CN101079105A - Human face identification method based on manifold learning - Google Patents
Human face identification method based on manifold learning Download PDFInfo
- Publication number
- CN101079105A CN101079105A CNA2007100419744A CN200710041974A CN101079105A CN 101079105 A CN101079105 A CN 101079105A CN A2007100419744 A CNA2007100419744 A CN A2007100419744A CN 200710041974 A CN200710041974 A CN 200710041974A CN 101079105 A CN101079105 A CN 101079105A
- Authority
- CN
- China
- Prior art keywords
- point
- matrix
- face
- data
- dimensional space
- 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
- G06V40/168—Feature extraction; Face representation
- G06V40/169—Holistic features and representations, i.e. based on the facial image taken as a whole
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention relates to a face identification method based on the manifold learning, which comprises the following steps: offering partial category relation saving the face data from the data partial structure; converting the optimum partial of every point to the total optimum with the arrangement technique; getting the data projection matrix from higher dimensional to lower dimensional by proceeding characterized dissociation for arrangement matrix, the original training sample and transposed product; mapping the identifying face image to the lower dimensional space; classifying with the nearest neighbor method in the lower dimensional space. The invention provides the higher face identification ratio, which provides higher identification ratio than the identification with principal component analytical method.
Description
Technical field
What the present invention relates to is a kind of method of technical field of image processing, specifically is a kind of face identification method based on manifold learning.
Background technology
Arrival along with the information age, data set had had significant variation more in the past, its principal feature can reduce: high data volume, high dimension, high Data Growth rate, destructuring and can not be by people's perception individual processing, and the knowledge quantity that can obtain is bigger on the one hand; Still be faced with arduous problem on the other hand, promptly can not from data, find needed knowledge rationally, effectively.The method for expressing of a lot of problems makes information very sparse, is the problem of a difficulty with denseization of information how.
Find through literature search prior art, people such as M.Turk are at " Journal of CognitiveNeuroscience " Vol.3, No.1,1991,71-86 (Cognitive Neuroscience, the 3rd volume, the 1st phase, 1991, the 71-86 page or leaf) on, set forth principal component method, and it has been applied on the dimensionality reduction of people's face data.Principal component analysis is sought the low-dimensional expression of high dimensional data by maximizing total dispersion matrix.It is optimum on the meaning of rebuilding.Discovering in recent years, the variation of facial image can come parametrization by some continuous variablees, and these variablees can be posture, and illumination, and expression that is to say that facial image belongs to the submanifold of low-dimensional in essence.Traditional dimension reduction method can not be found the essential structure of higher-dimension people face data such as principal component method etc.What principal component analysis was handled is the global structure of data, does not well utilize the local geometric information of data.
Summary of the invention
The objective of the invention is to overcome deficiency of the prior art, propose a kind of face identification method, make its method that adopts the local dimensionality reduction of preserving identification people face,, improve discrimination to realize better people's face manifold learning based on manifold learning.
The present invention is achieved by the following technical solutions, and the present invention proposes the local classification relation of depositary's face data from the partial structurtes of data; Utilize permutation technology then, the local optimum of each point is converted into global optimum; Carry out feature decomposition by product, obtain data by the projection matrix of higher-dimension to low-dimensional to permutation matrix and original training sample and transposition thereof; By this projection matrix, facial image to be identified is projected to lower dimensional space; At last, finish recognition of face at lower dimensional space with the nearest neighbor method classification.
Concrete steps of the present invention are as follows:
(1) original facial image windrow is stacked as a M dimension long vector, thereby all data constitutes a matrix X, as training sample.
(2) for everyone face data point x
i, search out its k similar point, form a part.For this data point x
iWith its k similar point, make them in the point of the lower dimensional space minimum of adjusting the distance, and obtain the expression of corresponding suboptimization.
(3) for the somebody of institute face data point, (2) described optimization all in steps, thus the optimization of each part is stacked up with permutation technology, obtain permutation matrix L, thereby obtain the optimization on the global sense.
(4) raw data set X be multiply by permutation matrix L, multiply by the transposition X of raw data set again
TThereby, obtain: XLX
TIn fact this realized optimizing the structure of matrix.To XLX
TCarry out feature decomposition, suppose that the lower dimensional space of asking is the d dimension, d was individual before the ascending arrangement of the eigenwert of trying to achieve was got, and its pairing d proper vector is then formed M * d projection matrix A.
