CN108717527A - Face alignment method based on posture priori - Google Patents
Face alignment method based on posture priori Download PDFInfo
- Publication number
- CN108717527A CN108717527A CN201810458899.XA CN201810458899A CN108717527A CN 108717527 A CN108717527 A CN 108717527A CN 201810458899 A CN201810458899 A CN 201810458899A CN 108717527 A CN108717527 A CN 108717527A
- Authority
- CN
- China
- Prior art keywords
- face
- model
- facial image
- posture
- apparent
- 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
-
- 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
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- 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
Abstract
The invention belongs to mode identification technology, specially a kind of face alignment method based on posture priori includes the following steps：Using active apparent model, on the face database under unlimited environment, the shape and apparent model of facial image are established；Training stage establishes positive face model, left avertence model and right avertence model respectively according to the characteristic point of every width facial image according to the difference of human face posture；Test phase-utilizes the feature triangle in test set facial image, selects suitable model as the initial model of face；The residual error between model and test set facial image is minimized, warp parameters and apparent parameter are updated come alternating iteration according to inverse composition algorithm SIC simultaneously, realize matching；The present invention establishes positive face model, left avertence model and right avertence model respectively in the training stage according to the difference of posture；In the search phase, suitable model is automatically selected as the initial model of face using feature triangle, the interference so as to avoid attitudes vibration to initial model.
Description
Technical field
The invention belongs to mode identification technologies；Specially a kind of face alignment method based on posture priori.
Background technology
Face alignment refers to automatic positioning human face characteristic point (such as the key points such as nose, the corners of the mouth, eye center and chin)
Process.It is widely used in multiple application fields such as recognition of face, Face datection, feature extraction and Attitude estimation.In crowd
In the algorithm research of more automatic study face informations, the algorithm based on model is proved to be most effective.The early stage of this kind of algorithm
Research includes deformable template and active contour model, mainly extracts face characteristic and people by solely research characteristic point
Face profile.Therefore, matching effect has certain limitation.Recent decades, a variety of including active appearance models AAM are based on
Model and method based on texture are suggested.Wherein, AAM is by shape and the display model (picture in one region of research
Plain intensity) synthesis model, be widely used in facial image, medical image and characteristic point detection etc. fields.Due to these
Model itself has high efficiency and accuracy, and the improved methods about ASM and AAM a variety of in recent years is promoted to be suggested so that
It has further promotion in the accuracy rate of human face characteristic point calibration.
But in an actual situation, face but there is always attitudes vibrations.Therefore in the above improved method there are still
Shortcoming, which is AAM, to be judged with attitudes vibration, higher to the dependence of initialization model effect.In other words, originally
When beginning model and larger standard face difference, subsequent match can be caused to go to pot.At the same time, the matching of AAM is one
A iterative process, the parameter of model can be all evaluated in the update of iteration each time.In general, the iteration update of parameter utilizes minimum
Change the error function between input picture and model instance to realize.There are two types of the main methods of such issues that solution.The first
It is to be realized based on recurrence learning, pace of learning is fast, but is easily trapped into Local Minimum.The method of second of matching AAM is based on
Nonlinear least square method algorithm is realized.
In the prior art, research and improvement [D] the Northwestern Polytechnical Universitys of training inverse composition image alignment algorithms are opened,
The improvement of the algorithm of inverse composition alignment image is proposed in 2007., but this method can be because of the attitudes vibration and light of face
According to and the factors such as block, Initial Face model is interfered；To cause have certain error in face alignment.
Invention content
Present invention seek to address that the above problem of the prior art.It is special to propose accurate face under a kind of non-limiting environmental
Sign point scaling method, the specially face alignment method based on posture priori, including：
S1, using active apparent model, on the face database under unlimited environment, establish facial image shape and
Apparent model；
S2, training stage, according to the characteristic point of every width facial image, different faces was established according to the difference of human face posture
Model；The face initial model includes：Positive face model, left avertence model, right avertence model；
S3, test phase select suitable model as face using the feature triangle in test set facial image
Initial model；Test set facial image is mapped on face initial model；
S4, the test set facial image after mapping and the residual error between template facial image are minimized, according to reversed simultaneously
Composition algorithm SIC alternating iterations update warp parameters and apparent parameter, realize active apparent model matching.
Further, the formula for establishing face shape model is：
S=s0+Sb；
Wherein, the face shape that s is, s0For the average shape of face, when S is using principal component analysis PCA dimensionality reductions
The matrix that obtained feature vector is formed, the vector that b is made of form parameter.
Further, the feature triangle in the facial image using test set, determines suitable face initial model
Including：Two centers and nose, constitutive characteristic triangle are positioned using Adaboost iterative algorithms；Using plane analytic geometry and
The cosine law finds out the three corresponding diagonal ∠ A in α, β, γ and three, ∠ B, ∠ C of feature triangle；If ∠ A>
∠ B judge face edge run-out to the left, then select left avertence model for suitable face initial model；If ∠ B>∠ A, then judge
Face selects right avertence model for suitable face initial model to right avertence；If ∠ A=∠ B, it is suitable to select positive face model
Face initial model；Wherein, when α indicates the company at right eye center and nose；When β indicates the company at left eye center and nose；
When γ indicates left eye center with right oculocentric company.
Further, described to establish face apparent model, specially：
I=A0+Ac
Wherein, I indicates test set facial image, A0Indicate that average texture vector, A are to utilize principal component analysis PCA dimensionality reductions
When the obtained matrix that is formed of feature texture vector, the vector that c is made of apparent parameter.
Further, the residual error formula between the test set facial image minimized after mapping and template facial image
For：
Wherein, | | | | indicate L2 norms；I indicates test set facial image, I (W (X；P) test set after distortion) is indicated
Facial image；W(X；P) warp function is indicated；X indicates pixel point coordinates；A0For the average texture vector of face, A is to utilize master
The matrix that the feature texture vector obtained when constituent analysis PCA dimensionality reductions is formed；P indicates the vector that warp parameters are formed；C tables
Show the vector that apparent parameter is formed.
Further, by it is described minimize residual error formula handled in the way of inverse composition including：
Wherein, Δ p indicates increment warp parameters；M indicates the quantity of apparent parameter, also illustrates that the number of feature vector；ci
Indicate i-th of apparent parameter；ΔciIndicate the increment of i-th of apparent parameter；AiIndicate ith feature texture.
Further, the acquisition update mode of the warp parameters is：
Wherein,Indicate synthesis operation；Δ p indicates increment warp parameters.
Further, the acquisition update mode of the apparent parameter is：
c←c+Δc
Wherein, Δ c indicates the increment of apparent parameter.
Further, the mode of the inverse composition is subjected to first order Taylor expansion, and ignores second order term and obtains：
Wherein,Indicate the gradient of the average texture vector of face；Indicate partial derivative of the distortion at p.
Further, the incremental update formula of apparent parameter is：
Further, the incremental update formula of warp parameters is：
Δ p=(JTJ)-1JT(I(W(X；p))-A0)
Wherein,P=E-SST；E indicates unit matrix.
Beneficial effects of the present invention：Present invention is generally directed to active apparent model AAM is high to initial model dependency degree, and
It the problem of factors influence such as is highly susceptible to posture, illumination and blocks, devise a kind of robust essence based on posture priori
Accurate face alignment method.Positive face model, left avertence model and right avertence mould are established according to the difference of posture respectively in the training stage
Type.In the search phase, suitable model is automatically selected as the initial model of face, so as to avoid appearance using feature triangle
State changes the interference to initial model.In addition, using inverse composition algorithm simultaneously, robust can be realized and accurately match effect
Fruit.
Description of the drawings
Fig. 1 is the flow chart of the present invention；
Fig. 2 is the exemplary plot of the carried face characteristic triangle of the present invention；
Fig. 3 is the face samples show of outdoor face database LFPW；
Fig. 4 is four kinds of distinct methods calibration result display diagrams on LFPW face databases：
Fig. 5 is the enlarged drawing of encircled in Fig. 4.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, below in conjunction with attached drawing to of the invention real
The technical solution applied in example is clearly and completely described, it is clear that described embodiment is only that a present invention part is implemented
Example, instead of all the embodiments.
As shown in Figure 1, the face alignment method based on posture priori of the present invention, including：
S1, using active apparent model, on the face database under unlimited environment, establish facial image shape and
Apparent model；
S2, training stage, according to the characteristic point of every width facial image, different faces was established according to the difference of human face posture
Model；The face initial model includes：Positive face model, left avertence model, right avertence model；
S3, test phase-select suitable model as face using the feature triangle in test set facial image
Initial model；Test set facial image is mapped on face initial model；
S4, the test set facial image after mapping and the residual error between template facial image are minimized, according to reversed simultaneously
Composition algorithm SIC alternating iterations update warp parameters and apparent parameter, realize active apparent model matching.
The foundation of face shape model includes：S=s0+Sb；
Wherein, the face shape that s is, s0For the average shape of face, when S is using principal component analysis PCA dimensionality reductions
The matrix that obtained feature vector is formed, the vector that b is made of form parameter.
Preferably, apparent model (texture variations model) is established using the texture of training image.It will be instructed by fractionation radiations
Practice and everyone face image is concentrated to be mapped to formula s=s0The average shape s of face in+Sb0, realize the normalization to texture,
PCA dimensionality reductions, the example that an apparent model can be obtained is recycled to be represented by：I=A0+ Ac, c=AT(I-A0)；
Wherein, I indicates test set facial image, A0Indicate that average texture vector, A are to utilize principal component analysis PCA dimensionality reductions
When the obtained matrix that is formed of feature texture vector, the vector that c is made of apparent parameter.
1) characteristic point detects
As shown in Fig. 2, positioning human eye center and nose with Adaboost algorithm, it is assumed that right eye coordinate is A (a1,a2), it is left
Eye coordinates are B (b1,b2), nose coordinate is C (c1,c2)。
2) feature triangle
Using the method for plane analytic geometry and the cosine law, seek feature triangle three divide in α, β, γ and three
Not corresponding diagonal ∠ A, ∠ B, ∠ C：
∠ C=π-∠ A- ∠ B * MERGEFORMAT (6)
3) posture is classified
According to the difference of posture, face can substantially be divided into left avertence, front and right avertence face.Utilize feature triangle needle
Positive surface model, left avertence model and right avertence model are respectively trained to different postures.As shown in Figure 2, it can be deduced that：If ∠ A>
∠ B judge face edge run-out to the left, then select left avertence model for suitable face initial model；If ∠ B>∠ A, then judge
Face selects right avertence model for suitable face initial model to right avertence；If ∠ A=∠ B, it is suitable to select positive face model
Face initial model；Wherein, when α indicates the company at right eye center and nose；When β indicates the company at left eye center and nose；
When γ indicates left eye center with right oculocentric company.
The matching of AAM is realized using inverse composition algorithm simultaneously：
1) test set facial image I is given, residual error is minimized and is also equivalent to minimize formula (7)：
Wherein, | | | | indicate L2 norms；I indicates test set facial image, I (W (X；P) test set after distortion) is indicated
Facial image；W(X；P) warp function is indicated；X indicates pixel point coordinates；A0For the average texture vector of face, A is to utilize master
The matrix that the feature texture vector obtained when constituent analysis PCA dimensionality reductions is formed；P indicates the vector that warp parameters are formed；C tables
Show the vector that apparent parameter is formed.
2) mode of inverse composition is used to handle formula (7)：
Wherein, Δ p indicates increment warp parameters；M indicates the quantity of apparent parameter, also illustrates that the number of feature vector；ci
Indicate i-th of apparent parameter；ΔciIndicate the increment of i-th of apparent parameter；AiIndicate ith feature texture.
3) according to formula (8), first order Taylor expansion is carried out：
4) according to formula (9), ignoring second order term can be reduced to：
5) according to formula (10), the update mode of the increment of apparent parameter can be obtained：
6) projector space P=E-AA is definedT, wherein E is unit matrix, since A is an orthogonal intersection space, then optimizes formula
(10) it is equal to optimization following equation (12)：
Then：
Δ p=H-1JT(I(W(X；p))-A0)\*MERGEFORMAT(13)
Wherein, H=JTJ, Indicate AiGradient.
Δ p and Δ c is updated being optimal by alternating iteration, realizes the accurate AAM matchings of robust.Wherein to distortion
Parameter is using reversed newer mode, to apparent parameter using the preceding update mode to addition, i.e.,
Wherein,Indicate synthesis operation.
c←c+Δc \*MERGEFORMAT(15)
One embodiment of the present of invention is as follows：
Using LFPW face databases as experimental data base.
Image in LFPW databases has random variation in the conditions such as posture, expression and illumination, as shown in Figure 3.
1) point-to-point flat by being carried out with existing mainstream alignment schemes in order to effectively assess the effect of face alignment
Equal error (being standardized using pupil spacing) is compared.As it can be seen from table 1 method proposed by the invention is in LFPW face databases
On performance be better than most of existing excellent algorithms, including mixing Tree-structure Model TSPM, the differentiation of local restriction
The cascade posture of response model DRMF, robust return RCPR, there is the gradient descent method SDM of supervision, Gauss-Newton Deformable model
GN-DPM, self-encoding encoder neural network CFAN from thick to thin, face pair of the cascade local binary feature on 3000FPS
Together, related local regression device learns JLRL.
The comparison of 1 distinct methods of table
2) on LFPW databases, comparison exhibition has been carried out from visual effect using the uncalibrated image of four kinds of distinct methods
Show.
Specific alignment effect is as shown in Figure 4 and Figure 5, wherein a is the mark of active direction model AOMs methods in Fig. 4~5
Determining image, b is the uncalibrated image of the depth convolutional network TCDCN of multitask concatenated convolutional network MTCNN junction belt task restrictions,
C is the uncalibrated image of the partial model frame CLM-framework for the belt restraining that Cambridge University builds, and d is the side of being carried of the invention
The uncalibrated image of method.If Fig. 4 can be seen that, for general simple posture (face for being illustrated in front two row in picture), institute
Methodical human face characteristic point matching result is all very effectively.However, in the case where face blocks or on special efficacy image, this
Inventing institute's extracting method has preferably performance.Contrast effect is emphasized to be marked with circle in Fig. 4.Fig. 5 is in Fig. 4
The enlarged drawing of circled region, it can be seen that when face, eyes and facial contour are blocked, the present invention has compared to other methods
There is superior calibration effect.In addition, for deep learning model, it is that training forms in advance with a large amount of external data sources,
Five key points of their main locating human faces (eyes center, nose, the corners of the mouth) compared to traditional algorithm, multiple spot detection result is not
Show very excellent.
One of ordinary skill in the art will appreciate that all or part of step in the various methods of above-described embodiment is can
It is completed with instructing relevant hardware by program, which can be stored in a computer readable storage medium, storage
Medium may include：ROM, RAM, disk or CD etc..
Embodiment provided above has carried out further detailed description, institute to the object, technical solutions and advantages of the present invention
It should be understood that embodiment provided above is only the preferred embodiment of the present invention, be not intended to limit the invention, it is all
Any modification, equivalent substitution, improvement and etc. made for the present invention, should be included in the present invention within the spirit and principles in the present invention
Protection domain within.
Claims (10)
1. a kind of face alignment method based on posture priori, which is characterized in that include the following steps：
S1, using active apparent model, on the face database under unlimited environment, establish the shape of facial image and apparent
Model；
S2, training stage, according to the characteristic point of every width facial image, different faceforms was established according to the difference of human face posture；
The face initial model includes：Positive face model, left avertence model, right avertence model；
S3, test phase select suitable model as the initial of face using the feature triangle in test set facial image
Model；Test set facial image is mapped on face initial model；
S4, the test set facial image after mapping and the residual error between template facial image are minimized, according to inverse composition simultaneously
Algorithm SIC alternating iterations update warp parameters and apparent parameter, realize active apparent model matching.
2. a kind of face alignment method based on posture priori according to claim 1, which is characterized in that establish face shape
The formula of shape model is：
S=s0+Sb；
Wherein, the face shape that s is, s0For the average shape of face, obtained when S is using principal component analysis PCA dimensionality reductions
The matrix that feature vector is formed, the vector that b is made of form parameter.
3. a kind of face alignment method based on posture priori according to claim 1, which is characterized in that described to utilize survey
Feature triangle in examination collection facial image, determines that suitable face initial model includes：It is fixed using Adaboost iterative algorithms
Two centers in position and nose, constitutive characteristic triangle；Using plane analytic geometry and the cosine law, the three of feature triangle are found out
Diagonal ∠ Α corresponding in α, β, γ and three, ∠ B, ∠ C；If ∠ Α>∠ B judge face edge run-out to the left, then
Select left avertence model for suitable face initial model；If ∠ B>∠ Α then judge face to right avertence, select right avertence model for
Suitable face initial model；If ∠ Α=∠ B select positive face model for suitable face initial model；Wherein, side α
Indicate the company side at right eye center and nose；When β indicates the company at left eye center and nose；Side γ is indicated in left eye center and right eye
The company side of the heart.
4. a kind of face alignment method based on posture priori according to claim 1, which is characterized in that establish face table
See model formula be：
I=A0+Ac
Wherein, I indicates test set facial image, A0Indicate that average texture vector, A obtain when being using principal component analysis PCA dimensionality reductions
The matrix that is formed of feature texture vector, the vector that c is made of apparent parameter.
5. a kind of face alignment method based on posture priori according to claim 1, which is characterized in that the minimum
Residual error formula between test set facial image after mapping and template facial image is：
Wherein, | | | | indicate L2 norms；I indicates test set facial image, I (W (X；P) the test set face after distortion) is indicated
Image；W(X；P) warp function is indicated；X indicates pixel point coordinates；A0For the average texture vector of face, A is to utilize principal component
Analyze the matrix that the feature texture vector obtained when PCA dimensionality reductions is formed；P indicates the vector that warp parameters are formed；C indicates table
See the vector that parameter is formed.
6. a kind of face alignment method based on posture priori according to claim 5, which is characterized in that the basis is same
When inverse composition algorithm SIC alternating iterations update warp parameters and apparent parameter include：Using inverse composition algorithm simultaneously by institute
It states the test set facial image after mapping and the residual error between template facial image minimizes；Specially：
Wherein, Δ p indicates increment warp parameters；M indicates the quantity of apparent parameter, also illustrates that the number of feature vector；ciIndicate the
I apparent parameters；ΔciIndicate the increment of i-th of apparent parameter；AiIndicate ith feature texture.
7. a kind of face alignment method based on posture priori according to claim 6, which is characterized in that the distortion ginseng
Several acquisition update modes are：
The acquisition update mode of the apparent parameter is：
c←c+Δc；
Wherein,Indicate synthesis operation.
8. a kind of face alignment method based on posture priori according to claim 7, which is characterized in that by the formula
First order Taylor expansion is carried out, and ignores second order term and obtains：
Wherein, ▽ A0Indicate the gradient of the average texture vector of face；Indicate partial derivative of the distortion at p.
9. a kind of face alignment method based on posture priori according to claim 8, which is characterized in that apparent parameter
Incremental update formula is：
。
10. a kind of face alignment method based on posture priori according to claim 8, which is characterized in that warp parameters
Incremental update formula be：
Δ p=(JTJ)-1JT(I(W(X；p))-A0)
Wherein,P=E-SST；E indicates unit matrix；▽AiIndicate AiGradient.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810458899.XA CN108717527A (en) | 2018-05-15 | 2018-05-15 | Face alignment method based on posture priori |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810458899.XA CN108717527A (en) | 2018-05-15 | 2018-05-15 | Face alignment method based on posture priori |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108717527A true CN108717527A (en) | 2018-10-30 |
Family
ID=63899928
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810458899.XA Pending CN108717527A (en) | 2018-05-15 | 2018-05-15 | Face alignment method based on posture priori |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108717527A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109360270A (en) * | 2018-11-13 | 2019-02-19 | 盎锐（上海）信息科技有限公司 | 3D human face posture alignment algorithm and device based on artificial intelligence |
| CN109584347A (en) * | 2018-12-18 | 2019-04-05 | 重庆邮电大学 | A kind of augmented reality mutual occlusion processing method based on active apparent model |
| CN109635749A (en) * | 2018-12-14 | 2019-04-16 | 网易（杭州）网络有限公司 | Image processing method and device based on video flowing |
| CN109685013A (en) * | 2018-12-25 | 2019-04-26 | 上海智臻智能网络科技股份有限公司 | The detection method and device of header key point in human body attitude identification |
| CN112434678A (en) * | 2021-01-27 | 2021-03-02 | 成都无糖信息技术有限公司 | Face measurement feature space searching system and method based on artificial neural network |
| TWI768913B (en) * | 2021-05-20 | 2022-06-21 | 國立中正大學 | Eye center localization method and localization system thereof |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1687957A (en) * | 2005-06-02 | 2005-10-26 | 上海交通大学 | Man face characteristic point positioning method of combining local searching and movable appearance model |
| CN104516489A (en) * | 2013-09-28 | 2015-04-15 | 南京专创知识产权服务有限公司 | Eye control mouse based on matlab |
| CN104794441A (en) * | 2015-04-15 | 2015-07-22 | 重庆邮电大学 | Human face feature extracting method based on active shape model and POEM (patterns of oriented edge magnituedes) texture model in complicated background |
| CN105117681A (en) * | 2015-06-29 | 2015-12-02 | 电子科技大学 | Multi-characteristic fatigue real-time detection method based on Android |
| CN106295549A (en) * | 2016-08-05 | 2017-01-04 | 深圳市鹰眼在线电子科技有限公司 | Multi-orientation Face collecting method and device |
| CN107103309A (en) * | 2017-05-31 | 2017-08-29 | 天津中科智能识别产业技术研究院有限公司 | A kind of sitting posture of student detection and correcting system based on image recognition |
| CN107909019A (en) * | 2017-11-07 | 2018-04-13 | 重庆邮电大学 | It is a kind of based on the face automatic aligning of TI SPCA and recognition methods |
-
2018
- 2018-05-15 CN CN201810458899.XA patent/CN108717527A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1687957A (en) * | 2005-06-02 | 2005-10-26 | 上海交通大学 | Man face characteristic point positioning method of combining local searching and movable appearance model |
| CN104516489A (en) * | 2013-09-28 | 2015-04-15 | 南京专创知识产权服务有限公司 | Eye control mouse based on matlab |
| CN104794441A (en) * | 2015-04-15 | 2015-07-22 | 重庆邮电大学 | Human face feature extracting method based on active shape model and POEM (patterns of oriented edge magnituedes) texture model in complicated background |
| CN105117681A (en) * | 2015-06-29 | 2015-12-02 | 电子科技大学 | Multi-characteristic fatigue real-time detection method based on Android |
| CN106295549A (en) * | 2016-08-05 | 2017-01-04 | 深圳市鹰眼在线电子科技有限公司 | Multi-orientation Face collecting method and device |
| CN107103309A (en) * | 2017-05-31 | 2017-08-29 | 天津中科智能识别产业技术研究院有限公司 | A kind of sitting posture of student detection and correcting system based on image recognition |
| CN107909019A (en) * | 2017-11-07 | 2018-04-13 | 重庆邮电大学 | It is a kind of based on the face automatic aligning of TI SPCA and recognition methods |
Non-Patent Citations (2)
| Title |
|---|
| 朱晓琪: "人脸标记点跟踪快速对齐算法及其树莓派应用", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 赵磊等: "基于ASM局部定位和特征三角形的列车驾驶员头部姿态估计", 《铁道学报》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109360270A (en) * | 2018-11-13 | 2019-02-19 | 盎锐（上海）信息科技有限公司 | 3D human face posture alignment algorithm and device based on artificial intelligence |
| CN109360270B (en) * | 2018-11-13 | 2023-02-10 | 盎维云(深圳)计算有限公司 | 3D face pose alignment method and device based on artificial intelligence |
| CN109635749A (en) * | 2018-12-14 | 2019-04-16 | 网易（杭州）网络有限公司 | Image processing method and device based on video flowing |
| CN109635749B (en) * | 2018-12-14 | 2021-03-16 | 网易（杭州）网络有限公司 | Image processing method and device based on video stream |
| CN109584347A (en) * | 2018-12-18 | 2019-04-05 | 重庆邮电大学 | A kind of augmented reality mutual occlusion processing method based on active apparent model |
| CN109584347B (en) * | 2018-12-18 | 2023-02-21 | 重庆邮电大学 | Augmented reality virtual and real occlusion processing method based on active appearance model |
| CN109685013A (en) * | 2018-12-25 | 2019-04-26 | 上海智臻智能网络科技股份有限公司 | The detection method and device of header key point in human body attitude identification |
| CN109685013B (en) * | 2018-12-25 | 2020-11-24 | 上海智臻智能网络科技股份有限公司 | Method and device for detecting head key points in human body posture recognition |
| CN112434678A (en) * | 2021-01-27 | 2021-03-02 | 成都无糖信息技术有限公司 | Face measurement feature space searching system and method based on artificial neural network |
| TWI768913B (en) * | 2021-05-20 | 2022-06-21 | 國立中正大學 | Eye center localization method and localization system thereof |
| US11769347B2 (en) | 2021-05-20 | 2023-09-26 | National Chung Cheng University | Eye center localization method and localization system thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108717527A (en) | Face alignment method based on posture priori | |
| CN103810490B (en) | A kind of method and apparatus for the attribute for determining facial image | |
| CN104572804B (en) | A kind of method and its system of video object retrieval | |
| CN104317391B (en) | A kind of three-dimensional palm gesture recognition exchange method and system based on stereoscopic vision | |
| KR100647322B1 (en) | Apparatus and method of generating shape model of object and apparatus and method of automatically searching feature points of object employing the same | |
| Xiong et al. | Supervised descent method for solving nonlinear least squares problems in computer vision | |
| Alnajar et al. | Calibration-free gaze estimation using human gaze patterns | |
| CN106570480B (en) | A kind of human action classification method based on gesture recognition | |
| CN104123749A (en) | Picture processing method and system | |
| US8768100B2 (en) | Optimal gradient pursuit for image alignment | |
| US10860755B2 (en) | Age modelling method | |
| CN105139007B (en) | Man face characteristic point positioning method and device | |
| CN111127668B (en) | Character model generation method and device, electronic equipment and storage medium | |
| CN109034131A (en) | A kind of semi-automatic face key point mask method and storage medium | |
| Pound et al. | A patch-based approach to 3D plant shoot phenotyping | |
| CN111695431A (en) | Face recognition method, face recognition device, terminal equipment and storage medium | |
| Saval-Calvo et al. | 3D non-rigid registration using color: color coherent point drift | |
| CN108960093A (en) | The recognition methods and equipment of face's rotational angle | |
| Kroon et al. | Eye localization in low and standard definition content with application to face matching | |
| Fan et al. | Convex hull aided registration method (CHARM) | |
| CN109948541A (en) | A kind of facial emotion recognition methods and system | |
| Chen et al. | 2d face alignment and pose estimation based on 3d facial models | |
| Kaiser et al. | Towards using covariance matrix pyramids as salient point descriptors in 3D point clouds | |
| Li et al. | Face aging effect simulation model based on multilayer representation and shearlet transform | |
| CN100418106C (en) | Method and apparatus for detecting human face |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181030 |