CN107341457A - Method for detecting human face and device - Google Patents
Method for detecting human face and device Download PDFInfo
- Publication number
- CN107341457A CN107341457A CN201710476106.2A CN201710476106A CN107341457A CN 107341457 A CN107341457 A CN 107341457A CN 201710476106 A CN201710476106 A CN 201710476106A CN 107341457 A CN107341457 A CN 107341457A
- Authority
- CN
- China
- Prior art keywords
- face
- npd
- human
- combinations
- trained
- 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
- G06V40/164—Detection; Localisation; Normalisation using holistic features
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The disclosure is directed to a kind of method for detecting human face and device, to improve the speed of human world Face datection and the degree of accuracy.Methods described includes：The first NPD for being used to represent face characteristic carried according to the human-face detector trained is combined, and extracts the 2nd NPD combinations of image to be detected；Recurrence processing is carried out to the 2nd NPD combinations by human-face detector, obtains the human face region in image to be detected；The N-dimensional characteristic vector of human face region is extracted, N is natural number；Recurrence calculating is carried out to N-dimensional characteristic vector by the logistic regression grader trained, obtains the probability that human face region has face；Determine face in human face region be present when the probability that face be present is more than predetermined threshold value.Disclosed technique scheme can improve speed and the degree of accuracy of Face datection.
Description
Technical field
This disclosure relates to image identification technical field, more particularly to a kind of method for detecting human face and device.
Background technology
Human face detection tech refers to the technology that human face region is identified from the image including face, is relatively conventional at present
A kind of image processing techniques.
In correlation technique, recognition of face can be used for the fields such as secure payment, authentication.Because these fields are to safety
The requirement of property is higher, so the speed and the degree of accuracy to recognition of face just propose higher requirement.For example, when recognition of face should
,, will be serious if recognition of face will spend 2 seconds on the basis of ensuring that recognition of face is accurate with when being unlocked on mobile phone
Influence customer experience, if it is possible to shorten to 0.2 second, the sensation of client can be entirely different.
The content of the invention
The embodiment of the present disclosure provides a kind of method for detecting human face and device, to improve the speed of human world Face datection and standard
Exactness.
According to the first aspect of the embodiment of the present disclosure, there is provided a kind of method for detecting human face, including：
The the first normalization pixel value difference NPD groups for being used to represent face characteristic carried according to the human-face detector trained
Close, extract the 2nd NPD combinations of image to be detected；
Recurrence processing is carried out to the 2nd NPD combinations by the human-face detector, obtained in described image to be detected
Human face region；
The N-dimensional characteristic vector of the human face region is extracted, N is natural number；
Recurrence calculating is carried out to the N-dimensional characteristic vector by the logistic regression grader trained, obtains the face
The probability of face be present in region；
Determine face be present in the human face region when the probability that face be present is more than predetermined threshold value.
In one embodiment, the human-face detector trained can carry the NPD orders in the first NPD combinations；
What the human-face detector that the basis has been trained carried is used to represent the first NPD combinations of face characteristic, and extraction is treated
The 2nd NPD combinations of detection image, it may include：
According to the first NPD combinations and NPD orders, the 2nd NPD combinations are extracted；
It is described that recurrence processing is carried out to the 2nd NPD combinations by the human-face detector, obtain the mapping to be checked
Human face region as in, it may include：
NPD orders are based on by the human-face detector recurrence processing is carried out to the 2nd NPD combinations, obtain institute
State the human face region in image to be detected.
In one embodiment, what the human-face detector that the basis has been trained carried is used to represent the first of face characteristic
NPD is combined, and before the 2nd NPD combinations for extracting image to be detected, methods described may also include：
It is determined that the first training set that the first positive sample including face forms with not including the first negative sample of face；
Determine first positive sample and each self-corresponding NPD of first negative sample；
Learn to obtain multiple depth Quadratic Finite Elements based on first positive sample and each self-corresponding NPD of first negative sample
Tree；Optimal N PD of the depth Quadratic Finite Element tree study of Confucian classics acquistion to each minor matters point；
Based on the human-face detector trained described in the multiple depth Quadratic Finite Element tree structure；The multiple depth Quadratic Finite Element
The optimal N PD that tree includes forms the first NPD combinations.
In one embodiment, it is described determine first positive sample with before each self-corresponding NPD of first negative sample,
Methods described may also include：
By each self-corresponding NPD storages of first positive sample and first negative sample in a lookup table；
Determination first positive sample and each self-corresponding NPD of first negative sample, including：
First positive sample and each self-corresponding NPD of first negative sample are determined by accessing the look-up table.
In one embodiment, the logistic regression grader by having trained returns to the N-dimensional characteristic vector
Calculate, before obtaining the probability that the human face region has face, methods described may also include：
The second positive sample including face is inputted not with not including the second training set that the second negative sample of face forms
The logistic regression grader of training；
Self-corresponding parameter each to N-dimensional characteristic vector in the anticipation function of the untrained logistic regression grader is carried out
Training；Second negative sample is the negative sample that face failure is detected based on NPD；
When it is determined that each self-corresponding parameter of N-dimensional characteristic vector meets preparatory condition in the anticipation function, deconditioning
The logistic regression grader, obtain the logistic regression grader trained.
In one embodiment, it is described to determine that each self-corresponding parameter of N-dimensional characteristic vector meets default in the anticipation function
Condition, it may include：
Determine whether the damage function value of the anticipation function reaches minimum value；
When the value of the loss function of the anticipation function reaches minimum value, determine in the anticipation function N-dimensional feature to
Measure each self-corresponding parameter and meet the preparatory condition.
According to the second aspect of the embodiment of the present disclosure, there is provided a kind of human face detection device, including：
Fisrt feature extraction module, it is configured as according to the human-face detector carrying trained for representing face characteristic
The first normalization pixel value difference NPD combination, extract the 2nd NPD combinations of image to be detected；
Processing module is returned, is configured as carrying out recurrence processing to the 2nd NPD combinations by the human-face detector,
Obtain the human face region in described image to be detected；
Second feature extraction module, is configured as extracting the N-dimensional characteristic vector of the human face region, and N is natural number；
Computing module is returned, is configured as carrying out the N-dimensional characteristic vector by the logistic regression grader trained
Return and calculate, obtain the probability that the human face region has face；
First determining module, it is configured as determining the face area when the probability that face be present is more than predetermined threshold value
Face in domain be present.
In one embodiment, the human-face detector trained can carry the NPD orders in the first NPD combinations；
The fisrt feature extraction module, it is also configured to according to the first NPD combinations and NPD orders,
Extract the 2nd NPD combinations；
The recurrence processing module, it is also configured to be based on NPD orders to described by the human-face detector
2nd NPD combinations carry out recurrence processing, obtain the human face region in described image to be detected.
In one embodiment, described device, may also include：
Second determining module, it is configured to determine that the first positive sample including face and does not include the first negative sample of face
First training set of composition；
3rd determining module, it is configured to determine that first positive sample and each self-corresponding NPD of first negative sample；
Study module, it is configured as learning based on first positive sample and each self-corresponding NPD of first negative sample
Obtain multiple depth Quadratic Finite Element trees；Optimal N PD of the depth Quadratic Finite Element tree study of Confucian classics acquistion to each minor matters point；
Module is built, is configured as based on the human-face detector trained described in the multiple depth Quadratic Finite Element tree structure；
The optimal N PD that the multiple depth Quadratic Finite Element tree includes forms the first NPD combinations.
In one embodiment, described device, may also include：
Memory module, it is configured as first positive sample and each self-corresponding NPD of first negative sample being stored in
In look-up table；
3rd determining module, be additionally configured to by access the look-up table determine first positive sample with it is described
Each self-corresponding NPD of first negative sample.
In one embodiment, described device, may also include：
Input module, it is configured as form the second positive sample including face with not including the second negative sample of face
Second training set inputs untrained logistic regression grader；
Training module, it is configured as N-dimensional characteristic vector in the anticipation function to the untrained logistic regression grader
Each self-corresponding parameter is trained；Second negative sample is the negative sample that face failure is detected based on NPD；
Control module, be configured as it is determined that in the anticipation function each self-corresponding parameter of N-dimensional characteristic vector meet it is pre-
If during condition, logistic regression grader described in deconditioning, obtain the logistic regression grader trained.
In one embodiment, described device, may also include：
4th determining module, is additionally configured to determine whether the damage function value of the anticipation function reaches minimum
Value；
5th determining module, it is additionally configured to when the value of the loss function of the anticipation function reaches minimum value,
Determine that each self-corresponding parameter of N-dimensional characteristic vector meets the preparatory condition in the anticipation function.
According to the third aspect of the embodiment of the present disclosure, there is provided a kind of human face detection device, described device include：
Processor；
For storing the memory of processor-executable instruction；
Wherein, the processor is configured as：
The the first normalization pixel value difference NPD groups for being used to represent face characteristic carried according to the human-face detector trained
Close, extract the 2nd NPD combinations of image to be detected；
Recurrence processing is carried out to the 2nd NPD combinations by the human-face detector, obtained in described image to be detected
Human face region；
The N-dimensional characteristic vector of the human face region is extracted, N is natural number；
Recurrence calculating is carried out to the N-dimensional characteristic vector by the logistic regression grader trained, obtains the face
The probability of face be present in region；
Determine face be present in the human face region when the probability that face be present is more than predetermined threshold value.
According to the fourth aspect of the embodiment of the present disclosure, there is provided a kind of computer-readable recording medium, be stored thereon with calculating
Machine program, the computer program realize following steps when being executed by processor：
The the first normalization pixel value difference NPD groups for being used to represent face characteristic carried according to the human-face detector trained
Close, extract the 2nd NPD combinations of image to be detected；
Recurrence processing is carried out to the 2nd NPD combinations by the human-face detector, obtained in described image to be detected
Human face region；
The N-dimensional characteristic vector of the human face region is extracted, N is natural number；
Recurrence calculating is carried out to the N-dimensional characteristic vector by the logistic regression grader trained, obtains the face
The probability of face be present in region；
Determine face be present in the human face region when the probability that face be present is more than predetermined threshold value.
The technical scheme provided by this disclosed embodiment can include the following benefits：First basis is based on NPD
What is carried in the human-face detector trained of (Normalized Pixel Difference, normalizing pixel value difference) is used for
The first NPD combinations of face characteristic are represented, the 2nd NPD combinations of image to be detected are extracted, by human-face detector to the 2nd NPD
Combination carries out recurrence processing, the human face region that can be quickly obtained in image to be detected；Then, obtained human face region is extracted
N-dimensional characteristic vector, recurrence calculating is carried out to N-dimensional characteristic vector by the logistic regression grader trained, can be rapidly
Obtain human face region and the probability of face be present, and the probability of face based on obtained human face region be present, can examine what is obtained
It whether there is face in human face region, wherein, obtained human face region is determined when the probability that face be present is more than predetermined threshold value
In face be present, so, can rapidly by obtained human face region be not present face human face region filter out, reduce miss
Inspection rate.The technical scheme of the disclosure, the speed of Face datection can be not only improved, the degree of accuracy of Face datection can also be improved.
It should be appreciated that the general description and following detailed description of the above are only exemplary and explanatory, not
The disclosure can be limited.
Brief description of the drawings
Accompanying drawing herein is merged in specification and forms the part of this specification, shows the implementation for meeting the present invention
Example, and for explaining principle of the invention together with specification.
Figure 1A is the flow chart of the method for detecting human face according to an exemplary embodiment.
Figure 1B is the NPD combination diagrams that the human-face detector trained according to an exemplary embodiment carries.
Fig. 1 C are the scene graph of the method for detecting human face according to an exemplary embodiment.
Fig. 2A is the flow chart of the method for detecting human face according to an exemplary embodiment one.
Fig. 2 B are the structural representations of the depth Quadratic Finite Element tree for Face datection according to an exemplary embodiment one
Figure.
Fig. 3 is the flow chart to human-face detector training according to an exemplary embodiment two.
Fig. 4 is the flow chart of the method for detecting human face according to an exemplary embodiment three.
Fig. 5 is the flow chart of the method for detecting human face according to an exemplary embodiment four.
Fig. 6 is a kind of block diagram of human face detection device according to an exemplary embodiment.
Fig. 7 A are the block diagrams of another human face detection device according to an exemplary embodiment.
Fig. 7 B are the block diagrams of another human face detection device according to an exemplary embodiment.
Fig. 8 A are the block diagrams of another human face detection device according to an exemplary embodiment.
Fig. 8 B are the block diagrams of another human face detection device according to an exemplary embodiment.
Fig. 9 is a kind of block diagram of human face detection device according to an exemplary embodiment.
Embodiment
Here exemplary embodiment will be illustrated in detail, its example is illustrated in the accompanying drawings.Following description is related to
During accompanying drawing, unless otherwise indicated, the same numbers in different accompanying drawings represent same or analogous key element.Following exemplary embodiment
Described in embodiment do not represent and the consistent all embodiments of the present invention.On the contrary, they be only with it is such as appended
The example of the consistent apparatus and method of some aspects being described in detail in claims, of the invention.
Figure 1A is the flow chart of the method for detecting human face according to an exemplary embodiment, and Figure 1B is exemplary according to one
The NPD combination diagrams that the human-face detector trained for implementing to exemplify carries, Fig. 1 C are shown according to an exemplary embodiment
The scene graph of the method for detecting human face gone out；The method for detecting human face can apply terminal device (such as：Smart mobile phone, flat board
Computer), in camera device or the human-face detection equipment that is connected with camera device.As shown in Figure 1A, the method for detecting human face
Comprise the following steps S101-S105：
In step S101, according to the first NPD groups for being used to represent face characteristic for the human-face detector carrying trained
Close, extract the 2nd NPD combinations of image to be detected.
In one embodiment, the first NPD combinations include several NPD, and several NPD combine and can represent people
The essential structure of face.In one exemplary embodiment, as shown in Figure 1B, the first NPD combinations can include f1, f2, f3, f4 table
Show that 4 NPD of face characteristic, 4 NPD can represent the essential structure of face.For example, f1 can represent the feature of eyes,
F2 can represent the feature of eyebrow, and f3 can represent the feature of nose, and f4 can represent the feature of mouth.
The NPD that first NPD combinations include determines according to the training result of human-face detector.Wherein, NPD is two pixels
Pixel value difference is normalized, can be obtained according to Weber (weber) rule, can be specifically calculated according to equation below：
Wherein, x, y are respectively the brightness value of two pixels.In exemplary embodiments mentioned above, as shown in Figure 1B, f2 is
Pixel P1, P2 normalization pixel value difference, x are the brightness value of pixel P2 pixel, and y is pixel P1 pixel
Brightness value.
When needing to carry out Face datection to image to be detected, processing is first zoomed in and out to image to be detected so that scaling
The resolution ratio of image to be detected after processing is identical with the input dimension of human-face detector.For example, it is 20*20 for input dimension
Human-face detector, if the video image of camera device is 800*600, the resolution ratio of the image collected is also 800*
600, now need the image for 20*20 by the image normalization of the 800*600 resolution ratio.
After processing is zoomed in and out to image to be detected, according to the human-face detector carrying trained for representing face
The first NPD combinations of feature, the 2nd NPD combinations of image to be detected after extraction scaling processing.In one embodiment, according to
The 2nd NPD combinations are extracted in the position of pixel corresponding to the NPD that first NPD combinations include.In exemplary embodiments mentioned above
In, as shown in Figure 1B, according to corresponding to f1, f2, f3, f4 pixel position extract the 2nd NPD combination include NPD (f1 ',
F2 ', f3 ', f4 '), wherein, f1 ', f2 ', f3 ', f4 ' corresponding pixel position pixel corresponding with f1, f2, f3, f4
The position of point corresponds.For example, the position of pixel corresponding to the f2 ' position correspondence with P1, P2 respectively.
In step s 102, recurrence processing is carried out to the 2nd NPD combinations by human-face detector, obtained in image to be detected
Human face region.
In one embodiment, recurrence processing can be carried out to the 2nd NPD combinations by human-face detector, obtained to be detected
Human face region in image, and human face region is irised out by square frame or oval frame, to show human face region that detection obtains.
In step s 103, the N-dimensional characteristic vector of human face region is extracted.Wherein N is natural number.
In one embodiment, N-dimensional characteristic vector can be N-dimensional HOG (Histogram of Oriented
Gradient, histograms of oriented gradients) characteristic vector.In another embodiment, N-dimensional characteristic vector can be N-dimensional LBP
(Local Binary Pattern, local binary patterns) characteristic vector.In a further embodiment, N-dimensional characteristic vector can be with
It is NPD characteristic vectors, can so saves the Face datection time, improves Face datection efficiency.
In step S104, recurrence calculating is carried out to N-dimensional characteristic vector by the logistic regression grader trained, obtained
The probability of face be present in human face region.
In one embodiment, logistic regression grader is returned by following formula (2), (3) to N-dimensional characteristic vector
Return calculating, obtain the probability that human face region has face.
Wherein, θ0、θ1、…、θnIt is characterized vector x1、x2、…、xnCorresponding parameter, i are natural number, θTFor the square of parameter θ
Formation formula, x are characterized vector matrix, hθ(x) it is anticipation function, hθ(x) functional value is the probability that human face region has face.
It can draw the probability of face be present in the N-dimensional characteristic vector calculating human face region based on extraction by (2) formula
When, once judge to need N number of multiplying to complete, calculating speed is fast, improves the efficiency of Face datection.
In step S105, determine face in human face region be present when the probability that face be present is more than predetermined threshold value.Base
The probability of face be present in human face region obtained above, whether there is face in the human face region that can examine to obtain, wherein,
Face be present in the human face region for determining to obtain when the probability that face be present is more than predetermined threshold value, when the probability that face be present is small
Face is not present in the human face region for determining to obtain when predetermined threshold value.
As an exemplary scenario, when user logs in certain application using mobile phone 11 as shown in Figure 1 C, the application allows
Brush face logs in.When user logs in application using the login mode of brush face, the application can be by calling camera 12 to face
Taken pictures or imaged, device 13 realizes recognition of face by performing the method for detecting human face that the disclosure provides.Device 13 can
To be a part for mobile phone 11 or the device independently of mobile phone.Specifically, image capture module 14 can be by default
The image that collects of sampling period acquisition camera 12, pretreatment module 15 is according to the input dimension of human-face detector to image
The image that acquisition module collects zooms in and out processing so that scaling processing after image resolution ratio and human-face detector it is defeated
Enter that dimension is identical, the image after pretreatment module 15 handles scaling is inputted to the human-face detector 16 trained, by having instructed
Experienced human-face detector 16 extracts the region that face in image be present, and human-face detector 16 is by the region of the presence face extracted
Output to the logistic regression grader 17 trained, logistic regression grader 17 whether there is to examining in above-mentioned human face region
Face.Just human face region is exported to follow-up face recognition module 18 after it is determined that face be present in human face region and carries out face
Identification, to determine whether that user logs in, the image of collection can not be exported when face is not present in the image of collection
To face recognition module 18.The technical scheme of the disclosure, use the method for dual Face datection not only can be with Face datection
The degree of accuracy of Face datection is improved, while ensure that the speed of Face datection, is favorably improved Consumer's Experience.
In another exemplary scenario, the technical scheme of the disclosure can use the inspection of default dimension (such as 20*20)
Survey window and Face datection is carried out to image to be detected including multiple faces, be not limited to image to be detected including individual human face
Carry out Face datection.
In the present embodiment, first according to based on NPD (Normalized Pixel Difference, normalizing pixel value difference)
The human-face detector trained in the first NPD combinations for being used to represent face characteristic that carry, extract the of image to be detected
Two NPD are combined, and are carried out recurrence processing to the 2nd NPD combinations by human-face detector, can be quickly obtained in image to be detected
Human face region；Then, the obtained N-dimensional characteristic vector of human face region is extracted, by the logistic regression grader trained to N
Dimensional feature vector carries out recurrence calculating, can be quickly obtained the probability that human face region has face, and based on obtained face
There is the probability of face in region, whether there is face in the human face region that can examine to obtain, wherein, when the probability that face be present
Face in the human face region for determining to obtain during more than predetermined threshold value be present, so, can be rapidly by obtained human face region
Filtered out in the absence of the human face region of face, reduce false drop rate.The technical scheme of the disclosure, it can not only improve Face datection
Speed, the degree of accuracy of Face datection can also be improved.
In one embodiment, the NPD that the human-face detector trained can be carried in the first NPD combinations is suitable
Sequence；
What the human-face detector that the basis has been trained carried is used to represent the first NPD combinations of face characteristic, and extraction is treated
The 2nd NPD combinations of detection image, it may include：
According to the first NPD combinations and NPD orders, the 2nd NPD combinations are extracted；
It is described that recurrence processing is carried out to the 2nd NPD combinations by the human-face detector, obtain the mapping to be checked
Human face region as in, it may include：
NPD orders are based on by the human-face detector recurrence processing is carried out to the 2nd NPD combinations, obtain institute
State the human face region in image to be detected.
In one embodiment, what the human-face detector that the basis has been trained carried is used to represent the first of face characteristic
NPD is combined, and before the 2nd NPD combinations for extracting image to be detected, methods described may also include：
It is determined that the first training set that the first positive sample including face forms with not including the first negative sample of face；
Determine first positive sample and each self-corresponding NPD of first negative sample；
Learn to obtain multiple depth Quadratic Finite Elements based on first positive sample and each self-corresponding NPD of first negative sample
Tree；Optimal N PD of the depth Quadratic Finite Element tree study of Confucian classics acquistion to each minor matters point；
Based on the human-face detector trained described in the multiple depth Quadratic Finite Element tree structure；The multiple depth Quadratic Finite Element
The optimal N PD that tree includes forms the first NPD combinations.
In one embodiment, it is described determine first positive sample and each self-corresponding NPD of first negative sample it
Before, methods described may also include：
By each self-corresponding NPD storages of first positive sample and first negative sample in a lookup table；
It is described to determine first positive sample and each self-corresponding NPD of first negative sample, it may include：
First positive sample and each self-corresponding NPD of first negative sample are determined by accessing the look-up table.
In one embodiment, the logistic regression grader by having trained returns to the N-dimensional characteristic vector
Return calculating, before obtaining the probability that the human face region has face, methods described may also include：
The second positive sample including face is inputted not with not including the second training set that the second negative sample of face forms
The logistic regression grader of training；
Self-corresponding parameter each to N-dimensional characteristic vector in the anticipation function of the untrained logistic regression grader is carried out
Training；Second negative sample is the negative sample that face failure is detected based on NPD；
When it is determined that each self-corresponding parameter of N-dimensional characteristic vector meets preparatory condition in the anticipation function, deconditioning
The logistic regression grader, obtain the logistic regression grader trained.
In one embodiment, it is described to determine that each self-corresponding parameter of N-dimensional characteristic vector meets pre- in the anticipation function
If condition, including：
Determine whether the damage function value of the anticipation function reaches minimum value；
When the value of the loss function of the anticipation function reaches minimum value, determine in the anticipation function N-dimensional feature to
Measure each self-corresponding parameter and meet the preparatory condition.
Specifically how face is detected, refer to subsequent embodiment.
So far, the above method that the embodiment of the present disclosure provides, it is first quick according to the human-face detector trained based on NPD
Ground obtains the human face region in image to be detected；Then, obtained people can be examined by the logistic regression grader trained
It whether there is face in face region.So, the technical scheme of the disclosure, the speed of Face datection can be not only improved, can be with
Improve the degree of accuracy of Face datection.
The technical scheme of embodiment of the present disclosure offer is provided with specific embodiment below.
Fig. 2A is the flow chart of the method for detecting human face according to an exemplary embodiment one；Fig. 2 B are according to an example
The structural representation of the depth Quadratic Finite Element tree (DQT) for Face datection shown in property embodiment one；The present embodiment utilizes this public affairs
The above method that embodiment provides is opened, with NPD order extraction the 2nd NPD combinations in the first NPD combinations, and is based on being somebody's turn to do
NPD orders carry out illustrative exemplified by recurrence processing to the 2nd NPD combinations, as shown in Figure 2 A, comprise the following steps：
In step s 201, according to the first NPD combinations and NPD orders, extraction the 2nd NPD combinations.
In one embodiment, the human-face detector trained can carry the NPD orders in the first NPD combinations.Wherein,
NPD orders are to the order of NPD processing when carrying out Face datection.In one exemplary embodiment, as shown in Figure 2 B, instructed
Experienced human-face detector can include multiple depth Quadratic Finite Element trees (DQT), and depth Quadratic Finite Element tree (DQT) is used to carry out face inspection
Survey.Depth Quadratic Finite Element tree (DQT) first first judges f1 in root node 21 when carrying out Face datection, specifically judges f1
Whether [θ 11, θ 12] is belonged to, when judged result when being, to judge in minor matters point 22 f3, when judged result is no,
Minor matters point 23 is judged f2.When minor matters point 22 is judged f3, specifically judge whether f3 belongs to [θ 31, θ 32],
When judged result is to be, judged result corresponding to output (for example not including face), when judged result is no, in leaf section
24 couples of f4 of point judge.When minor matters point 23 is judged f2, specifically judge whether f2 belongs to [θ 21, θ 22], when sentencing
Disconnected result is when being, judges f4 in leaf node 24, when judged result is no, (the ratio of judged result corresponding to output
If do not included face).When leaf node 24 is judged f4, specifically judge whether f4 belongs to [θ 41, θ 42], works as judgement
As a result for when being, judged result (such as including face) corresponding to output, when judged result is no, output is corresponding to judge knot
Fruit (for example not including face).
When extracting the 2nd NPD combinations, the position of pixel determines second corresponding to NPD in not only being combined according to the first NPD
The position of pixel corresponding to NPD in NPD combinations, moreover, it is also possible to which NPD orders determine the 2nd NPD groups in being combined according to the first NPD
NPD sequence of extraction in conjunction.In exemplary embodiments mentioned above, the first NPD combinations include f1, f2, f3, f4 etc. 4
NPD, the 2nd NPD combination include 4 NPD such as f1 ', f2 ', f3 ', f4 ', if the NPD orders of the first NPD combinations are followed successively by
F1, f2, f3, f4, then NPD sequence of extraction is followed successively by f1 ', f2 ', f3 ', f4 ' in the 2nd NPD combinations.2nd NPD groups of extraction
Conjunction can be stored according to NPD sequence of extraction.
In step S202, NPD orders are based on by human-face detector recurrence processing is carried out to the 2nd NPD combinations, obtained
Human face region in image to be detected.
, can be based on the NPD orders that the first NPD is combined when human-face detector return processing to the 2nd NPD combinations
Recurrence processing is carried out to the 2nd NPD combinations.Because the 2nd NPD combinations are according to the NPD order extraction NPD of the first NPD combinations, and
NPD orders during Face datection again according to the first NPD combinations carry out recurrence processing to the 2nd NPD combinations, so, can be extracted with side
NPD sides carry out Face datection, i.e. feature extraction is carried out simultaneously with Face datection, can shorten the Face datection time, improve face
Detection efficiency.
In step S203, the N-dimensional characteristic vector of human face region is extracted.Wherein N is natural number.
In step S204, recurrence calculating is carried out to N-dimensional characteristic vector by the logistic regression grader trained, obtained
The probability of face be present in human face region.
In step S205, determine face in human face region be present when the probability that face be present is more than predetermined threshold value.
Step S203~S205 in the present embodiment is similar with above-mentioned step S103~S105, will not be repeated here.
In the present embodiment, by NPD order extraction the 2nd NPD combinations in being combined according to the first NPD, and the NPD is based on
Order carries out recurrence processing to the 2nd NPD combinations, can further improve the speed of Face datection.
Fig. 3 is the flow chart to human-face detector training according to an exemplary embodiment two；The present embodiment utilizes
The above method that the embodiment of the present disclosure provides, it is illustrative exemplified by how being trained to human-face detector, such as Fig. 3
It is shown, comprise the following steps：
In step S301, it is determined that the first positive sample including face formed with not including the first negative sample of face the
One training set.
In one embodiment, the first positive sample and the first negative sample can be the samples of not carrier's work tag along sort,
Cost of labor can so be saved.
In step s 302, the first positive sample and each self-corresponding NPD of the first negative sample are determined.
In one embodiment, can be according to formula (1) by the first positive sample being calculated with the first negative sample each
Corresponding NPD.
In step S303, learn to obtain multiple depth based on the first positive sample and each self-corresponding NPD of the first negative sample
Quadratic Finite Element tree.Wherein optimal N PD of the depth Quadratic Finite Element tree study of Confucian classics acquistion to each node.
In exemplary embodiments mentioned above, as shown in Figure 1B, f1, f2, f3, f4 are respectively the root section of depth Quadratic Finite Element tree
Point 21, minor matters point 23, minor matters point 22, leaf node 24 learn obtained optimal N PD, [θ 11, θ 12], [θ 21, θ 22], [θ 31, θ
32], [θ 41, θ 42] is the division threshold value that study obtains.Depth Quadratic Finite Element tree can be with the optimal N PD of study of Confucian classics acquistion to each node
And division threshold value.Each depth Quadratic Finite Element tree can be used as a Weak Classifier, and each iteration can obtain a weak typing
Device (depth Quadratic Finite Element tree), multiple Weak Classifiers that multiple secondary iteration obtain may be constructed a strong classifier.
In one exemplary embodiment, AdaBoost Algorithm for Training graders can be used, study, which most has, distinguishes power
Feature, strong classifier is built, and the depth Quadratic Finite Element tree based on NPD is learnt using Gentle AdaBoost algorithms.So may be used
To improve the degree of accuracy of Face datection.
In another exemplary embodiment, a cascade classifier is further learnt to be used for Face datection.Moreover, also
Come Fast Learning and negative sample can be excluded with soft cascade structure.Soft cascade is considered as an AdaBoost grader, each
Weak Classifier has a terminal.In each iterative process, a depth Quadratic Finite Element tree is learnt as Weak Classifier, currently
The threshold value of AdaBoost graders is also learnt.The threshold value of AdaBoost graders is used to refuse the negative sample for not including face.
Multiple Weak Classifiers may be constructed a strong classifier.
In step s 304, the human-face detector trained based on multiple depth Quadratic Finite Element trees structure.Multiple depth are secondary
The optimal N PD that member tree includes forms the first NPD combinations.
When training terminates, obtained multiple depth Quadratic Finite Element trees can will be trained to be configured to human-face detector, wherein, it is above-mentioned
The optimal N PD that learn of each node of multiple depth Quadratic Finite Element trees form above-mentioned the first NPD combinations.
In the present embodiment, by the first positive sample including face with not including the first negative sample of face to Face datection
Device is trained, and is available for detecting the optimal N PD combinations of face, can improve the degree of accuracy of Face datection.
Fig. 4 is the flow chart of the method for detecting human face according to an exemplary embodiment three；The present embodiment utilizes this public affairs
The above method of embodiment offer is opened, to determine the first positive sample and each self-corresponding NPD of first negative sample by tabling look-up
Exemplified by it is illustrative, as shown in figure 4, comprising the following steps：
In step S401, by each self-corresponding NPD storages of the first positive sample and the first negative sample in a lookup table.
In one embodiment, the first positive sample can be completed to calculate in advance with each self-corresponding NPD of the first negative sample, and
Look-up table is stored in, for being called during training human-face detector.
In step S402, it is determined that the first positive sample including face formed with not including the first negative sample of face the
One training set.Step S402 is similar to above-mentioned step S301, will not be repeated here.
In step S403, the first positive sample and each self-corresponding NPD of the first negative sample are determined by accessing look-up table.
, can be true by accessing look-up table it needs to be determined that during each self-corresponding NPD of the first positive sample and the first negative sample
It is fixed.So, the speed of training human-face detector can be improved.
In step s 404, learn to obtain multiple depth based on the first positive sample and each self-corresponding NPD of the first negative sample
Quadratic Finite Element tree.Optimal N PD of the depth Quadratic Finite Element tree study of Confucian classics acquistion to each minor matters point；
In step S405, the human-face detector trained is built based on multiple depth Quadratic Finite Element trees.The multiple depth
The optimal N PD that Quadratic Finite Element tree includes forms the first NPD combinations.
Step S404~S405 in the present embodiment is similar with above-mentioned step S303~S304, will not be repeated here.
In the present embodiment, the first positive sample and each self-corresponding NPD of the first negative sample are determined by tabling look-up, people can be improved
The training speed of face detector.
Fig. 5 is the flow chart of the method for detecting human face according to an exemplary embodiment four；The present embodiment utilizes this public affairs
The above method of embodiment offer is provided, it is illustrative exemplified by how being trained to logistic regression grader, such as Fig. 5
It is shown, comprise the following steps：
In step S501, do not include the second positive sample including face and the second negative sample of face forms second
Training set inputs untrained logistic regression grader.
In step S502, N-dimensional characteristic vector in the anticipation function of untrained logistic regression grader is each corresponded to
Parameter be trained.Second negative sample be based on NPD detect face failure negative sample, for example, such as by recognition of face into
Sample non-face, that non-face mistake is identified as to face.
In the above-described embodiment, as shown in formula (2), (3), hθ(x) it is anticipation function, wherein, θ0、θ1、…、θnFor
Characteristic vector x1、x2、…、xnCorresponding parameter, parameter θ0、θ1、…、θnValue need by train determine.
In the present embodiment, using the negative sample based on NPD detection face failures, training effectiveness can be not only improved, and
And the defects of face is detected based on NPD can also be overcome, improve the degree of accuracy of Face datection.
In step S503, determine whether the damage function value of anticipation function reaches minimum value.If so, then perform step
504, if it is not, then performing step 502.Specifically, when it is determined that the damage function value of anticipation function reaches minimum value, step is performed
Rapid 504, when it is determined that the damage function value of anticipation function is not up to minimum value, perform step 502.
In step S504, when the value of the loss function of anticipation function reaches minimum value, determine that N-dimensional is special in anticipation function
Each self-corresponding parameter of sign vector meets preparatory condition.
In one embodiment, in training process, θ can be weighed using the damage function value of anticipation function0、θ1、…、
θnWhether preparatory condition is met.When damage function value reaches minimum value, θ0、θ1、…、θnValue be optimum parameter value, be considered as symbol
Preparatory condition is closed, otherwise, continues to train, until damage function value reaches minimum value.
In step S505, when it is determined that each self-corresponding parameter of N-dimensional characteristic vector meets preparatory condition in anticipation function,
Deconditioning logistic regression grader, the logistic regression grader trained.
In the present embodiment, logistic regression grader is trained using the negative sample based on NPD detection face failures, no
But training effectiveness can be improved, but also the defects of face is detected based on NPD can be overcome, improves the degree of accuracy of Face datection.
Fig. 6 is a kind of block diagram of human face detection device according to an exemplary embodiment, as shown in fig. 6, face is examined
Surveying device includes：
Fisrt feature extraction module 61, it is configured as being used to represent face spy according to what the human-face detector trained carried
First normalization pixel value difference NPD combinations of sign, extract the 2nd NPD combinations of image to be detected；
Processing module 62 is returned, is configured as carrying out at recurrence the 2nd NPD combinations by the human-face detector
Reason, obtains the human face region in described image to be detected；
Second feature extraction module 63, is configured as extracting the N-dimensional characteristic vector of the human face region, and N is natural number；
Computing module 64 is returned, is configured as entering the N-dimensional characteristic vector by the logistic regression grader trained
Row, which returns, to be calculated, and obtains the probability that the human face region has face；
First determining module 65, it is configured as determining the face when the probability that face be present is more than predetermined threshold value
Face in region be present.
In one embodiment, the NPD that the human-face detector trained can be carried in the first NPD combinations is suitable
Sequence；
The fisrt feature extraction module, it is also configured to according to the first NPD combinations and NPD orders,
Extract the 2nd NPD combinations；
The recurrence processing module, it is also configured to be based on NPD orders to described by the human-face detector
2nd NPD combinations carry out recurrence processing, obtain the human face region in described image to be detected.
Fig. 7 A are the block diagrams of another human face detection device according to an exemplary embodiment, as shown in Figure 7 A,
On the basis of above-mentioned embodiment illustrated in fig. 6, in one embodiment, device may also include：
Second determining module 71, it is configured to determine that the first positive sample including face and does not include the first negative sample of face
First training set of this composition；
3rd determining module 72, it is configured to determine that first positive sample and first negative sample are each self-corresponding
NPD；
Study module 73, it is configured as based on first positive sample and each self-corresponding NPD of first negative sample
Acquistion is to multiple depth Quadratic Finite Element trees；Optimal N PD of the depth Quadratic Finite Element tree study of Confucian classics acquistion to each minor matters point；
Module 74 is built, is configured as based on the Face datection trained described in the multiple depth Quadratic Finite Element tree structure
Device；The optimal N PD that the multiple depth Quadratic Finite Element tree includes forms the first NPD combinations.
Fig. 7 B are the block diagrams of another human face detection device according to an exemplary embodiment, as shown in Figure 7 B,
On the basis of above-mentioned Fig. 7 A illustrated embodiments, in one embodiment, device may also include：
Memory module 75, it is configured as storing each self-corresponding NPD of first positive sample and first negative sample
In a lookup table.
In the present embodiment, the 3rd determining module 72, it is additionally configured to by accessing described in the look-up table determination
First positive sample and each self-corresponding NPD of first negative sample.
Fig. 8 A are the block diagrams of another human face detection device according to an exemplary embodiment, as shown in Figure 8 A,
On the basis of above-mentioned embodiment illustrated in fig. 6, in one embodiment, device may also include：
Input module 81, it is configured as forming the second positive sample including face with not including the second negative sample of face
The second training set input untrained logistic regression grader；
Training module 82, be configured as in the anticipation function to the untrained logistic regression grader N-dimensional feature to
Each self-corresponding parameter is measured to be trained；Second negative sample is the negative sample that face failure is detected based on NPD；
Control module 83, it is configured as it is determined that each self-corresponding parameter of N-dimensional characteristic vector meets in the anticipation function
During preparatory condition, logistic regression grader described in deconditioning, the logistic regression grader trained is obtained.
Fig. 8 B are the block diagrams of another human face detection device according to an exemplary embodiment, as shown in Figure 8 B,
On the basis of above-mentioned Fig. 8 A illustrated embodiments, in one embodiment, device may also include：
4th determining module 84, is additionally configured to determine whether the damage function value of the anticipation function reaches minimum
Value；
5th determining module 85, it is additionally configured to reach minimum value in the value of the loss function of the anticipation function
When, determine that each self-corresponding parameter of N-dimensional characteristic vector meets the preparatory condition in the anticipation function.
On the device in above-described embodiment, wherein modules perform the concrete mode of operation in relevant this method
Embodiment in be described in detail, explanation will be not set forth in detail herein.
Fig. 9 is a kind of block diagram of human face detection device according to an exemplary embodiment.For example, device 900 can be with
It is mobile phone, computer, digital broadcast terminal, messaging devices, game console, tablet device, Medical Devices, body-building
Equipment, personal digital assistant etc..
Reference picture 9, device 900 can include following one or more assemblies：Processing component 902, memory 904, power supply
Component 906, multimedia groupware 908, audio-frequency assembly 910, the interface 912 of input/output (I/O), sensor cluster 914, and
Communication component 916.
The integrated operation of the usual control device 900 of processing component 902, such as communicated with display, call, data, phase
The operation that machine operates and record operation is associated.Treatment element 902 can refer to including one or more processors 920 to perform
Order, to complete all or part of step of above-mentioned method.In addition, processing component 902 can include one or more modules, just
Interaction between processing component 902 and other assemblies.For example, processing component 902 can include multi-media module, it is more to facilitate
Interaction between media component 908 and processing component 902.
Memory 904 is configured as storing various types of data to support the operation in equipment 900.These data are shown
Example includes the instruction of any application program or method for being operated on device 900, contact data, telephone book data, disappears
Breath, picture, video etc..Memory 904 can be by any kind of volatibility or non-volatile memory device or their group
Close and realize, as static RAM (SRAM), Electrically Erasable Read Only Memory (EEPROM) are erasable to compile
Journey read-only storage (EPROM), programmable read only memory (PROM), read-only storage (ROM), magnetic memory, flash
Device, disk or CD.
Electric power assembly 906 provides electric power for the various assemblies of device 900.Electric power assembly 906 can include power management system
System, one or more power supplys, and other components associated with generating, managing and distributing electric power for device 900.
Multimedia groupware 908 is included in the screen of one output interface of offer between described device 900 and user.One
In a little embodiments, screen can include liquid crystal display (LCD) and touch panel (TP).If screen includes touch panel, screen
Curtain may be implemented as touch-screen, to receive the input signal from user.Touch panel includes one or more touch sensings
Device is with the gesture on sensing touch, slip and touch panel.The touch sensor can not only sensing touch or sliding action
Border, but also detect and touched or the related duration and pressure of slide with described.In certain embodiments, more matchmakers
Body component 908 includes a front camera and/or rear camera.When equipment 900 is in operator scheme, such as screening-mode or
During video mode, front camera and/or rear camera can receive outside multi-medium data.Each front camera and
Rear camera can be a fixed optical lens system or have focusing and optical zoom capabilities.
Audio-frequency assembly 910 is configured as output and/or input audio signal.For example, audio-frequency assembly 910 includes a Mike
Wind (MIC), when device 900 is in operator scheme, during such as call model, logging mode and speech recognition mode, microphone by with
It is set to reception external audio signal.The audio signal received can be further stored in memory 904 or via communication set
Part 916 is sent.In certain embodiments, audio-frequency assembly 910 also includes a loudspeaker, for exports audio signal.
I/O interfaces 912 provide interface between processing component 902 and peripheral interface module, and above-mentioned peripheral interface module can
To be keyboard, click wheel, button etc..These buttons may include but be not limited to：Home button, volume button, start button and lock
Determine button.
Sensor cluster 914 includes one or more sensors, and the state for providing various aspects for device 900 is commented
Estimate.For example, sensor cluster 914 can detect opening/closed mode of equipment 900, and the relative positioning of component, for example, it is described
Component is the display and keypad of device 900, and sensor cluster 914 can be with 900 1 components of detection means 900 or device
Position change, the existence or non-existence that user contacts with device 900, the orientation of device 900 or acceleration/deceleration and device 900
Temperature change.Sensor cluster 914 can include proximity transducer, be configured to detect in no any physical contact
The presence of neighbouring object.Sensor cluster 914 can also include optical sensor, such as CMOS or ccd image sensor, for into
As being used in application.In certain embodiments, the sensor cluster 914 can also include acceleration transducer, gyro sensors
Device, Magnetic Sensor, pressure sensor or temperature sensor.
Communication component 916 is configured to facilitate the communication of wired or wireless way between device 900 and other equipment.Device
900 can access the wireless network based on communication standard, such as WiFi, 2G or 3G, or combinations thereof.In an exemplary implementation
In example, communication component 916 receives broadcast singal or broadcast related information from external broadcasting management system via broadcast channel.
In one exemplary embodiment, the communication component 916 also includes near-field communication (NFC) module, to promote junction service.Example
Such as, in NFC module radio frequency identification (RFID) technology can be based on, Infrared Data Association (IrDA) technology, ultra wide band (UWB) technology,
Bluetooth (BT) technology and other technologies are realized.
In the exemplary embodiment, device 900 can be believed by one or more application specific integrated circuits (ASIC), numeral
Number processor (DSP), digital signal processing appts (DSPD), PLD (PLD), field programmable gate array
(FPGA), controller, microcontroller, microprocessor or other electronic components are realized, for performing the above method.
In the exemplary embodiment, a kind of non-transitorycomputer readable storage medium including instructing, example are additionally provided
Such as include the memory 904 of instruction, above-mentioned instruction can be performed to complete the above method by the processor 920 of device 900.For example,
The non-transitorycomputer readable storage medium can be ROM, random access memory (RAM), CD-ROM, tape, floppy disk
With optical data storage devices etc..
Those skilled in the art will readily occur to the disclosure its after considering specification and putting into practice disclosure disclosed herein
Its embodiment.The application is intended to any modification, purposes or the adaptations of the disclosure, these modifications, purposes or
Person's adaptations follow the general principle of the disclosure and including the undocumented common knowledges in the art of the disclosure
Or conventional techniques.Description and embodiments are considered only as exemplary, and the true scope of the disclosure and spirit are by following
Claim is pointed out.
It should be appreciated that the precision architecture that the disclosure is not limited to be described above and is shown in the drawings, and
And various modifications and changes can be being carried out without departing from the scope.The scope of the present disclosure is only limited by appended claim.
Claims (14)
1. a kind of method for detecting human face, it is characterised in that methods described includes：
The the first normalization pixel value difference NPD combinations for being used to represent face characteristic carried according to the human-face detector trained,
Extract the 2nd NPD combinations of image to be detected；
Recurrence processing is carried out to the 2nd NPD combinations by the human-face detector, obtains the people in described image to be detected
Face region；
The N-dimensional characteristic vector of the human face region is extracted, N is natural number；
Recurrence calculating is carried out to the N-dimensional characteristic vector by the logistic regression grader trained, obtains the human face region
The probability of face be present；
Determine face be present in the human face region when the probability that face be present is more than predetermined threshold value.
2. according to the method for claim 1, it is characterised in that the human-face detector trained carries described first
NPD orders in NPD combinations；
What the human-face detector that the basis has been trained carried is used to represent the first NPD combinations of face characteristic, extracts to be detected
The 2nd NPD combinations of image, including：
According to the first NPD combinations and NPD orders, the 2nd NPD combinations are extracted；
It is described that recurrence processing is carried out to the 2nd NPD combinations by the human-face detector, obtain in described image to be detected
Human face region, including：
NPD orders are based on by the human-face detector recurrence processing is carried out to the 2nd NPD combinations, obtain described treat
Human face region in detection image.
3. according to the method for claim 1, it is characterised in that methods described also includes：
It is determined that the first training set that the first positive sample including face forms with not including the first negative sample of face；
Determine first positive sample and each self-corresponding NPD of first negative sample；
Learn to obtain multiple depth Quadratic Finite Element trees based on first positive sample and each self-corresponding NPD of first negative sample；
Optimal N PD of the depth Quadratic Finite Element tree study of Confucian classics acquistion to each minor matters point；
Based on the human-face detector trained described in the multiple depth Quadratic Finite Element tree structure；In the multiple depth Quadratic Finite Element tree
Including optimal N PD form the first NPD combination.
4. according to the method for claim 3, it is characterised in that methods described also includes：
By each self-corresponding NPD storages of first positive sample and first negative sample in a lookup table；
Determination first positive sample and each self-corresponding NPD of first negative sample, including：
First positive sample and each self-corresponding NPD of first negative sample are determined by accessing the look-up table.
5. according to the method for claim 1, it is characterised in that methods described also includes：
The second positive sample including face is not trained with not including the second training set input that the second negative sample of face forms
Logistic regression grader；
Each self-corresponding parameter of N-dimensional characteristic vector in the anticipation function of the untrained logistic regression grader is instructed
Practice；Second negative sample is the negative sample that face failure is detected based on NPD；
When it is determined that each self-corresponding parameter of N-dimensional characteristic vector meets preparatory condition in the anticipation function, described in deconditioning
Logistic regression grader, obtain the logistic regression grader trained.
6. according to the method for claim 5, it is characterised in that described to determine that N-dimensional characteristic vector is each in the anticipation function
Self-corresponding parameter meets preparatory condition, including：
Determine whether the damage function value of the anticipation function reaches minimum value；
When the value of the loss function of the anticipation function reaches minimum value, determine that N-dimensional characteristic vector is each in the anticipation function
Self-corresponding parameter meets the preparatory condition.
7. a kind of human face detection device, it is characterised in that described device includes：
Fisrt feature extraction module, it is configured as be used to represent face characteristic the carried according to the human-face detector trained
One normalization pixel value difference NPD combinations, extract the 2nd NPD combinations of image to be detected；
Processing module is returned, is configured as carrying out recurrence processing to the 2nd NPD combinations by the human-face detector, obtains
Human face region in described image to be detected；
Second feature extraction module, is configured as extracting the N-dimensional characteristic vector of the human face region, and N is natural number；
Computing module is returned, is configured as returning the N-dimensional characteristic vector by the logistic regression grader trained
Calculate, obtain the probability that the human face region has face；
First determining module, it is configured as determining in the human face region when the probability that face be present is more than predetermined threshold value
Face be present.
8. device according to claim 7, it is characterised in that the human-face detector trained carries described first
NPD orders in NPD combinations；
The fisrt feature extraction module, it is additionally configured to, according to the first NPD combinations and NPD orders, extract institute
State the 2nd NPD combinations；
The recurrence processing module, it is additionally configured to be based on NPD orders to the 2nd NPD by the human-face detector
Combination carries out recurrence processing, obtains the human face region in described image to be detected.
9. device according to claim 7, it is characterised in that described device, in addition to：
Second determining module, it is configured to determine that the first positive sample including face forms with not including the first negative sample of face
The first training set；
3rd determining module, it is configured to determine that first positive sample and each self-corresponding NPD of first negative sample；
Study module, it is configured as learning to obtain with each self-corresponding NPD of first negative sample based on first positive sample
Multiple depth Quadratic Finite Element trees；Optimal N PD of the depth Quadratic Finite Element tree study of Confucian classics acquistion to each minor matters point；
Module is built, is configured as based on the human-face detector trained described in the multiple depth Quadratic Finite Element tree structure；It is described
The optimal N PD that multiple depth Quadratic Finite Element trees include forms the first NPD combinations.
10. device according to claim 9, it is characterised in that described device, in addition to：
Memory module, it is configured as first positive sample and each self-corresponding NPD of first negative sample being stored in lookup
In table；
3rd determining module, it is additionally configured to determine first positive sample and described first by accessing the look-up table
Each self-corresponding NPD of negative sample.
11. device according to claim 7, it is characterised in that described device, in addition to：
Input module, it is configured as form the second positive sample including face with not including the second negative sample of face second
Training set inputs untrained logistic regression grader；
Training module, it is configured as in the anticipation function to the untrained logistic regression grader N-dimensional characteristic vector each
Corresponding parameter is trained；Second negative sample is the negative sample that face failure is detected based on NPD；
Control module, it is configured as it is determined that each self-corresponding parameter of N-dimensional characteristic vector meets default bar in the anticipation function
During part, logistic regression grader described in deconditioning, the logistic regression grader trained is obtained.
12. device according to claim 11, it is characterised in that described device, in addition to：
4th determining module, is additionally configured to determine whether the damage function value of the anticipation function reaches minimum value；
5th determining module, it is additionally configured to when the value of the loss function of the anticipation function reaches minimum value, it is determined that
Each self-corresponding parameter of N-dimensional characteristic vector meets the preparatory condition in the anticipation function.
13. a kind of human face detection device, it is characterised in that described device includes：
Processor；
For storing the memory of processor-executable instruction；
Wherein, the processor is configured as：
The the first normalization pixel value difference NPD combinations for being used to represent face characteristic carried according to the human-face detector trained,
Extract the 2nd NPD combinations of image to be detected；
Recurrence processing is carried out to the 2nd NPD combinations by the human-face detector, obtains the people in described image to be detected
Face region；
The N-dimensional characteristic vector of the human face region is extracted, N is natural number；
Recurrence calculating is carried out to the N-dimensional characteristic vector by the logistic regression grader trained, obtains the human face region
The probability of face be present；
Determine face be present in the human face region when the probability that face be present is more than predetermined threshold value.
14. a kind of computer-readable recording medium, is stored thereon with computer program, it is characterised in that the computer program
Following steps are realized when being executed by processor：
The the first normalization pixel value difference NPD combinations for being used to represent face characteristic carried according to the human-face detector trained,
Extract the 2nd NPD combinations of image to be detected；
Recurrence processing is carried out to the 2nd NPD combinations by the human-face detector, obtains the people in described image to be detected
Face region；
The N-dimensional characteristic vector of the human face region is extracted, N is natural number；
Recurrence calculating is carried out to the N-dimensional characteristic vector by the logistic regression grader trained, obtains the human face region
The probability of face be present；
Determine face be present in the human face region when the probability that face be present is more than predetermined threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710476106.2A CN107341457A (en) | 2017-06-21 | 2017-06-21 | Method for detecting human face and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710476106.2A CN107341457A (en) | 2017-06-21 | 2017-06-21 | Method for detecting human face and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107341457A true CN107341457A (en) | 2017-11-10 |
Family
ID=60221484
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710476106.2A Pending CN107341457A (en) | 2017-06-21 | 2017-06-21 | Method for detecting human face and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107341457A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107909034A (en) * | 2017-11-15 | 2018-04-13 | 清华大学深圳研究生院 | A kind of method for detecting human face, device and computer-readable recording medium |
| CN107992853A (en) * | 2017-12-22 | 2018-05-04 | 深圳市友信长丰科技有限公司 | Eye detection method, device, computer equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104915649A (en) * | 2015-06-04 | 2015-09-16 | 南京理工大学 | Living person detection method applied to face recognition |
| CN105550642A (en) * | 2015-12-08 | 2016-05-04 | 康佳集团股份有限公司 | Gender identification method and system based on multi-scale linear difference characteristic low-rank expression |
| CN106503617A (en) * | 2016-09-21 | 2017-03-15 | 北京小米移动软件有限公司 | Model training method and device |
| CN106845385A (en) * | 2017-01-17 | 2017-06-13 | 腾讯科技（上海）有限公司 | The method and apparatus of video frequency object tracking |
-
2017
- 2017-06-21 CN CN201710476106.2A patent/CN107341457A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104915649A (en) * | 2015-06-04 | 2015-09-16 | 南京理工大学 | Living person detection method applied to face recognition |
| CN105550642A (en) * | 2015-12-08 | 2016-05-04 | 康佳集团股份有限公司 | Gender identification method and system based on multi-scale linear difference characteristic low-rank expression |
| CN106503617A (en) * | 2016-09-21 | 2017-03-15 | 北京小米移动软件有限公司 | Model training method and device |
| CN106845385A (en) * | 2017-01-17 | 2017-06-13 | 腾讯科技（上海）有限公司 | The method and apparatus of video frequency object tracking |
Non-Patent Citations (3)
| Title |
|---|
| JIANGUO LI.ETC: "Learning SURF Cascade for Fast and Accurate Object Detection", 《2013 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION》 * |
| SHENGCAI LI.ETC: "A Fast and Accurate Unconstrained Face Detector", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 皇望: "基于逻辑回归改进的随机行走空谱融合分类", 《信息技术》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107909034A (en) * | 2017-11-15 | 2018-04-13 | 清华大学深圳研究生院 | A kind of method for detecting human face, device and computer-readable recording medium |
| CN107992853A (en) * | 2017-12-22 | 2018-05-04 | 深圳市友信长丰科技有限公司 | Eye detection method, device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109697416B (en) | Video data processing method and related device | |
| CN106295511B (en) | Face tracking method and device | |
| CN106548145A (en) | Image-recognizing method and device | |
| CN104700353B (en) | Image filters generation method and device | |
| CN107133576A (en) | Age of user recognition methods and device | |
| CN112036331B (en) | Living body detection model training method, device, equipment and storage medium | |
| CN110602527A (en) | Video processing method, device and storage medium | |
| CN107239535A (en) | Similar pictures search method and device | |
| CN107193983A (en) | Image search method and device | |
| CN106709399A (en) | Fingerprint identification method and device | |
| CN105426857A (en) | Training method and device of face recognition model | |
| CN107527053A (en) | Object detection method and device | |
| CN105844206A (en) | Identity authentication method and identity authentication device | |
| CN107832741A (en) | The method, apparatus and computer-readable recording medium of facial modeling | |
| CN106295515A (en) | Determine the method and device of human face region in image | |
| CN107545248A (en) | Biological characteristic biopsy method, device, equipment and storage medium | |
| WO2021047069A1 (en) | Face recognition method and electronic terminal device | |
| CN107766820A (en) | Image classification method and device | |
| CN106682736A (en) | Image identification method and apparatus | |
| CN105654033A (en) | Face image verification method and device | |
| CN106169075A (en) | Auth method and device | |
| CN106228556A (en) | Image quality analysis method and device | |
| CN106548468A (en) | The method of discrimination and device of image definition | |
| CN107886070A (en) | Verification method, device and the equipment of facial image | |
| CN107463903A (en) | Face key independent positioning method and device |
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 |
Application publication date: 20171110 |
|
| RJ01 | Rejection of invention patent application after publication |