input item, and the discrimination of different characteristic has difference
Different, give up the characteristic parameter item of negative contribution rate and zero contribution rate, thus filtering out best features parameter combination, as svm classifier
The input item of device.
Embodiment seven
The present embodiment is a kind of face characteristic identifying device based on SVM, and it is known with the described face characteristic based on SVM
Other method is corresponding, as shown in figure 5, it includes：
Obtain human face photo module 1, various comprise improper skin (speckle and black from network as obtained Baidu's picture
Nevuss) human face photo；
Image pre-processing module 2, by improper region on the human face photo of Face datection algorithm intercept network acquisition simultaneously
Setting picture size and resolution；
Separate face improper skin module 3, using K-means algorithm according to similarity size by the improper skin of face
Skin (as speckle and melanotic nevus) carries out separating：
Extract improper skin characteristic parameter module 4, extract color, shape and the textural characteristics parameter of speckle and melanotic nevus,
Generate brief effective characteristic parameter space：
Training sample photo module 5, carries out off-line training using SVM to the sample photo of input, improper skin is divided
For speckle and melanotic nevus, train and obtain SVM classifier；
Optimal characteristics space generation module 6, according to the contribution rate to recognition result for each characteristic parameter, selects contribution rate relatively
High set of characteristic parameters composition optimal characteristics space；
Face recognition module 7, obtains test sample photo, execution image pre-processing module, the improper skin of separation face
Module and improper skin characteristic parameter extraction module, are identified to it classifying using face identification device, to check this dress
The speed of service put and accuracy in detection.
Wherein, described image pretreatment module includes：
Human face photo input submodule, the human face photo that input obtains；
Face portion intercepts submodule, obtains face local by Face datection algorithm, intercepts face part；
Set face picture attribute submodule, set size and the resolution of picture.
Wherein, the described face's improper skin module that separates includes：
Initial cluster center generates submodule 31, and given sample packages contain N number of sample space data set, if iterationses are
R, it is intended that cluster numbers are K, generates K pixel at random as initial cluster center Y (k, r), wherein k=1,2 ..., K；R=1,
2,…,R；
Calculate similarity apart from submodule 32, calculate each data object and current cluster centre Y (k, r) in sample aperture
Similarity apart from D (xn,Y(k,r))；Wherein, n=1,2 ..., N, form cluster Wk, meet following formula,
Then xn∈Wk, xnIt is designated as ω, wherein ε is any positive number setting；
In order to reduce amount of calculation, using squared euclidean distance as similarity distance, formula is as follows：
D2(X, Y)=(x-x0)2+(y-y0)2
Wherein X (x, y) is the coordinate of arbitrary data object, Y (x0,y0) it is its place cluster centre coordinate；
New cluster centre calculating sub module 33, calculates K new cluster centre with following computing formula：
Wherein k represents place cluster, and n represents the data object number that this cluster comprises；
Whether rationally cluster judging submodule 34, judge cluster with equation below,
| E (r+1)-E (r) | ＜ ε
Wherein, the computing formula of clustering criteria functional value is as follows：
If cluster is rationally, stop iteration, otherwise continue to call initial cluster center to generate submodule., similarity distance
Computing module, new cluster centre computing module and cluster judge module.
Wherein, the improper skin characteristic parameter module 4 of described extraction includes：
Extract parameters for shape characteristic submodule 41, extract the characteristic parameter of speckle and melanotic nevus：Area, girth, circular similar
Degree；
Extract Color characteristics parameters submodule 42, extract the low order of tri- component gray values of R, G, B in RGB color system
Square and color component ratio, also in hsv color system tri- component gray values of H, S, V low-order moment；
Texture feature extraction parameter sub-module 43, the characteristic parameter of speckle and melanotic nevus in texture feature extraction：RGB color body
It is contrast, dependency and the energy of gray level co-occurrence matrixes；The contrast reflection clean mark degree of gray level co-occurrence matrixes, numerical value
Bigger, texture is deeper；The energy reflection gradation of image of gray level co-occurrence matrixes is evenly distributed degree and texture fineness；Gray scale is altogether
The dependency of raw matrix weigh be expert at or column direction on similarity degree, reflect local gray level dependency；
Characteristic parameter vector data is normalized operation submodule 44, carries out unifying dimension and parameter distribution is interval；
Characteristic parameter vector space carries out dimension-reduction treatment submodule 45, carries out dimension-reduction treatment to characteristic parameter vector space,
Avoid " dimension disaster ".
Characteristic parameter space generates submodule 46, generates brief effective characteristic parameter space.
Wherein, in training sample pictures module, the conventional linear kernel function of kernel function of SVM classifier, multinomial
Kernel function, Radial basis kernel function and sigmoid kernel function are it is contemplated that Radial basis kernel function classification performance is higher and relatively more steady
Fixed, the present invention uses Radial basis kernel function, and its function expression is as follows：
Wherein, in the generation module of optimal characteristics space, each characteristic parameter item has corresponding discrimination as input item,
The discrimination of different characteristic is variant, gives up the characteristic parameter item of negative contribution rate and zero contribution rate, thus filtering out best features
Parameter combination, as the input item of SVM classifier.
Based on the face characteristic identifying device of SVM, using SVM algorithm, problem is converted into quadratic form optimization problem, obtains
Global optimum, solves the local extremum problem that neural network algorithm is difficult to avoid that；SVM algorithm has strict theory and mathematics
Basis, based on structural risk minimization, generalization ability is higher than neutral net；By analyzing each feature in SVM classifier
The discrimination to sample classification result for the parameter, screening wherein contribution rate is the input item as SVM classifier for the positive characteristic parameter,
Improve classification accuracy.
The foregoing is only presently preferred embodiments of the present invention, be merely illustrative for the purpose of the present invention, and non-limiting
's.Those skilled in the art understands, it can be carried out in the spirit and scope that the claims in the present invention are limited with many changes,
Modification, in addition equivalent, but fall within protection scope of the present invention.
Claims (10)
1. a kind of face characteristic recognition methodss based on support vector machine are it is characterised in that include：
Step a, obtains the various human face photos comprising improper skin from network；
Step b, Image semantic classification, face part in photo is obtained by Face datection algorithm, intercepts the improper skin region of face
Domain simultaneously arranges picture size and resolution；
The improper skin of face is carried out separating by step c by K-means algorithm；
Step d, extracts color, shape and the textural characteristics parameter of improper skin speckle and melanotic nevus, characteristic parameter is carried out
Normalization operation, and reduce parameter dimension, obtain brief feature space；
Step e, carries out off-line training using SVM to the sample photo of input, improper skin is divided into speckle and melanotic nevus, training
Obtain SVM classifier；
Step f, according to the contribution rate to recognition result for each characteristic parameter, selects the set of characteristic parameters composition that contribution rate is higher
Optimal characteristics space；
Step g, obtains test sample photo, execution step b～step d, it is identified classify using face identification device,
To check the speed of service and the accuracy in detection of this device.
2. the face characteristic recognition methodss based on support vector machine according to claim 1 are it is characterised in that described step
C includes
Step c1, given sample packages contain N number of sample space data set, if iterationses are R it is intended that cluster numbers are K, random generation
K pixel is as initial cluster center；
Step c2, calculates the similarity distance of each data object and current cluster centre Y (k, r) in sample aperture；
Step c3, calculates K new cluster centre, and formula is as follows：
Whether rationally step c4, judge cluster, whether rationally to judge cluster with equation below,
| E (r+1)-E (r) | ＜ ε.
3. the face characteristic recognition methodss based on support vector machine according to claim 2 are it is characterised in that described step
Rapid d includes：
Step d1：Extract parameters for shape characteristic；
Step d2：Extract Color characteristics parameters；
Step d3：Texture feature extraction parameter；
Step d4：Characteristic parameter vector data is normalized with operation, unified dimension and parameter distribution are interval；
Step d5：Dimension-reduction treatment is carried out to characteristic parameter vector space；
Step d6：Obtain brief effective characteristic parameter space.
4. the face characteristic recognition methodss based on support vector machine according to claim 3 are it is characterised in that described step
In e, SVM classifier is Radial basis kernel function, and its function expression is as follows：
.
5. the face characteristic recognition methodss based on support vector machine according to claim 4 are it is characterised in that described step
In f, give up the characteristic parameter item of negative contribution rate and zero contribution rate, filter out best features parameter combination, as SVM classifier
Input item.
6. a kind of described face characteristic recognition methodss corresponding dress based on support vector machine arbitrary with claim 1-5
Put it is characterised in that described included based on the face characteristic identifying device of support vector machine：
Obtain human face photo module, from the various human face photo comprising improper skin of Network Capture；
Image pre-processing module, improper region picture attribute is set on the human face photo that intercept network obtains；
Separate face's improper skin module, according to similarity size, the improper skin of face is entered using K-means algorithm
Row separates：
Extract improper skin characteristic parameter module, extract the characteristic parameter of speckle and melanotic nevus, generate brief effective feature ginseng
Number space：
Training sample photo module, carries out off-line training using SVM to the sample photo of input, improper skin is divided into speckle
And melanotic nevus, train and obtain SVM classifier；
Optimal characteristics space generation module, according to the contribution rate to recognition result for each characteristic parameter, selects contribution rate higher
Set of characteristic parameters forms optimal characteristics space；
Face recognition module, obtain test sample photo, execution image pre-processing module, separate face's improper skin module and
Improper skin characteristic parameter extraction module, is identified to it classifying using face identification device, to check the fortune of this device
Scanning frequency degree and accuracy in detection.
7. the face characteristic identifying device based on support vector machine according to claim 6 is it is characterised in that described separation
Face's improper skin module includes：
Initial cluster center generates submodule, and given sample packages contain N number of sample space data set, if iterationses are R it is intended that gathering
Class number is K, generates K pixel at random as initial cluster center；
Calculate similarity apart from submodule, calculate the similar of each data object and current cluster centre Y (k, r) in sample aperture
Degree distance；
New cluster centre calculating sub module, calculates K new cluster centre with following computing formula：
Whether rationally cluster judging submodule, judge cluster with equation below,
| E (r+1)-E (r) | ＜ ε.
8. the face characteristic identifying device based on support vector machine according to claim 7 is it is characterised in that described extraction
Improper skin characteristic parameter module includes：
Extract parameters for shape characteristic submodule, extract the characteristic parameter of speckle and melanotic nevus：Area, girth, circular similarity；
Extract Color characteristics parameters submodule, extract the low-order moment of tri- component gray values of R, G, B and color in RGB color system
Component ratio, also in hsv color system tri- component gray values of H, S, V low-order moment；
Texture feature extraction parameter sub-module, the characteristic parameter of speckle and melanotic nevus in texture feature extraction：RGB colour system ash
The contrast of degree co-occurrence matrix, dependency and energy；
Characteristic parameter vector data is normalized operation submodule, carries out unifying dimension and parameter distribution is interval；
Characteristic parameter vector space carries out dimension-reduction treatment submodule, carries out dimension-reduction treatment to characteristic parameter vector space；
Characteristic parameter space generates submodule, generates brief effective characteristic parameter space.
9. the face characteristic identifying device based on support vector machine according to claim 8 is it is characterised in that described training
The kernel function that in sample pictures module, SVM classifier uses is Radial basis kernel function, and its function expression is as follows：
.
10. the face characteristic identifying device based on support vector machine according to claim 9 it is characterised in that described
In excellent feature space generation module, give up the characteristic parameter item of negative contribution rate and zero contribution rate, filter out best features parameter group
Close, as the input item of SVM classifier.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610798856.7A CN106407915A (en) | 2016-08-31 | 2016-08-31 | SVM (support vector machine)-based face recognition method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610798856.7A CN106407915A (en) | 2016-08-31 | 2016-08-31 | SVM (support vector machine)-based face recognition method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106407915A true CN106407915A (en) | 2017-02-15 |
Family
ID=58002113
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610798856.7A Pending CN106407915A (en) | 2016-08-31 | 2016-08-31 | SVM (support vector machine)-based face recognition method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106407915A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107967461A (en) * | 2017-12-08 | 2018-04-27 | 深圳云天励飞技术有限公司 | The training of SVM difference models and face verification method, apparatus, terminal and storage medium |
| CN108171650A (en) * | 2017-12-14 | 2018-06-15 | 华侨大学 | One kind has the function of that stroke optimizes Chinese flower ink and wash style stroke generation method |
| CN108460264A (en) * | 2018-03-03 | 2018-08-28 | 程爱军 | Relating computer file content maintaining method |
| CN108804929A (en) * | 2018-03-03 | 2018-11-13 | 程爱军 | Relating computer file content maintenance system |
| CZ307563B6 (en) * | 2017-08-25 | 2018-12-05 | Certicon A.S. | A method of dynamic distribution of computing processes to one or more graphics processing units (GPU) |
| CN109325468A (en) * | 2018-10-18 | 2019-02-12 | 广州智颜科技有限公司 | A kind of image processing method, device, computer equipment and storage medium |
| CN109753938A (en) * | 2019-01-10 | 2019-05-14 | 京东方科技集团股份有限公司 | Image-recognizing method and equipment and the training method of application, neural network |
| CN109961426A (en) * | 2019-03-11 | 2019-07-02 | 西安电子科技大学 | A kind of detection method of face skin skin quality |
| CN109977803A (en) * | 2019-03-07 | 2019-07-05 | 北京超维度计算科技有限公司 | A kind of face identification method based on Kmeans supervised learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745204A (en) * | 2014-01-17 | 2014-04-23 | 公安部第三研究所 | Method of comparing physical characteristics based on nevus spilus points |
| CN104318262A (en) * | 2014-09-12 | 2015-01-28 | 上海明穆电子科技有限公司 | Method and system for replacing skin through human face photos |
-
2016
- 2016-08-31 CN CN201610798856.7A patent/CN106407915A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103745204A (en) * | 2014-01-17 | 2014-04-23 | 公安部第三研究所 | Method of comparing physical characteristics based on nevus spilus points |
| CN104318262A (en) * | 2014-09-12 | 2015-01-28 | 上海明穆电子科技有限公司 | Method and system for replacing skin through human face photos |
Non-Patent Citations (2)
| Title |
|---|
| 吕峰等: "支持向量机在皮肤症状图像识别中的应用研究", 《计算机仿真》 * |
| 谢凤英等: "基于组合BP神经网络的皮肤肿瘤目标识别", 《中国体视学与图像分析》 * |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CZ307563B6 (en) * | 2017-08-25 | 2018-12-05 | Certicon A.S. | A method of dynamic distribution of computing processes to one or more graphics processing units (GPU) |
| CN107967461B (en) * | 2017-12-08 | 2020-07-24 | 深圳云天励飞技术有限公司 | SVM (support vector machine) differential model training and face verification method, device, terminal and storage medium |
| CN107967461A (en) * | 2017-12-08 | 2018-04-27 | 深圳云天励飞技术有限公司 | The training of SVM difference models and face verification method, apparatus, terminal and storage medium |
| CN108171650A (en) * | 2017-12-14 | 2018-06-15 | 华侨大学 | One kind has the function of that stroke optimizes Chinese flower ink and wash style stroke generation method |
| CN108171650B (en) * | 2017-12-14 | 2021-08-24 | 华侨大学 | Chinese flower water-ink painting style stroke generation method with stroke optimization function |
| CN108804929A (en) * | 2018-03-03 | 2018-11-13 | 程爱军 | Relating computer file content maintenance system |
| CN108460264A (en) * | 2018-03-03 | 2018-08-28 | 程爱军 | Relating computer file content maintaining method |
| CN109325468A (en) * | 2018-10-18 | 2019-02-12 | 广州智颜科技有限公司 | A kind of image processing method, device, computer equipment and storage medium |
| CN109753938A (en) * | 2019-01-10 | 2019-05-14 | 京东方科技集团股份有限公司 | Image-recognizing method and equipment and the training method of application, neural network |
| CN109753938B (en) * | 2019-01-10 | 2021-11-05 | 京东方科技集团股份有限公司 | Image recognition method and device, application of image recognition method and device, and training method of neural network |
| US11244151B2 (en) | 2019-01-10 | 2022-02-08 | Boe Technology Group Co., Ltd. | Computer-implemented method of recognizing facial expression, apparatus for recognizing facial expression, method of pre-training apparatus for recognizing facial expression, computer-program product for recognizing facial expression |
| CN109977803A (en) * | 2019-03-07 | 2019-07-05 | 北京超维度计算科技有限公司 | A kind of face identification method based on Kmeans supervised learning |
| CN109961426A (en) * | 2019-03-11 | 2019-07-02 | 西安电子科技大学 | A kind of detection method of face skin skin quality |
| CN109961426B (en) * | 2019-03-11 | 2021-07-06 | 西安电子科技大学 | Method for detecting skin of human face |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106407915A (en) | SVM (support vector machine)-based face recognition method and device | |
| He et al. | A fully convolutional neural network for wood defect location and identification | |
| US11417148B2 (en) | Human face image classification method and apparatus, and server | |
| Yi et al. | Text detection in natural scene images by stroke gabor words | |
| Varish et al. | Image retrieval scheme using quantized bins of color image components and adaptive tetrolet transform | |
| CN105005765B (en) | A kind of facial expression recognizing method based on Gabor wavelet and gray level co-occurrence matrixes | |
| CN106228137A (en) | A kind of ATM abnormal human face detection based on key point location | |
| CN109948566B (en) | Double-flow face anti-fraud detection method based on weight fusion and feature selection | |
| CN107862267A (en) | Face recognition features' extraction algorithm based on full symmetric local weber description | |
| CN105868796B (en) | The design method of linear discriminant rarefaction representation classifier based on nuclear space | |
| CN108960260B (en) | Classification model generation method, medical image classification method and medical image classification device | |
| CN105956570B (en) | Smiling face's recognition methods based on lip feature and deep learning | |
| CN115375690B (en) | Classification and identification method for greasy tongue coating | |
| CN110688888B (en) | Pedestrian attribute identification method and system based on deep learning | |
| CN110472652A (en) | A small amount of sample classification method based on semanteme guidance | |
| CN109740417A (en) | Invoice type recognition methods, device, storage medium and computer equipment | |
| CN110826408A (en) | Face recognition method by regional feature extraction | |
| Anggraini | Face recognition using principal component analysis and self organizing maps | |
| WO2020119624A1 (en) | Class-sensitive edge detection method based on deep learning | |
| CN114863125A (en) | Intelligent scoring method and system for calligraphy/fine art works | |
| CN110516661A (en) | U.S. pupil detection method and device applied to iris recognition | |
| CN109740429A (en) | Smiling face's recognition methods based on corners of the mouth coordinate mean variation | |
| CN109902692A (en) | A kind of image classification method based on regional area depth characteristic coding | |
| Lee et al. | Intelligent classification methods of grain kernels using computer vision analysis | |
| Qiao | Application of Gabor image recognition technology in intelligent clothing design |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170215 |
|
| RJ01 | Rejection of invention patent application after publication |