(5) for each people's face data x to be identified
i, with the transposition A of projection matrix A
TMultiply by x
i, just obtain its low-dimensional and express y
i=A
Tx
t, in lower dimensional space, with the arest neighbors method people's face data are classified and then to have finished recognition of face.
Described for this data point x
iWith its k similar point, make them in the point of the lower dimensional space minimum of adjusting the distance, be meant: establish x
iBe expressed as x
I0, x
I1, L, x
IkExpression x
iSimilar point, establish
For they at the pairing point of lower dimensional space.In order to obtain loyal mapping, the present invention expects that neighbour's point still keeps the neighbour, that is to say some y
I0Still be close to y
I1, L, y
IkSo such optimization is arranged:
The implication of this formula is that the point that minimizes neighborhood is adjusted the distance.To express in order providing more clearly, the optimization on this local sense to be changed into this form: min tr (Y
iL
iY
i T), L wherein
iBe local permutation matrix, the mark of tr (g) representing matrix.Like this, just finished the expression of suboptimization.
Described permutation technology is meant a kind ofly to be superposed to the technology of global optimization by local optimum that the present invention realizes with an iteration function: L (I
i, I
i) ← L (I
i, I
i)+L
i, i=1, L, N.I in the formula
iBe an index function, it has indicated each data point x
iWith its label of similar point, L
iBe the expression of local optimum, N represents to have N data point.With regard to the passable final global alignment matrix L that arrives, it has represented the optimization on the global sense like this.
Described to XLX
TCarry out feature decomposition, be meant that final optimization problem is converted into finding the solution an eigenvalue problem: XLX
TF=λ f, in the following formula, f represents proper vector, the λ representation eigenvalue.With the ascending arrangement of eigenwert, d is individual before getting, and projection matrix A is made up of its pairing proper vector.
Described arest neighbors method is meant some facial images to be identified are referred in the class under its nearest training sample.
Compared with prior art, the present invention can survey the low-dimensional manifold structure that is embedded in the higher-dimension people face data, can access higher recognition of face rate therefrom.Experiment showed, the present invention in the identification of the enterprising pedestrian's face of ORL database, discrimination can reach 97.42% for the situation of 5 training samples, the discrimination 88.15% that obtains apparently higher than principal component method.
Embodiment
Below embodiments of the invention are elaborated: present embodiment is being to implement under the prerequisite with the technical solution of the present invention, provided detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to following embodiment.
Embodiment has adopted a public face database: the ORL database.The ORL database comprises 40 people, everyone 10 images.The size of every figure is 40 * 40.At first facial image is stacked as 1600 dimension long vectors with its windrow, all then data are formed one 1600 * 400 matrix X=[x
1, L, x
400].For each people, select 5 sample training at random, remaining being used for discerned.For each given number of training, produce 20 groups of training at random---recognition sample collection, calculate average recognition rate on this basis.The present invention at first finds the similar point of 4 of each data point, is designated as
, and use I
i={ i
0, i
1, L, i
4As the record of index to them.Then, for each point and its neighborhood point, such optimization is arranged:
, being intended that of it allows these points still keep neighbor relationships at lower dimensional space.Subsequently, local optimum is converted into this form: min tr (Y
iL
iY
i T), like this, local permutation matrix just can extract.According to the method, the present invention can obtain 200 local optimums, and 200 local arrangements matrixes are arranged accordingly.Then, the present invention is according to permutation technology, and these 200 local optimums are converted into optimum on the global sense, according to such iterative formula: L (I
i, I
i) ← L (I
i, I
i)+L
i, i=1, L, 200, wherein, I
iBe the index that each data point noted earlier and its neighbour are ordered, notice that this patent is made as 0 with the initial value of L.Like this, just finished structure to permutation matrix.Finally, optimization problem is converted into finding the solution an eigenvalue problem: XLX
TF=λ f, wherein, L is the permutation matrix of trying to achieve.In fact the implication of following formula is exactly to XLX
TCarry out feature decomposition.If α
1, α
2Λ, α
dBe the proper vector of trying to achieve, it is corresponding to eigenvalue
1<λ
2<Λ<λ
dSo, projection matrix A is: A=(α
1, α
2Λ, α
d).Next, the present invention uses projection matrix A that facial image to be identified is projected to lower dimensional space, then with the comparison of the training facial image of low-dimensional, facial image to be identified is referred in the class under its nearest training sample, promptly classifies with nearest neighbor method.So far, the present invention has finished the overall process of recognition of face.
The present invention is for the situation of 5 training samples of ORL database, and discrimination can reach 97.42%, the discrimination 88.15% that obtains apparently higher than principal component method.
