CN105005768B - Dynamic percentage sample cuts AdaBoost method for detecting human face - Google Patents
Dynamic percentage sample cuts AdaBoost method for detecting human face Download PDFInfo
- Publication number
- CN105005768B CN105005768B CN201510391152.3A CN201510391152A CN105005768B CN 105005768 B CN105005768 B CN 105005768B CN 201510391152 A CN201510391152 A CN 201510391152A CN 105005768 B CN105005768 B CN 105005768B
- Authority
- CN
- China
- Prior art keywords
- sample
- training
- error rate
- weak classifier
- samples
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- 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)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of dynamic percentage samples to cut AdaBoost Face datection algorithms, specially:The percentage f of number of samples is cut needed for being determined first at the beginning of each iteration, each round crops the smaller sample of weight according to f, it is trained with remaining sample, when the best Weak Classifier error rate for the current iteration that training obtains is more than the error rate that random value generates, by reducing the constant f cut, enlarged sample collection quantity, training is re-started for current iteration.If when being trained using whole samples, error rate stops iteration still above 0.5.When the present invention is excessive suitable for the number of samples for participating in training, by selected part to the better sample of performance boost effect, to achieve the purpose that save the training time.
Description
Technical field
The present invention relates to a kind of dynamic percentage samples to cut AdaBoost method for detecting human face, belongs to mode identification technology
Field.
Background technology
Biometrics identification technology is to realize identity card by each individual exclusive physiological characteristic and behavioural characteristic
The purpose that real or individual differentiates.The features such as one kind of face as biological characteristic has and is easily obtained, and interface is friendly, compared to
Existing frequently-used mode, such as password, credit card, ID card have the advantages such as not reproducible, easy to carry, distinctive is strong.Cause
This has broad prospects in fields such as video monitoring, smart home and criminal investigations.As embedded device operational capability is got over
Come stronger, intelligent algorithm is increasingly being applied to embedded development field, realizes different functions.Wherein Face datection conduct
The basis of recognition of face becomes the research hotspot of artificial intelligence field.
Its core of AdaBoost algorithms is to extract classifying quality most from a large amount of Haar features by the method for iteration
Good feature is as Weak Classifier, and the strong classifier ultimately generated is made of a large amount of Weak Classifier.AdaBoost is practical
And it is simple, and the method for detecting human face based on AdaBoost algorithms not only has high inspection for the detection of single facial image
Precision is surveyed, and has detection speed quickly, therefore the face recognition technology based on the algorithm is widely used.
Work as training sample, sample characteristics, when Weak Classifier number is more, using the classification of AdaBoost algorithms training
Device can consume a large amount of training time.Characteristic Number determines that the iterations of algorithm, each iteration obtain individual features and instructing
Practice the error rate in sample set, best Weak Classifier is obtained finally by comparison error rate.A best weak typing is often trained
The weight of device, training sample can change accordingly, so if needing more Weak Classifiers, then need to repeat corresponding time
Several above-mentioned steps.It can be seen that work as training sample, and when sample characteristics number and Weak Classifier number increase, training time meeting
Increased with the three cubed order of magnitude.
Invention content
It is an object of the invention to overcome deficiency in the prior art, a kind of dynamic percentage sample cutting is provided
AdaBoost method for detecting human face solves to use the grader of AdaBoost algorithms training that can consume a large amount of instruction in the prior art
The technical issues of practicing the time.
In order to solve the above technical problems, the technical solution adopted in the present invention is:Dynamic percentage sample is cut
AdaBoost method for detecting human face, at the beginning of each iteration, it is first determined the required percentage f for cutting number of samples,
Each round crops the smaller sample of weight according to f, is trained with remaining sample;
When training the best Weak Classifier error rate of obtained current iteration to be more than the error rate that random value generates, by subtracting
The constant f of small cutting, enlarged sample collection quantity, training is re-started for current iteration;
If when being trained using whole samples, error rate then stops iteration still above 0.5;
Specific algorithm includes the following steps:
Step 1:If the training sample sum of input is N, wherein negative sample is m, and positive sample is n, training sample set
For S={ (x1,y1),...(xn,yn), wherein xiIndicate i-th of sample, yi={ 1,0 } is respectively used to identify positive negative sample;
Step 2:Initialization sample weight:
Step 3:Assuming that the sample percentage that each round is cast out is f, then the number of samples that each round participates in training is N
× (1-f), iterations t=1,2 ..., T;
Step 4:Optimal Weak Classifier is obtained, Weak Classifier h is acquiredtWeighting coefficient α in strong classifiert, method is such as
Under:
Step 401:Normalize the weighted value of sample:
Step 402:For each feature j, one simple Weak Classifier h of trainingj(x,fj,pj,θj):
Wherein, fj(x) value, p are characterizedjIndicate sign of inequality direction, θjFor Weak Classifier threshold value;
Step 403:Select the corresponding Weak Classifier h of minimal error ratet(x), wherein minimal error rate is defined as:
Step 404:If εt=0 or occur as soon as ε when training the first roundt>=0.5, then T=t-1 is enabled, step is jumped to
Six;If εt>=0.5 and be not the first round, then T=t-1 is enabled, judges whether f is more than 2/3, it is no if more than f=2 × f-1 is then enabled
Then f=f/2 is enabled to jump to step 5;
Step 405:Update sample weights:
As sample xiE when being classified by mistakei=0, on the contrary ei=1,
Step 406:Acquire Weak Classifier htWeighting coefficient in strong classifier:
Step 5:It to sample in training set, is arranged by weighted value, according to the percentage f of cutting, is cut from small to large
Fall the smaller preceding n × f sample of weight;
Step 6:Export strong classifier:
Compared with prior art, the advantageous effect of the invention reached is:Number of samples suitable for participating in training is excessive
When, by selected part to the better sample of performance boost effect, to achieve the purpose that save the training time.
Description of the drawings
Fig. 1 is the flow chart of the method for the present invention.
Fig. 2 is the flow chart for obtaining optimal Weak Classifier.
Specific implementation mode
The invention will be further described below in conjunction with the accompanying drawings.
Meaning in attached drawing represented by each function is as follows:
Function cvGetTickCount ():It returns from os starting to the millisecond number currently passed through, passes through calculating
The difference of two back amount can count the training spent time.
Function Single_Classifier (int i):For generating a strong classifier, incoming parameter indicates to constitute
The Weak Classifier number of this strong classifier.
Function Generate_AllFeatures (int count):Feature for generating all Haar-like, count
Indicate the quantity using characteristic type.The present invention has selected 5 kinds of common feature templates, therefore count values are 5.
Function Input_Samples ():Positive negative sample is read in from specified directory.
Function Select_WeakClassifier ():For obtaining optimal Weak Classifier.
Function Output_WeakClassifier ():For exporting the Weak Classifier generated.
Function Cal_HaarValue (j, k):J-th of feature for calculating k-th of sample.
Function qsort ():Sample is ranked up according to the size of characteristic value.
As shown in Figure 1, dynamic percentage sample cuts AdaBoost method for detecting human face, when each iteration starts
It waits, it is first determined the required percentage f for cutting number of samples, each round crops the smaller sample of weight according to f, with remaining sample
Originally it is trained;
When training the best Weak Classifier error rate of obtained current iteration to be more than the error rate that random value generates, by subtracting
The constant f of small cutting, enlarged sample collection quantity, training is re-started for current iteration;
If when being trained using whole samples, error rate then stops iteration still above 0.5;
Specific algorithm includes the following steps:
Step 1:If the training sample sum of input is N, wherein negative sample is m, and positive sample is n, training sample set
For S={ (x1,y1),...(xn,yn), wherein xiIndicate i-th of sample, yi={ 1,0 } is respectively used to identify positive negative sample;
Step 2:Initialization sample weight:
Step 3:Assuming that the sample percentage that each round is cast out is f, then the number of samples that each round participates in training is N
× (1-f), iterations t=1,2 ..., T;
Step 4:Optimal Weak Classifier is obtained, Weak Classifier h is acquiredtWeighting coefficient α in strong classifiert, such as Fig. 2 institutes
Show, method is as follows:
Step 401:Normalize the weighted value of sample:
Step 402:For each feature j, one simple Weak Classifier h of trainingj(x,fj,pj,θj):
Wherein, fj(x) value, p are characterizedjIndicate sign of inequality direction, θjFor Weak Classifier threshold value;
Step 403:Select the corresponding Weak Classifier h of minimal error ratet(x), wherein minimal error rate is defined as:
Step 404:If εt=0 or occur as soon as ε when training the first roundt>=0.5, then T=t-1 is enabled, step is jumped to
Six;If εt>=0.5 and be not the first round, then T=t-1 is enabled, judges whether f is more than 2/3, it is no if more than f=2 × f-1 is then enabled
Then f=f/2 is enabled to jump to step 5;
Step 405:Update sample weights:
As sample xiE when being classified by mistakei=0, on the contrary ei=1,
Step 406:Acquire Weak Classifier htWeighting coefficient in strong classifier:
Step 5:It to sample in training set, is arranged by weighted value, according to the percentage f of cutting, is cut from small to large
Fall the smaller preceding n × f sample of weight;
Step 6:Export strong classifier:
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, without departing from the technical principles of the invention, several improvement and deformations can also be made, these improvement and deformations
Also it should be regarded as protection scope of the present invention.
Claims (1)
1. dynamic percentage sample cuts AdaBoost method for detecting human face, which is characterized in that at the beginning of each iteration,
The percentage f of number of samples is cut needed for determining first, each round crops the smaller sample of weight according to f, with remaining sample
It is trained;
When training the best Weak Classifier error rate of obtained current iteration to be more than the error rate that random value generates, passes through to reduce and cut out
The constant f cut, enlarged sample collection quantity, training is re-started for current iteration;
If when being trained using whole samples, error rate then stops iteration still above 0.5;
Specific algorithm includes the following steps:
Step 1:If the training sample sum of input is N, wherein negative sample is m, and positive sample is n, training sample set S
={ (x1,y1),...(xn,yn), wherein xiIndicate i-th of sample, yi={ 1,0 } is respectively used to identify positive negative sample;
Step 2:Initialization sample weight:
Step 3:Assuming that the sample percentage that each round is cast out is f, then the number of samples that each round participates in training is N × (1-
F), iterations t=1,2 ..., T;
Step 4:Optimal Weak Classifier is obtained, Weak Classifier h is acquiredtWeighting coefficient α in strong classifiert, method is as follows:
Step 401:Normalize the weighted value of sample:
Step 402:For each feature j, one simple Weak Classifier h of trainingj(x,fj,pj,θj):
Wherein, fj(x) value, p are characterizedjIndicate sign of inequality direction, θjFor Weak Classifier threshold value;
Step 403:Select the corresponding Weak Classifier h of minimal error ratet(x), wherein minimal error rate is defined as:
Step 404:If εt=0 or occur as soon as ε when training the first roundt>=0.5, then T=t-1 is enabled, step 6 is jumped to;Such as
Fruit εt>=0.5 and be not the first round, then T=t-1 is enabled, judges whether f is more than 2/3, if more than f=2 × f-1 is then enabled, otherwise enables f
=f/2 jumps to step 5;
Step 405:Update sample weights:
As sample xiE when being classified by mistakei=0, on the contrary ei=1,
Step 406:Acquire Weak Classifier htWeighting coefficient in strong classifier:
Step 5:It to sample in training set, is arranged from small to large by weighted value, according to the percentage f of cutting, crops power
The smaller preceding n × f sample of weight;
Step 6:Export strong classifier:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510391152.3A CN105005768B (en) | 2015-07-06 | 2015-07-06 | Dynamic percentage sample cuts AdaBoost method for detecting human face |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510391152.3A CN105005768B (en) | 2015-07-06 | 2015-07-06 | Dynamic percentage sample cuts AdaBoost method for detecting human face |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105005768A CN105005768A (en) | 2015-10-28 |
CN105005768B true CN105005768B (en) | 2018-09-14 |
Family
ID=54378433
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510391152.3A Expired - Fee Related CN105005768B (en) | 2015-07-06 | 2015-07-06 | Dynamic percentage sample cuts AdaBoost method for detecting human face |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105005768B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106022225B (en) * | 2016-05-10 | 2019-03-05 | 中科天网(广东)科技有限公司 | A kind of Face datection classifier building method based on AdaBoost |
CN106951930A (en) * | 2017-04-13 | 2017-07-14 | 杭州申昊科技股份有限公司 | A kind of instrument localization method suitable for Intelligent Mobile Robot |
CN107477809A (en) * | 2017-09-20 | 2017-12-15 | 四川长虹电器股份有限公司 | Air conditioner energy source management system based on Adaboost |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101196984A (en) * | 2006-12-18 | 2008-06-11 | 北京海鑫科金高科技股份有限公司 | Fast face detecting method |
CN103116756A (en) * | 2013-01-23 | 2013-05-22 | 北京工商大学 | Face detecting and tracking method and device |
CN103605964A (en) * | 2013-11-25 | 2014-02-26 | 上海骏聿数码科技有限公司 | Face detection method and system based on image on-line learning |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7953253B2 (en) * | 2005-12-31 | 2011-05-31 | Arcsoft, Inc. | Face detection on mobile devices |
-
2015
- 2015-07-06 CN CN201510391152.3A patent/CN105005768B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101196984A (en) * | 2006-12-18 | 2008-06-11 | 北京海鑫科金高科技股份有限公司 | Fast face detecting method |
CN103116756A (en) * | 2013-01-23 | 2013-05-22 | 北京工商大学 | Face detecting and tracking method and device |
CN103605964A (en) * | 2013-11-25 | 2014-02-26 | 上海骏聿数码科技有限公司 | Face detection method and system based on image on-line learning |
Non-Patent Citations (2)
Title |
---|
基于AdaBoost算法的人脸检测研究;李长风;《中国优秀硕士学位论文全文数据库 信息科技辑》;20141015(第10期);全文 * |
基于动态权重裁剪的快速Adaboost训练算法;贾慧星;《计算机学报》;20090228;第32卷(第2期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN105005768A (en) | 2015-10-28 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Li et al. | Sequential three-way decision and granulation for cost-sensitive face recognition | |
Zhang et al. | Tiny YOLO optimization oriented bus passenger object detection | |
Li et al. | Nonparametric discriminant analysis for face recognition | |
US9251402B2 (en) | Association and prediction in facial recognition | |
Gualdi et al. | Multistage particle windows for fast and accurate object detection | |
CN110348416A (en) | Multi-task face recognition method based on multi-scale feature fusion convolutional neural network | |
Huang et al. | Unifying identification and context learning for person recognition | |
CN101661554B (en) | Front face human body automatic identity recognition method under long-distance video | |
CN110781829A (en) | Light-weight deep learning intelligent business hall face recognition method | |
CN105069396B (en) | Dynamic percentage feature cuts AdaBoost Face datection algorithms | |
CN105809125A (en) | Multi-core ARM platform based human face recognition system | |
Gunasekar et al. | Face detection on distorted images augmented by perceptual quality-aware features | |
CN104463128A (en) | Glass detection method and system for face recognition | |
CN105205460B (en) | Human face expression feature extraction and recognition methods based on maximum direction encoding | |
CN109977887A (en) | A kind of face identification method of anti-age interference | |
CN105005768B (en) | Dynamic percentage sample cuts AdaBoost method for detecting human face | |
Mady et al. | Efficient real time attendance system based on face detection case study “MEDIU staff” | |
CN110633689B (en) | Face recognition model based on semi-supervised attention network | |
CN113076919A (en) | Intelligent access control system based on face recognition | |
Tofighi et al. | A robust face recognition system in image and video | |
Yu et al. | A depth cascade face detection algorithm based on adaboost | |
Wan et al. | Face detection method based on skin color and adaboost algorithm | |
Wang et al. | Research on face detection based on fast Haar feature | |
Al-Dabbas et al. | Machine Learning Approach for Facial Image Detection System | |
Jiang et al. | A three-category face detector with contextual information on finding tiny faces |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180914 Termination date: 20210706 |