CN107615295A - For the apparatus and method for the facial key feature for positioning face-image - Google Patents
For the apparatus and method for the facial key feature for positioning face-image Download PDFInfo
- Publication number
- CN107615295A CN107615295A CN201580080396.8A CN201580080396A CN107615295A CN 107615295 A CN107615295 A CN 107615295A CN 201580080396 A CN201580080396 A CN 201580080396A CN 107615295 A CN107615295 A CN 107615295A
- Authority
- CN
- China
- Prior art keywords
- shape
- face
- image
- subregion
- facial
- 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.)
- Granted
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
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/755—Deformable models or variational models, e.g. snakes or active contours
- G06V10/7557—Deformable models or variational models, e.g. snakes or active contours based on appearance, e.g. active appearance models [AAM]
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses the system, apparatus and method of the facial key feature for positioning face-image.Method for positioning the facial key feature in face-image includes：Obtain one group of candidate shape respectively from predetermined shape area, each candidate shape demarcation has facial key feature；Accessed each candidate shape and face-image is set to align with the corresponding shape of aliging of acquisition；Alignment shape is stated to determine the subregion of shape area, therefrom to select to stay in the one group of candidate shape to be obtained of next stage behind the current generation according to what is obtained in the current generation in two or more stages；And repeated in two or more stages it is described obtain, the alignment and the step of the determination to position the facial key feature in the face-image.Using this method and system, it can prevent final solution from causing locally optimal solution due to the initialization (this is the FAQs that cascade homing method runs into) of clumsiness, and the robustness in terms of larger attitudes vibration is tackled can be improved.
Description
Technical field
This disclosure relates to face alignment, especially relates to be used for the facial key feature for positioning facial (face) image
The method, apparatus and system of (facial landmark).
Background technology
The purpose of face alignment is to be automatically positioned facial key point (key point).In many for face alignment
Among kind method, cascade returns (cascaded regression) method and rapidly becomes one of method most in fashion.Algorithm is usual
Since original shape (for example, average shape of training sample), and shape is improved by the recurrence device trained successively.
However, cascade homing method have its dependent on initialization process, this accepted extensively the shortcomings that.It is special
Not, if the shape of initialization is far from target shape, this deviation would be impossible to obtain by the successive iterations in cascade
Corrected to complete.Therefore, final solution is probably locally optimal solution.Existing method is often through using some heuristic hypothesis
Or strategy evades this problem, these heuristic hypothesis or strategy alleviates to some extent this problem but simultaneously non-fully
Solves the problem.
All above methods assume that (being usually average shape) provides original shape with some form.Assuming that test sample
The average posture for being distributed in training sample uses average shape in the case of nearby.This hypothesis is not always set up, especially
For the face with larger attitudes vibration.Cao et al.Propose to run algorithm several times and by institute using different initialization
There is the intermediate value of prediction as final output.Burgos-Artizzu et al.By Intelligent restarting method come improvement strategy, but it is needed
Cross validation is wanted with threshold value and number of run.Usually, these strategies alleviate this problem to a certain extent, but still
The dependence to shape initialization is not completely eliminated.Zhang et al.Propose by from global image block prediction rough estimate value
(be still the autocoder Recurrent networks trained successively below) is initialized.
The content of the invention
The purpose of the application is at least one or more problem in solving face alignment, above-mentioned.According to the application's
Method is started with the coarse search carried out in the shape space comprising varied shapes, and using thick solution come to then thinner shape
Search for into row constraint (that is, " from coarse to fine " method).Unique sublevel stage progressive and adaptable search can be final i) to prevent
Solution locally optimal solution is caused due to the initialization of clumsiness (this is to cascade the FAQs that runs into of homing method)；And ii)
Improve the robustness in terms of larger attitudes vibration is tackled.
In addition, composite character setting (hybrid features setting) is proposed come real according to the equipment of the application
The speed of reality is showed.Due to unique tolerance in search mechanisms from coarse to fine, equipment can be in the different optimizing phases
Switch different types of recurrence feature, without making accuracy sacrifice too much.
In an aspect, a kind of method for being used to position the facial key feature of face-image is disclosed.Methods described
It may include：Obtain one group of candidate shape respectively from predetermined shape area, each candidate shape demarcation has facial key feature；
Accessed each candidate shape is set to be alignd with face-image with the corresponding shape of aliging of acquisition；According at two or two
Alignment shape obtained in current generation in the above stage is worked as to determine the subregion of shape area with therefrom selecting to stay in
Next stage behind last stage one group of candidate shape to be obtained；And repeat to obtain, align and determine in these stages
Processing to position the facial key feature in face-image.
In another aspect, a kind of equipment for being used to position the facial key feature of face-image is disclosed.It is described to set
It is standby to may include：Acquiring unit, it is used in one or more successive stages obtain one group of candidate from predetermined shape area
Shape, demarcation has facial key feature to each candidate shape in advance；Alignment unit, itself and acquiring unit telecommunication, and
Accessed each candidate shape and face-image is set to align with the corresponding shape of aliging of acquisition；And determining unit, its with
Alignment unit telecommunication, and according to the alignment shape obtained in the current generation in one or more successive stages come really
The subregion of shape area is determined, therefrom to select to stay in the one group of candidate's shape to be obtained of next stage behind the current generation
Shape.
In another aspect, a kind of system for being used to position the facial key feature of face-image is disclosed.The system
System may include：Memory, it is used to store executable part；And processor, it is executable to perform that it is electrically coupled to memory
Part is included with the operation of execution system, wherein these executable parts：Obtaining widget, it is used in one or more continuous ranks
One group of candidate shape is obtained in shape area predetermined Duan Zhongcong, demarcation has facial key feature to each candidate shape in advance；It is right
Neat part, it is used to making accessed each candidate shape and face-image to align with the corresponding shape of aliging of acquisition；And
Part is determined, it is used to determine the son of shape area according to the alignment shape obtained in the current generation in these stages
Region is therefrom to select to stay in one group of candidate shape that next stage behind the current generation is acquired.
Brief description of the drawings
The Exemplary, non-limiting embodiment of the present invention is described below with reference to accompanying drawing.Accompanying drawing is illustrative, and
Do not drawn in definite ratio typically.The same or like element on different figures is quoted with identical drawing reference numeral.
Fig. 1 illustrates setting for the facial key feature for being used to position face-image of an embodiment according to the application
It is standby.
Fig. 2 illustrates the schematic block diagram of the determining unit of the embodiment according to the application.
Fig. 3 illustrates the side for being used to position the facial key feature of face-image of an embodiment according to the application
Method.
Fig. 4 illustrates the schematic flow of the determination step of the method for an embodiment according to the application.
Fig. 5 is to illustrate to be used in three phases selection in 2D spaces by can according to the embodiment of the application
Depending on the figure for the process of subregion changed.
Fig. 6 is to be used to position facial key according to wherein the performing during three phases for an embodiment of the application
The example of the method for feature.
Fig. 7 illustrate according to the embodiment of the application by software come implement the present invention it is function, be used for
The system for positioning the facial key feature in face-image.
Embodiment
It is contemplated for carrying out the present invention's with detailed reference to some particulars of the present invention, including by inventor
Best mode.The example of these particulars is illustrated in accompanying drawing.Although describe this with reference to these particulars
Invention, it is to be understood that it, which is not intended as, limits the invention to described embodiment.On the contrary, it is intended to as can
Alternative, modification and the equivalent being included in the spirit and scope of the present invention as defined by the accompanying claims.Following
In description, numerous specific details are elaborated to provide thorough understanding of the present invention.Can be in these no specific details
The present invention is put into practice in the case of some or all.In other examples, be not described in detail well-known process operation so as not to
Necessarily obscure the present invention.
Term used herein is only used for describing the purpose of particular and being not intended to be limiting the present invention.Such as this
Used in text, unless the context clearly indicates otherwise, otherwise singulative " one " and " described/to be somebody's turn to do " are also intended to comprising plural shape
Formula.It will be further understood that, when used in this specification, term includes providing stated feature, integer, step, operation, member
Part and _/or part presence, but be not precluded from other one or more features, integer, step, operation, element, part and/or
The presence or addition of its group.
Hereinafter, shape space refers to 2n dimensional linears space, and wherein n refers to the number of key feature.In shape space
(x, y) coordinate of n facial key features of shape representation.Subregion refers to the subset of shape space rather than the space of facial zone
Concept.
Fig. 1 illustrates setting for the facial key feature for being used to position face-image of an embodiment according to the application
Standby 1000.As indicated, equipment 1000 includes acquiring unit 100, alignment unit 200 and determining unit 300.Using according to the application
Equipment, can with automatic detection face facial key feature (such as, eye pupil or the corners of the mouth etc.) position.
As shown in fig. 1, acquiring unit 100 can be used in one or more successive stages from predetermined shape area
One group of candidate shape is obtained, demarcation has facial key feature to each candidate shape in advance.In embodiments, these candidate shapes
Obtained from the collection for making to pre-process by Pu Luke (Procrustes) analyses.Shape space S is fixed through whole process.
Alignment unit 200 can be with the telecommunication of acquiring unit 100.Alignment unit 200 can be used for making accessed each time
Form slection shape and face-image align with the corresponding shape of aliging of acquisition.In embodiments, alignment unit 200 can be further from face
Facial characteristics (facial feature) and the face spy that will be extracted by using at least one recurrence device are extracted in portion's image
Sign is mapped as shape residual error (shape residual) so that obtains alignment shape by using the shape residual error.In difference
Different numbers and different types of facial characteristics can be extracted in stage.For example, SIFT (Scale are used in all stages
Invariant Feature Transform；Scale invariant features transform) feature to be to obtain optimal accuracy.In realization side
In formula, BRIEF (Binary Robust Independent Elementary Features are used in the first two stage；Two
System robust isolated footing feature) feature, and SIFT feature is used in the last stage.It should be understood that the application is not
It is limited to this, these features can be any of feature.
Determining unit 300 can with the telecommunication of alignment unit 200, and for according in current generation in these stages
The alignment shape obtained determines the subregion of shape area, therefrom to select to stay in next stage behind the current generation
One group of candidate shape being acquired.According to embodiment, determining unit 300 can further comprise that center is inferred unit 301 and fitted
Suitable property infers unit 302, and the center infers that unit and the suitability infer that unit is shown in Figure 2 and will be subject to later in detail
Thin description.
Fig. 3 illustrates the side for being used to position the facial key feature of face-image of an embodiment according to the application
Method 2000.Fig. 4 illustrates the schematic flow of the determination performed by determining unit 300.It will be described in detail referring to figs. 1 to Fig. 4
The configuration of the element of equipment 1000 and the process of method 2000 and function.
As shown in Figure 3, in step S100, one group of candidate shape can be obtained respectively from predetermined shape area, each
Candidate shape demarcation has facial key feature.In step S200, accessed each candidate shape and face-image pair can be made
Together with shape of being alignd corresponding to acquisition.In S300, according to the current rank at step S200 in two or more stages
Alignment shape obtained in section, determines the subregion of shape area, next behind the current generation therefrom to select to stay in
One group of candidate shape that stage is acquired.
Then, in S400, it is determined that whether completing step S100-S300 in all stages.In embodiments, complete
The stage can of predetermined number represents that processing procedure is completed.It should be noted that the application is not limited to this, it is of the prior art any
Known method is available.If being "Yes" at step S400, process terminates and last rank in these stages
The center for the subregion that section is inferred is confirmed as the oriented facial key feature of face-image, and this will be retouched later
State.If " non-", then process proceeds to step S100.Method 2000 is to carry out in the shape space comprising varied shapes
Coarse search starts, and enters row constraint using then thinner search of the thick result to shape.Application method 2000, can be exactly
Position the facial key feature of face-image.
Hereinafter, description is obtained into N number of candidate shape simultaneously wherein in l=1 ..., during L stage from shape space
They are expressed as S={ s1, s2..., sN(N ＞ ＞ 2n) example.Fig. 6 illustrates wherein existing according to embodiment
The exemplary embodiment of method 2000 is performed during three phases.As seen from Figure 6, by for positioning shape from coarse to fine
The method of shape, nose and these key features of face can be overcome to cause part most due to clumsy initialization of the prior art
The problem of excellent solution.In the implementation according to the present processes, 35fps real-time performances are realized on monokaryon i5-4590.
Compared with routinely cascading recurrence, evaluated error is only 12.04.It should be understood that embodiment be only exemplary and the application simultaneously
Not limited to this.
The candidate shape in S is obtained from predetermined shape space.In the first stage, (such as) be based on being uniformly distributed from shape
Randomly being obtained in space S includesJ=1,2 ... one group of NlIndividual candidate shape.
Alignment unit 200 can make NlIndividual candidate shape is alignd with the face-image of iteration several times.For iteration k=1,
2 ..., K, by local form's patternCalculating is characterized f.Then, by using KlIndividual recurrence device reg (k) is by feature
F is mapped as shape residual delta x=Mreg(k)(f).In the case of K iteration, pass throughTo be alignd
Shape
After shape of aliging is obtained, the center of the subregion of the deducibility shape space of unit 31 is inferred at center.At l-th
In stage, byThe subregion of shape space is represented, whereinThe center of subregion is represented, andRepresent fixed
Justice surrounds centerSubregion scope suitability probability.
According to an embodiment, by linearly combining all alignment shapes as follows for being jointly inferred to sub-district
Domain center determines the center of subregion：
In equation (1), weight vectors w is used.The weight vectors can be determined by using diversity method is dominated.It is more accurate
Say that the weight at each edge in construction non-directed graph G={ V, E }, wherein E is represented by the similarity being defined as below in ground：
By representing all elements a in the matrix formpqTo form similar matrix A, and A diagonal element is set as zero
To avoid self-loop.
For t=1 ..., T
WhereinRepresent by element vector multiplication；And
Weight vectors can be thereby determined that.Different from wherein aliging what shapes were averaged to all by fixed weight
Conventional method, thus, the sensitiveness of the alignment shape of a small amount of mistake is inhibited as caused by locally optimal solution.
After inferring that unit 301 is inferred to the center of subregion according to equation (1) such as above at center, determining unit
300 can be with it is thus determined that subregion so that will obtain one group of candidate shape from subregion according to suitability probability.
According to another embodiment, suitability infers that unit 302 can be according to the center inferred of subregion and face
Local form's pattern of image infers the suitability probability for each candidate shape for being suitable for face-image, to determine shape area
The subregion in domain.In embodiments, suitability infers that unit 302 is further used for：According to determined by institute's subregion
The heart stays in the adjustable probability for the scope that immediate vicinity adjusts to calculate；Facial figure is calculated according to the identified center of subregion
The facial parts likelihood probability of multiple facial parts of picture；And by adjustment probability is multiplied with facial parts likelihood probability come
Obtain suitability.
Specifically, for subregion x(l)Center and shape space { si, adjustable Probability p is calculated by below equationi：
The purpose of adjustable probability is approx to be depicted in x(l)Neighbouring acquisition scope, and generally for the stage below
For suitability more concentrate.
It is in addition, similar to calculate facial parts based on the local form pattern φ extracted from face-image by below equation
Probability pi：
(Hough returns ballot) is mapped by discriminate to represent component behind in equation (5), and divided by different face
Portion part r.The purpose of facial likelihood probability is by individually considering local form from each facial parts to guide shape
Shift to more reasonably shape area.
It is similar to the facial parts inferred by equation (5) general by the adjustment probability that equation (4) is inferred by as above making
Rate is multiplied to calculate suitability.
After these processes continue to that all stage L terminate, by last subregion in the last stage
Center is defined as net shape, you can to accurately determine the coordinate of the facial key feature of face-image.
Hereinbefore, the method for the facial key feature for positioning face-image has been described referring to figs. 1 to Fig. 4.Can
Subregion x is inferred to train by training algorithm(l)Center and infer suitabilityProcess.Training is listed in table 1 to calculate
Method.
Training algorithm from coarse to fine table 1-
In training program, pass through given suitabilityTo train the subregion x in l stages(l)Center.It is specific
Ground, each candidate shapeJ=1,2 ... it is revert to and is closer to actual conditions shape x*Shape.
For iteration k=1,2 ..., K, first by local form's informationCalculating is characterized；Then, under
Formula trains recurrence deviceFinally, pass throughTo updateTo obtainJ=1,2 ....
Then, for i-th of training sample, by below equation come training area domain center
For weight vectors wi, the summit of construction non-directed graph and the figure is alignment shape.According to being defined as following formula
Similarity carrys out each Weighted Edges concentrated to edge：
Then, weight vectors w is optimized by below equationi：
In another training program, pass through the subregion x in l-th of stage(l)Given center train suitability
For the adjustable Probability p such as represented by equation (4)i, pass through actual conditions shape x*With subregion x(l)Center practise
Obtain covariance matrix.∑ is the covariance matrix x throughout all training samples(l)-x*And it is restricted to diagonal.
For the facial parts likelihood probability p such as represented by equation (5)i, divided by different facial parts.For facial portion
Divide r, acquistion is mapped by discriminate
Then, suitability probability is trained by below equation
Such as by it will be apparent to those skilled in the art that system, method or computer program product can the invention is embodied as.Cause
This, the present invention can use complete hardware embodiment and hardware aspect (it can be typically referred to as to " unit ", " electricity herein
Road ", " module " or " system ").The major part and many invention principles of the function of the present invention are when realizing most preferably by integrated electricity
Road (IC) support, such as digital signal processor and therefore software or application-specific integrated circuit.While it may be possible to pay great efforts and permitted
More design alternatives by (such as) available time, current technology and economic consideration drive, but still expect ordinary skill
It will readily be able under the guiding of personnel's concept disclosed herein and principle and produce IC with minimum experiment.Therefore, for letter
Any risk of fuzzy principles and concepts according to the present invention is simultaneously preferably minimized by clean property, to such software and IC (if
Words) be discussed further be limited to regard to the principle as used in preferred embodiment and the key element for concept.
In addition, the present invention can use complete software embodiment (including firmware, resident software, microcode etc.) or with reference to software
Embodiment.In addition, the present invention can use the form for the computer program product being embodied in any tangible performance media,
The performance media have the computer usable program code being embodied in the media.Fig. 7 illustrates one according to the application
The system 3000 for being used to position the facial key feature of face-image of embodiment, wherein implementing the present invention's by software
Function.With reference to figure 7, system 3000 includes：Memory 3001, it stores executable part；And processor 3002, it is electrically connected
Executable part is performed to memory 3001, with the operation of execution system 3000.These executable parts may include：Acquisition unit
Part 3003, it is used in one or more successive stages obtain one group of candidate shape, Mei Gehou from predetermined shape area
Demarcation has facial key feature to form slection shape in advance；Align unit 3004, it is used to make accessed each candidate shape and face
Portion's image alignment is with alignment shape corresponding to obtaining；Part 3005 is determined, it was used for according to the current generation in these stages
Obtained in alignment shape determine the subregion of shape area, therefrom to select to stay in next rank behind the current generation
One group of candidate shape that section is acquired.The function of part 3003 to 3005 is analogous respectively to the function of unit 100 to 300, and because
This omits its detailed description herein.
Although having been described for the preferred exemplary of the present invention, those skilled in the art can know that basic invention is general
At once these examples are made with change or modification after thought.Appended claims are intended to be considered as including preferred exemplary and all changes
Change or modification is all fallen within the scope of the present invention.
Obviously, those skilled in the art can make to the present invention without departing from the spirit and scope of the present invention
Change or modification.Thus, if these changes or modification belong to claim and the scope of equivalence techniques, they can also fall into
In the scope of the present invention.
Claims (20)
1. a kind of method for being used to position the facial key feature in face-image, it includes：
Obtain one group of candidate shape respectively from predetermined shape area, each candidate shape demarcation has facial crucial special
Sign；
Accessed each candidate shape and the face-image is set to align with the corresponding shape of aliging of acquisition；
The shape area is determined according to the alignment shape obtained in the current generation in two or more stages
The subregion in domain, therefrom to select to stay in the one group of candidate shape to be obtained of next stage behind the current generation；With
And
Repeated in described two or two or more stage it is described obtain, the alignment and the step of the determination to position the face
Facial key feature in portion's image.
2. according to the method for claim 1, wherein, the determination further comprises：
Pushed away according to the alignment shape obtained in the current generation and local form's pattern of the face-image
The center of the disconnected subregion.
3. the method according to claim 11, wherein, the last stage from the described two or two or more stage
The center of subregion inferred, described is crucial special to determine to have determined that the face of position in the face-image
Sign.
4. equipment according to claim 2, wherein, the determination further comprises：
Each time is inferred according to the center and local form's pattern of the face-image being inferred to for the subregion
Form slection shape is suitable for the suitable probability of the face-image, to determine the subregion of the shape area.
5. equipment according to claim 4, wherein, infer that the suitable probability is further performed by following steps：
The adjustable probability for the scope for treating to be adjusted around the center is calculated according to the center determined for the subregion；With
And
The facial phase of facial parts in the face-image is calculated according to local form's pattern of the face-image
Like probability, with by making the adjustment probability be multiplied with the facial likelihood probability to obtain the suitable probability.
6. equipment according to claim 1, wherein, the alignment further comprises：
Facial characteristics is extracted from the face-image；And
The facial characteristics extracted is mapped as by shape residual error by using at least one recurrence device so that residual by the shape
Difference obtains the alignment shape.
7. equipment according to claim 6, wherein, different numbers and different types of face can be extracted in different phase
Portion's feature.
8. equipment according to claim 7, wherein, the facial characteristics extracted in the first phase is SIFT, and
The facial characteristics extracted in other stages is SIFT and BRIEF.
9. a kind of equipment for being used to position the facial key feature of face-image, it includes：
Acquiring unit, it is used in one or more successive stages obtain one group of candidate shape from predetermined shape area,
Demarcation has facial key feature to each candidate shape in advance；
Alignment unit, itself and the acquiring unit telecommunication, and accessed each candidate shape is schemed with the face
Picture aligns to obtain corresponding alignment shape；And
Determining unit, itself and the alignment unit telecommunication, and according to current in one or more of successive stages
The alignment shape obtained in stage determines the subregion of the shape area, therefrom to select to stay in the current rank
The one group of candidate shape to be obtained of next stage behind section.
10. equipment according to claim 9, wherein, the determining unit further comprises：
Unit is inferred at center, and it is used for according to the alignment shape and the part of the face-image obtained in the current generation
Topographic pattern infers the center of the subregion.
11. equipment according to claim 10, wherein, last rank from one or more of successive stages
The center of subregion that section is inferred, described is crucial special to determine to have determined that the face of position in the face-image
Sign.
12. equipment according to claim 10, wherein, the determining unit further comprises：
Suitability infers unit, and it is used for according to the center being inferred to for the subregion and the part of the face-image
Topographic pattern infers that each candidate shape is suitable for the suitable probability of the face-image, to determine the shape area
The subregion.
13. equipment according to claim 12, wherein, the suitability infers that unit is further used for：
The adjustable probability for the scope for treating to be adjusted around the center is calculated according to the center determined for the subregion；With
And
Calculated according to local form's pattern of the face-image facial facial similar general in the face-image
Rate, with by making the adjustment probability be multiplied with the facial likelihood probability to obtain the suitable probability.
14. equipment according to claim 9, wherein, the alignment unit is further used for：
Facial characteristics is extracted from the face-image；And
The facial characteristics extracted is mapped as by shape residual error by using at least one recurrence device so that by using the shape
Shape residual error obtains the alignment shape.
15. equipment according to claim 14, wherein, different numbers and different types of can be extracted in different phase
Facial characteristics.
16. equipment according to claim 15, wherein, extracted in two earliest stages described in be characterized in BRIEF,
And it is characterized in SIFT described in being extracted in other stages.
17. a kind of system for being used to position the facial key feature in face-image, it includes：
Image capturing unit, it is used to capture the face-image；
Acquiring unit, it is used in one or more successive stages obtain one group of candidate shape from predetermined shape area,
Demarcation has facial key feature to each candidate shape in advance；
Alignment unit, itself and the acquiring unit telecommunication, and accessed each candidate shape is schemed with the face
Picture aligns to obtain corresponding alignment shape；And
Determining unit, itself and the alignment unit telecommunication, and according to current in one or more of successive stages
The alignment shape obtained in stage determines the subregion of the shape area, therefrom to select to stay in the current rank
The one group of candidate shape to be obtained of next stage behind section.
18. system according to claim 17, wherein the determining unit further comprises：
Unit is inferred at center, and it is used for according to the alignment shape obtained in the current generation and the office of the face-image
Portion's topographic pattern infers the center of the subregion；And
Suitability infers unit, and it is used for according to the center being inferred to for the subregion and the office of the face-image
Portion's topographic pattern infers that each candidate shape is suitable for the suitable probability of the face-image, to determine the shape area
The subregion in domain.
19. system according to claim 18, it further comprises：
Training unit, it is used to train the center to infer unit with given suitability, and with the given of the subregion
Center train the suitability to infer unit, to change the parameter used by the determining unit.
20. system according to claim 19, wherein, the last stage from one or more of stages is
The center that the subregion is inferred to determines to have determined that the facial key feature of position in the face-image.
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| PCT/CN2015/079429 WO2016183834A1 (en) | 2015-05-21 | 2015-05-21 | An apparatus and a method for locating facial landmarks of face image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107615295A true CN107615295A (en) | 2018-01-19 |
| CN107615295B CN107615295B (en) | 2020-09-25 |
Family
ID=57319095
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201580080396.8A Active CN107615295B (en) | 2015-05-21 | 2015-05-21 | Apparatus and method for locating key features of face image |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN107615295B (en) |
| WO (1) | WO2016183834A1 (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020248789A1 (en) * | 2019-06-11 | 2020-12-17 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and system for facial landmark detection using facial component-specific local refinement |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101877055A (en) * | 2009-12-07 | 2010-11-03 | 北京中星微电子有限公司 | Method and device for positioning key feature point |
| CN103377382A (en) * | 2012-04-27 | 2013-10-30 | 通用电气公司 | Optimum gradient pursuit for image alignment |
| US20140147022A1 (en) * | 2012-11-27 | 2014-05-29 | Adobe Systems Incorporated | Facial Landmark Localization By Exemplar-Based Graph Matching |
| US20140185924A1 (en) * | 2012-12-27 | 2014-07-03 | Microsoft Corporation | Face Alignment by Explicit Shape Regression |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8027521B1 (en) * | 2008-03-25 | 2011-09-27 | Videomining Corporation | Method and system for robust human gender recognition using facial feature localization |
| CN103824050B (en) * | 2014-02-17 | 2017-03-15 | 北京旷视科技有限公司 | A kind of face key independent positioning method returned based on cascade |
-
2015
- 2015-05-21 WO PCT/CN2015/079429 patent/WO2016183834A1/en active Application Filing
- 2015-05-21 CN CN201580080396.8A patent/CN107615295B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101877055A (en) * | 2009-12-07 | 2010-11-03 | 北京中星微电子有限公司 | Method and device for positioning key feature point |
| CN103377382A (en) * | 2012-04-27 | 2013-10-30 | 通用电气公司 | Optimum gradient pursuit for image alignment |
| US20140147022A1 (en) * | 2012-11-27 | 2014-05-29 | Adobe Systems Incorporated | Facial Landmark Localization By Exemplar-Based Graph Matching |
| US20140185924A1 (en) * | 2012-12-27 | 2014-07-03 | Microsoft Corporation | Face Alignment by Explicit Shape Regression |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107615295B (en) | 2020-09-25 |
| WO2016183834A1 (en) | 2016-11-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6832504B2 (en) | Object tracking methods, object tracking devices and programs | |
| US20240168563A1 (en) | Information processing device and method, program and recording medium for identifying a gesture of a person from captured image data | |
| CN103136504B (en) | Face identification method and device | |
| CN101558431B (en) | Face authentication device | |
| CN110569731B (en) | Face recognition method and device and electronic equipment | |
| US9984280B2 (en) | Object recognition system using left and right images and method | |
| CN105224947B (en) | classifier training method and system | |
| JP5235691B2 (en) | Information processing apparatus and information processing method | |
| CN107851192B (en) | Apparatus and method for detecting face part and face | |
| JP5166102B2 (en) | Image processing apparatus and method | |
| US9262672B2 (en) | Pattern recognition apparatus and pattern recognition method that reduce effects on recognition accuracy, and storage medium | |
| CN109117825A (en) | Lane line treating method and apparatus | |
| CN107423306B (en) | Image retrieval method and device | |
| Barman et al. | Shape: A novel graph theoretic algorithm for making consensus-based decisions in person re-identification systems | |
| US20190066311A1 (en) | Object tracking | |
| JP6997369B2 (en) | Programs, ranging methods, and ranging devices | |
| CN109146918B (en) | Self-adaptive related target positioning method based on block | |
| JP2009026326A (en) | Group study device and method | |
| JP2014102810A (en) | Subject recognition device, subject recognition method, and subject recognition program | |
| CN113297963A (en) | Multi-person posture estimation method and device, electronic equipment and readable storage medium | |
| JP2015204030A (en) | Authentication device and authentication method | |
| CN112488062B (en) | Image identification method, device, equipment and medium | |
| CN107615295A (en) | For the apparatus and method for the facial key feature for positioning face-image | |
| Li et al. | Robust visual tracking by exploiting the historical tracker snapshots | |
| Makris et al. | Robust 3d human pose estimation guided by filtered subsets of body keypoints |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |