CN107729827A - A kind of man face characteristic point positioning method and device - Google Patents
A kind of man face characteristic point positioning method and device Download PDFInfo
- Publication number
- CN107729827A CN107729827A CN201710908134.7A CN201710908134A CN107729827A CN 107729827 A CN107729827 A CN 107729827A CN 201710908134 A CN201710908134 A CN 201710908134A CN 107729827 A CN107729827 A CN 107729827A
- Authority
- CN
- China
- Prior art keywords
- mrow
- facial image
- cascade
- face characteristic
- characteristic point
- 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
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of man face characteristic point positioning method and device, comprise the following steps：Gather facial image；Key point mark processing is carried out to facial image, obtains marking the original shape of key point, using the facial image of all mark original shapes as training set；Using training set to default recurrence device training, the cascade for obtaining location feature point returns device；Facial image to be detected is inputted to cascade and returns device, obtains corresponding human face characteristic point position mark.The invention provides the detection speed of characteristic point, elapsed time in Face datection are short, computing cost is small, accuracy rate is high and stably.
Description
Technical field
The present invention relates to image identification technical field, and in particular to a kind of man face characteristic point positioning method and device.
Background technology
The purpose of facial modeling be on the basis of Face datection, further non-determined facial feature points (research,
Eyebrow, nose, face, face's outline) position.Can be recognition of face, face appearance by facial modeling technology
The research work such as state Expression analysis, beautifying faces and face tracking provide corresponding basic data, thus with importantly
Position.Facial modeling includes,, 194 points of localization method, 194 human face characteristic points of existing positioning at 5 points at 27 points at 68 points
Technology mainly include deep learning and cascade regression model.Wherein, deep learning model very deep network or cascade model
More structure has very high accuracy rate to facial modeling, but needs longer time to be calculated, and speed is slow；
If network, in the case that depth or cascade model be not less, although speed improves, but performance have dropped, and accuracy rate is low；
And cascade regression model and initialization shape matching is relied on, initialization shape can cause very big influence to the precision of model, and
And speed is slower, elapsed time length, computing cost is big and accuracy rate is unstable.
The content of the invention
It is an object of the invention to provide the performance of extract facial feature, reduces expense, and it is accurate to improve locating speed science and technology
Rate.
In order to solve above-mentioned purpose, the invention provides a kind of man face characteristic point positioning method, comprise the following steps：
Gather facial image；
Key point mark processing is carried out to facial image, obtains marking the original shape of key point, all marks are initial
The facial image of shape is as training set；
Using training set to default recurrence device training, the cascade for obtaining location feature point returns device；
Facial image to be detected is inputted to cascade and returns device, obtains corresponding human face characteristic point position mark.
Further, described that default recurrence device is trained using training set, obtaining cascade recurrence device includes
Calculate the difference of the original shape of the shape for returning device prediction and face images
Obtain differenceThe absolute value for making the value γ differences of loss function minimization with returning device, obtains likelihood minimum value,
Initialized according to likelihood minimum value and return device；
UtilizeWith the negative gradient r of the mathematic interpolation current predictive facial image of the recurrence device of last round of iterationik；
The approximation of all differences is fitted, is updated according to approximation iteration and returns device, cascade is obtained and returns device
Further, sample initialization is carried out according to the following equation:
Wherein,
What is represented is difference of original 194 points with returning point out, and γ makes loss function minimization
Constant.
Further, the calculation formula of difference is：
It is difference of the point of current recurrence out with original point,
What is represented is the γ values of last round of iterative calculation out.
Further, iterative calculation formula is：
ν is constant,For linear approximation.
Further, the depth that cascade returns device K is 6.
Present invention also offers a kind of facial modeling device, including
Acquisition module, for gathering facial image；
Pretreatment module, for carrying out key point mark processing to facial image, obtain marking the original shape of key point,
Using the facial image of all mark original shapes as training set；
Training module, for, to default recurrence device training, the cascade for obtaining location feature point to return device using training set；
Locating module, device is returned for facial image to be detected to be inputted to cascade, obtain corresponding human face characteristic point position
Tagging.
In the above-mentioned technical solutions, the present invention uses depth to lift decision-making for the gradient of 6 194 human face characteristic points of positioning
Tree-model, to realize 194 human face characteristic points of positioning, real-time high-efficiency, speed is fast, and accuracy rate is high, improves special in Face datection
Levy the detection speed of point, elapsed time is short, computing cost is small, accuracy rate is high and stably, positions 194 point features real-time high-precision
Point., it is necessary to real-time high-precision ground locating human face's characteristic point in some scenes, in order to which follow-up recognition of face etc. operates.
Brief description of the drawings
, below will be to institute in embodiment in order to illustrate more clearly of the embodiment of the present application or technical scheme of the prior art
The accompanying drawing needed to use is briefly described, it should be apparent that, drawings in the following description are only one described in the present invention
A little embodiments, for those of ordinary skill in the art, other accompanying drawings can also be obtained according to these accompanying drawings.
Fig. 1 is the flowage structure schematic diagram of man face characteristic point positioning method of the present invention；
Fig. 2 be man face characteristic point positioning method of the present invention in obtain cascade return device schematic flow sheet；
Fig. 3 is the structure of block diagram schematic diagram of facial modeling device of the present invention；
Fig. 4 is the result schematic diagram of application facial modeling device of the present invention.
Embodiment
In order that those skilled in the art more fully understands technical scheme, below in conjunction with accompanying drawing to this hair
It is bright to be further detailed.
As shown in figure 1, the invention provides a kind of man face characteristic point positioning method, comprise the following steps：
S101, collection facial image；
Specifically, facial image is used to train as training sample returns device, and then obtain cascade and return device.Facial image
Face image data collection (such as helen data sets) can be derived from, photo, the video of picture pick-up device collection can also be derived from
Deng.By taking helen data sets as an example, by the use of 2000 pictures of the data set as training set, 330 are used as test set, and
These pictures can obtain 2330 pictures by mirror image change, so final training set is 4000 pictures, 660 pictures
For test set.
S102, key point mark processing is carried out to facial image, obtain marking the original shape of key point, by all marks
The facial image of original shape is as training set；
S103, using training set to it is default recurrence device training, obtain location feature point cascade return device；
The recurrence device that the present invention utilizes is that the gradient that depth is 6 lifts decision-tree model.There are original shape, original shape
It is just as, during splay tree, it is current shape that the input of use, which is not, but is extracted according to current shape from the picture
Feature.For every piece image, original shape is although identical, but each width picture spies that are all different, therefore extracting
Sign is also just different, and using pixel difference as feature, the splitting operation of node is carried out according to feature, until reaching the leaf section of tree
Point.When N pictures are all inputted this one tree by we, naturally each pictures finally can all fall into one of leaf section
Point, such as the 1st pictures fall into the 3rd leaf node, and the 2nd pictures fall into the 1st leaf node, and the 3rd pictures fall
The 3rd leaf node etc. is entered.So, there can be picture to fall into each leaf node, may also not have certainly, this
It doesn't matter.At this moment, we will calculate residual error, calculate the difference of the current shape and true shape of each picture, afterwards,
The difference of all pictures in same leaf node is averaged, and is exactly the residual error that the leaf node should preserve.When all leaves
After child node all saves residual error, one tree also finishes with regard to construction.Before second tree of construction, we are every figure
The current shape of piece does a renewal, that is, current shape is updated to：Current shape+residual error.One tree is corresponded to,
Be that original shape adds residual error, the current shapes of so each pictures just from original shape become original shape add it is residual
Difference, apart from true shape again closer to a step.Second tree is built with same method again afterwards, node point is carried out according to feature
Split, until leaf node.The difference of the current shape of each pictures and true shape is calculated in leaf node, is then averaged,
This average value is stored in the leaf node, as residual error.The current shape of each pictures is updated afterwards, i.e., by leaf
The residual error preserved in node adds its current shape, and as new current shape, then can establishes the 3rd tree.Until
Untill establishing enough trees, the depth at this moment set is 6, current shape representation true shape that can be last.
The present invention is to lift decision tree by gradient, and projected depth is 6 tree-model, is instructed on Helen data sets
Practice, by the study to each recurrence device, cascade returns device and obtains a model that can position 194 human face characteristic points,
The model can position 194 human face characteristic points to the face picture real-time high-precision in actual scene.
As shown in Fig. 2 in step s 103, the process that training returns device specifically includes following steps：
The residual error of S1031, the shape predicted according to recurrence device and face images original shape
S1032, calculate residual errorWith the absolute value for the increment γ differences for returning device prediction, likelihood minimum value is obtained, according to
The initialization of likelihood minimum value returns device；
Specifically, initialization returns device and carried out according to the following equation:
Wherein,
What △ Si (t) were represented is difference of original 194 points with returning point out.
Understood according to the formula,Negative gradient is referred to, what γ was represented is whatIt is the increment of predictionγ is to make
Return out point with original point as close possible to constant value, for making loss function minimization.
Specifically, the calculation formula of difference is：
It is difference of the point of current recurrence out with original point,
What is represented is the γ values of last round of iterative calculation out.
S1033, utilize the residual error using current face's image and the difference of the recurrence device of initialization；
S1034, all differences of fitting approximation, update according to approximation iteration and return device, obtain cascade and return device
Specifically, the iterative calculation formula of cascade recurrence device is：
ν is constant,For linear approximation
Value.
Further, the depth that cascade returns device is 6.The experiment proved that, the effect that the cascade that depth is 6 returns device is best,
Have the beneficial effect that：Depth is used effectively to prevent from causing precision because of the difference of initialization shape for 6 recurrence device
Decline；Using depth, speed can reach 1000fps for 6 recurrence device, sufficiently high in precision while ensure actual effect.
S104, facial image to be detected inputted to cascade return device, obtain corresponding to human face characteristic point position mark.
The present invention completes the structure of cascade recurrence device by S101~S103, and it is complete finally to return device using the cascade built
Into the facial modeling to facial image to be detected.
As shown in figure 3, present invention also offers a kind of facial modeling device, including acquisition module 10, pretreatment
Module 20, training module 30 and locating module 40.
Wherein, acquisition module 10, for gathering facial image；Pretreatment module 20, it is crucial for being carried out to facial image
Point mark processing, obtain marking the original shape of key point, using the facial image of all mark original shapes as training set；Instruction
Practice module 30, for, to default recurrence device training, the cascade for obtaining location feature point to return device using training set；Locating module
40, device is returned for facial image to be detected to be inputted to cascade, obtains corresponding human face characteristic point position mark.
During concrete application, after left figure in Fig. 4 being inputted into the facial modeling device, the result of right figure is obtained.
Some one exemplary embodiments of the present invention are only described by way of explanation above, undoubtedly, for ability
The those of ordinary skill in domain, without departing from the spirit and scope of the present invention, can be with a variety of modes to institute
The embodiment of description is modified.Therefore, above-mentioned accompanying drawing and description are inherently illustrative, should not be construed as to the present invention
The limitation of claims.
Claims (7)
1. a kind of man face characteristic point positioning method, it is characterised in that comprise the following steps：
Gather facial image；
Key point mark processing is carried out to facial image, obtains marking the original shape of key point, by all mark original shapes
Facial image as training set；
Using training set to default recurrence device training, the cascade for obtaining location feature point returns device；
Facial image to be detected is inputted to cascade and returns device, obtains corresponding human face characteristic point position mark.
2. man face characteristic point positioning method according to claim 1, it is characterised in that described to utilize training set to default
Device training is returned, obtaining cascade recurrence device includes
Calculate the difference of the original shape of the shape for returning device prediction and face images
Obtain differenceThe absolute value for making the value γ differences of loss function minimization with returning device, obtains likelihood minimum value, according to
The initialization of likelihood minimum value returns device；
UtilizeWith the negative gradient r of the mathematic interpolation current predictive facial image of the recurrence device of last round of iterationik；
The approximation of all differences is fitted, is updated according to approximation iteration and returns device, cascade is obtained and returns device
3. man face characteristic point positioning method according to claim 1, it is characterised in that return device initialization according to following public affairs
Formula is carried out:
Wherein,
What is represented is original 194 points with returning the difference of point out, and γ is make loss function minimization normal
Number.
4. man face characteristic point positioning method according to claim 1, it is characterised in that the calculation formula of difference is：
It is difference of the point of current recurrence out with original point,Represent be
The γ values of last round of iterative calculation out.
5. man face characteristic point positioning method according to claim 1, it is characterised in that iterating to calculate formula is：
<mrow>
<msub>
<mi>f</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>,</mo>
<msup>
<mover>
<mi>S</mi>
<mo>^</mo>
</mover>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msup>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msub>
<mi>f</mi>
<mrow>
<mi>k</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>,</mo>
<msup>
<mover>
<mi>S</mi>
<mo>^</mo>
</mover>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msup>
<mo>)</mo>
</mrow>
<mo>+</mo>
<msub>
<mi>vg</mi>
<mi>k</mi>
</msub>
<mrow>
<mo>(</mo>
<mi>I</mi>
<mo>,</mo>
<msup>
<mover>
<mi>S</mi>
<mo>^</mo>
</mover>
<mrow>
<mo>(</mo>
<mi>t</mi>
<mo>)</mo>
</mrow>
</msup>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
ν is constant,For linear approximation.
6. man face characteristic point positioning method according to claim 1, it is characterised in that the depth that cascade returns device K is 6.
A kind of 7. facial modeling device, it is characterised in that including
Acquisition module, for gathering facial image；
Pretreatment module, for carrying out key point mark processing to facial image, obtain marking the original shape of key point, by institute
There is the facial image of mark original shape as training set；
Training module, for, to default recurrence device training, the cascade for obtaining location feature point to return device using training set；
Locating module, device is returned for facial image to be detected to be inputted to cascade, obtain corresponding human face characteristic point position mark
Note.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710908134.7A CN107729827A (en) | 2017-09-29 | 2017-09-29 | A kind of man face characteristic point positioning method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710908134.7A CN107729827A (en) | 2017-09-29 | 2017-09-29 | A kind of man face characteristic point positioning method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107729827A true CN107729827A (en) | 2018-02-23 |
Family
ID=61208820
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710908134.7A Pending CN107729827A (en) | 2017-09-29 | 2017-09-29 | A kind of man face characteristic point positioning method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107729827A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109241881A (en) * | 2018-08-22 | 2019-01-18 | 东北大学 | A kind of estimation method of human posture |
| CN109376659A (en) * | 2018-10-26 | 2019-02-22 | 北京陌上花科技有限公司 | Training method, face critical point detection method, apparatus for face key spot net detection model |
| CN113065552A (en) * | 2021-03-29 | 2021-07-02 | 天津大学 | Method for automatically positioning head shadow measurement mark point |
| CN114947902A (en) * | 2022-05-16 | 2022-08-30 | 天津大学 | X-ray head shadow measurement mark point automatic positioning method based on reinforcement learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824050A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascade regression-based face key point positioning method |
| CN105426870A (en) * | 2015-12-15 | 2016-03-23 | 北京文安科技发展有限公司 | Face key point positioning method and device |
| CN106022215A (en) * | 2016-05-05 | 2016-10-12 | 北京海鑫科金高科技股份有限公司 | Face feature point positioning method and device |
-
2017
- 2017-09-29 CN CN201710908134.7A patent/CN107729827A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103824050A (en) * | 2014-02-17 | 2014-05-28 | 北京旷视科技有限公司 | Cascade regression-based face key point positioning method |
| CN105426870A (en) * | 2015-12-15 | 2016-03-23 | 北京文安科技发展有限公司 | Face key point positioning method and device |
| CN106022215A (en) * | 2016-05-05 | 2016-10-12 | 北京海鑫科金高科技股份有限公司 | Face feature point positioning method and device |
Non-Patent Citations (1)
| Title |
|---|
| VAHID KAZEMI,JOSEPHINE SULLIVAN KTH: ""One Millisecond Face Alignment with an Ensemble of Regression Trees"", 《2014 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109241881A (en) * | 2018-08-22 | 2019-01-18 | 东北大学 | A kind of estimation method of human posture |
| CN109376659A (en) * | 2018-10-26 | 2019-02-22 | 北京陌上花科技有限公司 | Training method, face critical point detection method, apparatus for face key spot net detection model |
| CN113065552A (en) * | 2021-03-29 | 2021-07-02 | 天津大学 | Method for automatically positioning head shadow measurement mark point |
| CN114947902A (en) * | 2022-05-16 | 2022-08-30 | 天津大学 | X-ray head shadow measurement mark point automatic positioning method based on reinforcement learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106980858B (en) | Language text detection and positioning system and language text detection and positioning method using same | |
| CN103824050B (en) | A kind of face key independent positioning method returned based on cascade | |
| CN105205475B (en) | A kind of dynamic gesture identification method | |
| CN107944559B (en) | Method and system for automatically identifying entity relationship | |
| CN106022353B (en) | A kind of linguistic indexing of pictures method based on super-pixel segmentation | |
| CN107729827A (en) | A kind of man face characteristic point positioning method and device | |
| CN105825191B (en) | Gender identification method and system based on face multi-attribute information and shooting terminal | |
| CN104142995B (en) | The social event recognition methods of view-based access control model attribute | |
| CN107895160A (en) | Human face detection and tracing device and method | |
| CN103455794B (en) | A kind of dynamic gesture identification method based on frame integration technology | |
| CN106469298A (en) | Age recognition methodss based on facial image and device | |
| CN106767812A (en) | A kind of interior semanteme map updating method and system based on Semantic features extraction | |
| CN107766320A (en) | A kind of Chinese pronoun resolution method for establishing model and device | |
| Ding et al. | Research on daily objects detection based on deep neural network | |
| CN110909736A (en) | Image description method based on long-short term memory model and target detection algorithm | |
| CN106127222B (en) | A kind of the similarity of character string calculation method and similitude judgment method of view-based access control model | |
| CN109241995A (en) | A kind of image-recognizing method based on modified ArcFace loss function | |
| CN106803084B (en) | Facial feature point positioning method based on end-to-end circulation network | |
| CN104966052A (en) | Attributive characteristic representation-based group behavior identification method | |
| CN107679469A (en) | A kind of non-maxima suppression method based on deep learning | |
| Balasuriya et al. | Learning platform for visually impaired children through artificial intelligence and computer vision | |
| CN110533661A (en) | Adaptive real-time closed-loop detection method based on characteristics of image cascade | |
| CN111444850A (en) | Picture detection method and related device | |
| CN115187704A (en) | Virtual anchor generation method, device, equipment and storage medium | |
| CN114742224A (en) | Pedestrian re-identification method and device, computer equipment and storage medium |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180223 |