CN106250843B - A kind of method for detecting human face and system based on forehead region - Google Patents
A kind of method for detecting human face and system based on forehead region Download PDFInfo
- Publication number
- CN106250843B CN106250843B CN201610607532.0A CN201610607532A CN106250843B CN 106250843 B CN106250843 B CN 106250843B CN 201610607532 A CN201610607532 A CN 201610607532A CN 106250843 B CN106250843 B CN 106250843B
- Authority
- CN
- China
- Prior art keywords
- classifier
- forehead
- region
- feature
- cascade
- 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.)
- Active
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
- G06V40/165—Detection; Localisation; Normalisation using facial parts and geometric relationships
-
- 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
-
- 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
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Human Computer Interaction (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Geometry (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of method for detecting human face and system based on forehead region.This method comprises: being traversed by the way of sliding window to image to be detected；The variance of each sliding window to be detected is obtained, if the variance of the sliding window to be detected is greater than preset threshold, extracts the Haar-Like feature and multichannel color frequency feature of the extension of sliding window to be detected；The target forehead region in described image to be detected is obtained according to trained cascade forehead classifier；By Face Detection and edge detection, the corresponding human face region in each target forehead region is obtained using improved least square method.The embodiment of the present invention is by extracting the Haar-Like feature and multichannel color frequency feature that extend, the color development in description forehead region more abundant and the Probability Characteristics of the colour of skin, the detection to face is realized using the forehead region with abundant shape and color characteristic, improves the precision of Face datection.
Description
Technical field
The present invention relates to technical field of image processing, and in particular to a kind of method for detecting human face based on forehead region and is
System.
Background technique
With the raising of demand for security, the commercial value of the technologies such as people flow rate statistical, personnel characteristics' identification, recognition of face is
Through starting to appear, and gradually start to apply, and primary link of the Face datection as these tasks, it has very important effect
And meaning, in recent years, researcher has put into a large amount of time and efforts in this field, is dedicated to developing a kind of quick standard
True method for detecting human face.
The existing method for detecting human face based on the colour of skin, is mainly transformed into different color spaces from rgb space for face
(such as YCbCr, HSV, LUV etc.) then trains corresponding gaussian probability mould according to the distribution situation of area of skin color in space
Type is predicted to treat the point in sample one's respective area according to gaussian probability model, judges whether to belong to human face region, finally
Minimum circumscribed rectangle judgement is carried out to human face region.The existing method for detecting human face based on the colour of skin ignores the hair in forehead region
The probability density characteristics of color and the colour of skin, Face datection effect are poor.
Under true environment, the position of camera head monitor is often higher, has certain angle with face, it is more difficult to collect
Front face；Positional relationship under more people's occasions simultaneously, so that face is easy to that partial occlusion phenomenon occurs, it is clear to be not easy to obtain
The characteristic areas such as eye, nose, mouth；The face under different distance has different resolution ratio simultaneously.Three above problem for
There is very big challenge, so that existing method for detecting human face detection performance declines for Face datection.
Summary of the invention
The technical problems to be solved by the present invention are: how to provide the method that a kind of pair of face is accurately detected.
In order to solve the above technical problems, this is based on the invention proposes a kind of method for detecting human face based on forehead region
The method for detecting human face in forehead region includes:
Image to be detected is traversed by the way of sliding window；
Obtain the variance of each sliding window to be detected, if the variance of the sliding window to be detected be greater than preset threshold, extract to
Detect the Haar-Like feature and multichannel color frequency feature of the extension of sliding window；
The target forehead region in described image to be detected is obtained according to trained cascade forehead classifier；
By Face Detection and edge detection, it is corresponding that each target forehead region is obtained using improved least square method
Human face region；
Wherein, the cascade forehead classifier of training includes preceding to classifier and backward classifier；The forward direction classifier
Including cascade Adaboost classifier；The backward classifier includes classifier based on cascade structure and based on Multiple Kernel Learning
Classifier.
Optionally, the method also includes:
Forehead sample is obtained, according to the cascade forehead classifier of forehead sample training.
Optionally, the acquisition forehead sample includes: according to the cascade forehead classifier of forehead sample training
The Haar-Like feature for extracting the extension of forehead sample, by cascade Adaboost algorithm to the extension
Haar-Like feature is learnt, and training obtains the forward direction classifier；
The multichannel color frequency feature for the forehead sample classified through the forward direction classifier is longitudinally extracted, warp is laterally extracted
The multichannel color frequency feature of the forehead sample of the forward direction classifier classification；
The multichannel color frequency feature longitudinally extracted is learnt by Cascade algorithms, passes through Multiple Kernel Learning algorithm pair
The multichannel color frequency feature laterally extracted is learnt, and is trained after obtaining to classifier；
Wherein, the forehead sample includes positive sample and negative sample；The Haar-Like feature of extension includes description level side
To the Haar-Like feature of curved features and the Haar-Like feature of description vertical direction curved features.
Optionally, the multichannel color frequency for longitudinally extracting the forehead sample classified through the forward direction classifier is special
Sign, laterally extracts the multichannel color frequency feature for the forehead sample classified through the forward direction classifier, comprising:
In the longitudinal direction of N number of different Color Channel, it is various forms of that each position of the forehead sample areas corresponds to N kind
Mapping, obtains the various forms of mapping values of N kind of each position；
In the transverse direction of N number of different Color Channel, the corresponding channel of each dimension, the N kind for obtaining each channel is different
The colouring information of type；
Wherein, N is the integer greater than 3.
It is optionally, described that the multichannel color frequency feature longitudinally extracted is learnt by Cascade algorithms, comprising:
Obtain the corresponding multiple Weak Classifiers of multichannel color frequency feature longitudinally extracted；
Each Weak Classifier is learnt by the GentleAdaboost algorithm based on CART tree, obtains longitudinal extract
The corresponding strong classifier of multichannel color frequency feature.
Optionally, described that the multichannel color frequency feature laterally extracted is learnt by Multiple Kernel Learning algorithm, packet
It includes:
The multichannel color frequency feature laterally extracted for obtaining each pixel, by Multiple Kernel Learning algorithm to each pixel
The multichannel color frequency feature laterally extracted learnt, it is corresponding to obtain the multichannel color frequency feature laterally extracted
Strong classifier.
Optionally, the target forehead area obtained according to trained cascade forehead classifier in described image to be detected
Domain, comprising:
The forehead region in described image to be detected is obtained by the forward direction classifier；
It is obtained by the backward classifier through the preceding target forehead region into classifier acquisition forehead region.
Optionally, described to be obtained by the backward classifier through the preceding target forehead into classifier acquisition forehead region
Region includes:
It is obtained respectively through preceding to classifier by the classifier based on cascade structure and the judgement of the classifier based on Multiple Kernel Learning
Whether the forehead region obtained is forehead region；
Classifier based on cascade structure and the classifier based on Multiple Kernel Learning are judged before to classifier acquisition
The correct region in forehead region is determined as target forehead region.
Optionally, described by Face Detection and edge detection, each target volume is obtained using improved least square method
The corresponding human face region of head region, comprising:
Each target forehead region is extended to obtain target person head region；
Area of skin color segmentation is carried out to the target person head region, obtains face complexion area；
Multiple profile points of face complexion area are obtained using edge detection algorithm；
It is selected at random from the multiple profile point, obtains the elliptical profile point of local optimum；
Least square fitting is carried out to the local optimum ellipse and obtains global optimum's ellipse, the optimal ellipse of whole is right
Answer the human face region.
The invention also provides a kind of face detection systems based on forehead region, comprising:
Image traversal unit, for being traversed by the way of sliding window to image to be detected；
Feature extraction unit, for obtaining the variance of each sliding window to be detected, if the variance of the sliding window to be detected is greater than
Preset threshold then extracts the Haar-Like feature and multichannel color frequency feature of the extension of sliding window to be detected；
Target forehead area acquisition unit, for obtaining described image to be detected according to trained cascade forehead classifier
In target forehead region；
Human face region acquiring unit, for being obtained using improved least square method by Face Detection and edge detection
The corresponding human face region in each target forehead region；
Wherein, the cascade forehead classifier of training includes preceding to classifier and backward classifier；The forward direction classifier
Including cascade Adaboost classifier；The backward classifier includes classifier based on cascade structure and based on Multiple Kernel Learning
Classifier.
Method for detecting human face and system provided by the invention based on forehead region, extracts the extension of sliding window to be detected
Haar-Like feature and multichannel color frequency feature；It is single since the external expressive form of different type feature is not consistent
Feature learning algorithm be difficult well to learn feature, therefore the present invention is obtained according to trained cascade forehead classifier
Obtain the target forehead region in described image to be detected；By Face Detection and edge detection, using improved least square method
Obtain the corresponding human face region in each target forehead region.The present invention is by extracting the Haar-Like feature extended and multichannel
Color frequency feature, the color development in description forehead region more abundant and the Probability Characteristics of the colour of skin, enrich shape using having
The detection to face is realized in the forehead region of shape and color characteristic, improves the precision of Face datection.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is the present invention
Some embodiments for those of ordinary skill in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.
Fig. 1 is the flow diagram of the method for detecting human face based on forehead region of one embodiment of the invention；
Fig. 2 is the schematic diagram of the Haar-Like feature of one embodiment of the invention；
Fig. 3 is the schematic diagram of 8 Color Channels of the forehead area image of one embodiment of the invention；
Fig. 4 is being carried out by Cascade algorithms to the multichannel color frequency feature longitudinally extracted for one embodiment of the invention
The schematic diagram of study；
Fig. 5 be one embodiment of the invention by Multiple Kernel Learning algorithm to the multichannel color frequency feature laterally extracted
The schematic diagram learnt；
Fig. 6 is the structural schematic diagram of the face detection system based on forehead region of one embodiment of the invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, the technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Fig. 1 is the flow diagram of the method for detecting human face based on forehead region of one embodiment of the invention.Such as Fig. 1 institute
Show, the method for detecting human face based on forehead region of the embodiment includes:
S11: image to be detected is traversed by the way of sliding window；
S12: obtaining the variance of each sliding window to be detected, if the variance of the sliding window to be detected is greater than preset threshold, mentions
Take the Haar-Like feature and multichannel color frequency feature of the extension of sliding window to be detected；
S13: the target forehead region in described image to be detected is obtained according to trained cascade forehead classifier；
S14: by Face Detection and edge detection, each target forehead region pair is obtained using improved least square method
The human face region answered；
Wherein, the cascade forehead classifier of training includes preceding to classifier and backward classifier；The forward direction classifier
Including cascade Adaboost classifier；The backward classifier includes classifier based on cascade structure and based on Multiple Kernel Learning
Classifier.
In practical applications, to be checked to whole by the way of sliding window for given image to be detected in detection-phase
Altimetric image is traversed, and determines the variance of each sliding window to be detected first, if it is less than preset threshold (can be set to 50), then mistake
It filters, otherwise extracts the Haar-Like feature and multichannel color frequency feature of extension to sliding window to be detected, then utilize training
Cascade forehead classifier it is detected.The forehead region that can obtain a large amount of candidates in this way, is then pressed down by maximum value
The method of system merges candidate region, obtains final forehead region.
The method for detecting human face based on forehead region of the embodiment of the present invention, extracts the Haar- of the extension of sliding window to be detected
Like feature and multichannel color frequency feature；Due to not consistent, the single spy of the external expressive form of different type feature
Sign learning algorithm is difficult well to learn feature, therefore the present invention obtains institute according to trained cascade forehead classifier
State the target forehead region in image to be detected；By Face Detection and edge detection, obtained using improved least square method
The corresponding human face region in each target forehead region.The present invention is by extracting the Haar-Like feature and multichannel color that extend
Frequecy characteristic, the color development in description forehead region more abundant and the Probability Characteristics of the colour of skin, using have abundant shape and
The detection to face is realized in the forehead region of color characteristic, improves the precision of Face datection.
In a kind of preferred embodiment of the embodiment of the present invention, the method also includes:
Forehead sample is obtained, according to the cascade forehead classifier of forehead sample training.
Further, the process of the cascade forehead classifier of training includes:
The Haar-Like feature for extracting the extension of forehead sample, by cascade Adaboost algorithm to the extension
Haar-Like feature is learnt, and training obtains the forward direction classifier；
The multichannel color frequency feature for the forehead sample classified through the forward direction classifier is longitudinally extracted, warp is laterally extracted
The multichannel color frequency feature of the forehead sample of the forward direction classifier classification；
The multichannel color frequency feature longitudinally extracted is learnt by Cascade algorithms, passes through Multiple Kernel Learning algorithm pair
The multichannel color frequency feature laterally extracted is learnt, and is trained after obtaining to classifier；
Wherein, the forehead sample includes positive sample and negative sample；The Haar-Like feature of extension includes description level side
To the Haar-Like feature of curved features and the Haar-Like feature of description vertical direction curved features.
It should be noted that the embodiment of the present invention uses Haar-Like feature mainly to describe the shape feature of forehead.It is former
The 4 class Haar-Like features to begin mainly describe the shape feature of face in the method that V&J is proposed, for example, linear character, in
Heart feature, edge feature and to corner characteristics.Haar-Like feature was extended later, proposes the Haar- of 45 degree of rotations
Like feature and center ring around Haar-Like feature.Since forehead region is for human face region, the edge of arc is special
Levy obvious, therefore the embodiment of the present invention extends Haar-Like feature for forehead, and 2 kinds of description arcs are proposed
The Haar-Like feature at shape edge is respectively intended to describe the curved features in horizontal and vertical direction.As shown in Fig. 2, 1-16 is to expand
The Haar-Like feature of the description human face region of exhibition, 17-18 are the Haar- for the description curved features that the embodiment of the present invention proposes
Like feature.Table 1 shows the corresponding quantity of every class Haar-Like feature in the forehead region of the embodiment of the present invention.
The corresponding quantity of the every class Haar-Like feature of table 1
It should be noted that forward direction classifier is mainly using classical cascade Adaboost algorithm to forehead region
Haar-Like feature is learnt, so that filtering out non-forehead region to the maximum extent, and detects the almost forehead areas having more
Domain.In Face datection, V&J algorithm, which trains the faceform come, has good effect for the detection of front face, no
When excessive human face posture has greatly changed, which is difficult to be accurately detected human face region.But correlative study shows
The main reason for V&J algorithm declines the performance that multi-pose Face detects is that traditional Haar-Like feature can not be fine
Ground captures the face invariant features in the case of multi-pose, rather than caused by the deficiency of Adaboost algorithm.Therefore the present invention
Embodiment specially proposes 2 classes and describes forehead arc spy to preferably capture the corresponding constant shape feature in forehead region
The Haar-Like feature of sign, and with classical cascade Adaboost algorithm to the Haar-Like feature in forehead region
It practises.
It is further, described to be learnt by Haar-Like feature of the cascade Adaboost algorithm to the extension,
Training obtains the forward direction classifier and includes:
For given training sample (positive sample P, negative sample M), then the Haar-Like feature square of corresponding extension
Battle array is respectively as follows: P × 105888 and M × 105888.Learnt by cascade AdaBoost algorithm, controls each layer of positive sample
The probability and negative sample that are divided into positive sample are divided into the probability of negative sample, and what is set here is respectively 0.999 and 0.5, simultaneously
The termination condition of cascade classifier is that negative sample is divided into the probability of negative sample less than 10-6If Weak Classifier in each layer
Number is more than 500 training for also terminating cascade classifier.It is various due to being contained in training sample used in the embodiment of the present invention
Posture is blocked, illumination, the colour of skin, color development and situations such as resolution ratio, and the number of strong classifier is 5 in obtained cascade classifier, together
When every layer of corresponding Weak Classifier number be respectively that 55,85,135,140,331 namely cascade classifier negative sample are divided
Although the probability for negative sample is 0.0306. not up to defined 10-6, but classifier is filtered out almost
0.97 × M negative sample, while positive sample almost all is passed through.
Further, the multichannel color frequency for longitudinally extracting the forehead sample classified through the forward direction classifier is special
Sign, laterally extracts the multichannel color frequency feature for the forehead sample classified through the forward direction classifier, comprising:
In the longitudinal direction of N number of different Color Channel, it is various forms of that each position of the forehead sample areas corresponds to N kind
Mapping, obtains the various forms of mapping values of N kind of each position；
In the transverse direction of N number of different Color Channel, the corresponding channel of each dimension, the N kind for obtaining each channel is different
The colouring information of type；
Wherein, N is the integer greater than 3.
Particularly, N=8.
It should be noted that in order to fully describe the color characteristic of forehead, the embodiment of the present invention proposes multi-pass
The color frequency feature in road, experiment show that this feature can effectively capture the constant color characteristic of forehead.It is traditional based on
Face is mainly transformed into different color spaces (such as YCbCr, HSV, LUV from rgb space by the method for detecting human face of the colour of skin
Deng), corresponding gaussian probability model is then trained according to the distribution situation of area of skin color in space, thus according to gaussian probability
Model is treated the point in sample one's respective area and is predicted, judges whether to belong to human face region, finally carries out to human face region minimum
Boundary rectangle determines.Unlike conventional method, in order to more galore describe the probability of color development and the colour of skin in forehead region
Forehead is extended to multiple color spaces (YCbCr, HSV and LUV color space) from rgb space first herein by distribution character,
As shown in figure 3, then the information in 8 in three spaces different channels is effectively learnt with different learning algorithms,
And then obtain corresponding distribution character under different directions.
It is further, described that the multichannel color frequency feature longitudinally extracted is learnt by Cascade algorithms, comprising:
Obtain the corresponding multiple Weak Classifiers of multichannel color frequency feature longitudinally extracted；
Each Weak Classifier is learnt by the Gentle Adaboost algorithm based on CART tree, obtains longitudinal mention
The corresponding strong classifier of multichannel color frequency feature taken.
It should be noted that longitudinal extracting mode of color characteristic is in multiple Color Channels, in the longitudinal direction in different channels,
Each position in forehead region can have 8 kinds of various forms of projected forms, for forehead region, 8 kinds of differences of each position
The mapping of form has similar distribution, extracts 8 kinds of various forms of mapping values to each position.In multiple color spaces
In the different channels of (YCBCR, HSV, LUV), each position in forehead region corresponds to 8 relevant data, by N number of
The study of 8 related datas in training sample, the available one forehead region Weak Classifier based on more Color Channels, for
Entire forehead region, it will have 384 Weak Classifiers, then by SoftCasacde structure to 384 Weak Classifiers
It practises, so that it may the corresponding strong classifier of color frequency feature longitudinally extracted.In this configuration, Weak Classifier mainly passes through
Gentle Adaboost algorithm based on CART tree learns, and in order to improve detection rates, the depth of CART tree is no more than 2, together
When Gentle Adaboost algorithm in CART tree number cannot be more than 8.For each Weak Classifier, training sample is carried out
Detection, obtains the corresponding classification accuracy of 384 Weak Classifiers, then borrows SoftCascade structure to 384 weak point
Class device is trained, the corresponding strong classifier of color frequency feature longitudinally extracted.
It is further, described that the multichannel color frequency feature laterally extracted is learnt by Multiple Kernel Learning algorithm,
Include:
The multichannel color frequency feature laterally extracted for obtaining each pixel, by Multiple Kernel Learning algorithm to each pixel
The multichannel color frequency feature laterally extracted learnt, it is corresponding to obtain the multichannel color frequency feature laterally extracted
Strong classifier.
It should be noted that the lateral extracting mode of color frequency feature is in multiple Color Channels, in different channels
Laterally, every dimension all corresponds to a channel, and there are 384 pixels in each channel to describe the colouring information in forehead region, right
There is certain correlation between forehead region, 8 channels, by being together in series the corresponding pixel in each channel as class
Feature describes the colouring information in forehead region, then 8 channels then correspond to 8 kinds of different types of colouring informations.In multiple colors
In the different channels in space (YCBCR, HSV, LUV), forehead region has unique distribution in each channel, while each logical
There is very strong correlation between road, how the relational learning between them to be come out and the forehead identification of multichannel is extremely closed
Key.For each training sample, the corresponding length of each channel characteristics is 384, then the corresponding characteristic length in 8 channels is 8
× 384=3072.The embodiment of the present invention effectively merges the feature of different interchannels using Multiple Kernel Learning algorithm, obtains
One corresponding strong classifier of color frequency feature laterally extracted.In Multiple Kernel Learning, the feature pair extracted in each channel
The core answered is Gaussian kernel, while in order to increase the generalization ability of classifier, the penalty coefficient of Multiple Kernel Learning classifier is set as
16 (can voluntarily adjust according to demand), by the training of Multiple Kernel Learning classifier, an available color laterally extracted
The corresponding strong classifier of frequecy characteristic.
Further, described that target forehead in described image to be detected is obtained according to trained cascade forehead classifier
Region, comprising:
The forehead region in described image to be detected is obtained by the forward direction classifier；
It is obtained by the backward classifier through the preceding target forehead region into classifier acquisition forehead region.
It should be noted that a large amount of non-forehead sample has been filtered out, then by the forehead of erroneous detection to classifier by preceding
Sample collection plays the sample as new negative sample and forehead sample together as backward classifier.
Further, described to be obtained by the backward classifier through the preceding target volume into classifier acquisition forehead region
Head region includes:
It is obtained respectively through preceding to classifier by the classifier based on cascade structure and the judgement of the classifier based on Multiple Kernel Learning
Whether the forehead region obtained is forehead region；
Classifier based on cascade structure and the classifier based on Multiple Kernel Learning are judged before to classifier acquisition
The correct region in forehead region is determined as target forehead region.
It should be noted that, in order to increase the generalization ability of classifier, SoftCascade divides due to when training
Related coefficient (depth of CART and Gentle Adaboost in SoftCascade structure in class device and Multiple Kernel Learning classifier
The number of middle tree；The penalty coefficient of Multiple Kernel Learning) setting it is all smaller so that the false detection rate of classifier is relatively high.Therefore originally
Inventive embodiments construct the corresponding strong classifier of multichannel color frequency feature using voting mechanism, if being based on cascade structure
Classifier and classifier based on Multiple Kernel Learning be all detected as forehead, then it is forehead that the sample to be examined is corresponding, otherwise remaining
The case where be non-forehead region.
Further, described by Face Detection and edge detection, each target is obtained using improved least square method
The corresponding human face region in forehead region, comprising:
Each target forehead region is extended to obtain target person head region；
Area of skin color segmentation is carried out to the target person head region, obtains face complexion area；
Multiple profile points of face complexion area are obtained using edge detection algorithm；
It is selected at random from the multiple profile point, obtains the elliptical profile point of local optimum；
Least square fitting is carried out to the local optimum ellipse and obtains global optimum's ellipse, the optimal ellipse of whole is right
Answer the human face region.
It should be noted that the profile of face or part face under forehead region is one or partial ellipse.As a result,
Number of people region and its edge key point are obtained using priori knowledge and Skin Color Information, constrained ellipse fitting algorithm is recycled to obtain
To face；Meanwhile the gross alignment of face may be implemented according to elliptic parameter.
In order to detect human face region, author is first with priori knowledge, and the forehead region that will test is horizontal and vertical
Histogram obtains number of people region to respectively with the ratio of 3:2 and 3:1 into extension.Then the gaussian probability based on Skin Color Information is utilized
Model carries out area of skin color segmentation to the people's head region, obtains face complexion area therein.Then, edge detection algorithm is utilized
The profile point of the human face region, referred to as edge key point are obtained, (x is denoted asi, yi), (i=1,2 ..., k).Finally, being based on this
A little edges key point, is fitted face using improved least square method, while obtaining the inclined of face according to elliptic parameter
Gyration.In view of edge contour point is easy the interference by noise, deviation is brought to elliptical fitting, in the mistake of ellipse fitting
Minimum error of fitting constraint is introduced in journey.
Specifically, it was known that elliptic equation is defined as follows:
Wherein a=[A B C D E]TFor elliptical parameter vector,It is about marginal point (xi,
yi) vector.Consider noise spot and edge key point, above elliptical equation redefinable are as follows:
Wherein X=[X1X2...Xk]T∈RI×5It is the matrix that edge key point is constituted, b is two obtained by random process
Value sparse vector.λ is the penalty coefficient of sparse vector, and wherein f is defined as follows:
Wherein εdIt is an empirical value, X` indicates that the subset of edge key point, subset are represented by edge key dot matrix
With sparse vector b product:
X '=Xb
Detailed algorithm is as follows
Experiment discovery, cycle-index T directly affect fitting and obtain elliptical accuracy and calculate the duration (embodiment of the present invention
Experiment takes T=5000).Compared with tradition is based on the ellipse fitting method of whole profile points, the error of fitting of this method is smaller.By
This, face can be obtained by accurate positioning, and can obtain the angle excursion of face according to elliptic parameter simultaneously.
Fig. 2 is the structural schematic diagram of the face detection system based on forehead region of one embodiment of the invention.Such as Fig. 2 institute
Show, the face detection system based on forehead region of the embodiment of the present invention includes: image traversal unit 61, feature extraction unit
62, target forehead area acquisition unit 63 and human face region acquiring unit 64；Specifically:
Image traversal unit 61, for being traversed by the way of sliding window to image to be detected；
Feature extraction unit 62, for obtaining the variance of each sliding window to be detected, if the variance of the sliding window to be detected is big
In preset threshold, then the Haar-Like feature and multichannel color frequency feature of the extension of sliding window to be detected are extracted；
Target forehead area acquisition unit 63, for obtaining the mapping to be checked according to trained cascade forehead classifier
Target forehead region as in；
Human face region acquiring unit 64, for being obtained using improved least square method by Face Detection and edge detection
Obtain the corresponding human face region in each target forehead region；
Wherein, the cascade forehead classifier of training includes preceding to classifier and backward classifier；The forward direction classifier
Including cascade Adaboost classifier；The backward classifier includes classifier based on cascade structure and based on Multiple Kernel Learning
Classifier.
The face detection system based on forehead region of the present embodiment can be used for executing above method embodiment, principle
Similar with technical effect, details are not described herein again.
Method for detecting human face and system provided by the invention based on forehead region, extracts the extension of sliding window to be detected
Haar-Like feature and multichannel color frequency feature；It is single since the external expressive form of different type feature is not consistent
Feature learning algorithm be difficult well to learn feature, therefore the present invention is obtained according to trained cascade forehead classifier
Obtain the target forehead region in described image to be detected；By Face Detection and edge detection, using improved least square method
Obtain the corresponding human face region in each target forehead region.The present invention is by extracting the Haar-Like feature extended and multichannel
Color frequency feature, the color development in description forehead region more abundant and the Probability Characteristics of the colour of skin, enrich shape using having
The detection to face is realized in the forehead region of shape and color characteristic, improves the precision of Face datection.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, it wherein includes the computer of computer usable program code that the present invention, which can be used in one or more,
The computer program implemented in usable storage medium (including but not limited to magnetic disk storage, CD-ROM, optical memory etc.) produces
The form of product.
It should be noted that the terms "include", "comprise" or its any other variant are intended to the packet of nonexcludability
Contain, so that the process, method, article or equipment for including a series of elements not only includes those elements, but also including
Other elements that are not explicitly listed, or further include for elements inherent to such a process, method, article, or device.
In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including the element
Process, method, article or equipment in there is also other identical elements.
In specification of the invention, numerous specific details are set forth.Although it is understood that the embodiment of the present invention can
To practice without these specific details.In some instances, well known method, structure and skill is not been shown in detail
Art, so as not to obscure the understanding of this specification.Similarly, it should be understood that disclose in order to simplify the present invention and helps to understand respectively
One or more of a inventive aspect, in the above description of the exemplary embodiment of the present invention, each spy of the invention
Sign is grouped together into a single embodiment, figure, or description thereof sometimes.However, should not be by the method solution of the disclosure
Release is in reflect an intention that i.e. the claimed invention requires more than feature expressly recited in each claim
More features.More precisely, as the following claims reflect, inventive aspect is less than single reality disclosed above
Apply all features of example.Therefore, it then follows thus claims of specific embodiment are expressly incorporated in the specific embodiment,
It is wherein each that the claims themselves are regarded as separate embodiments of the invention.
The above examples are only used to illustrate the technical scheme of the present invention, rather than its limitations；Although with reference to the foregoing embodiments
Invention is explained in detail, those skilled in the art should understand that: it still can be to aforementioned each implementation
Technical solution documented by example is modified or equivalent replacement of some of the technical features；And these are modified or replace
It changes, the spirit and scope for technical solution of various embodiments of the present invention that it does not separate the essence of the corresponding technical solution.
Claims (10)
1. a kind of method for detecting human face based on forehead region characterized by comprising
Image to be detected is traversed by the way of sliding window；
The variance of each sliding window to be detected is obtained, if the variance of the sliding window to be detected is greater than preset threshold, is extracted to be detected
The Haar-Like feature and multichannel color frequency feature of the extension of sliding window；
The target forehead region in described image to be detected is obtained according to trained cascade forehead classifier；
By Face Detection and edge detection, the corresponding face in each target forehead region is obtained using improved least square method
Region；
Wherein, the cascade forehead classifier of training includes preceding to classifier and backward classifier；The forward direction classifier includes
Cascade Adaboost classifier；The backward classifier includes the classifier based on cascade structure and point based on Multiple Kernel Learning
Class device.
2. the method for detecting human face according to claim 1 based on forehead region, which is characterized in that the method is also wrapped
It includes:
Forehead sample is obtained, according to the cascade forehead classifier of forehead sample training.
3. the method for detecting human face according to claim 2 based on forehead region, which is characterized in that the acquisition forehead sample
This, include: according to the cascade forehead classifier of forehead sample training
The Haar-Like feature for extracting the extension of forehead sample, by cascade Adaboost algorithm to the Haar- of the extension
Like feature is learnt, and training obtains the forward direction classifier；
The multichannel color frequency feature for the forehead sample classified through the forward direction classifier is longitudinally extracted, is laterally extracted described in warp
The multichannel color frequency feature of the forehead sample of forward direction classifier classification；
The multichannel color frequency feature longitudinally extracted is learnt by Cascade algorithms, by Multiple Kernel Learning algorithm to transverse direction
The multichannel color frequency feature of extraction is learnt, and is trained after obtaining to classifier；
Wherein, the forehead sample includes positive sample and negative sample；The Haar-Like feature of extension includes description horizontal direction arc
The Haar-Like feature of shape feature and the Haar-Like feature of description vertical direction curved features.
4. the method for detecting human face according to claim 3 based on forehead region, which is characterized in that longitudinal extract passes through
The multichannel color frequency feature of the forehead sample of the forward direction classifier classification, laterally extracts and classifies through the forward direction classifier
Forehead sample multichannel color frequency feature, comprising:
In the longitudinal direction of N number of different Color Channel, each position of the forehead sample areas corresponds to that N kind is various forms of to reflect
It penetrates, obtains the various forms of mapping values of N kind of each position；
In the transverse direction of N number of different Color Channel, the corresponding channel of each dimension obtains the N kind different type in each channel
Colouring information；
Wherein, N is the integer greater than 3.
5. the method for detecting human face according to claim 3 based on forehead region, which is characterized in that described to be calculated by cascade
Method learns the multichannel color frequency feature longitudinally extracted, comprising:
Obtain the corresponding multiple Weak Classifiers of multichannel color frequency feature longitudinally extracted；
Each Weak Classifier is learnt by the GentleAdaboost algorithm based on CART tree, acquisition is longitudinally extracted more
The corresponding strong classifier of channel color frequency feature.
6. the method for detecting human face according to claim 3 based on forehead region, which is characterized in that described by multicore
Algorithm is practised to learn the multichannel color frequency feature laterally extracted, comprising:
The multichannel color frequency feature laterally extracted for obtaining each pixel, by Multiple Kernel Learning algorithm to the cross of each pixel
Learnt to the multichannel color frequency feature of extraction, it is strong point corresponding to obtain the multichannel color frequency feature laterally extracted
Class device.
7. the method for detecting human face according to claim 1 based on forehead region, which is characterized in that described according to trained
Cascade forehead classifier obtains the target forehead region in described image to be detected, comprising:
The forehead region in described image to be detected is obtained by the forward direction classifier；
It is obtained by the backward classifier through the preceding target forehead region into classifier acquisition forehead region.
8. the method for detecting human face according to claim 7 based on forehead region, which is characterized in that described by after described
It is obtained to classifier and includes: through the preceding target forehead region into classifier acquisition forehead region
It is obtained before to classifier by the classifier based on cascade structure and the judgement of the classifier based on Multiple Kernel Learning respectively
Whether forehead region is forehead region；
Classifier based on cascade structure and the classifier based on Multiple Kernel Learning are judged through the preceding forehead obtained to classifier
The correct region in region is determined as target forehead region.
9. the method for detecting human face according to claim 1 based on forehead region, which is characterized in that described to be examined by the colour of skin
Survey and edge detection obtain the corresponding human face region in each target forehead region using improved least square method, comprising:
Each target forehead region is extended to obtain target person head region；
Area of skin color segmentation is carried out to the target person head region, obtains face complexion area；
Multiple profile points of face complexion area are obtained using edge detection algorithm；
It is selected at random from the multiple profile point, obtains the elliptical profile point of local optimum；
Least square fitting is carried out to the local optimum ellipse and obtains global optimum's ellipse, global optimum's ellipse corresponds to institute
State human face region.
10. a kind of face detection system based on forehead region characterized by comprising
Image traversal unit, for being traversed by the way of sliding window to image to be detected；
Feature extraction unit is preset for obtaining the variance of each sliding window to be detected if the variance of the sliding window to be detected is greater than
Threshold value then extracts the Haar-Like feature and multichannel color frequency feature of the extension of sliding window to be detected；
Target forehead area acquisition unit, for being obtained in described image to be detected according to trained cascade forehead classifier
Target forehead region；
Human face region acquiring unit, for being obtained using improved least square method each by Face Detection and edge detection
The corresponding human face region in target forehead region；
Wherein, the cascade forehead classifier of training includes preceding to classifier and backward classifier；The forward direction classifier includes
Cascade Adaboost classifier；The backward classifier includes the classifier based on cascade structure and point based on Multiple Kernel Learning
Class device.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610607532.0A CN106250843B (en) | 2016-07-28 | 2016-07-28 | A kind of method for detecting human face and system based on forehead region |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610607532.0A CN106250843B (en) | 2016-07-28 | 2016-07-28 | A kind of method for detecting human face and system based on forehead region |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106250843A CN106250843A (en) | 2016-12-21 |
| CN106250843B true CN106250843B (en) | 2019-04-26 |
Family
ID=57603993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610607532.0A Active CN106250843B (en) | 2016-07-28 | 2016-07-28 | A kind of method for detecting human face and system based on forehead region |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106250843B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108280388A (en) * | 2017-01-06 | 2018-07-13 | 富士通株式会社 | The method and apparatus and type of face detection method and device of training face detection model |
| CN109002799B (en) * | 2018-07-19 | 2021-08-24 | 苏州市职业大学 | Face recognition method |
| CN112053381A (en) * | 2020-07-13 | 2020-12-08 | 北京迈格威科技有限公司 | Image processing method, image processing device, electronic equipment and storage medium |
| CN113536917B (en) * | 2021-06-10 | 2024-06-07 | 浙江大华技术股份有限公司 | Dressing recognition method, system, electronic device and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093250A (en) * | 2013-02-22 | 2013-05-08 | 福建师范大学 | Adaboost face detection method based on new Haar- like feature |
| CN104715244A (en) * | 2015-04-01 | 2015-06-17 | 华中科技大学 | Multi-viewing-angle face detection method based on skin color segmentation and machine learning |
| CN105335684A (en) * | 2014-06-25 | 2016-02-17 | 小米科技有限责任公司 | Face detection method and device |
-
2016
- 2016-07-28 CN CN201610607532.0A patent/CN106250843B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093250A (en) * | 2013-02-22 | 2013-05-08 | 福建师范大学 | Adaboost face detection method based on new Haar- like feature |
| CN105335684A (en) * | 2014-06-25 | 2016-02-17 | 小米科技有限责任公司 | Face detection method and device |
| CN104715244A (en) * | 2015-04-01 | 2015-06-17 | 华中科技大学 | Multi-viewing-angle face detection method based on skin color segmentation and machine learning |
Non-Patent Citations (3)
| Title |
|---|
| AdaBoost Face Detection on the GPU using Haar-like Features;M.Martinez-Zarzuela等;《ResearchGate》;20110531 |
| FACE RECOGNITION ALGORITHM BASED ON HAAR-LIKE FEATURES AND GENTLE ADABOOST FEATURE SELECTION VIA SPARSE REPRESENTATION;Qingwei Wang等;《Applied Mechanics and Materials》;20150316;第742卷 |
| FACIAL FEATURE DETECTION USING HAAR CLASSIFIERS;Phillip Ian Wilson等;《Journal of Computing Science in Colleges》;20060430;第21卷(第4期) |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106250843A (en) | 2016-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101383001B (en) | Quick and precise front human face discriminating method | |
| CN103400110B (en) | Abnormal face detecting method before ATM cash dispenser | |
| CN102194108B (en) | Smile face expression recognition method based on clustering linear discriminant analysis of feature selection | |
| CN108268859A (en) | A kind of facial expression recognizing method based on deep learning | |
| CN106250843B (en) | A kind of method for detecting human face and system based on forehead region | |
| Roomi et al. | Race classification based on facial features | |
| CN103632132A (en) | Face detection and recognition method based on skin color segmentation and template matching | |
| Zheng et al. | Attention-based spatial-temporal multi-scale network for face anti-spoofing | |
| CN106355138A (en) | Face recognition method based on deep learning and key features extraction | |
| Abate et al. | BIRD: Watershed based iris detection for mobile devices | |
| Gupta et al. | Face detection using modified Viola jones algorithm | |
| El Maghraby et al. | Detect and analyze face parts information using Viola-Jones and geometric approaches | |
| CN104021384B (en) | A kind of face identification method and device | |
| CN108446642A (en) | A kind of Distributive System of Face Recognition | |
| Özbudak et al. | Effects of the facial and racial features on gender classification | |
| CN107862298A (en) | It is a kind of based on the biopsy method blinked under infrared eye | |
| Szczepański et al. | Pupil and iris detection algorithm for near-infrared capture devices | |
| Anantharajah et al. | Quality based frame selection for video face recognition | |
| Méndez-Llanes et al. | On the use of local fixations and quality measures for deep face recognition | |
| Yi et al. | Face detection method based on skin color segmentation and facial component localization | |
| Rifaee et al. | Unconstrained iris recognition | |
| Xu et al. | Patch-wise skin segmentation of human body parts via deep neural networks | |
| El Maghraby et al. | Detecting and tracking of multiple people in video based on hybrid detection and human anatomy body proportion | |
| Das et al. | A novel approach towards detecting faces and gender using skin segmentation and template matching | |
| Qian et al. | A new preprocessing algorithm and local binary pattern based facial expression recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |