CN106022269A - Face registering method and device - Google Patents
Face registering method and device Download PDFInfo
- Publication number
- CN106022269A CN106022269A CN201610344806.1A CN201610344806A CN106022269A CN 106022269 A CN106022269 A CN 106022269A CN 201610344806 A CN201610344806 A CN 201610344806A CN 106022269 A CN106022269 A CN 106022269A
- Authority
- CN
- China
- Prior art keywords
- resolution
- template
- grid
- registration
- facial image
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
Abstract
The invention provides a face registering method and device. A low-resolution grid template and a high-resolution grid template are set in advance; the low-resolution grid template is used to register an object face image, and a low-resolution registering result is obtained; a high-resolution registering reference value is determined via the low-resolution registering result and a grid conversion matrix, and the conversion matrix realizes conversion between the low-resolution grid template and the high-resolution grid template; and on the basis o the high-resolution registering reference value, the high-resolution grid template is used to register the object face image, and a final registering result is obtained. The low-resolution grid template is used to find an approximate area of the registering position, so that when the high-resolution grid template is used for registering, other local extreme points are not optimized, calculation results far from the reference value can be eliminated on the basis of the reference value, and an accurate registering result is obtained finally.
Description
Technical field
The present invention relates to technical field of image processing, particularly relate to a kind of face method for registering and device.
Background technology
Face registration is computer vision and the study hotspot in image perception field, and it is important as facial image analysis
Step, provides strong guarantee for follow-up recognition of face, Attitude estimation, Expression analysis, human face animation etc..
The essence of face method for registering is, finds the corresponding relation of faceform and benchmark model.People of the prior art
Face method for registering mainly includes two kinds, and the first is similarity transformation, and the second is affine transformation.Wherein, similarity transformation can be protected
The aspect ratio of witness's face does not changes, its displacement in scaled size, the anglec of rotation, horizontal direction and Vertical Square
These four enterprising line translations of degree of freedom of displacement upwards；And affine transformation is compared with similarity transformation, add two degree of freedom, i.e.
In the horizontal direction from carry out different scalings and angle in vertical direction and rotate.
But, although similarity transformation ensure that the ratio between shape and the facial characteristics of face does not changes, but
It is that, owing to the facial ratio of different people is the most inconsistent, this can cause the face of different people to be difficult to be registrated to a unified mould
Plate is up；Affine transformation has carried out comprehensive conversion to the ratio of face, although characteristic point can be reflected by it with minimum error
It is mapped to template up, but may cause being optimized to when optimizing minimum error in the Local Extremum of mistake.
Summary of the invention
In view of this, the invention provides a kind of face method for registering and device, use affine in order to solve prior art
Conversion carries out, in the scheme of face registration, to be optimized to the problem in the Local Extremum of mistake when optimizing minimum error,
Its technical scheme is as follows:
A kind of face method for registering, presets low resolution grid template and fine-resolution meshes template, described method
Including:
Utilize described low resolution grid template that target facial image is registrated, it is thus achieved that low resolution registration result；
By described low resolution registration result and grid conversion matrix, determine that high-resolution registrates reference value, described net
Lattice transition matrix is for realizing the conversion between described low resolution grid template and described fine-resolution meshes template；
Registrate reference value based on described high-resolution, utilize described fine-resolution meshes template to described target facial image
Registrate, it is thus achieved that the registration result of described target facial image.
Wherein, obtain the process of described grid conversion matrix, including:
By described low resolution grid template coordinate under geocentric coordinate system, and, described fine-resolution meshes mould
Plate coordinate under described geocentric coordinate system calculates described grid conversion matrix.
Preferably, the grid in described low resolution grid template is irregular triangle, described fine-resolution meshes
Grid in template is the triangle of rule.
Wherein, described utilize described low resolution grid template that target facial image is registrated, it is thus achieved that low resolution
Registration result, including:
Utilize described low resolution grid template, use Lucas-Kanade algorithm that described target facial image is joined
Accurate, it is thus achieved that described low resolution registration result；
Accordingly, described based on described high-resolution registration reference value, utilize described fine-resolution meshes template to described
Target facial image registrates, it is thus achieved that the registration result of described target facial image, including:
Registrate reference value based on described high-resolution, utilize described fine-resolution meshes template, use and comprise grid constraint
Lucas-Kanade algorithm described target facial image is registrated, it is thus achieved that the registration result of described target facial image.
Wherein, described based on described high-resolution registrate reference value, utilize described fine-resolution meshes template, employing comprises
Described target facial image is registrated by the Lucas-Kanade algorithm of grid constraint, particularly as follows:
Registrate reference value based on described high-resolution, utilize described fine-resolution meshes template, use grid constraint (XD-λAXA-λBXBλCXC)2+(YD-λAYA-λBYB-λCYC)2=0 combines described Lucas-Kanade algorithm enters described target facial image
Row registration；
Wherein, XA、XB、XC、XDFor described low resolution grid template intermediate cam shape ABC and the coordinate of neighbor point D, YA、YB、
YC、YDFor described fine-resolution meshes template intermediate cam shape ABC and the coordinate of neighbor point D, λA、λB、λCFor each vertex correspondence it is
Number.
A kind of face registration apparatus, described device includes: template sets module, the first registration module, determine module and the
Two registration module；
Described template sets module, is used for setting low resolution grid template and fine-resolution meshes template；
Described first registration module, is used for utilizing described template sets module described low resolution grid mould set in advance
Target facial image is registrated by plate, it is thus achieved that low resolution registration result；
Described determine module, for by the described low resolution registration result of described first registration module and grid conversion
Matrix, determines that high-resolution registrates reference value, and described grid conversion matrix is used for realizing described low resolution grid template and institute
State the conversion between fine-resolution meshes template；
Described second registration module, for registrating reference value based on the described described high-resolution determining that module is determined,
Utilize described template sets module described fine-resolution meshes template set in advance that described target facial image is registrated,
Obtain the registration result of described target facial image.
Wherein, described face registration apparatus also includes: acquisition module；
Described acquisition module, for by described low resolution grid template coordinate under geocentric coordinate system, and, institute
State fine-resolution meshes template coordinate under described geocentric coordinate system and calculate described grid conversion matrix.
Wherein, described template sets module, specifically for setting the grid low resolution grid mould as sealene triangle
Plate, and, grid is the fine-resolution meshes template of regular triangle.
Wherein, described first registration module includes the first registration submodule；
Described first registration submodule, is used for utilizing described low resolution grid template, uses Lucas-Kanade algorithm
Described target facial image is registrated, it is thus achieved that described low resolution registration result；
Wherein, described second registration module includes the second registration submodule；
Described second registration submodule, for registrating reference value based on described high-resolution, utilizes described high-resolution net
Grid template, described target facial image is registrated by the Lucas-Kanade algorithm comprising grid constraint described in employing, it is thus achieved that
The registration result of described target facial image.
Wherein, described second registration submodule, specifically for registrating reference value based on described high-resolution, utilize described height
Resolution grid template, retrains (X by gridD-λAXA-λBXBλCXC)2+(YD-λAYA-λBYB-λCYC)2=0 combination is described
Described target facial image is registrated by Lucas-Kanade algorithm；
Wherein, XA、XB、XC、XDFor described low resolution grid template intermediate cam shape ABC and the coordinate of neighbor point D, YA、YB、
YC、YDFor described fine-resolution meshes template intermediate cam shape ABC and the coordinate of neighbor point D, λA、λB、λCFor each vertex correspondence it is
Number.
Technique scheme has the advantages that
The face method for registering that the present invention provides, joins target facial image first with low resolution grid template
Standard, this registration process can find a registration position region substantially, so that using fine-resolution meshes template to carry out
During registration, it is unlikely to be optimized in other Local Extremum, utilizing low resolution grid template, target facial image is entered
After row registration, can determine that high-resolution registrates reference value by low resolution registration result and grid conversion matrix, and then
Based on this high-resolution registration reference value, fine-resolution meshes template is utilized target facial image to be carried out accuracy registration, in profit
During carrying out target facial image is registrated by fine-resolution meshes template, some deviations can be got rid of based on reference value
The result of calculation of the local minimum that reference value is bigger, further avoid the situation being optimized to other Local Extremum, finally
Obtain accurate registration result.
Accompanying drawing explanation
In order to be illustrated more clearly that the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
In having technology to describe, the required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is only this
Inventive embodiment, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to according to
The accompanying drawing provided obtains other accompanying drawing.
The schematic flow sheet of the face method for registering that Fig. 1 provides for the embodiment of the present invention；
Another schematic flow sheet of the face method for registering that Fig. 2 provides for the embodiment of the present invention；
Another schematic flow sheet of the face method for registering that Fig. 3 provides for the embodiment of the present invention；
One structural representation of the face registration apparatus that Fig. 4 provides for the embodiment of the present invention；
Another structural representation of the face registration apparatus that Fig. 5 provides for the embodiment of the present invention.
Detailed description of the invention
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on
Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under not making creative work premise
Embodiment, broadly falls into the scope of protection of the invention.
Embodiments providing a kind of face method for registering, the method carries out face based on graded mesh model and joins
Standard, wherein, graded mesh model includes low resolution grid template set in advance and fine-resolution meshes template, refers to figure
1, it is shown that the schematic flow sheet of this face method for registering, may include that
Step S101: utilize low resolution grid template that target facial image is registrated, it is thus achieved that low resolution registrates
Result.
Step S102: by low resolution registration result and grid conversion matrix, determines that high-resolution registrates reference value.
Wherein, grid conversion matrix is for realizing turning between low resolution grid template and fine-resolution meshes template
Change.
Step S103: registrate reference value based on high-resolution, utilize fine-resolution meshes template that target facial image is entered
Row registration, it is thus achieved that the registration result of target facial image.