Claims (5)
1. face identification method based on manifold learning is characterized in that concrete steps are as follows:
(1) original facial image windrow is stacked as a M dimension long vector, thereby all data constitutes a matrix X, as training sample;
(2) for everyone face data point x
i, search out its k similar point, form a part, for this data point x
iWith its k similar point, make them in the point of the lower dimensional space minimum of adjusting the distance, and obtain the expression of corresponding suboptimization;
(3) for the somebody of institute face data point, all through the described optimization of step (2), thereby the optimization of each part is stacked up with permutation technology, obtain permutation matrix L, thereby obtain the optimization on the global sense;
(4) raw data set X be multiply by permutation matrix L, multiply by the transposition X of raw data set again
TThereby, obtain: XLX
T, realized optimizing the structure of matrix;
(5) to XLX
TCarry out feature decomposition, suppose that the lower dimensional space of asking is the d dimension, d was individual before the ascending arrangement of the eigenwert of trying to achieve was got, and its pairing d proper vector is then formed M * d projection matrix A;
(6) for each people's face data x to be identified
t, with the transposition A of projection matrix A
TMultiply by x
t, just obtain its low-dimensional and express y
t=A
Tx
t, in lower dimensional space, with the arest neighbors method people's face data are classified and then to have finished recognition of face.
2. according to the face identification method based on manifold learning of claim 1, it is characterized in that, described for this data point x
iWith its k similar point, make them in the point of the lower dimensional space minimum of adjusting the distance, be meant: establish x
iBe expressed as x
I0, x
I1, L, x
IkExpression x
iSimilar point, establish
For they at the pairing point of lower dimensional space, it is optimized:
The implication of this formula is that the point that minimizes neighborhood is adjusted the distance.
4. the face identification method based on manifold learning according to claim 1 is characterized in that, described permutation technology is meant a kind ofly to be superposed to the technology of global optimization by local optimum, realizes with an iteration function: L (I
i, I
i) ← L (I
i, I
i)+L
i, i=1, L, N, the I in the formula
iBe an index function, it has indicated each data point x
iWith its label of similar point, L
iBe the expression of local optimum, N represents to have N data point, so just obtains final global alignment matrix L, and it has represented the optimization on the global sense.
5. the face identification method based on manifold learning according to claim 1 is characterized in that, and is described to XLX
TCarry out feature decomposition, be meant that final optimization problem is converted into finding the solution an eigenvalue problem: XLX
TF=λ f, in the following formula, f represents proper vector, the λ representation eigenvalue, with the ascending arrangement of eigenwert, d is individual before getting, and projection matrix A is made up of its pairing proper vector.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNA2007100419744A CN101079105A (en) | 2007-06-14 | 2007-06-14 | Human face identification method based on manifold learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CNA2007100419744A CN101079105A (en) | 2007-06-14 | 2007-06-14 | Human face identification method based on manifold learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN101079105A true CN101079105A (en) | 2007-11-28 |
Family
ID=38906574
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CNA2007100419744A Pending CN101079105A (en) | 2007-06-14 | 2007-06-14 | Human face identification method based on manifold learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101079105A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101369316A (en) * | 2008-07-09 | 2009-02-18 | 东华大学 | Image characteristics extraction method based on global and local structure amalgamation |
CN102156878A (en) * | 2011-04-08 | 2011-08-17 | 南京邮电大学 | Sparse embedding with manifold information-based human face identification method |
CN102867171A (en) * | 2012-08-23 | 2013-01-09 | 山东师范大学 | Label propagation and neighborhood preserving embedding-based facial expression recognition method |
CN103336960A (en) * | 2013-07-26 | 2013-10-02 | 电子科技大学 | Human face identification method based on manifold learning |
CN103514445A (en) * | 2013-10-15 | 2014-01-15 | 武汉科技大学 | Strip steel surface defect identification method based on multiple manifold learning |
CN103927522A (en) * | 2014-04-21 | 2014-07-16 | 内蒙古科技大学 | Face recognition method based on manifold self-adaptive kernel |
CN105678265A (en) * | 2016-01-06 | 2016-06-15 | 广州洪森科技有限公司 | Manifold learning-based data dimensionality-reduction method and device |
CN106599833A (en) * | 2016-12-12 | 2017-04-26 | 武汉科技大学 | Field adaptation and manifold distance measurement-based human face identification method |
-
2007
- 2007-06-14 CN CNA2007100419744A patent/CN101079105A/en active Pending
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101369316B (en) * | 2008-07-09 | 2011-08-31 | 东华大学 | Image characteristics extraction method based on global and local structure amalgamation |
CN101369316A (en) * | 2008-07-09 | 2009-02-18 | 东华大学 | Image characteristics extraction method based on global and local structure amalgamation |
CN102156878A (en) * | 2011-04-08 | 2011-08-17 | 南京邮电大学 | Sparse embedding with manifold information-based human face identification method |
CN102867171B (en) * | 2012-08-23 | 2014-11-19 | 山东师范大学 | Label propagation and neighborhood preserving embedding-based facial expression recognition method |
CN102867171A (en) * | 2012-08-23 | 2013-01-09 | 山东师范大学 | Label propagation and neighborhood preserving embedding-based facial expression recognition method |
CN103336960A (en) * | 2013-07-26 | 2013-10-02 | 电子科技大学 | Human face identification method based on manifold learning |
CN103514445B (en) * | 2013-10-15 | 2016-09-14 | 武汉科技大学 | Strip steel surface defect identification method based on multiple manifold study |
CN103514445A (en) * | 2013-10-15 | 2014-01-15 | 武汉科技大学 | Strip steel surface defect identification method based on multiple manifold learning |
CN103927522A (en) * | 2014-04-21 | 2014-07-16 | 内蒙古科技大学 | Face recognition method based on manifold self-adaptive kernel |
CN103927522B (en) * | 2014-04-21 | 2017-07-07 | 内蒙古科技大学 | A kind of face identification method based on manifold self-adaptive kernel |
CN105678265A (en) * | 2016-01-06 | 2016-06-15 | 广州洪森科技有限公司 | Manifold learning-based data dimensionality-reduction method and device |
CN105678265B (en) * | 2016-01-06 | 2019-08-20 | 广州洪森科技有限公司 | Method of Data with Adding Windows and device based on manifold learning |
CN106599833A (en) * | 2016-12-12 | 2017-04-26 | 武汉科技大学 | Field adaptation and manifold distance measurement-based human face identification method |
CN106599833B (en) * | 2016-12-12 | 2019-06-25 | 武汉科技大学 | A kind of face identification method adapted to based on field and manifold distance is measured |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101079105A (en) | Human face identification method based on manifold learning | |
CN111695467B (en) | Spatial spectrum full convolution hyperspectral image classification method based on super-pixel sample expansion | |
Liu et al. | Non-negative matrix factorization based methods for object recognition | |
CN105894045B (en) | A kind of model recognizing method of the depth network model based on spatial pyramid pond | |
CN109858557B (en) | Novel semi-supervised classification method for hyperspectral image data | |
CN100483445C (en) | Method for identifying human face subspace based on geometry preservation | |
Liu et al. | Sparse representation using nonnegative curds and whey | |
CN111695456A (en) | Low-resolution face recognition method based on active discriminability cross-domain alignment | |
CN105046272B (en) | A kind of image classification method based on succinct non-supervisory formula convolutional network | |
CN110210534B (en) | Multi-packet fusion-based high-resolution remote sensing image scene multi-label classification method | |
Yin et al. | Feature combination using boosting | |
CN103020640A (en) | Facial image dimensionality reduction classification method based on two-dimensional principal component analysis | |
CN109919242A (en) | A kind of images steganalysis method based on depth characteristic and joint sparse | |
CN108388904B (en) | Dimensionality reduction method based on convolutional neural network and covariance tensor matrix | |
CN112115780A (en) | Semi-supervised pedestrian re-identification method based on deep multi-model cooperation | |
CN111695455A (en) | Low-resolution face recognition method based on coupling discrimination manifold alignment | |
CN117576467B (en) | Crop disease image identification method integrating frequency domain and spatial domain information | |
Zhang et al. | Integrating leaf and flower by local discriminant CCA for plant species recognition | |
CN102779271B (en) | Expression semanteme sparse quantization method based on spectrum sparse canonical correlation analysis | |
CN112966735A (en) | Supervision multi-set correlation feature fusion method based on spectral reconstruction | |
CN112364902B (en) | Feature selection learning method based on self-adaptive similarity | |
CN105718858A (en) | Pedestrian recognition method based on positive-negative generalized max-pooling | |
CN102831425A (en) | Rapid feature extraction method for facial images | |
CN115527064A (en) | Toxic mushroom fine-grained image classification method based on multi-stage ViT and contrast learning | |
CN115457509A (en) | Traffic sign image segmentation algorithm based on improved space-time image convolution |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
WD01 | Invention patent application deemed withdrawn after publication |