CN107909034A - A kind of method for detecting human face, device and computer-readable recording medium - Google Patents
A kind of method for detecting human face, device and computer-readable recording medium Download PDFInfo
- Publication number
- CN107909034A CN107909034A CN201711132667.7A CN201711132667A CN107909034A CN 107909034 A CN107909034 A CN 107909034A CN 201711132667 A CN201711132667 A CN 201711132667A CN 107909034 A CN107909034 A CN 107909034A
- Authority
- CN
- China
- Prior art keywords
- image
- face
- sample
- positive
- negative
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Abstract
The invention discloses method for detecting human face, device and computer-readable recording medium, method includes model training and Face datection；Training process is to use positive and negative samples image and its standard pixel difference feature, to build the secondary tree of depth, then by the secondary tree composition random forest of multiple depth carries out Face datection tracking.All tested after a random forest is often obtained in training process, whether see can be correctly detecting face, and the image update of detection mistake is continued constantly training study into corresponding training set, until obtaining detection result meets expected random forest.The feature of image to be detected is input to random forest during detection, is judged respectively by each of which secondary tree, then the judging result of all secondary trees is counted, facial image is determine whether according to statistical result.Have the advantages that to obtain the detection model of degree of precision using less training sample, be especially suitable for the occasion to there is the face blocked to be detected.
Description
Technical field
The present invention relates to computer vision and image processing field, and the people based on random forest is utilized more particularly, to a kind of
Face detector carrys out in detection image whether to include the method for detecting human face of face.
Background technology
With the fast development of image processing techniques and machine learning method, the application based on recognition of face is more and more wider
It is general, such as the field such as authentication, intelligent transportation, security protection.First step of the Face datection as recognition of face, it is accurate
True property is most important.
Human face detection tech is the feature based on facial image, to the image or video flowing of input, is judged whether
The process of face.First, a kind of characteristics of image is selected, to carrying out feature extraction for trained face sample image；Secondly, base
In the sample image feature extracted, training grader, for judging to whether there is face in image, to be further based on people
The technology application of face detection is prepared.
Traditional method for detecting human face, usually utilizes and is based on Haar features, HOG features, LBP features or assemblage characteristic
Deng training grader as Expressive Features.And in some special occasions, there is certain obstacle, background are multiple for Face datection
Miscellaneous, face form difference is larger or face wears ornaments etc. and causes face partial occlusion, to design a kind of pervasive Face datection
Algorithm is more difficult.In addition to reach preferable detection result, it is necessary to substantial amounts of face picture is trained algorithm, and it is special
It is unpractical to determine the Face datection in field to obtain large-scale face picture collection.
The disclosure of background above technology contents is only used for inventive concept and the technical solution that auxiliary understands the present invention, it is not
The prior art of present patent application is necessarily belonged to, shows the applying date of the above in present patent application in no tangible proof
In the case of disclosed, above-mentioned background technology should not be taken to the novelty and creativeness of evaluation the application.
The content of the invention
It is a primary object of the present invention to propose a kind of method for detecting human face, by face it is sample registered and constantly update
Method obtains the face sample image under specific occasion, and constructs the random forest based on the secondary tree of depth as Face datection
Device detects the face blocked.Compared with conventional method, can utilize less sample image realize have accuracy of detection it is high,
Generalization ability is strong, and can effectively reduce the human-face detector of over-fitting.
The technical solution that the present invention is proposed for the above-mentioned purpose is as follows：
A kind of method for detecting human face, including the training step of Face datection model and Face datection step：
Wherein, the training step comprises the following steps：
A1, obtain multiple facial images and inhuman face image, to establish positive and negative samples training set respectively；
A2, all images to positive and negative samples training set carry out the extraction of standard pixel difference feature respectively, obtain each figure
As respective feature vector；To between all standard pixel difference feature normalizations processing in all feature vectors to 0~255；
A3, randomly select several positive sample image sum number negative sample images from positive and negative samples training set respectively, and divides
Several features are not randomly selected from the feature vector of each positive sample image and each negative sample image, obtain each positive sample image
Positive sample feature vector and each negative sample image negative sample feature vector；
A4, based on the step A3 positive sample images extracted and its positive sample feature vector, and negative sample image and its
Negative sample feature vector, build the secondary tree of a depth；
A5, repeat step A3 and A4, obtain the secondary tree of multiple and different depth, form random forest；
A6, using obtained random forest be made whether test image to include the detection of face, if containing in test image
Have face but and be not detected at, then the test image is added in positive sample training set, to update positive sample training set；If
Face is free of in test image but detects face, then the test image is added in negative sample training set, to update negative sample
This training set；
A7, using renewal the continuous repeat step A2 to A6 of positive and negative samples training set until obtain detection result meet it is pre-
If it is required that random forest；
Face datection step includes：
B1, receive image to be detected, by the secondary tree of each depth of random forest come respectively to described image to be detected
It is made whether the detection for including face；
B2, the judging result to the secondary tree of all depth count, if judging to include face in image to be detected
The ratio of the secondary tree of depth is more than a preset value, then the judging result of the random forest is to contain someone in described image to be detected
Face.
Above-mentioned method for detecting human face provided by the invention, especially has advantage for the Face datection of special occasions, special
Occasion for example has the Face datection blocked, this kind of facial image is often difficult to collect, that is, is difficult to obtain plurality when training
The training set of amount, and the method for the present invention can use small number of training set in advance, be utilized in training and test process
The image for detection mistake occur updates training set as new samples, so not only causes training set in the training process increasingly
Greatly, more there is the wrong image of detection it is essential that training may all use when being tested in preceding a training process each time, with
This improves the accuracy of detection, so that the higher Face datection model (random forest) of accuracy of detection may finally be obtained.
Another embodiment of the present invention also proposed a kind of human face detection device, including image acquisition units, storage list
Member and model training program and Face datection program；
Described image collecting unit is used to obtain different facial images；The storage unit storage facial image and collection
The inhuman face image arrived, wherein, facial image and inhuman face image are divided into positive and negative samples training set；
The model training program is used to train to obtain a Face datection model based on positive and negative samples training set；
The Face datection program is used to carry out feature extraction to image to be detected of input, and is input to Face datection mould
The detection for including face is made whether in type；
Wherein, the model training program includes：
Feature extraction program, standard pixel difference feature is carried out for all images to positive and negative samples training set respectively
Extraction, obtains the respective feature vector of each image；At all standard pixel difference feature normalizations in all feature vectors
Manage between 0~255；
Stochastical sampling program, for randomly selecting several positive sample image sum numbers from positive and negative samples training set respectively
Negative sample image, and several features are randomly selected from the feature vector of each positive sample image and each negative sample image respectively, obtain
To the positive sample feature vector of each positive sample image and the negative sample feature vector of each negative sample image；
Secondary tree construction procedures, for the positive sample image and its positive sample feature being drawn into according to stochastical sampling program
Vector, and negative sample image and its negative sample feature vector, build the secondary tree of a depth；The secondary tree structure is performed a plurality of times
Program, obtains the secondary tree of multiple depth, forms random forest, i.e. Face datection model；
Test program, is made whether test image using obtained random forest to include the detection of face, if test chart
As in containing face but and be not detected at, then the test image is added in positive sample training set, with update positive sample instruction
Practice collection；If without face but detecting face in test image, which is added in negative sample training set, with more
New negative sample training set.
On the other hand, the invention also provides a kind of computer-readable recording medium, a computer program is stored thereon with,
The step of computer program realizes foregoing method for detecting human face when being executed by processor.
Brief description of the drawings
Fig. 1 is the method for detecting human face flow chart of one embodiment of the invention；
Fig. 2 is an exemplary secondary tree schematic diagram of depth constructed in training step of the present invention.
Embodiment
The invention will be further described with specific embodiment below in conjunction with the accompanying drawings.
The embodiment of the present invention proposes a kind of method for detecting human face, the face available for some special occasions
Detection, for example can be directed to has the facial image blocked to carry out Face datection tracking.The method for detecting human face includes Face datection
The training step and Face datection step of model, with reference to figure 1：
The training step comprises the following steps A1 to A7：
A1, obtain multiple facial images and inhuman face image, to establish positive and negative samples training set respectively.Special occasions example
If any the facial image blocked, if collecting ready-made image deficiency, can also voluntarily shoot, such as, set using camera center as
Origin, one suitable region of delimitation is face sample collection region, and is marked with red block.Secondly, user puts face
Put to indicia framing position, move back and forth and make it that indicia framing size and face size are basically identical, and rotate face by a small margin, collect
Different angle facial image increases the diversity of sample.Negative sample image is the image without face.
A2, all images to positive and negative samples training set carry out the extraction of standard pixel difference feature respectively, obtain each figure
As respective feature vector；To between all standard pixel difference feature normalizations processing in all feature vectors to 0~255.
Extracted in the present invention using standard pixel difference as a kind of characteristics of image, standard pixel difference feature can be carried with following formula (1)
Take calculating：
Wherein, p, q are respectively the pixel value of two pixels in same image.Above formula (1) is not applied in p=q=0,
Therefore f (0,0)=0 is defined, in addition, working asWhen represent that this two do not have difference between pixel p and q.Mark
What quasi- pixel difference was weighed is the relative difference between two pixels, its symbol (+or -) represents the ordering relationship of two pixels.
With pixel value absolute difference | p-q | compared with, standard pixel difference has scale invariability, this with the present invention for have the face blocked into
The algorithm of row detection is adapted.
Preferably, in order to enable feature to calculate simple and perform well in the threshold learning of the secondary tree of depth, the present invention
Linear transformation is carried out to the standard pixel difference feature of image, by the above-mentioned computational methods of pixel difference featureBecome
It is changed toWhen p and q are all higher than 0, p=q=0, the standard pixel difference between p and q is defined as 0.
A3, randomly select several positive sample image sum number negative sample images from positive and negative samples training set respectively, and divides
Several features are not randomly selected from the feature vector of each positive sample image and each negative sample image, obtain each positive sample image
Positive sample feature vector and each negative sample image negative sample feature vector.That is, instructed from the positive sample that size is N
Practice concentration and randomly select n (0 ＜ n ＜ N) facial images, while it is non-face that (S-n) is randomly selected from negative sample training set
Image, obtains S positive and negative sample images altogether；Then, to each sample image, and respectively from the feature vector that its size is M
In randomly select m (0 ＜ m ＜ M) a feature (follow-up feature each means standard pixel difference feature).
A4, based on the step A3 positive sample images extracted and its positive sample feature vector, and negative sample image and its
Negative sample feature vector, build the secondary tree of a depth；
A5, repeat step A3 and A4, obtain the secondary tree of multiple and different depth, form random forest；One is obtained
Random forest Face datection model.
A6, using obtained random forest be made whether test image to include the detection of face, if containing in test image
Have face but and be not detected at, then the test image is added in positive sample training set, to update positive sample training set；If
Face is free of in test image but detects face, then the test image is added in negative sample training set, to update negative sample
This training set；
A7, using renewal the continuous repeat step A2 to A6 of positive and negative samples training set until obtain detection result meet it is pre-
If it is required that random forest.If for example, there are face in test image, and detection model fails to examine side and, user can be manual
Mark face location simultaneously goes renewal positive sample training set, continuous repetitive learning process, until Face datection tracking knot with the image
Untill fruit reaches satisfied.
Face datection step includes：
B1, receive image to be detected, by the secondary tree of each depth of random forest come respectively to described image to be detected
It is made whether the detection for including face.Specifically, image to be detected of input, carries out feature extraction in advance, extracts very much
A standard pixel difference feature, forms a feature vector to be detected, and then to each secondary tree input, this is to be checked respectively
The feature vector of survey, makes each secondary tree separately be made whether to include the detection of face, and obtains a testing result
(detect face or be not detected by face).
B2, the judging result to the secondary tree of all depth count, if judging to include face in image to be detected
The ratio of the secondary tree of depth is more than a preset value, then the judging result of the random forest is to contain someone in described image to be detected
Face.
In one embodiment, the secondary tree of a depth is built in step A4 to specifically include：First, traveled through by the method for exhaustion
All sample characteristics in positive and negative samples feature vector, assigned to by calculating left and right branch sample characteristics weight it is cumulative,
To obtain the error in classification of each sample characteristics, the pixel value corresponding to the error in classification of minimum is selected to classify as root node
Threshold value；Secondly, reuse the method for exhaustion and travel through respectively and assign to all sample characteristics of left and right two, obtain the optimal of the second layer
Classification thresholds；The rest may be inferred, stops division when the secondary tree of depth reaches the threshold condition of setting, obtains the secondary tree of a depth.
In a preferred scheme, step A4 can construct the secondary tree of a depth by the following method：
A41, based on the step A3 positive sample images extracted and its positive sample feature vector, and negative sample image and its
Negative sample feature vector, establishes the initialization weight histogram of positive and negative samples feature respectively, and respectively obtains positive and negative samples spy
The sample matrix number by characteristic value classification of sign.Weight histogram includes 256 columns, numbering 0,1,2 ..., 255, each
The value (relative altitude) of column adds up for the weight that normalization characteristic value is equal to all sample characteristics that the column is numbered, and (sample characteristics are weighed
1/S is initialized as again, S is the sum of positive and negative sample characteristics).Sample matrix number is 1 × 256 matrix, some member in matrix
The sum of the sample number of the value of element by characteristic value for some value, such as first element are characterized the total a of the sample characteristics that value is 0
Number, adding the sample characteristics that characteristic value is 0 has 8, then first element of matrix is 8.
A42, two layers of traversal (realization of two layers of traversal can be the nested circulation of ectonexine by algorithm to realize) positive sample are special
All sample characteristics in vector sum negative sample feature vector are levied, when outer layer each time travels through, the root node right side is assigned in initialization
The sample number of branch, and pass through the sample number of the sample number matrix update right branch when internal layer travels through；At the same time according to described
Weight histogram is initialized, error in classification and positive sample ratio fl, fr of left and right branch are calculated using weighted mean square error；When
When error in classification is less than setting value, using the bound of threshold value of the inside and outside layer period as each node at this time, and, root
The positive sample ratio fit of the leaf node for Face datection is calculated according to positive sample the ratio fl and fr of left and right branch.
By performing step an A41 and A42, the decision threshold condition of the root node of secondary tree is can obtain, is [θ11,
θ12], as shown in Figure 2.Then constantly circulation step A41 and A42, to carry out the downward division of secondary tree, obtains downward every
Decision threshold condition [the θ of one layer of node21,θ22]、[θ31,θ32] etc., until stopping division when reaching preset stopping condition, obtain
To the secondary tree of a depth, such as the secondary tree shown in Fig. 2.By the way that step A3 and A4 is performed a plurality of times, multiple secondary trees, group can obtain
Into the random forest for detecting face.
During the secondary tree of depth is built, such splitting rule af can be based on2+ bf+c=t, wherein f are represented
Standard pixel difference feature, t represent division threshold value, and a, b, c are the constants on feature f.This is relative to original linear division f ＜
For t, secondary split strategy considers the single order and second order information of f at the same time.In addition, the present invention can learn three kinds of target knots
Structure：
Wherein, θ represents the value of some standard pixel difference feature, θ1＜ 0, θ2＞ 0；Formula (2) represents target pixel value p ratio pictures
Plain value q is small, and formula (3) represents that target pixel value p is bigger than pixel value q；If formula (4) is invalid, represent pixel p and pixel q it
Between have obvious edge or difference.
After random forest has been obtained, when there is a unknown images input to be detected, multiple spies for extracting the image
Levy f1、f2、f3Deng obtaining a feature vector [f1,f2,f3...], the secondary tree of each depth allowed in random forest all carries out
Judge, by taking Fig. 2 as an example, judgement is proceeded by from the root node of secondary tree, until leaf node, so as to obtain the judgement of secondary tree
As a result.When judging that as soon as input picture is more than setting value (be greater than setting 0.5) for the ratio of the secondary tree of face, recognize
Image for input is facial image.
Another embodiment of the present invention proposes a kind of human face detection device, including image acquisition units, storage unit and
Model training program and Face datection program；Image acquisition units are used to obtain different facial images, and example is different as the aforementioned
The face shooting of angle；Storage unit storage facial image and the inhuman face image being collected into, wherein, facial image and non-
Facial image is divided into positive and negative samples training set.The model training program is used to train based on positive and negative samples training set
Obtain a Face datection model；The Face datection program is used to carry out feature extraction to image to be detected of input, and inputs
The detection for including face is made whether into Face datection model.
Wherein, the model training program includes：
Feature extraction program, standard pixel difference feature is carried out for all images to positive and negative samples training set respectively
Extraction, obtains the respective feature vector of each image；At all standard pixel difference feature normalizations in all feature vectors
Manage between 0~255；
Stochastical sampling program, for randomly selecting several positive sample image sum numbers from positive and negative samples training set respectively
Negative sample image, and several features are randomly selected from the feature vector of each positive sample image and each negative sample image respectively, obtain
To the positive sample feature vector of each positive sample image and the negative sample feature vector of each negative sample image；
Secondary tree construction procedures, for the positive sample image and its positive sample feature being drawn into according to stochastical sampling program
Vector, and negative sample image and its negative sample feature vector, build the secondary tree of a depth；The secondary tree structure is performed a plurality of times
Program, obtains the secondary tree of multiple depth, forms random forest, i.e. Face datection model；
Test program, is made whether test image using obtained random forest to include the detection of face, if test chart
As in containing face but and be not detected at, then the test image is added in positive sample training set, with update positive sample instruction
Practice collection；If without face but detecting face in test image, which is added in negative sample training set, with more
New negative sample training set.
The Face datection program includes statistics program and decision procedure, and statistics program is used for each in random forest
The secondary tree of depth counts the judging result of image to be detected, to draw the depth for judging that face is included in image to be detected
The ratio of secondary tree；Whether decision procedure is used for by the ratio compared with preset value, and wrapped according to comparative result output
Testing result containing face.
The human face detection device that above-described embodiment provides can be used for the Face datection of the different different occasions of equipment progress, if
The face for having often and blocking, the then facial image blocked when being trained using having are for example detected dedicated for special occasions
As positive sample training set.
Another embodiment of the present invention additionally provides a kind of computer-readable recording medium, is stored thereon with a computer journey
Sequence, can be realized when which is executed by processor in previous embodiment be related to method for detecting human face the step of.
Above content is that a further detailed description of the present invention in conjunction with specific preferred embodiments, it is impossible to is assert
The specific implementation of the present invention is confined to these explanations.For those skilled in the art, do not taking off
On the premise of from present inventive concept, some equivalent substitutes or obvious modification can also be made, and performance or purposes are identical, all should
When being considered as belonging to protection scope of the present invention.
Claims (9)
1. a kind of method for detecting human face, including the training step of Face datection model and Face datection step：
Wherein, the training step comprises the following steps：
A1, obtain multiple facial images and inhuman face image, to establish positive and negative samples training set respectively；
A2, all images to positive and negative samples training set carry out the extraction of standard pixel difference feature respectively, and it is each to obtain each image
From feature vector；To between all standard pixel difference feature normalizations processing in all feature vectors to 0~255；
A3, randomly select several positive sample image sum numbers negative sample images from positive and negative samples training set respectively, and respectively from
Several features are randomly selected in the feature vector of each positive sample image and each negative sample image, are obtaining each positive sample image just
The negative sample feature vector of sampling feature vectors and each negative sample image；
A4, based on the step A3 positive sample images extracted and its positive sample feature vector, and negative sample image and its negative sample
Eigen vector, builds the secondary tree of a depth；
A5, repeat step A3 and A4, obtain the secondary tree of multiple and different depth, form random forest；
A6, using obtained random forest be made whether test image to include the detection of face, if containing someone in test image
Face but and be not detected at, then the test image is added in positive sample training set, to update positive sample training set；If test
Face is free of in image but detects face, then the test image is added in negative sample training set, to update negative sample instruction
Practice collection；
A7, using renewal the continuous repeat step A2 to A6 of positive and negative samples training set until obtain detection result meet it is default will
The random forest asked；
Face datection step includes：
B1, receive image to be detected, and described image to be detected is carried out respectively by the secondary tree of each depth of random forest
Whether the detection of face is included；
B2, the judging result to the secondary tree of all depth count, if judging to include the depth of face in image to be detected
The ratio of secondary tree is more than a preset value, then the judging result of the random forest is to contain face in described image to be detected.
2. method for detecting human face as claimed in claim 1, it is characterised in that：Step A2 Plays pixel difference featuresP, q is respectively the pixel value of two pixels in same image.
3. method for detecting human face as claimed in claim 1, it is characterised in that：The specific bag of the secondary tree of a depth is built in step A4
Include：
First, all sample characteristics in positive and negative samples feature vector are traveled through by the method for exhaustion, left and right point is assigned to by calculating
The sample characteristics weight of branch adds up, and to obtain the error in classification of each sample characteristics, selects the error in classification institute of minimum right
The pixel value answered is as root node classification thresholds；
Secondly, reuse the method for exhaustion and travel through respectively and assign to all sample characteristics of left and right two, obtain the most optimal sorting of the second layer
Class threshold value；The rest may be inferred, stops division when the secondary tree of depth reaches the threshold condition of setting, obtains the secondary tree of a depth.
4. method for detecting human face as claimed in claim 1, it is characterised in that：The specific bag of the secondary tree of a depth is built in step A4
Include：
A41, based on the step A3 positive sample images extracted and its positive sample feature vector, and negative sample image and its negative sample
Eigen vector, establishes the initialization weight histogram of positive and negative samples feature, and respectively obtain positive and negative samples feature respectively
The sample matrix number classified by characteristic value；
All sample characteristics in A42, two layers of traversal positive sample feature vector and negative sample feature vector, in outer layer time each time
Last, the sample number of root node right branch is assigned in initialization, and is divided when internal layer is traveled through by the way that the sample number matrix update is right
The sample number of branch；At the same time according to the initialization weight histogram, error in classification and left and right point are calculated using weighted mean square error
Positive sample ratio fl, fr of branch；When error in classification is less than setting value, using inside and outside layer period at this time as each node
Threshold value bound, and, according to positive sample the ratio fl and fr of left and right branch calculate for Face datection leaf node
Positive sample ratio fit；
Circulation step A41 and A42, to carry out the downward division of secondary tree, until stopping division when reaching preset stopping condition, obtain
To the secondary tree of a depth.
5. method for detecting human face as claimed in claim 1, it is characterised in that：The secondary tree of the depth that is built in step A4 meet with
Lower splitting rule：af2+ bf+c=t, wherein f represent standard pixel difference feature, and t represents division threshold value, and a, b, c are on feature f
Constant.
6. method for detecting human face as claimed in claim 1, it is characterised in that：Step B1 in Face datection step is specifically wrapped
Include：Receive image to be detected and the extraction of standard pixel difference feature is carried out to it, obtain feature vector to be detected；Treated described
The feature vector of detection is input in random forest, the judgement for being made whether to include face respectively by the secondary tree of each depth.
A kind of 7. human face detection device, it is characterised in that：Including image acquisition units, storage unit and model training program and
Face datection program；
Described image collecting unit is used to obtain different facial images；Storage unit storage facial image and it is collected into
Inhuman face image, wherein, facial image and inhuman face image are divided into positive and negative samples training set；
The model training program is used to train to obtain a Face datection model based on positive and negative samples training set；
The Face datection program is used to carry out feature extraction to image to be detected of input, and is input in Face datection model
It is made whether the detection for including face；
Wherein, the model training program includes：
Feature extraction program, the extraction of standard pixel difference feature is carried out for all images to positive and negative samples training set respectively,
Obtain the respective feature vector of each image；To all standard pixel difference feature normalizations processing in all feature vectors to 0
Between~255；
Stochastical sampling program, for randomly selecting the several negative samples of positive sample image sum number from positive and negative samples training set respectively
This image, and several features are randomly selected from the feature vector of each positive sample image and each negative sample image respectively, obtain every
The negative sample feature vector of the positive sample feature vector of one positive sample image and each negative sample image；
Secondary tree construction procedures, for the positive sample image that is drawn into according to stochastical sampling program and its positive sample feature to
Amount, and negative sample image and its negative sample feature vector, build the secondary tree of a depth；The secondary tree structure journey is performed a plurality of times
Sequence, obtains the secondary tree of multiple depth, forms random forest, i.e. Face datection model；
Test program, is made whether test image using obtained random forest to include the detection of face, if in test image
Containing face but and be not detected at, then the test image is added in positive sample training set, to update positive sample training set；
If without face but detecting face in test image, which is added in negative sample training set, it is negative to update
Sample training collection.
8. human face detection device as claimed in claim 7, it is characterised in that：The Face datection program include statistics program and
Decision procedure, statistics program are used for tree secondary to each depth in random forest and unite to the judging result of image to be detected
Meter, to draw the ratio for judging the secondary tree of depth comprising face in image to be detected；Decision procedure be used for by the ratio with
Preset value is compared, and the testing result of face whether is included according to comparative result output.
9. a kind of computer-readable recording medium, is stored thereon with a computer program, it is characterised in that：The computer program
The step of claim 1 to 6 any one of them method for detecting human face is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711132667.7A CN107909034A (en) | 2017-11-15 | 2017-11-15 | A kind of method for detecting human face, device and computer-readable recording medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711132667.7A CN107909034A (en) | 2017-11-15 | 2017-11-15 | A kind of method for detecting human face, device and computer-readable recording medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107909034A true CN107909034A (en) | 2018-04-13 |
Family
ID=61845767
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711132667.7A Pending CN107909034A (en) | 2017-11-15 | 2017-11-15 | A kind of method for detecting human face, device and computer-readable recording medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107909034A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108446682A (en) * | 2018-05-11 | 2018-08-24 | 深圳尊豪网络科技股份有限公司 | A kind of recognition of face calibration detection method of full-automation |
| CN108833348A (en) * | 2018-05-08 | 2018-11-16 | 北京奇艺世纪科技有限公司 | A kind of method for detecting abnormality and device based on the modeling of log figure |
| CN108960047A (en) * | 2018-05-22 | 2018-12-07 | 中国计量大学 | Face De-weight method in video monitoring based on the secondary tree of depth |
| CN110399826A (en) * | 2019-07-22 | 2019-11-01 | 清华大学深圳研究生院 | A kind of end-to-end human face detection and recognition method |
| CN111291657A (en) * | 2020-01-21 | 2020-06-16 | 同济大学 | Crowd counting model training method based on difficult case mining and application |
| CN111652046A (en) * | 2020-04-17 | 2020-09-11 | 济南浪潮高新科技投资发展有限公司 | Safe wearing detection method, equipment and system based on deep learning |
| CN112348765A (en) * | 2020-10-23 | 2021-02-09 | 深圳市优必选科技股份有限公司 | Data enhancement method and device, computer readable storage medium and terminal equipment |
| CN112668385A (en) * | 2020-08-17 | 2021-04-16 | 紫光云技术有限公司 | Method for marking human face |
| CN113343295A (en) * | 2021-06-07 | 2021-09-03 | 支付宝(杭州)信息技术有限公司 | Image processing method, device, equipment and storage medium based on privacy protection |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101510254A (en) * | 2009-03-25 | 2009-08-19 | 北京中星微电子有限公司 | Method for updating gender classifier in image analysis and the gender classifier |
| CN102063642A (en) * | 2010-12-30 | 2011-05-18 | 上海电机学院 | Selection method for fuzzy neural network sample on basis of active learning |
| CN102169545A (en) * | 2011-04-25 | 2011-08-31 | 中国科学院自动化研究所 | Detection method for changes of high-resolution remote sensing images |
| CN103679209A (en) * | 2013-11-29 | 2014-03-26 | 广东领域安防有限公司 | Sparse theory based character recognition method |
| CN106056752A (en) * | 2016-05-25 | 2016-10-26 | 武汉大学 | Banknote authentication method based on random forest |
| CN106557750A (en) * | 2016-11-22 | 2017-04-05 | 重庆邮电大学 | It is a kind of based on the colour of skin and the method for detecting human face of depth y-bend characteristics tree |
| CN106599918A (en) * | 2016-12-13 | 2017-04-26 | 开易（深圳）科技有限公司 | Vehicle tracking method and system |
| CN106980825A (en) * | 2017-03-15 | 2017-07-25 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of face posture sorting technique based on normalization pixel difference feature |
| CN107341457A (en) * | 2017-06-21 | 2017-11-10 | 北京小米移动软件有限公司 | Method for detecting human face and device |
-
2017
- 2017-11-15 CN CN201711132667.7A patent/CN107909034A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101510254A (en) * | 2009-03-25 | 2009-08-19 | 北京中星微电子有限公司 | Method for updating gender classifier in image analysis and the gender classifier |
| CN102063642A (en) * | 2010-12-30 | 2011-05-18 | 上海电机学院 | Selection method for fuzzy neural network sample on basis of active learning |
| CN102169545A (en) * | 2011-04-25 | 2011-08-31 | 中国科学院自动化研究所 | Detection method for changes of high-resolution remote sensing images |
| CN103679209A (en) * | 2013-11-29 | 2014-03-26 | 广东领域安防有限公司 | Sparse theory based character recognition method |
| CN106056752A (en) * | 2016-05-25 | 2016-10-26 | 武汉大学 | Banknote authentication method based on random forest |
| CN106557750A (en) * | 2016-11-22 | 2017-04-05 | 重庆邮电大学 | It is a kind of based on the colour of skin and the method for detecting human face of depth y-bend characteristics tree |
| CN106599918A (en) * | 2016-12-13 | 2017-04-26 | 开易（深圳）科技有限公司 | Vehicle tracking method and system |
| CN106980825A (en) * | 2017-03-15 | 2017-07-25 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of face posture sorting technique based on normalization pixel difference feature |
| CN107341457A (en) * | 2017-06-21 | 2017-11-10 | 北京小米移动软件有限公司 | Method for detecting human face and device |
Non-Patent Citations (2)
| Title |
|---|
| SHENGCAI LIAO 等: "A Fast and Accurate Unconstrained Face Detector", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 朱玟谦: "一种收敛性随机森林在人脸检测中的应用研究", 《中国优秀硕士学位论文全文数据库-信息科技辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108833348A (en) * | 2018-05-08 | 2018-11-16 | 北京奇艺世纪科技有限公司 | A kind of method for detecting abnormality and device based on the modeling of log figure |
| CN108833348B (en) * | 2018-05-08 | 2021-01-05 | 北京奇艺世纪科技有限公司 | Abnormity detection method and device based on log graph modeling |
| CN108446682A (en) * | 2018-05-11 | 2018-08-24 | 深圳尊豪网络科技股份有限公司 | A kind of recognition of face calibration detection method of full-automation |
| CN108960047A (en) * | 2018-05-22 | 2018-12-07 | 中国计量大学 | Face De-weight method in video monitoring based on the secondary tree of depth |
| CN108960047B (en) * | 2018-05-22 | 2021-09-14 | 中国计量大学 | Face duplication removing method in video monitoring based on depth secondary tree |
| CN110399826A (en) * | 2019-07-22 | 2019-11-01 | 清华大学深圳研究生院 | A kind of end-to-end human face detection and recognition method |
| CN110399826B (en) * | 2019-07-22 | 2021-06-15 | 清华大学深圳研究生院 | End-to-end face detection and identification method |
| CN111291657A (en) * | 2020-01-21 | 2020-06-16 | 同济大学 | Crowd counting model training method based on difficult case mining and application |
| CN111652046A (en) * | 2020-04-17 | 2020-09-11 | 济南浪潮高新科技投资发展有限公司 | Safe wearing detection method, equipment and system based on deep learning |
| CN112668385A (en) * | 2020-08-17 | 2021-04-16 | 紫光云技术有限公司 | Method for marking human face |
| CN112348765A (en) * | 2020-10-23 | 2021-02-09 | 深圳市优必选科技股份有限公司 | Data enhancement method and device, computer readable storage medium and terminal equipment |
| CN113343295A (en) * | 2021-06-07 | 2021-09-03 | 支付宝(杭州)信息技术有限公司 | Image processing method, device, equipment and storage medium based on privacy protection |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107909034A (en) | A kind of method for detecting human face, device and computer-readable recording medium | |
| CN108898047B (en) | Pedestrian detection method and system based on blocking and shielding perception | |
| CN109948425A (en) | A kind of perception of structure is from paying attention to and online example polymerize matched pedestrian's searching method and device | |
| CN109815801A (en) | Face identification method and device based on deep learning | |
| CN104134071B (en) | A kind of deformable part model object detecting method based on color description | |
| CN106845357A (en) | A kind of video human face detection and recognition methods based on multichannel network | |
| CN105096307B (en) | The method of detection object in paired stereo-picture | |
| CN104992142A (en) | Pedestrian recognition method based on combination of depth learning and property learning | |
| CN105069481B (en) | Natural scene multiple labeling sorting technique based on spatial pyramid sparse coding | |
| CN108334848A (en) | A kind of small face identification method based on generation confrontation network | |
| CN106529448A (en) | Method for performing multi-visual-angle face detection by means of integral channel features | |
| CN107437099A (en) | A kind of specific dress ornament image recognition and detection method based on machine learning | |
| CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
| CN104598885B (en) | The detection of word label and localization method in street view image | |
| CN104134059B (en) | Keep the bad image detecting method under the mixing deformation model of colouring information | |
| CN106127164A (en) | The pedestrian detection method with convolutional neural networks and device is detected based on significance | |
| CN110110650A (en) | Face identification method in pedestrian | |
| CN108961675A (en) | Fall detection method based on convolutional neural networks | |
| CN109190643A (en) | Based on the recognition methods of convolutional neural networks Chinese medicine and electronic equipment | |
| CN108921107A (en) | Pedestrian's recognition methods again based on sequence loss and Siamese network | |
| CN104598889B (en) | The method and apparatus of Human bodys' response | |
| CN110503000B (en) | Teaching head-up rate measuring method based on face recognition technology | |
| CN108205684A (en) | Image disambiguation method, device, storage medium and electronic equipment | |
| CN106897681A (en) | A kind of remote sensing images comparative analysis method and system | |
| CN101667245A (en) | Human face detection method by cascading novel detection classifiers based on support vectors |
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: 20180413 |
|
| RJ01 | Rejection of invention patent application after publication |