The face method for registering that the embodiment of the present invention provides, first with low resolution grid template to target facial image
Registrating, this registration process can find a registration position region substantially, so that using fine-resolution meshes mould
When plate registrates, it is unlikely to be optimized in other Local Extremum, is utilizing low resolution grid template to target face
After image registrates, can determine that high-resolution registrates reference by low resolution registration result and grid conversion matrix
Value, and then registrate reference value based on this high-resolution, utilize fine-resolution meshes template that target facial image is accurately joined
Standard, during registrating target facial image utilizing fine-resolution meshes template to carry out, can get rid of based on reference value
The result of calculation of the local minimum that some deviation reference values are bigger, further avoid the feelings being optimized to other Local Extremum
Condition, finally gives accurate registration result.
Refer to Fig. 2, it is shown that another schematic flow sheet of the face method for registering that the embodiment of the present invention provides, the method
Carrying out face registration based on graded mesh model, wherein, graded mesh model includes low resolution grid template set in advance
With fine-resolution meshes template, it is preferred that the grid of low resolution grid template is irregular triangle gridding, compared to rule
Grid is more flexible, and it can avoid the region that the grey scale change such as eyes, the wing of nose, eyebrow are violent, the net of fine-resolution meshes template
Lattice are the triangle gridding of rule, and it is difficult to drain message, for face is carried out more accurate affine transformation, the embodiment of the present invention
The face method for registering provided may include that
Step S201: utilize low resolution grid template that target facial image is registrated, it is thus achieved that low resolution registrates
Result.
Concrete, utilize low resolution grid template that target facial image is registrated, it is thus achieved that low resolution registration knot
Really, including: utilize low resolution grid template, use Lucas-Kanade algorithm that target facial image is registrated, it is thus achieved that
Low resolution registration result.
It should be noted that when using low resolution grid template that target facial image is registrated, one can be found
The individual alignment band of position substantially so that when using fine-resolution meshes template to registrate, be unlikely to be optimized to other
In Local Extremum.
Step S202: by low resolution grid template coordinate under geocentric coordinate system, and fine-resolution meshes mould
Plate coordinate under geocentric coordinate system calculates grid conversion matrix.
Exemplary, low resolution grid template is Mesh1, and fine-resolution meshes template is Mesh2.In center-of-mass coordinate
Under, there is pair relationhip in Mesh1 Yu Mesh2, available grids transfer equation M2=TM1 describes, and wherein, T is grid conversion matrix,
It can be tried to achieve by Mesh1 and Mesh2 coordinate under geocentric coordinate system.
Step S203: by low resolution registration result and grid conversion matrix, determines that high-resolution registrates reference value.
Low resolution matching result is obtained target image being carried out coupling by low resolution grid template Mesh1
Mesh1 ＇, and, after being calculated grid conversion matrix T, can be calculated by Mesh1 ＇ and grid conversion matrix T
Mesh2 ＇.
It should be noted that low resolution grid template Mesh1 and fine-resolution meshes template Mesh2 are by replacing matrix
T changes, i.e. the two exists the relation of M2=TM1, and target image is mated the low resolution obtained by Mesh1
Matching result Mesh1 ＇, and by Mesh2 target image mated the low resolution matching result Mesh2 ＇ that obtains it
Between change again by grid conversion matrix T, i.e. replace matrix constant, Mesh1 ＇ and Mesh2 ＇ equally exists following grid
Transfer equation: M2 ＇=TM1 ＇.
In the present embodiment, at the Lucas-Kanade algorithm utilizing Mesh1 use to comprise grid constraint, face is carried out
After registration, it is thus achieved that low resolution registration result, low resolution registration result combines grid conversion matrix T can calculate use
Mesh2 carries out the registration result registrated, and this registration result remains low resolution, can not realize accuracy registration, but its energy
Enough for accuracy registration offer reference value.
Step S204: registrate reference value based on high-resolution, utilize fine-resolution meshes template that target facial image is entered
Row registration, it is thus achieved that the registration result of target facial image.
Concrete, registrate reference value based on high-resolution, utilize fine-resolution meshes template that target facial image is carried out
Registration, it is thus achieved that the registration result of target facial image, including: registrate reference value based on high-resolution, utilize fine-resolution meshes
Template, uses the Lucas-Kanade algorithm comprising grid constraint to registrate target facial image, it is thus achieved that target face figure
The registration result of picture.
In the present embodiment, under grid retrains, by fine-resolution meshes template and target facial image being carried out
The error function joined asks for local minimum, the matching scheme that available error is minimum, i.e. best matching result.Concrete, adopt
When registrating target facial image with the Lucas-Kanade algorithm comprising grid constraint, target setting function is:
E (s)=∑ (T (x；Δs)-I(W(x；Δs)))2+λsSTKs (1)
Grid is constrained to:
(XD-λAXA-λBXBλCXC)2+(YD-λAYA-λBYB-λCYC)2=0 (2)
Wherein, the T in formula (1) is fine-resolution meshes template, and I is target facial image, and S is grid vertex A, B, C
Coordinate, W (x；Δ s) represent image affine transformation, Δ s is the parameter of piecemeal affine transformation, x be the affine triangle of xth to
Amount, λ s is the coefficient lambda that summit A, B, C are correspondingA、λB、λCVector.X in formula (2)A、XB、XC、XDFor low resolution grid template
Intermediate cam shape ABC and the coordinate of neighbor point D, YA、YB、YC、YDFor fine-resolution meshes template intermediate cam shape ABC and neighbor point D
Coordinate, λA、λB、λCFor the corresponding coefficient in each summit.
Wherein, the meaning of grid constraint is, when locally there is affine transformation, deformation is 0 to the punishment of object function, i.e.
When locally there is affine transformation, when some triangle block carries out affine transformation, the error function of full face is not changed in, thus not
Affecting the affine transformation in other region, when solving direct application affine transformation coupling triangle piecemeal, easily there is mistake in other parts
The problem of deformation by mistake.
The face method for registering that the embodiment of the present invention provides, uses Lucas-first with low resolution grid template
Target facial image is registrated by Kanade algorithm, and this registration process can find a registration position region substantially, thus
Make, when using fine-resolution meshes template to registrate, to be unlikely to be optimized in other Local Extremum, low utilizing
After target facial image is registrated by resolution grid template, by low resolution registration result and grid conversion matrix
Can determine that high-resolution registrates reference value, so based on this high-resolution registration reference value, utilize fine-resolution meshes template, adopt
With the Lucas-Kanade algorithm comprising grid constraint, target facial image is carried out accuracy registration, utilize fine-resolution meshes
During template carries out registrating target facial image, can get rid of, based on reference value, the office that some deviation reference values are bigger
The minimizing result of calculation in portion, further avoid the situation being optimized to other Local Extremum, and final acquisition registrates accurately
Result.Registrate it addition, the embodiment of the present invention combines grid constraint, solve directly application affine transformation coupling triangle piecemeal
Time, the problem that other parts are susceptible to mistake deformation.
In the face method for registering that above-described embodiment provides, the calculation procedure of grid conversion matrix is utilizing low resolution
Performing after the step for that target facial image being registrated by net template, the present embodiment is not limited to this, grid conversion
As long as the calculating of matrix by low resolution registration result and grid conversion matrix determine high-resolution registration reference value this
Perform before step.
Refer to Fig. 3, it is shown that another schematic flow sheet of the face method for registering that the embodiment of the present invention provides, the method
Carrying out face registration based on graded mesh model, wherein, graded mesh model includes low resolution grid template set in advance
With fine-resolution meshes template, it is preferred that the grid of low resolution grid template is irregular triangle gridding, compared to rule
Grid is more flexible, and it can avoid the region that the grey scale change such as eyes, the wing of nose, eyebrow are violent, the net of fine-resolution meshes template
Lattice are the triangle gridding of rule, and it is difficult to drain message, and for face is carried out more accurate affine transformation, the method can be wrapped
Include:
Step S301: by low resolution grid template coordinate under geocentric coordinate system, and, fine-resolution meshes mould
Plate coordinate under geocentric coordinate system calculates grid conversion matrix.
Concrete, utilize low resolution grid template that target facial image is registrated, it is thus achieved that low resolution registration knot
Really, including: utilize low resolution grid template, use Lucas-Kanade algorithm that target facial image is registrated, it is thus achieved that
Low resolution registration result.
Step S302: utilize low resolution grid template that target facial image is registrated, it is thus achieved that low resolution registrates
Result.
Step S303: by low resolution registration result and grid conversion matrix, determines that high-resolution registrates reference value.
Step S304: registrate reference value based on high-resolution, utilize fine-resolution meshes template that target facial image is entered
Row registration, it is thus achieved that the registration result of target facial image.
Concrete, registrate reference value based on high-resolution, utilize fine-resolution meshes template that target facial image is carried out
Registration, it is thus achieved that the registration result of target facial image, including: registrate reference value based on high-resolution, utilize fine-resolution meshes
Template, uses the Lucas-Kanade algorithm comprising grid constraint to registrate target facial image, it is thus achieved that target face figure
The registration result of picture.
In the present embodiment, the object function and the grid constraint that comprise the Lucas-Kanade algorithm setting of grid constraint please
See formula (1) and formula (2).
It should be noted that the face that the face method for registering of the present embodiment offer and above-described embodiment provide registrates not
The execution sequence the step for of being only that the calculating of grid conversion matrix with part, the specific implementation of its each step and right
Explanation in each step can be found in above-described embodiment, and therefore not to repeat here.
The face method for registering that the embodiment of the present invention provides, uses Lucas-first with low resolution grid template
Target facial image is registrated by Kanade algorithm, and this registration process can find a registration position region substantially, thus
Make, when using fine-resolution meshes template to registrate, to be unlikely to be optimized in other Local Extremum, low utilizing
After target facial image is registrated by resolution grid template, by low resolution registration result and grid conversion matrix
Can determine that high-resolution registrates reference value, so based on this high-resolution registration reference value, utilize fine-resolution meshes template, adopt
With the Lucas-Kanade algorithm comprising grid constraint, target facial image is carried out accuracy registration, utilize fine-resolution meshes
During template carries out registrating target facial image, can get rid of, based on reference value, the office that some deviation reference values are bigger
The minimizing result of calculation in portion, further avoid the situation being optimized to other Local Extremum, and final acquisition registrates accurately
Result.Registrate it addition, the face method for registering that the embodiment of the present invention provides combines grid constraint, solve directly application imitative
When penetrating Transformation Matching triangle piecemeal, the problem that other parts are susceptible to mistake deformation.
Corresponding with said method, the embodiment of the present invention additionally provides a kind of face registration apparatus, and this device is based on layering
Grid model carries out face registration, and wherein, graded mesh model includes low resolution grid template set in advance and high-resolution
Rate net template, refers to Fig. 4, it is shown that the structural representation of this device, this device may include that template sets module 400,
First registration module 401, determine module 402 and the second registration module 403.Wherein:
Template sets module 400, is used for presetting low resolution grid template and fine-resolution meshes template.
First registration module 401, is used for utilizing template sets module 400 low resolution grid set in advance template to mesh
Mark facial image registrates, it is thus achieved that low resolution registration result.
Determine module 402, be used for the low resolution registration result by the first registration module 401 and grid conversion matrix,
Determine that high-resolution registrates reference value.
Wherein, grid conversion matrix is for realizing turning between low resolution grid template and fine-resolution meshes template
Change.
Second registration module 403, for based on a determination that the high-resolution that module 402 is determined registrates reference value, utilizing mould
Target facial image is registrated by plate setting module 400 fine-resolution meshes set in advance template, it is thus achieved that target face figure
The registration result of picture.
The face registration apparatus that the embodiment of the present invention provides, first with low resolution grid template to target facial image
Registrating, this registration process can find a registration position region substantially, so that using fine-resolution meshes mould
When plate registrates, it is unlikely to be optimized in other Local Extremum, is utilizing low resolution grid template to target face
After image registrates, can determine that high-resolution registrates reference by low resolution registration result and grid conversion matrix
Value, and then registrate reference value based on this high-resolution, utilize fine-resolution meshes template that target facial image is accurately joined
Standard, during registrating target facial image utilizing fine-resolution meshes template to carry out, can get rid of based on reference value
The result of calculation of the local minimum that some deviation reference values are bigger, further avoid the feelings being optimized to other Local Extremum
Condition, finally gives accurate registration result.
In the above-described embodiments, grid conversion matrix can be obtained by acquisition module, it is thus achieved that the acquisition of grid conversion matrix
Module, for by low resolution grid template coordinate under geocentric coordinate system, and, fine-resolution meshes template is at barycenter
Coordinate under coordinate system calculates grid conversion matrix.
Refer to Fig. 5, it is shown that the structural representation of the another kind of face registration apparatus that the embodiment of the present invention provides, this dress
Putting and carry out face registration based on graded mesh model, wherein, graded mesh model includes low resolution grid mould set in advance
Plate and fine-resolution meshes template, this device may include that template sets module the 500, first registration module 501, computing module
502, module 503 and the second registration module 504 are determined.Wherein:
Template sets module 500, is used for presetting low resolution grid template and fine-resolution meshes template.
First registration module 501, is used for utilizing template sets module 500 low resolution grid set in advance template to mesh
Mark facial image registrates, it is thus achieved that low resolution registration result.
Computing module 502, for by low resolution grid template coordinate under geocentric coordinate system, and, high-resolution
Rate net template coordinate under geocentric coordinate system calculates grid conversion matrix.
Determine module 503, based on by the low resolution registration result of the first registration module 501 and computing module 502
The grid conversion matrix obtained, determines that high-resolution registrates reference value.
Second registration module 504, for based on a determination that the high-resolution that module 503 is determined registrates reference value, utilizing mould
Target facial image is registrated by plate setting module 500 fine-resolution meshes set in advance template, it is thus achieved that target face figure
The registration result of picture.
The face registration apparatus that the embodiment of the present invention provides, first with low resolution grid template to target facial image
Registrating, this registration process can find a registration position region substantially, so that using fine-resolution meshes mould
When plate registrates, it is unlikely to be optimized in other Local Extremum, is utilizing low resolution grid template to target face
After image registrates, can determine that high-resolution registrates reference by low resolution registration result and grid conversion matrix
Value, and then registrate reference value based on this high-resolution, utilize fine-resolution meshes template that target facial image is accurately joined
Standard, during registrating target facial image utilizing fine-resolution meshes template to carry out, can get rid of based on reference value
The result of calculation of the local minimum that some deviation reference values are bigger, further avoid the feelings being optimized to other Local Extremum
Condition, finally gives accurate registration result.
Preferably, the grid in low resolution grid template in any of the above-described embodiment is irregular triangle, high
Grid in resolution grid template is the triangle of rule.
In any of the above-described embodiment, the first registration module includes the first registration submodule.
First registration submodule, is used for utilizing low resolution grid template, uses Lucas-Kanade algorithm to target person
Face image registrates, it is thus achieved that low resolution registration result.
In any of the above-described embodiment, the second registration module includes the second registration submodule.
Second registration submodule, for registrating reference value based on high-resolution, utilizes fine-resolution meshes template, uses bag
Target facial image is registrated by the Lucas-Kanade algorithm containing grid constraint, it is thus achieved that the registration knot of target facial image
Really.
In any of the above-described embodiment, the second registration submodule uses the Lucas-Kanade algorithm pair comprising grid constraint
When target facial image registrates, target setting function is:
E (s)=∑ (T (x；Δs)-I(W(x；Δs)))2+λsSTKs
Grid is constrained to:
(XD-λAXA-λBXBλCXC)2+(YD-λAYA-λBYB-λCYC)2=0
Wherein, T is fine-resolution meshes template, and I is target facial image, and S is the coordinate of grid vertex A, B, C, W (x；
Δ s) represents the affine transformation of image, and Δ s is the parameter of piecemeal affine transformation, and x is the vector of the affine triangle of xth, and λ s is top
The coefficient lambda that some A, B, C is correspondingA、λB、λCVector.XA、XB、XC、XDFor low resolution grid template intermediate cam shape ABC and neighbouring
The coordinate of some D, YA、YB、YC、YDFor fine-resolution meshes template intermediate cam shape ABC and the coordinate of neighbor point D, λA、λB、λCFor respectively
The corresponding coefficient in summit.
In this specification, each embodiment uses the mode gone forward one by one to describe, and what each embodiment stressed is and other
The difference of embodiment, between each embodiment, identical similar portion sees mutually.
In several embodiments provided herein, it should be understood that disclosed method, device and equipment, permissible
Realize by another way.Such as, device embodiment described above is only schematically, such as, and described unit
Dividing, be only a kind of logic function and divide, actual can have other dividing mode, the most multiple unit or assembly when realizing
Can in conjunction with or be desirably integrated into another system, or some features can be ignored, or does not performs.Another point, shown or
The coupling each other discussed or direct-coupling or communication connection can be by between some communication interfaces, device or unit
Connect coupling or communication connection, can be electrical, machinery or other form.
The described unit illustrated as separating component can be or may not be physically separate, shows as unit
The parts shown can be or may not be physical location, i.e. may be located at a place, or can also be distributed to multiple
On NE.Some or all of unit therein can be selected according to the actual needs to realize the mesh of the present embodiment scheme
's.It addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, it is also possible to be each
Unit is individually physically present, it is also possible to two or more unit are integrated in a unit.
If described function is using the form realization of SFU software functional unit and as independent production marketing or use, permissible
It is stored in a computer read/write memory medium.Based on such understanding, technical scheme is the most in other words
The part contributing prior art or the part of this technical scheme can embody with the form of software product, this meter
Calculation machine software product is stored in a storage medium, including some instructions with so that a computer equipment (can be individual
People's computer, server, or the network equipment etc.) perform all or part of step of method described in each embodiment of the present invention.
And aforesaid storage medium includes: USB flash disk, portable hard drive, read only memory (ROM, Read-Only Memory), random access memory are deposited
The various media that can store program code such as reservoir (RAM, Random Access Memory), magnetic disc or CD.
Described above to the disclosed embodiments, makes professional and technical personnel in the field be capable of or uses the present invention.
Multiple amendment to these embodiments will be apparent from for those skilled in the art, as defined herein
General Principle can realize without departing from the spirit or scope of the present invention in other embodiments.Therefore, the present invention
It is not intended to be limited to the embodiments shown herein, and is to fit to and principles disclosed herein and features of novelty phase one
The widest scope caused.
Claims (10)
1. a face method for registering, it is characterised in that preset low resolution grid template and fine-resolution meshes template,
Described method includes:
Utilize described low resolution grid template that target facial image is registrated, it is thus achieved that low resolution registration result；
By described low resolution registration result and grid conversion matrix, determining that high-resolution registrates reference value, described grid turns
Change matrix for realizing the conversion between described low resolution grid template and described fine-resolution meshes template；
Registrate reference value based on described high-resolution, utilize described fine-resolution meshes template that described target facial image is carried out
Registration, it is thus achieved that the registration result of described target facial image.
Face method for registering the most according to claim 1, it is characterised in that obtain the process of described grid conversion matrix,
Including:
By described low resolution grid template coordinate under geocentric coordinate system, and, described fine-resolution meshes template exists
Coordinate under described geocentric coordinate system calculates described grid conversion matrix.
Face method for registering the most according to claim 1, it is characterised in that the grid in described low resolution grid template
For irregular triangle, the grid in described fine-resolution meshes template is the triangle of rule.
Face method for registering the most as claimed in any of claims 1 to 3, it is characterised in that described utilize described low
Target facial image is registrated by resolution grid template, it is thus achieved that low resolution registration result, including:
Utilize described low resolution grid template, use Lucas-Kanade algorithm that described target facial image is registrated,
Obtain described low resolution registration result；
Accordingly, described based on described high-resolution registration reference value, utilize described fine-resolution meshes template to described target
Facial image registrates, it is thus achieved that the registration result of described target facial image, including:
Registrate reference value based on described high-resolution, utilize described fine-resolution meshes template, use and comprise grid constraint
Described target facial image is registrated by Lucas-Kanade algorithm, it is thus achieved that the registration result of described target facial image.
Face method for registering the most according to claim 4, it is characterised in that described based on the registration reference of described high-resolution
Value, utilizes described fine-resolution meshes template, uses the Lucas-Kanade algorithm comprising grid constraint to described target face
Image registrates, particularly as follows:
Registrate reference value based on described high-resolution, utilize described fine-resolution meshes template, use grid constraint (XD-λAXA-λBXBλCXC)2+(YD-λAYA-λBYB-λCYC)2=0 combines described Lucas-Kanade algorithm joins described target facial image
Accurate；
Wherein, XA、XB、XC、XDFor described low resolution grid template intermediate cam shape ABC and the coordinate of neighbor point D, YA、YB、YC、YD
For described fine-resolution meshes template intermediate cam shape ABC and the coordinate of neighbor point D, λA、λB、λCCoefficient for each vertex correspondence.
6. a face registration apparatus, it is characterised in that described device includes: template sets module, the first registration module, determine
Module and the second registration module；
Described template sets module, is used for setting low resolution grid template and fine-resolution meshes template；
Described first registration module, is used for utilizing described template sets module described low resolution grid template pair set in advance
Target facial image registrates, it is thus achieved that low resolution registration result；
Described determine module, for by the described low resolution registration result of described first registration module and grid conversion square
Battle array, determines that high-resolution registrates reference value, and described grid conversion matrix is used for realizing described low resolution grid template with described
Conversion between fine-resolution meshes template；
Described second registration module, for based on the described described high-resolution registration reference value determining that module is determined, utilizing
Described target facial image is registrated by described template sets module described fine-resolution meshes template set in advance, it is thus achieved that
The registration result of described target facial image.
Face registration apparatus the most according to claim 6, it is characterised in that described face registration apparatus also includes: obtain
Module；
Described acquisition module, for by described low resolution grid template coordinate under geocentric coordinate system, and, described height
Resolution grid template coordinate under described geocentric coordinate system calculates described grid conversion matrix.
Face registration apparatus the most according to claim 6, it is characterised in that described template sets module, specifically for setting
Determine the low resolution grid template that grid is sealene triangle, and, grid is the fine-resolution meshes mould of regular triangle
Plate.
9. according to the face registration apparatus described in any one in claim 6 to 8, it is characterised in that
Described first registration module includes the first registration submodule；
Described first registration submodule, is used for utilizing described low resolution grid template, uses Lucas-Kanade algorithm to institute
State target facial image to registrate, it is thus achieved that described low resolution registration result；
Described second registration module includes the second registration submodule；
Described second registration submodule, for registrating reference value based on described high-resolution, utilizes described fine-resolution meshes mould
Plate, described target facial image is registrated by the Lucas-Kanade algorithm comprising grid constraint described in employing, it is thus achieved that described
The registration result of target facial image.
Face registration apparatus the most according to claim 9, it is characterised in that described second registration submodule, specifically for
Registrate reference value based on described high-resolution, utilize described fine-resolution meshes template, retrain (X by gridD-λAXA-λBXBλCXC)2+(YD-λAYA-λBYB-λCYC)2=0 combines described Lucas-Kanade algorithm registrates described target facial image；
Wherein, XA、XB、XC、XDFor described low resolution grid template intermediate cam shape ABC and the coordinate of neighbor point D, YA、YB、YC、YD
For described fine-resolution meshes template intermediate cam shape ABC and the coordinate of neighbor point D, λA、λB、λCCoefficient for each vertex correspondence.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610344806.1A CN106022269B (en) | 2016-05-23 | 2016-05-23 | A kind of face method for registering and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610344806.1A CN106022269B (en) | 2016-05-23 | 2016-05-23 | A kind of face method for registering and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106022269A true CN106022269A (en) | 2016-10-12 |
| CN106022269B CN106022269B (en) | 2019-04-16 |
Family
ID=57097062
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610344806.1A Expired - Fee Related CN106022269B (en) | 2016-05-23 | 2016-05-23 | A kind of face method for registering and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106022269B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210209732A1 (en) * | 2020-06-17 | 2021-07-08 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Face super-resolution realization method and apparatus, electronic device and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101216939A (en) * | 2008-01-04 | 2008-07-09 | 江南大学 | A multi-resolution medical image registration method based on quantum behaviors particle swarm algorithm |
| CN102262781A (en) * | 2011-05-11 | 2011-11-30 | 浙江工业大学 | Method for registration of ink-jet printing texture image based on unit decomposition optical flow field |
| CN102651127A (en) * | 2012-04-01 | 2012-08-29 | 深圳市万兴软件有限公司 | Image processing method and image processing system for super-resolution reconstruction |
| US20150304634A1 (en) * | 2011-08-04 | 2015-10-22 | John George Karvounis | Mapping and tracking system |
| CN105190692A (en) * | 2013-05-09 | 2015-12-23 | 皇家飞利浦有限公司 | Method and system for mesh segmentation and mesh registration |
-
2016
- 2016-05-23 CN CN201610344806.1A patent/CN106022269B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101216939A (en) * | 2008-01-04 | 2008-07-09 | 江南大学 | A multi-resolution medical image registration method based on quantum behaviors particle swarm algorithm |
| CN102262781A (en) * | 2011-05-11 | 2011-11-30 | 浙江工业大学 | Method for registration of ink-jet printing texture image based on unit decomposition optical flow field |
| US20150304634A1 (en) * | 2011-08-04 | 2015-10-22 | John George Karvounis | Mapping and tracking system |
| CN102651127A (en) * | 2012-04-01 | 2012-08-29 | 深圳市万兴软件有限公司 | Image processing method and image processing system for super-resolution reconstruction |
| CN105190692A (en) * | 2013-05-09 | 2015-12-23 | 皇家飞利浦有限公司 | Method and system for mesh segmentation and mesh registration |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210209732A1 (en) * | 2020-06-17 | 2021-07-08 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Face super-resolution realization method and apparatus, electronic device and storage medium |
| US11710215B2 (en) * | 2020-06-17 | 2023-07-25 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Face super-resolution realization method and apparatus, electronic device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106022269B (en) | 2019-04-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11514642B2 (en) | Method and apparatus for generating two-dimensional image data describing a three-dimensional image | |
| CN102663820B (en) | Three-dimensional head model reconstruction method | |
| CN103824050B (en) | A kind of face key independent positioning method returned based on cascade | |
| CN101404091B (en) | Three-dimensional human face reconstruction method and system based on two-step shape modeling | |
| CN100430690C (en) | Method for making three-dimensional measurement of objects utilizing single digital camera to freely shoot | |
| Tseng et al. | Semiautomated building extraction based on CSG model-image fitting | |
| US11610373B2 (en) | Method of generating three-dimensional model data of object | |
| JP4785880B2 (en) | System and method for 3D object recognition | |
| CN110832501A (en) | System and method for pose-invariant face alignment | |
| CN104933755A (en) | Static object reconstruction method and system | |
| CN102722697B (en) | Unmanned aerial vehicle autonomous navigation landing visual target tracking method | |
| CN105654492A (en) | Robust real-time three-dimensional (3D) reconstruction method based on consumer camera | |
| CN102750704B (en) | Step-by-step video camera self-calibration method | |
| Bednarik et al. | Shape reconstruction by learning differentiable surface representations | |
| CN103649998A (en) | Method for determining a parameter set designed for determining the pose of a camera and/or for determining a three-dimensional structure of the at least one real object | |
| CN103838437A (en) | Touch positioning control method based on projection image | |
| CN109635752A (en) | Localization method, face image processing process and the relevant apparatus of face key point | |
| CN105139007B (en) | Man face characteristic point positioning method and device | |
| CN107862707A (en) | A kind of method for registering images based on Lucas card Nader's image alignment | |
| CN108122280A (en) | The method for reconstructing and device of a kind of three-dimensional point cloud | |
| CN112132739A (en) | 3D reconstruction and human face posture normalization method, device, storage medium and equipment | |
| CN106127743B (en) | The method and system of automatic Reconstruction bidimensional image and threedimensional model accurate relative location | |
| CN107330980A (en) | A kind of virtual furnishings arrangement system based on no marks thing | |
| CN106547724A (en) | Theorem in Euclid space coordinate transformation parameter acquisition methods based on minimum point set | |
| CN103854271B (en) | A kind of planar pickup machine scaling method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Kong Lingsheng Inventor after: Liu Xiaofeng Inventor after: Diao Zhihui Inventor after: Yan Junliang Inventor after: Jia Ping Inventor after: Cao Xinyu Inventor before: Kong Lingsheng Inventor before: Liu Xiaofeng Inventor before: Diao Zhihui Inventor before: Yan Junliang Inventor before: Jia Ping |
|
| CB03 | Change of inventor or designer information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190416 Termination date: 20210523 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |