CN106570459A - Face image processing method - Google Patents
Face image processing method Download PDFInfo
- Publication number
- CN106570459A CN106570459A CN201610905128.1A CN201610905128A CN106570459A CN 106570459 A CN106570459 A CN 106570459A CN 201610905128 A CN201610905128 A CN 201610905128A CN 106570459 A CN106570459 A CN 106570459A
- Authority
- CN
- China
- Prior art keywords
- face
- characteristics vector
- coordinate
- state
- singularity characteristics
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/169—Holistic features and representations, i.e. based on the facial image taken as a whole
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a face image processing method which processes the face image to obtain a formal model and includes four main steps. The face image is processed through the steps, in this process, the correction problem of the image position is considered, and the formal model of the face image is generated to embody the feature of the face image and to facilitate the image data storage. The formal model can strictly express the face image and can process the subsequent face image.
Description
Technical field
The present invention relates to a kind of image processing field, is related to a kind of processing method of facial image.
Background technology
Face can provide important information as the most important expression organ of the mankind to us：Such as sex, race, feelings
Thread, age and personality etc., therefore face image processing technology also just necessarily becomes the important means of man-machine interaction.Additionally, from
Substantially studying the rule of face can help us to deepen the grasp to the general consciousness cooked mode rule of the mankind, can also have
Effect promotes the research and development of machine vision.Existing face image processing process is only from image, to image
Quality be processed, make the quality of image purer, but existing image procossing mode more needs to be to pass through
Process can strictly represent facial image and pictures subsequent can be facilitated to process.
The content of the invention
To solve problem above, the present invention provides a kind of processing method of facial image.To reach above-mentioned technical proposal
Effect, the technical scheme is that：The step of a kind of processing method of facial image, the method, is as follows：
1) training set of a colorized face images is given, using HLS colour model transfer algorithms, HLS colour models turn
Scaling method is nonlinear, has the advantages that edge brightness noise is few, smooth effect is good, by colorized face images training set
Each width colorized face images are converted into Gray Face image, by the corresponding Gray Face image training of Gray Face image construction
Collection；
2) Lycoperdon polymorphum Vitt facial image training set is sampled, obtains sample window set, sample window set digitized is obtained
Sample window matrix, gathers sample window matrix conversion into singularity characteristics vector according to singular value decomposition theorem, singularity characteristics vector
Set is into the average mark implantation of calculating characteristic point, calculating formula is as follows by feature point group one by one：
In formula：I=1,2 ..., M, siFor the coordinate figure of the characteristic point, P is characterized average mark implantation a little, and variable M is used
In the quantity of recording feature point；
3) according to the average mark implantation of characteristic point, using calculated coordinate figure as new coordinate axess origin, in order to disappear
Except the difference of face location, carry out the translation of face and carry out rotation transformation, each is strange during first is gathered singularity characteristics vector
Different characteristic vector all deducts new coordinate axess origin value, the vector set of singularity characteristics after being translated, and then observer is on the face
The coordinate of eyebrow, by the coordinate (x of two eyebrows on face1, y1)、(x2, y2) connect the tilt angle alpha for obtaining face into a line,
The computing formula of tilt angle alpha is as follows：
Singularity characteristics vector set after the translation is surrounded around the new zero rotationAngle, rotation parameter
ForThe coordinate of the singularity characteristics vector set after all translations is all multiplied by rotation parameter, is rotated
Singularity characteristics vector set after conversion；
4) by the singularity characteristics vector set even partition after rotation transformation, it is corresponding to set up state matrix, state matrix
Middle state is corresponded with the singularity characteristics vector set coordinate after rotation transformation, and calculates turning between state in state matrix
Probability is changed, Probability State model is built by this process, the composition ultimate unit of Probability State model is in state matrix
State, and each state corresponds to the transition probability of other states, and the Probability State model for finally giving is a kind of formalization
Model.
Specific embodiment
In order that the technical problem to be solved, technical scheme and beneficial effect become more apparent, below tie
Embodiment is closed, the present invention will be described in detail.It should be noted that specific embodiment described herein is only to explain
The present invention, is not intended to limit the present invention, and the product that can realize said function belongs to equivalent and improvement, is all contained in this
Within bright protection domain.Concrete grammar is as follows：
Embodiment one：Real facial image is colored, and these colors can provide more richer than Gray Face image
Rich information.However, because gray level image has the characteristics of being easily handled, and most of classical image processing methods are all based on
Gray level image, therefore, if one width coloured image is converted into into gray level image through certain conversion.The first step of this method just will
Colorized face images are converted to Gray Face image, make in the gray level image comprising the most features in original color image
Information, then, subsequent treatment can just adopt classical image processing method, greatly reduce amount of calculation.
Principal component feature gray level image in order to obtain colorized face images, can be simulated using the optimal base for proposing
Karhunen-Loeve transformation method, so as to obtain the principal component feature image of the new most characteristic informations for containing coloured image.It is unusual
It is a kind of algebraic characteristic extracting method of effective image that value is decomposed.Singular value features are stable when image is described, and
With critical natures such as transposition invariance, rotational invariance, shift invariant, mirror transformation invariance, therefore singular value features
Can describe as a kind of effective algebraic characteristic of image.Singularity value decomposition is at Image Data Compression, signal
It is widely applied in reason and pattern analyses.After having obtained singular value, need to be analyzed face, using principal component analysiss
Face identification method.The method by facial image by row or) launch, define a high dimension vector, be seen as it is a kind of with
Machine vector, therefore its orthogonal K-L substrate can be obtained using Karhunen-Loeve transformation.Have corresponding to the substrate of wherein larger eigenvalue
With the shape of human face similarity, therefore eigenface is called.Thereafter, face is described using relatively small Eigenface collection, is made per width
Facial image then corresponds to a relatively low weight vector of dimension, therefore, what recognition of face can be after dimensionality reduction is spatially carried out.So
And, it is optimal Expressive Features that the shortcoming of the method is the feature for obtaining, rather than optimal classification feature.Feature is further
Projection, makes scatter matrix between the class of the pattern sample after projection maximum using linear discriminant method, and while spreading in its class
Matrix is minimum, after projection Assured Mode sample have simultaneously in new space in the class of maximum between class distance and minimum away from
From that is, the pattern sample has within this space optimal separability.The set of eigenvectors that Fisher linear discriminant analysiss are extracted
What is paid attention to is the difference of different faces, rather than the change in lighting condition, human face expression and direction.Thus, using the method pair
The change of illumination condition, human face posture etc. is less sensitive, so as to be favorably improved recognition effect.However, due in normal condition
Human face recognizes a problem always small sample problem, therefore scatter matrix always becomes for singular matrix the solution of the method in its class
Obtain highly difficult.
Embodiment two：Due to the interference of the extraneous factors such as shooting angle, distance, sitting posture, and manual feature point for calibration is deposited
In certain error, the Lycoperdon polymorphum Vitt facial image training set of acquisition is likely to appear in different positions, with different sizes and rotation
Gyration, it is therefore desirable to the face PDM models for further being obtained by manual fixed point, wherein containing a lot " non-shapes "
Factor, further comprises many redundancies in addition to the shape of face, and the factor of these redundancies can be to face below
Feature extraction produces deleterious effect.In order to eliminate the impact of this redundancy factor, train in the facial image to gray scale
Collection is carried out before statistical analysiss, it is necessary to it is initialized, that is, is alignd.The method that can be used to align is a lot, by suitable
When translation, rotation, scaling conversion, snap to same framework on the basis of global shape for not changing points distribution models
Under, so as to change the rambling state of the initial data of acquisition, reduce redundancy Factors on Human face points distribution models number
According to the impact of analysis.
Carry out in discrete space in the feature point search of face, the point for searching not is truly continuous
Extreme point in space, so that the discrete space obtained using oneself clicks through the extreme point that row interpolation obtains continuous space, this
Plant extreme point and be frequently found in sub- metric space plane.D (x) is expanded to into quadratic term according to Taylor series：
X=(x, y, σ) is space-yardstick coordinate in formula, and D (x) is the value of approaching of continuous space extreme point, quadratic each
Term coefficient can pass through oneself the adjacent discrete scale-space point Difference Calculation of acquisition and approximately obtain.Then D is asked for) extreme value (x),
MeetSo as to obtain the extreme point that degree of accuracy is sub-pix-Asia yardstick level：
Determine reference direction by being characterized, can L justice construction and directional correlation feature description vector so that feature
Possesses rotational invariance.The direction of characteristic point determines according to the Gradient distribution characteristic of its local neighborhood pixel, if feature point scale
For σ, corresponding Gauss scalogram picture is L (x, y, σ), such as uses finite difference, is calculated centered on characteristic point, is with the σ of 3x 1.5
The gradient magnitude and deflection of image in radius region.Amplitude m (x, y, σ) and deflectionBe calculated as follows formula：
Statistical analysiss are carried out to the Gradient distribution characteristic in neighborhood window using rectangular histogram, gradient direction angle is divided into into 36
Interval, an interval coordinate as rectangular histogram transverse axis per 10 degree, the longitudinal axis is the corresponding gradient magnitude of gradient direction angle from this
Weighted cumulative value.Here the variance of weight Gaussian window is characterized 1.5 times of point scale, and addition Gaussian window can be attached with Enhanced feature point
Near gradient magnitude affects, and the histogrammic main peak value for so generating reflects local neighborhood image gradient around this feature point
The principal direction of Main way, i.e. this feature point.When there are the minor peaks that another is more than the energy of main peak value 80%, it is believed that this
Individual direction is the auxiliary direction of this feature point, and it is identical with yardstick that such a characteristic point may produce coordinate position, and direction is not
Same characteristic point, to strengthen the robustness of matching.
The present invention will be described in detail for above-described embodiment.It should be noted that specific embodiment described herein
Only to explain the present invention, it is not intended to limit the present invention, the product that can realize said function belongs to equivalent and improvement,
It is included within protection scope of the present invention.
The invention has the beneficial effects as follows：The present invention is processed facial image, in this course not only image position
The Correction Problemss put are taken into account, and generate the formalized model of facial image to embody feature and the side of facial image
Just the data storage of image, the formalized model can strictly represent facial image, and can be used for the place of follow-up facial image
Reason.
Claims (1)
1. a kind of processing method of facial image, it is characterised in that as follows the step of the method：
1) a colorized face images training set is given, the CPU in computer is counted using HLS colour model transfer algorithms
Calculate, the HLS colour models transfer algorithm is nonlinear, can make that edge brightness noise is few, smooth effect is good, by the colour
Each width colorized face images in facial image training set are converted into Gray Face image, by the Gray Face image construction
Corresponding Gray Face training set of images；
2) the Lycoperdon polymorphum Vitt facial image training set is sampled, obtains sample window set, by the sample window set digitized,
Sample window matrix is obtained, is gathered the sample window matrix conversion into singularity characteristics vector according to singular value decomposition theorem, it is described
Singularity characteristics vector set be by feature point group one by one into, calculate average mark implantation P of the characteristic point, calculating formula is such as
Under：
In formula：I=1,2 ..., M, siFor the coordinate figure of the characteristic point, P is the average mark implantation of the characteristic point, and variable M is used
In the quantity for recording the characteristic point；
3) according to the average mark implantation of the characteristic point, using calculated coordinate figure as new coordinate axess origin, in order to disappear
Except the difference of face location, carry out the translation of face and carry out rotation transformation, first will be every in singularity characteristics vector set
Individual singularity characteristics vector all deducts the new coordinate axess origin value, the singularity characteristics vector set after being translated, and then sees
The coordinate of eyebrow on face is examined, by the coordinate (x of two eyebrows on face1, y1)、(x2, y2) even into a line obtain inclining for face
Rake angle α, the computing formula of the tilt angle alpha is as follows：
Singularity characteristics vector set after the translation is surrounded around the new zero rotationAngle, rotation parameter isThe coordinate of the singularity characteristics vector set after all translations is all multiplied by the rotation parameter, obtains
Singularity characteristics vector set to after rotation transformation；
4) by the singularity characteristics vector set even partition after the rotation transformation, state matrix, the shape are accordingly set up
State in state matrix is corresponded with the singularity characteristics vector set coordinate after the rotation transformation, and calculates the state square
The transition probability between state in battle array, by this process Probability State model, the composition of the Probability State model are built
Ultimate unit is the state in the state matrix, and each state both corresponds to the transition probability with other states, finally
The Probability State model for obtaining is a kind of formalized model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610905128.1A CN106570459A (en) | 2016-10-11 | 2016-10-11 | Face image processing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610905128.1A CN106570459A (en) | 2016-10-11 | 2016-10-11 | Face image processing method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106570459A true CN106570459A (en) | 2017-04-19 |
Family
ID=58533067
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610905128.1A Pending CN106570459A (en) | 2016-10-11 | 2016-10-11 | Face image processing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106570459A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650606A (en) * | 2016-10-21 | 2017-05-10 | 江苏理工学院 | Matching and processing method of face image and face image model construction system |
| CN108073914A (en) * | 2018-01-10 | 2018-05-25 | 成都品果科技有限公司 | A kind of animal face key point mask method |
| CN109948397A (en) * | 2017-12-20 | 2019-06-28 | Tcl集团股份有限公司 | A kind of face image correcting method, system and terminal device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101593272A (en) * | 2009-06-18 | 2009-12-02 | 电子科技大学 | A kind of human face characteristic positioning method based on the ASM algorithm |
| CN105718906A (en) * | 2016-01-25 | 2016-06-29 | 宁波大学 | Living body face detection method based on SVD-HMM |
-
2016
- 2016-10-11 CN CN201610905128.1A patent/CN106570459A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101593272A (en) * | 2009-06-18 | 2009-12-02 | 电子科技大学 | A kind of human face characteristic positioning method based on the ASM algorithm |
| CN105718906A (en) * | 2016-01-25 | 2016-06-29 | 宁波大学 | Living body face detection method based on SVD-HMM |
Non-Patent Citations (1)
| Title |
|---|
| 白冬辉: "人脸识别技术的研究与应用", 《中国优秀博硕士学位论文全文数据库（硕士）信息科技辑》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106650606A (en) * | 2016-10-21 | 2017-05-10 | 江苏理工学院 | Matching and processing method of face image and face image model construction system |
| CN109948397A (en) * | 2017-12-20 | 2019-06-28 | Tcl集团股份有限公司 | A kind of face image correcting method, system and terminal device |
| CN108073914A (en) * | 2018-01-10 | 2018-05-25 | 成都品果科技有限公司 | A kind of animal face key point mask method |
| CN108073914B (en) * | 2018-01-10 | 2022-02-18 | 成都品果科技有限公司 | Animal face key point marking method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10929649B2 (en) | Multi-pose face feature point detection method based on cascade regression | |
| CN104834922B (en) | Gesture identification method based on hybrid neural networks | |
| Rekha et al. | Shape, texture and local movement hand gesture features for indian sign language recognition | |
| CN110458192B (en) | Hyperspectral remote sensing image classification method and system based on visual saliency | |
| CN108681725A (en) | A kind of weighting sparse representation face identification method | |
| CN103207986A (en) | Face recognition method based on local binary pattern-histogram Fourier (LBP-HF) | |
| CN107341510A (en) | Image clustering method based on sparse orthogonal digraph Non-negative Matrix Factorization | |
| CN107862680B (en) | Target tracking optimization method based on correlation filter | |
| CN107886558A (en) | A kind of human face expression cartoon driving method based on RealSense | |
| Saval-Calvo et al. | 3D non-rigid registration using color: color coherent point drift | |
| CN106570459A (en) | Face image processing method | |
| CN103714331A (en) | Facial expression feature extraction method based on point distribution model | |
| CN102495999A (en) | Face recognition method | |
| CN108154176B (en) | 3D human body posture estimation algorithm aiming at single depth image | |
| CN109543637A (en) | A kind of face identification method, device, equipment and readable storage medium storing program for executing | |
| CN106599833A (en) | Field adaptation and manifold distance measurement-based human face identification method | |
| CN105844667A (en) | Structural target tracking method of compact color coding | |
| CN105740838A (en) | Recognition method in allusion to facial images with different dimensions | |
| CN110598719A (en) | Method for automatically generating face image according to visual attribute description | |
| Avraam | Static gesture recognition combining graph and appearance features | |
| CN111783526B (en) | Cross-domain pedestrian re-identification method using posture invariance and graph structure alignment | |
| Wang et al. | The facial expression recognition based on KPCA | |
| CN112800882A (en) | Mask face posture classification method based on weighted double-flow residual error network | |
| CN106952287A (en) | A kind of video multi-target dividing method expressed based on low-rank sparse | |
| CN109918998A (en) | A kind of big Method of pose-varied face |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170419 |
|
| WD01 | Invention patent application deemed withdrawn after publication | ||
| DD01 | Delivery of document by public notice |
Addressee: He Zhengdi Document name: Notification that Application Deemed to be Withdrawn |
|
| DD01 | Delivery of document by public notice |