CN109948519A - It is a kind of based on the sparse face group recognition methods with weighted residual in region - Google Patents
It is a kind of based on the sparse face group recognition methods with weighted residual in region Download PDFInfo
- Publication number
- CN109948519A CN109948519A CN201910201677.4A CN201910201677A CN109948519A CN 109948519 A CN109948519 A CN 109948519A CN 201910201677 A CN201910201677 A CN 201910201677A CN 109948519 A CN109948519 A CN 109948519A
- Authority
- CN
- China
- Prior art keywords
- sample
- region
- face
- test sample
- feature
- 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
- 238000000034 method Methods 0.000 title claims abstract description 31
- 238000012360 testing method Methods 0.000 claims abstract description 50
- 238000012549 training Methods 0.000 claims abstract description 30
- 230000001815 facial effect Effects 0.000 claims abstract description 25
- 238000005286 illumination Methods 0.000 claims abstract description 20
- 238000013480 data collection Methods 0.000 claims abstract description 10
- 230000008569 process Effects 0.000 claims abstract description 7
- 239000003086 colorant Substances 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 239000000203 mixture Substances 0.000 claims description 2
- 241000208340 Araliaceae Species 0.000 claims 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 claims 1
- 235000003140 Panax quinquefolius Nutrition 0.000 claims 1
- 210000004709 eyebrow Anatomy 0.000 claims 1
- 235000008434 ginseng Nutrition 0.000 claims 1
- 239000011159 matrix material Substances 0.000 description 8
- 238000005516 engineering process Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 239000000284 extract Substances 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 210000000887 face Anatomy 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000002156 mixing Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Abstract
The invention discloses a kind of based on the sparse face group recognition methods with weighted residual in region, specifically includes the following steps: step 1, national human face data collection of the input with national label；Step 2, the national human face data collection of input is subjected to facial image registration process；Step 3, the national human face data collection of input is subjected to facial image illumination normalized；Step 4, whole face geometrical combination features are screened, filters out the region T；Step 5, the region T of original sample concentration training sample and test sample image is extracted respectively；Step 6, face group recognizer model is constructed；Step 7, the region T of the training sample in step 5 and test sample image is input in step 6 in face group recognizer model, according to the output of model, obtains the classification results of test sample.Test sample is carried out rarefaction representation on multiple random spaces using multiple accidental projection method by the application, and weighted residual is more stable, and False Rate is small.
Description
Technical field
It is specifically a kind of to be known based on the sparse face group with weighted residual in region the invention belongs to field of face identification
Other method.
Background technique
China is a multi-ethnic country, and the research for different nationalities face characteristic can not only recognize well
The process they different nationalities procreation and evolved, more can be saved the face characteristic of different nationalities by technological means, be the people
The wealth of next record preciousness is stayed in race, anthropology correlative study in future.The identification of face nationality can also expand simultaneously and enrich existing
Face datection and field of face identification.
Face recognition technology is to pass through and calculate as the identification technology based on physiological characteristic a kind of in field of biological recognition
Machine extracts the feature of face, and a kind of technology of authentication is carried out according to these features.Face and fingerprint, iris, sound etc.
It is equally inherent, uniqueness possessed by them and be not easy the superperformance being replicated be identity identify provide it is necessary
Premise；Compared with other biological identification technology, there is face recognition technology easy to operate, visual result, concealment to avoid directly
The superiority of contact.In recent years, group's analysis of face is increasingly becoming one of field of face identification hot spot with identification.Permitted both at home and abroad
More scholars are from different approaches such as geometrical characteristic, features of skin colors, global feature, local feature and assemblage characteristics to face group spy
Sign has carried out research work within the past some time, and rarefaction representation is initially to be made to solve many field of signal processing
The problem of, such as image denoising, compression of images and image recovery etc., pass through the exploration of numerous researchers in recent years, finds sparse
Indicate that all there is preferable application in terms of recognition of face, expression and age identification.
The rarefaction representation of face is based on illumination model.That is a facial image can use same person institute in database
The linear combination of somebody's face image indicates.And for the face of people other in database, the coefficient of linear combination is theoretically
Zero.Due to generally there is multiple images of very multiple and different faces in database, if the line of image all in database
Property combination to indicate that this given test face, coefficient vector are sparse.Because in addition to the people of this and the same person
Outside face image combination coefficient is not zero, other coefficients are all zero.The rarefaction representation for how defining image is exactly to this image
Using the expression most approached based on sparsity, while guaranteeing its parsimony.Rarefaction representation has identification, it is meant that it can be with
A kind of expression coefficient the most sparse is selected to express original image well, exactly because this characteristic, leads in recognition of face
Domain, rarefaction representation using extremely successful.In the classification method based on rarefaction representation, we pass through from training sample concentration
Acquistion is to an excessively complete dictionary, atom in the training sample i.e. dictionary, includes a certain classification in the same linear subspaces
All training samples.The purpose of rarefaction representation is intended to allow certain class testing sample only by the training sample similar with its Lai linear
It indicates, is in other words exactly the only expression system of the training sample generic with test sample in the coefficient matrix of rarefaction representation
Number is not zero, other are zero.
The rarefaction representation of traditional face refers to linear group directly by test sample global feature for all training samples
It closes, and the size according to difference between the weighted sum test sample of all training samples of each classification, passes through minimal reconstruction residual error
Judge test sample generic, but under non-ideal conditions, the face of especially same people is under different illumination conditions
Difference is possible to be greater than difference of inhomogeneity face under the conditions of same light is shone, to belong to the sample of the i-th class in other classifications
On non-zero code coefficient can greatly increase, rely solely at this time single reconstructed residual be difficult to carry out face sample it is accurate
Judgement, in addition the influence of the disturbing factors such as noise, further increases the non-zero code coefficient in other classifications, therefore increase knowledge
False Rate when other.
Summary of the invention
To solve the prior art there are the above problem, the application provides a kind of based on the sparse face with weighted residual in region
Test sample is carried out rarefaction representation on multiple random spaces using multiple accidental projection method by group's recognition methods, weighting
Residual error is more stable, and False Rate is small.
Before group's analysis, need to carry out sample the pretreatment of face alignment and unitary of illumination；Because acquiring
Facial image is influenced by the problems such as posture of people, shooting angle, light, camera distance in journey.Face alignment can
To correct human face region, position is uncertain in the picture, the inconsistent problem of size, the specific steps are as follows:
(1) coordinate of eyes is obtained by eye detection method, eyes coordinate uses E respectively1And ErIt indicates；
(2) it in order to make eyes be maintained at same horizontal line, needs to adjust image direction, can be rotated, make two eye coordinates
ElAnd ErLine ElrWith horizontal direction parallel；
(3) according to eyes distance and facial image other parts proportionate relationship is specified, facial image is cut；
(4) image for cutting previous step image zooms to specified pixel size；
There are the different colours of skin and textures due to geographical location etc. for different groups, using unitary of illumination (SSR) side
Method handles human face data collection；Illumination pretreatment is carried out to the facial image of acquisition, the reflection of image is obtained from image
Property；And assume irradiate light component be spatially smooth and human eye perception object brightness depend on environment illumination
Reflection with body surface to irradiation light.
S (x, y)=R (x, y) L (x, y)
Wherein S (x, y) represents the picture signal that observed or camera receives；L (x, y) represents the sub-irradiation of environment light
Amount；R (x, y) indicates to carry the reflecting component of the target object of image detail information；In order to easily isolate reflecting component and
Irradiation component, both sides take logarithm, then the property that can cast aside incident light obtains the style of object:
R (x, y) corresponds to the high frequency section of image, and L (x, y) corresponds to the low frequency part of image；Using gaussian filtering G
(x, y) estimates illumination component L (x, y) available reflecting component R (x, y):
R (x, y)=log [R (x, y)]=log [S (x, y)]-log [S (x, y) × G (x, y)]
Wherein G (x, y) is surround function, is generally indicated with Gaussian functionC is Gauss ring
Around scale, x is the size of core, and y is that the mean square deviation sigma, K of Gaussian Profile meet the constant of ∫ ∫ F (x, y) dxdy=1；Pass through
SSR algorithm carries out unitary of illumination to face difference group's sample.
The group of face is analyzed in the description of region rapid sparse and weighted residual based on k nearest neighbor, passes through neighbour
Effective selection is carried out to sample, and then constructs group's pedigree expression of individual；The rarefaction representation of face group includes following
Two steps:
The first step determines the k nearest neighbor of test sample from entire training set；
Second step utilizes the k nearest neighbor description determined and class test sample；Based on the description of k nearest neighbor by test sample
It is expressed as the linear combination of its k nearest neighbor；It includes that k nearest neighbor determines step, linear description and classifying step that rapid sparse, which describes method,；
K nearest neighbor determines that step picks out K nearest training sample of distance test sample from training set, and records these training samples
Class label；Assuming that a shared L classification, if a training sample belongs to jth class (j=1,2 ..., L), then by digital j
As its class label；Enabling K neighbour of test sample y is X1..., XKAnd the class label composition collection of these neighbours is combined into C
={ c1,c2,…,cdObviously, the number of element must be less than or equal to L and K in C, and in other words, C is one for gathering { 1,2 ..., L }
Subset；
Test sample: being expressed as the linear combination of K determining neighbour by linear description, and is carried out accordingly to test sample
Classification；The step assumes that following formula approximation is set up,
Y=a1x1+…+akxk
In formula: ai, i=1,2 ..., K are coefficient, and each neighbour contributes the expression of test sample in above formula；Its
In, the contribution margin of i-th of neighbour is aixi；The said firm can be replaced with form the following, be shown below:
Y=XA
A=[a in formula1,……,aK]T, X=[x1,……,xk], it is desirable that there is minimum deflection between XA and test sample y, and
The norm of solution vector A is smaller, using following formula function minimization as objective function, is shown below:
L (A)=| y-X4 ‖2+μ||A‖2=(y-XA)T(y-X4)+μATA
Wherein μ indicates a positive constant；According to Lagrangian method, optimal A meetsTherefore, rapid sparse
The optimal solution of description method is shown below:
A=(XTX+μI)-1XTy
Wherein I indicates unit matrix；
The summation for the contribution expressed from all kinds of k nearest neighbors test sample having determined that is calculated, then according to contribution
Size judges the classification of test sample；If belonging to r in the k nearest neighbor of test image, all neighbours of (r ∈ C) class are
xs..., xr, then there is r class contribution margin to calculate gained by following formula:
gr=asxs+…+atxr,
grDeviation calculation formula as follows between test image:
er=| | y-gr||2, r ∈ c
Work as er=| | y-gr||2Value get over hour, r class expression test sample played in effect it is bigger；Therefore, survey
This y of sample is classified into that maximum class of contribution；But under non-ideal conditions, the face of especially same people is not being shared the same light
Difference according under the conditions of is possible to be greater than difference of inhomogeneity face under the conditions of same light is shone, to belong to the sample of the i-th class
Non-zero code coefficient in other classifications can greatly increase, and rely solely on single reconstructed residual at this time and be difficult to face sample
Accurately judged, in addition the influence of the disturbing factors such as noise, further increases the non-zero code coefficient in other classifications, is increased
False Rate when big identification, in view of the above-mentioned problems, using multiple accidental projection method by test sample in multiple random spaces
Upper carry out rarefaction representation.Accidental projection is a kind of simple and highly efficient space projection method, it is independent of data sample and energy
Enough to retain main information while projecting dimensionality reduction, the intrinsic propesties of the bigger initial data of projected dimensions can more remain.
Random space when due to projecting every time is all different, is equivalent to and implements different feature space mappings to primitive character, and
Since different projector spaces is extracted the feature of facial image with level from different angles, the identification of sample characteristics is enhanced
Ability, this is created condition for the safety of biometric templates；Secondly repeatedly accidental projection enriches face characteristic, obtains
More illumination invariant features, this is created condition for the completeness of biometric templates；It is obtained by accidental projection multiple heavy
Structure residual error, then the weighted residual using test sample in each classification carries out recognition of face, and weighted residual comprehensive utilization is more
The distinguishing ability of a residual error, stability and better reliability.
Specific practice is as follows: assuming that facial image marks human face characteristic point using STASM algorithm, being mentioned by positioning feature point
The sample characteristics collection for taking the "T"-shaped region of face is
X=[X1, X2..., Xk]∈Rp×1
Wherein k is classification number,Indicate the i-th class sample；
It using Normal Distribution and respectively arranges mutually orthogonal random matrix and carries out accidental projection, can effectively keep original
The intrinsic propesties of sample.Therefore setting accidental projection maximum number of iterations is Tn(Tn>=2), if when the t times projection of a generates at random
Matrix Rt∈Rq×p, for some test sample y ∈ Rp, by x and yiIn RtUpper projection can obtain RtRandom spy spatially
Sign, it may be assumed that
Ψt=RtX
yt=Rty
At this timeWhereinIt is the i-th class training sample in RtFeature set spatially, together
Manage ytIt is y in space RtOn feature.
It willAs rarefaction representation institute dictionary,For the sub- dictionary of the i-th class, according to sparse
Representation theory chooses a test sample yt, it is expressed as the linear combination of all training samples as shown by the equation:
Wherein, aI, jFor code coefficient, α={ a1,1, a1,2..., aI, 1, aI, 2..., aK, mIt is coefficient vector.According to dilute
Representation theory is dredged, if y belongs to the i-th class sample in the ideal case, the coefficient in above formula only hasIt is not zero,
Other coefficients are as small as possible or are zero.Usually require that obtain a sparse coefficient vector for meeting above-mentioned conditionL can be used0Norm carrys out approximate solution；
Wherein, λ is Lagrangian constant, for balancing reconstructed error, if the code coefficient vector solved isCoding vector of the sample on the sub- dictionary of the i-th class be
The t times accidental projection is calculated by following formula, reconstructs y using the sub- dictionary of the i-th class and coding vectortIt is residual
Difference:
At this point, the t times accidental projection and rarefaction representation have been completed.
In traditional sparse representation theory, ideally, according to the characteristic of sparse signal representation, if yiFrom i-th
Class sample, then yiIt is expressed as the linear combination of the sub- dictionary atom of the i-th class, bothIt sets up, and sparse coding vector is other
ComponentIt is 0 or is close to 0, so if i-th of residual error is minimum, so that it may by ytIt is determined as the i-th class face.But
It is that under non-ideal conditions, due to can also have certain similitude between different classes of face sample, will increase other
Non-zero code coefficient in classification makes algorithm increase to the False Rate of test sample generic
In iterative optimization procedure, it is assumed that special using the i-th class training sample feature reconstruction test sample after t iteration
The weighted residual of sign is
Wherein i=1,2 ..., K, ω=[ω1, ω2..., ωt] be t accidental projection and rarefaction representation after, add
The weight vector of power t residual error of fusion, s (s=1,2 ..., t) secondary blending weight ωsThe following formula of calculation:
WhereinAfter t iteration, the energy summation for the test sample that all secondary iteration generate, E (ys) it is s
The energy of test sample when secondary iteration.The classification of face sample differentiated by minimum weight residual error,Value get over the i-th class of hour
Group's effect played in expression test sample is bigger, therefore y just belongs to such group.
The present invention due to using the technology described above, can obtain following technical effect: the algorithm is first from original instruction
Practice the region T for extracting sample in sample, then devises the face group recognizer mould based on the sparse description in the neighboring regions k
The training sample of extraction and the region test sample T are finally input in algorithm, take full advantage of the local feature of sample by type.
Using weighted residual to traditional rarefaction representation, classifier is improved using the classification of single reconstructed residual.Weighted residual is more steady
Fixed, False Rate is small, enhances the recognition accuracy of entire algorithm to a certain extent.
Detailed description of the invention
Fig. 1 is a kind of based on the sparse face group recognition methods flow chart with weighted residual in region；
Specific embodiment
The present invention is described in further detail in the following with reference to the drawings and specific embodiments: doing as example to the application
Further description explanation.
The present invention is "T"-shaped by the part to face according to the face group characteristic remarkable region of geometrical characteristic difference weight
Region carries out the identification of group's feature.In the image vector for extracting the region image difference T, described in algorithm using rapid sparse, root
K nearest sample of distance test sample is calculated according to different T shape regions, carries out the sparse description of group's feature of face；Face
Sparse description during, only extract T shape region facial image feature, the determination in these regions is according to Face detection feature
Point is calculated；Test sample is carried out on multiple random spaces by rarefaction representation using multiple accidental projection method.Due to every
Random space when secondary projection is all different, is equivalent to and implements different feature space mappings to primitive character, and due to not
Same projector space is extracted the feature of facial image with level from different angles, enhances the distinguishing ability of sample characteristics,
This is created condition for the safety of biometric templates, and secondly repeatedly accidental projection enriches face characteristic, is obtained more
Illumination invariant feature, this is created condition for the completenesses of biometric templates；It is residual that multiple reconstruct are obtained by accidental projection
Difference, then the weighted residual using test sample in each classification carries out recognition of face, and weighted residual comprehensive utilization is multiple residual
The distinguishing ability of difference, stability and better reliability.The specific steps of which are as follows:
S801, national human face data collection of the input with national label；
S802, due in collection process facial image by the posture of people, shooting angle, distance of camera etc.
The influence of problem, leading to human face region, position is uncertain in the picture, and size is inconsistent.It is aligned by face and solves people
The problems such as face size is inconsistent；
S803, in image acquisition process, facial image will receive the influence such as illumination, camera parameters, lead to face figure
Image brightness and contrast are different, are handled using unitary of illumination, and the illumination of facial image can be adjusted to unification
Standard；
The national facial image of input is proportionally divided into training set and test set by S804；
Face images are marked human face characteristic point using STASM algorithm, extract face by positioning feature point by S805
"T"-shaped region；
Facial image in each group is divided into training sample and test sample two parts, is denoted as X=respectively by S806
[x1, x2..., xm] and Y=[y1, y2..., yn] wherein xiAnd yiBe unit composed by the region face extraction T arrange to
Amount；
S807 generates q × p random matrix by random function；
S808, orthogonalization random matrix using Normal Distribution and respectively arrange mutually orthogonal random matrix and carry out at random
Projection, can effectively keep the intrinsic propesties of original sample；
S809, it is T that accidental projection maximum number of iterations, which is arranged, in wen(Tn>=2), if when a the t times projection generates at random
Matrix Rt∈Rq×p, for some test sample y ∈ Rp, by x and yiIn RtUpper projection, so that it may obtain RtRandom spy spatially
Sign, i.e.,
Ψt=RtX
yt=Rty
At this timeWhereinIt is the i-th class training sample in RtSpy spatially
It collects, similarly ytIt is y in space RtOn feature；
S810, willAs rarefaction representation institute dictionary,For the sub- dictionary of the i-th class, according to
Sparse representation theory chooses a test sample yt, it is expressed as the linear combination of all training samples as shown by the equation:
Wherein, aI, jFor code coefficient, α={ a1,1, a1,2..., aI, 1, aI, 2..., aK, mIt is coefficient vector；
S811 calculates the t times accidental projection by following formula, reconstructs y using the sub- dictionary of the i-th class and coding vectort
Residual error:
At this point, the t times accidental projection and rarefaction representation have been completed；
Test sample y is categorized into the classification of corresponding minimum deflection, works as e by S812r=| | y-gr||2Value get over hour,
The effect played in expression test sample of r class group is bigger, therefore y just belongs to such group, judges the mark of test sample
It whether consistent signs the label predicted with algorithm, if it is inconsistent, algorithm goes to S807 step, re-starts calculating, if
The label predicted is consistent with former label, then algorithm terminates.
Random space when due to projecting every time is all different, is equivalent to and implements different feature spaces to primitive character and reflect
It penetrates, and since different projector spaces is extracted the feature of facial image with level from different angles, enhances sample spy
The distinguishing ability of sign, secondly this creates condition repeatedly accidental projection for the safety of biometric templates and enriches face spy
Sign, obtains more illumination invariant features, and " this is created condition for the completeness of biometric templates；It is obtained by accidental projection
Multiple reconstructed residuals are obtained, then the weighted residual using test sample in each classification carries out recognition of face, and weighted residual is comprehensive
Close the distinguishing ability using multiple residual errors, stability and better reliability.
The preferable specific embodiment of the above, only the invention, but the protection scope of the invention is not
It is confined to this, anyone skilled in the art is in the technical scope that the invention discloses, according to the present invention
The technical solution of creation and its inventive concept are subject to equivalent substitution or change, should all cover the invention protection scope it
It is interior.
Claims (5)
1. a kind of based on the sparse face group recognition methods with weighted residual in region, which is characterized in that specifically include following step
It is rapid:
Step 1, national human face data collection of the input with national label；
Step 2, the national human face data collection of input is subjected to facial image registration process；
Step 3, the national human face data collection of input is subjected to facial image illumination normalized；
Step 4, whole face geometrical combination features are screened, filters out the region T；
Step 5, the region T of original sample concentration training sample and test sample image is extracted respectively；
Step 6, it constructs based on the sparse face group recognizer model with weighted residual in region；
Step 7, the region T of the training sample in step 5 and test sample image face group identification in step 6 is input to calculate
In method model, according to the output of model, the classification results of test sample are obtained.
2. a kind of based on the sparse face group recognition methods with weighted residual in region according to claim 1, feature exists
In the national human face data collection of input, which is carried out facial image registration process, in step 2 is specifically: being obtained by eye detection method
To the coordinate of eyes, eyes coordinate is used into E respectively1And ErIt indicates, makes two eye coordinates ElAnd ErLine ElrIt is flat with horizontal direction
Row, cuts facial image, and the image that cutting is obtained zooms to specified pixel size.
3. a kind of based on the sparse face group recognition methods with weighted residual in region according to claim 1, feature exists
In the national human face data collection of input, which is carried out facial image illumination normalized, in step 3 is specifically: different groups due to
There are the different colours of skin and textures for geographical location reason；In image acquisition process, facial image will receive illumination, video camera ginseng
Number influences, and causes facial image brightness and contrast different, is handled using unitary of illumination, by the illumination of facial image
It is adjusted to seek unity of standard.
4. a kind of based on the sparse face group recognition methods with weighted residual in region according to claim 1, feature exists
In being screened specifically in step 4 to Face geometric eigenvector: maximum statistics based on mutual information relies on criterion, passes through maximum
Change the correlation between feature and classified variable, minimizes the correlation between feature and feature to obtain preferable feature；
Define maximal correlation are as follows:
Define minimal redundancy are as follows:
Wherein, F is Face geometric eigenvector, and c is sample group attribute classification, I (fr, c) and indicate feature frIt is mutual between classification c
Information, I (fr, f0) indicate feature frWith feature frBetween mutual information；
Two stochastic variables x and y are given, if their probability density is respectively p (x), p (y), joint density is p (x, y), it
Between mutual information are as follows:
Human face characteristic point subset can be acquired by following formula:
195 length characteristics are filtered out from 2926 special types that 77 characteristic points include by above-mentioned formula, and using mRMR
Information, wherein weight highest zone is the "T"-shaped region of nose, eye and eyebrow composition.
5. a kind of based on the sparse face group recognition methods with weighted residual in region according to claim 1, feature exists
In, when predicting test sample generic, it is compared with training sample, finds the most similar K training sample,
And there is most labels as final prediction label using in this K training sample；Specific step is as follows:
Step 1, initialization distance is maximum value；
Step 2, the distance of unknown sample and each training sample, range formula are calculated are as follows:
WhereinIndicate i-th point of l dimension coordinate；
Step 3, current k are obtained closest to the maximum distance maxdist in sample；
Step 4, if dist is less than maxdist, using the training sample as K nearest samples；
Step 5, step 2,3,4 are repeated, until unknown sample and the distance of all training samples are all complete；
Step 6, the number that each class label occurs in K nearest samples is counted；
Step 7, class label of the maximum class label of the frequency of occurrences as unknown sample is selected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910201677.4A CN109948519A (en) | 2019-03-18 | 2019-03-18 | It is a kind of based on the sparse face group recognition methods with weighted residual in region |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910201677.4A CN109948519A (en) | 2019-03-18 | 2019-03-18 | It is a kind of based on the sparse face group recognition methods with weighted residual in region |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109948519A true CN109948519A (en) | 2019-06-28 |
Family
ID=67008850
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910201677.4A Pending CN109948519A (en) | 2019-03-18 | 2019-03-18 | It is a kind of based on the sparse face group recognition methods with weighted residual in region |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109948519A (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108319935A (en) * | 2018-03-23 | 2018-07-24 | 大连民族大学 | Based on the face group recognizer that region is sparse |
-
2019
- 2019-03-18 CN CN201910201677.4A patent/CN109948519A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108319935A (en) * | 2018-03-23 | 2018-07-24 | 大连民族大学 | Based on the face group recognizer that region is sparse |
Non-Patent Citations (1)
| Title |
|---|
| 李燕等: "基于随机投影与加权稀疏表示残差的光照鲁棒人脸识别方法", 《计算机工程与科学》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102844766B (en) | Human eyes images based multi-feature fusion identification method | |
| US8472681B2 (en) | Iris and ocular recognition system using trace transforms | |
| CN107145842A (en) | With reference to LBP characteristic patterns and the face identification method of convolutional neural networks | |
| KR20050025927A (en) | The pupil detection method and shape descriptor extraction method for a iris recognition, iris feature extraction apparatus and method, and iris recognition system and method using its | |
| Zhao et al. | Fingerprint image synthesis based on statistical feature models | |
| CN105956515A (en) | Stereo-hyperspectral human face recognition method based on auroral imaging | |
| JP2017016593A (en) | Image processing apparatus, image processing method, and program | |
| CN103268485A (en) | Sparse-regularization-based face recognition method capable of realizing multiband face image information fusion | |
| CN109255319A (en) | For the recognition of face payment information method for anti-counterfeit of still photo | |
| CN106599785A (en) | Method and device for building human body 3D feature identity information database | |
| CN106022241A (en) | Face recognition method based on wavelet transformation and sparse representation | |
| Chen et al. | Spherical-patches extraction for deep-learning-based critical points detection in 3D neuron microscopy images | |
| Russ et al. | 3D facial recognition: a quantitative analysis | |
| Pathak et al. | Multimodal eye biometric system based on contour based E-CNN and multi algorithmic feature extraction using SVBF matching | |
| GB2471192A (en) | Iris and Ocular Recognition using Trace Transforms | |
| Kakani et al. | Segmentation-based ID preserving iris synthesis using generative adversarial networks | |
| Scherhag | Face Morphing and Morphing Attack Detection | |
| Sun et al. | 3D ear shape matching using joint α-entropy | |
| CN109948519A (en) | It is a kind of based on the sparse face group recognition methods with weighted residual in region | |
| Park | Face Recognition: face in video, age invariance, and facial marks | |
| CN114627493A (en) | Gait feature-based identity recognition method and system | |
| Moradi et al. | I-gans for synthetical infrared images generation | |
| CN110334698A (en) | Glasses detection system and method | |
| Lal et al. | MHA_VGG19: Multi-Head Attention with VGG19 Backbone Classifier-based Face Recognition for Real-Time Security Applications | |
| Hahmann et al. | Combination of facial landmarks for robust eye localization using the Discriminative Generalized Hough Transform |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190628 |