CN1633944A - Quick human face detection method based on one kind supporting vector quantities - Google Patents
Quick human face detection method based on one kind supporting vector quantities Download PDFInfo
- Publication number
- CN1633944A CN1633944A CNA2003101160476A CN200310116047A CN1633944A CN 1633944 A CN1633944 A CN 1633944A CN A2003101160476 A CNA2003101160476 A CN A2003101160476A CN 200310116047 A CN200310116047 A CN 200310116047A CN 1633944 A CN1633944 A CN 1633944A
- Authority
- CN
- China
- Prior art keywords
- rule
- face
- people
- support vector
- class support
- 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
- 239000013598 vector Substances 0.000 title claims abstract description 31
- 238000001514 detection method Methods 0.000 title abstract description 10
- 238000012795 verification Methods 0.000 claims abstract description 8
- 238000000034 method Methods 0.000 claims description 39
- 230000008569 process Effects 0.000 claims description 4
- 239000003086 colorant Substances 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 description 7
- 230000001815 facial effect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000005286 illumination Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000012880 independent component analysis Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a quick human face detection method based on one kind supporting vector quantities which comprises the steps of, (1) fining human face candidate regions through complexion characteristics, (2) performing mosaic graph verification so as to further minimize human face candidate region, (3) simplifying flow shape and reducing dimensionality, describing with class I support vectors of the human face modes, (4) categorizer decision-making to determine whether the candidate region is the human face region. The invention can be applied for improving detection efficiency and performance.
Description
Technical field
The present invention relates to the method that a kind of people's face detects, refer in particular to the thought that has proposed to adopt a quasi-mode analysis and solve people's face and detect problem, and a kind of method that detects based on the fast face of a class support vector is provided.
Background technology
Based on the human face analysis of image and video is one of research focus in computer vision and the area of pattern recognition in recent years, because it has broad application prospect, such as: biological characteristic authentication, information security, man-machine interaction and vision monitoring or the like.
It is one of committed step in the automatic human face analysis system that people's face detects.Its purpose is exactly to catch the position of finding people's face and existence thereof the next image from video camera, is further to analyze to do initial preparation.Therefore, the effect of people's face detection directly influences the performance of human face analysis system.
In recent years, people have proposed a large amount of method for detecting human face.Can divide two big classes substantially: one, early stage method is to utilize features of skin colors, and in conjunction with simple geometric feature etc.Its shortcoming is that in addition, the extraction of geometric properties is very sensitive to environmental factorss such as illumination to the background close with the colour of skin very robust not.Two, based on the method for learning.Utilize the method for statistical learning to find people's face pattern and non-face mode difference.Because it has good performance, now become the main stream approach that detects for people's face.
But mostly the detection of people's face is regarded as two class problems based on the method for detecting human face of study at present, promptly people's face pattern and non-face pattern are collected a large amount of people's face samples and non-face sample and are learnt the criterion of a classification.Such as: based on neural network method, based on the method for example study, based on the method for support vector machine with based on method of Adaboost or the like.We know that people's face pattern is easy to describe, but non-face pattern modeling is actually very difficult, all are non-face patterns because do not belong to the part of people's face pattern in image space.All be to produce non-face sample by the mode of booting in the present system, the quality of non-face sample is the performance of influence detection directly.
Summary of the invention
The objective of the invention is to, propose people's face detection problem is regarded as a class problem, promptly only be concerned about " people's face pattern ", and the method that provides a kind of fast face based on a class support vector to detect, avoided influencing the performance of detection, thereby reached better detection effect owing to the quality of non-face sample.
According to the analysis to background technology, the model of people's face pattern is set up than being easier to, but the model of setting up non-face pattern almost is impossible, because all patterns that do not belong to people's face all should belong to non-face pattern, is inexhaustible.Starting point of the present invention is that people's face detection problem is regarded as a class problem, promptly only to the modeling of people's face pattern.We propose with having classification and representing a class support vector of ability to describe people's face pattern.In order to obtain better and the support vector of a compact class people face pattern, at first facial image is compressed in a low-dimensional, the compact simplification stream shape space.In addition, we also combine the skin color feature, and pretreatment such as mosaic figure further improves efficient and the performance that detects.
Concretely, the present invention proposes to come analyst's face to detect problem with a quasi-mode, and a kind of method that detects based on the fast face of a class support vector is provided, and comprises the steps:
A) utilize features of skin colors, find people's face candidate region;
B) mosaic figure verification further reduces people's face candidate region;
C) simplify manifold dimension-reducing, and a class support vector of personnel selection face pattern is described;
D) grader decision-making judges whether the candidate region is human face region.
The described method that detects based on the fast face of a class support vector, wherein step a) is meant following steps: a1) set up the Gauss model of face complexion, and use it to carry out colour of skin check, can remove most non-face zone in the image to be detected; A2) adopt the computings such as open and close in the Flame Image Process that area of skin color is incorporated into together, be for further processing as the candidate region.
The described method that detects based on the fast face of a class support vector, step a1 wherein) is meant when carrying out colour of skin check, calculate the probit of each picture element in the coloured image to be detected, probit can be considered as colour of skin point greater than preset threshold, to filter remaining picture element then and be integrated into complete zone, 70% of image-region all be colour of skin point can be considered as people's face candidate region.
The described method that detects based on the fast face of a class support vector, wherein said threshold value is 0.5.
The described method that detects based on the fast face of a class support vector, wherein step b) is meant following steps:
B1) people's face is divided into nine average zones;
B2) use following 8 gray scale density rule and 10 marginal density rules to carry out verification,
Gray scale density rule definition is as follows:
Rule 1:(1)＜(2)
Rule 2:(3)＜(2)
Rule 3:(1)＜(4)
Rule 4:(3)＜(6)
Rule 5:(5)＜(4)
Rule 6:(5)＜(6)
Rule 7:(8)＜(9)
Rule 8:(8)＜(7)
The marginal density rule definition is as follows:
Rule 1:(4)＜(1)
Rule 2:(7)＜(1)
Rule 3:(6)＜(3)
Rule 4:(9)＜(3)
Rule 5:(2)＜(1)
Rule 6:(2)＜(3)
Rule 7:(4)＜(5)
Rule 8:(6)＜(5)
Rule 9:(7)＜(8)
Rule 10:(9)＜(8)
B3) adopt most mechanism of voting to determine the candidate region of people's face.
The described method that detects based on the fast face of a class support vector, wherein step c) is meant following steps: c1) adopt the LPP method that people's face sample is compressed in a stream shape compactness, low-dimensional the subspace; C2) learn to obtain a representative class support vector with a class support vector and describe people's face pattern.
Description of drawings
Fig. 1 is a flow chart of the present invention;
Fig. 2 is a mosaic figure verification sketch map of the present invention;
Fig. 3 is implementation result figure of the present invention.
The specific embodiment
As shown in Figure 1, the method that detects based on the fast face of a class support vector provided by the invention mainly comprises following step:
A) complexion model detects
Determine the position of people's face in image and video, the colour of skin of people's face is a very important information.At first use the color of skin that the zone of people's face to be detected is handled in our algorithm, filter out the zone that is in beyond the colour of skin space.Can improve speed and order of accuarcy that people's face detects like this.Concrete steps are as follows:
A1) set up the Gauss model of face complexion, and use it to carry out colour of skin check, can remove most non-face zone in the image to be detected;
A2) adopt the computings such as open and close in the Flame Image Process that area of skin color is incorporated into together, be for further processing as the candidate region.
We adopt normalized r, and the g color is set up face complexion bivariate Gauss model.
r＝R/(R+G+B)
g＝G/(R+G+B)
The average of Gauss model and variance are
Mean=E{x}, wherein x=(r, g)
T
var＝E{(x-m)(x-m)
T}
Any given 1 x=(r
x, g
x)
TProbit is
P(x)＝exp(-(x-mean)
Tvar
-1(x-mean)/2)
In coloured image to be detected, calculate the probit of each picture element, probit can be considered as colour of skin point greater than preset threshold (such as 0.5), will filter remaining picture element then and be integrated into complete zone, 70% of image-region all be colour of skin point can be considered as people's face candidate region.
B) mosaic figure verification
According to the structure of people's face, we have also adopted some simple rules to remove some candidate regions that obtain by the skin-color model.Such as, in general, the average gray value of people's eye position is less than cheek and the part above the bridge of the nose.Near eyes or the face gradient density is greater than the gradient density of cheek.By these simple pretreatment, can further improve the speed and the robustness of our algorithm.We have adopted the method for mosaic figure, as shown in Figure 2.Concrete steps are as follows:
B1) people's face is divided into nine average zones;
B2) use following 8 gray scale density rule and 10 marginal density rules to carry out verification,
Gray scale density rule definition is as follows:
Rule 1:(1)＜(2)
Rule 2:(3)＜(2)
Rule 3:(1)＜(4)
Rule 4:(3)＜(6)
Rule 5:(5)＜(4)
Rule 6:(5)＜(6)
Rule 7:(8)＜(9)
Rule 8:(8)＜(7)
The marginal density rule definition is as follows:
Rule 1:(4)＜(1)
Rule 2:(7)＜(1)
Rule 3:(6)＜(3)
Rule 4:(9)＜(3)
Rule 5:(2)＜(1)
Rule 6:(2)＜(3)
Rule 7:(4)＜(5)
Rule 8:(6)＜(5)
Rule 9:(7)＜(8)
Rule 10:(9)＜(8)
B3) adopt a most mechanism of voting, for example Shang Mian 18 rules satisfy the candidate region that then can think people's face more than 14.
C) simplify manifold dimension-reducing
Because image space is the space of higher-dimension, and in fact facial image only belongs to the subspace of a low-dimensional in the pattern space.Traditional dimension reduction method such as pivot analysis, independent component analysis all are based on the linear analysis method that the overall situation is considered, can not describe in the facial image because complicated nonlinear change such as illumination, expression, attitudes.Non-linearity manifold study has obtained people's extensive concern in recent years, but present manifold learning method as, local linear (LLE), ISOMap and the Laplace Eigenmap etc. of embedding can not provide good treatment to the sample of the unknown.
It is a kind of linear manifold learning method of simplification that the present invention adopts local retaining projection (LPP).Its thought is exactly to go approximate Laplace Eigenmap method with a kind of mode of linearity, can obtain and be similar to the characteristics (the local distance of protecting) of stream shape.Concrete steps are as follows:
C1) adopt the LPP method that people's face sample is compressed in a low-dimensional, the compact stream shape subspace;
C2) learn to obtain a representative class support vector with a class support vector and describe people's face pattern.
D) grader decision-making
In the process that detects, after pretreatment such as skin-color feature, Masic figure, judge with a class support vector that obtains whether the candidate region is human face region.
Implementation result
It is the image that the colour of some random collectings has people's face.Image source comprises the image of own shooting, the image that has people's face that obtains from news video and from the stage photo image above the Internet etc.
Claims (7)
1, a kind of method that detects based on the fast face of a class support vector is characterized in that, adopts a quasi-mode mode that people's face is detected.
2, the method that detects based on the fast face of a class support vector according to claim 1 is characterized in that, comprises the steps:
A) utilize features of skin colors, find people's face candidate region;
B) mosaic figure verification further reduces people's face candidate region;
C) simplify manifold dimension-reducing, and a class support vector of personnel selection face pattern is described;
D) grader decision-making judges whether the candidate region is human face region.
3, the method that detects based on the fast face of a class support vector according to claim 2 is characterized in that wherein step a) is meant following steps:
A1) set up the Gauss model of face complexion, and use it to carry out colour of skin check, can remove most non-face zone in the image to be detected;
A2) adopt the computings such as open and close in the Flame Image Process that area of skin color is incorporated into together, be for further processing as the candidate region.
4, the method that detects based on the fast face of a class support vector according to claim 3, it is characterized in that, step a1 wherein) is meant when carrying out colour of skin check, calculate the probit of each picture element in the coloured image to be detected, probit can be considered as colour of skin point greater than preset threshold, to filter remaining picture element then and be integrated into complete zone, 70% of image-region all be colour of skin point can be considered as people's face candidate region.
5, the method that detects based on the fast face of a class support vector according to claim 4 is characterized in that wherein said threshold value is 0.5.
6, the method that detects based on the fast face of a class support vector according to claim 2 is characterized in that wherein step b) is meant following steps:
B1) people's face is divided into nine average zones;
B2) use following 8 gray scale density rule and 10 marginal density rules to carry out verification, gray scale density rule definition is as follows:
Rule 1:(1)＜(2)
Rule 2:(3)＜(2)
Rule 3:(1)＜(4)
Rule 4:(3)＜(6)
Rule 5:(5)＜(4)
Rule 6:(5)＜(6)
Rule 7:(8)＜(9)
Rule 8:(8)＜(7)
The marginal density rule definition is as follows:
Rule 1:(4)＜(1)
Rule 2:(7)＜(1)
Rule 3:(6)＜(3)
Rule 4:(9)＜(3)
Rule 5:(2)＜(1)
Rule 6:(2)＜(3)
Rule 7:(4)＜(5)
Rule 8:(6)＜(5)
Rule 9:(7)＜(8)
Rule 10:(9)＜(8)
B3) adopt most mechanism of voting to determine the candidate region of people's face.
7, the method that detects based on the fast face of a class support vector according to claim 2 is characterized in that wherein step c) is meant following steps:
C1) adopt the LPP method that people's face sample is compressed in a low-dimensional, the compact stream shape subspace;
C2) learn to obtain a representative class support vector with a class support vector and describe people's face pattern.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2003101160476A CN1633944A (en) | 2003-12-30 | 2003-12-30 | Quick human face detection method based on one kind supporting vector quantities |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2003101160476A CN1633944A (en) | 2003-12-30 | 2003-12-30 | Quick human face detection method based on one kind supporting vector quantities |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN1633944A true CN1633944A (en) | 2005-07-06 |
Family
ID=34843541
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNA2003101160476A Pending CN1633944A (en) | 2003-12-30 | 2003-12-30 | Quick human face detection method based on one kind supporting vector quantities |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1633944A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100416592C (en) * | 2005-12-23 | 2008-09-03 | 北京海鑫科金高科技股份有限公司 | Human face automatic identifying method based on data flow shape |
| CN102156871A (en) * | 2010-02-12 | 2011-08-17 | 中国科学院自动化研究所 | Image classification method based on category correlated codebook and classifier voting strategy |
| CN102291520A (en) * | 2006-05-26 | 2011-12-21 | 佳能株式会社 | Image processing method and image processing apparatus |
| CN101609502B (en) * | 2009-07-24 | 2012-10-24 | 西安电子科技大学 | Human face detecting method based on sequence simplifying support vector |
| CN105046231A (en) * | 2015-07-27 | 2015-11-11 | 小米科技有限责任公司 | Face detection method and device |
-
2003
- 2003-12-30 CN CNA2003101160476A patent/CN1633944A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100416592C (en) * | 2005-12-23 | 2008-09-03 | 北京海鑫科金高科技股份有限公司 | Human face automatic identifying method based on data flow shape |
| CN102291520A (en) * | 2006-05-26 | 2011-12-21 | 佳能株式会社 | Image processing method and image processing apparatus |
| CN102291520B (en) * | 2006-05-26 | 2017-04-12 | 佳能株式会社 | Image processing method and image processing apparatus |
| CN101609502B (en) * | 2009-07-24 | 2012-10-24 | 西安电子科技大学 | Human face detecting method based on sequence simplifying support vector |
| CN102156871A (en) * | 2010-02-12 | 2011-08-17 | 中国科学院自动化研究所 | Image classification method based on category correlated codebook and classifier voting strategy |
| CN102156871B (en) * | 2010-02-12 | 2012-12-12 | 中国科学院自动化研究所 | Image classification method based on category correlated codebook and classifier voting strategy |
| CN105046231A (en) * | 2015-07-27 | 2015-11-11 | 小米科技有限责任公司 | Face detection method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Schwartz et al. | Face spoofing detection through partial least squares and low-level descriptors | |
| CN1975759A (en) | Human face identifying method based on structural principal element analysis | |
| US11263435B2 (en) | Method for recognizing face from monitoring video data | |
| CN1777915A (en) | Face image candidate area search method, face image candidate area search system, and face image candidate area search program | |
| CN1209073C (en) | Identity discriminating method based on living body iris | |
| CN1710593A (en) | Hand-characteristic mix-together identifying method based on characteristic relation measure | |
| CN1932847A (en) | Method for detecting colour image human face under complex background | |
| CN101059838A (en) | Human face recognition system and recognition method | |
| CN1801181A (en) | Robot capable of automatically recognizing face and vehicle license plate | |
| CN1794264A (en) | Method and system of real time detecting and continuous tracing human face in video frequency sequence | |
| CN1503194A (en) | Status identification method by using body information matched human face information | |
| Zang et al. | Traffic sign detection based on cascaded convolutional neural networks | |
| CN1685357A (en) | Method and apparatus for identifying a palmprint image | |
| CN1842125A (en) | Extraction and scaled display of objects in an image | |
| CN1731417A (en) | Method of robust human face detection in complicated background image | |
| CN104134061A (en) | Number gesture recognition method for support vector machine based on feature fusion | |
| CN1967561A (en) | Method for making gender recognition handler, method and device for gender recognition | |
| CN101923637A (en) | Mobile terminal as well as human face detection method and device thereof | |
| CN1776712A (en) | Human face recognition method based on human face statistics | |
| Kasim et al. | A study on facial expression recognition using local binary pattern | |
| CN1801180A (en) | Identity recognition method based on eyebrow recognition | |
| Satapathy et al. | A lite convolutional neural network built on permuted Xceptio-inception and Xceptio-reduction modules for texture based facial liveness recognition | |
| CN1633944A (en) | Quick human face detection method based on one kind supporting vector quantities | |
| CN101055617A (en) | Human face quick detection method based on local description | |
| CN1781122A (en) | Method, system and program for searching area considered to be face image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |