CN103530648A - Face recognition method based on multi-frame images - Google Patents

Face recognition method based on multi-frame images Download PDF

Info

Publication number
CN103530648A
CN103530648A CN201310477125.9A CN201310477125A CN103530648A CN 103530648 A CN103530648 A CN 103530648A CN 201310477125 A CN201310477125 A CN 201310477125A CN 103530648 A CN103530648 A CN 103530648A
Authority
CN
China
Prior art keywords
face
people
personnel
identified
image
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
Application number
CN201310477125.9A
Other languages
Chinese (zh)
Inventor
刘先勇
侯磊
凌霄
张亮
贺庆
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Sichuan Airport Consciousness Science And Technology Ltd
Original Assignee
Sichuan Airport Consciousness Science And Technology Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Sichuan Airport Consciousness Science And Technology Ltd filed Critical Sichuan Airport Consciousness Science And Technology Ltd
Priority to CN201310477125.9A priority Critical patent/CN103530648A/en
Publication of CN103530648A publication Critical patent/CN103530648A/en
Pending legal-status Critical Current

Links

Landscapes

  • Collating Specific Patterns (AREA)

Abstract

The invention relates to the field of security and protection, in particular to a face recognition method based on multi-frame images. The method concretely comprises the following steps of 1, personnel identity registration; 2, data collection of personnel to be recognized: multi-frame images containing faces of personnel to be recognized are collected to be used as original sample sequences to be recognized, wherein J is the face image collection times of the personnel to be recognized; 3, face recognition: after personnel to be recognized enter a recognition region for data collection, the original registration sample sequence is inquired according to the identity information of personnel to be recognized, and the registration sample sequence corresponding to the personnel to be recognized is obtained; the orthogonal subspace of the registration sample sequence and the orthogonal subspace of the sample sequence to be recognized are calculated, the similarity measure of the two subspaces is calculated, and whether the faces to be recognized are consistent with registration faces or not is judged according to the similarity measure of the two subspaces. The similarity measure of the two subspaces is compared, and the faces are recognized on the basis of multi-frame images.

Description

A kind of face identification method based on multiple image
Technical field
The present invention relates to technical field of security and protection, relate in particular to a kind of face identification method based on multiple image, be applicable to the higher occasion of accuracy of identification and stability requirement.
Background technology
What the personnel identity identification equipment of available technology adopting adopted conventionally is access card system, the access card that different personnel assignments is different, after access card confirmation is errorless, gate is opened and is allowed to pass through, it is not high that such mode is applicable to safety requirements, personnel and access card do not need occasion one to one, such as the mess card in mass transit card, dining room etc.But such mode can not be carried out corresponding one by one with personnel to be identified by access card, also just there is certain potential safety hazard, especially in the higher occasion of demand for security, above-mentioned mode obviously cannot solve the problem of borrowing card or stealing card, such as identification on airport or when the safe procuratorial work in railway station, bank debits number are larger etc., these have higher demand for security, the access card of only usining is obviously inadequate as the mode of identification, also needs the information in personnel to be identified and access card to mate.This also just needs a kind of method that people's face of registration in people's face of personnel to be identified and card is judged.
Recognition of face of the prior art mainly adopts following two kinds.The first, gather frame people face data, people's face of preserving in this frame people's face collecting and database is compared, judge the similarity between two two field pictures, thereby identify whether people's face of preserving in people's face of collection and database is same person.Yet while differing greatly when people's face angle to be identified, attitude, light or expression and registration, discrimination significantly reduces, and causes and repeatedly tests discrimination fluctuation greatly.For this problem, in prior art, occurred that another many camera lens gathers the method for people's face to be identified, camera lens is installed respectively in the positions such as front, rear, left and right personnel to be identified, adopt a plurality of camera lenses to obtain multiframe people's face of synchronization different angles, people's face that each camera lens is obtained respectively one by one with database in registrant's face of preserving compare one by one.Although it is a plurality of that sample point to be identified has, but still be comparing between sample point and sample point, lack the related information between sample point, the distributed intelligence of the attitude that is beyond expression, light, expression.And because the introducing of a plurality of camera lenses, hardware cost is higher, system stability reduces.
Summary of the invention
The discrimination occurring when personnel's to be identified attitude, light, the expression shape change for recognition of face of the prior art reduces, the technical matters of identification poor stability, the invention discloses a kind of face identification method based on multiple image.
The invention discloses a kind of face identification method based on multiple image, it specifically comprises the following steps:
Step 1, personnel identity registration: gather the image that multiframe comprises accredited personnel people's face, the image that comprises accredited personnel people's face collecting is carried out to the detection of people's face, extract the people's face in each image and preserve, be called accredited personnel's the bashful figure of people; By the bashful figure of a plurality of accredited personnel's people carry out yardstick, towards the normalized with illumination, the bashful figure of people after normalized is as original licensed sample sequence
Figure 591300DEST_PATH_IMAGE001
, wherein K is
Figure 2013104771259100002DEST_PATH_IMAGE002
the number of times of individual accredited personnel's man face image acquiring, the total quantity that N is accredited personnel;
Step 2, personnel's data acquisition to be identified: gather the image that multiframe comprises personnel people's face to be identified, the image that comprises personnel people's face to be identified collecting is carried out to the detection of people's face, extract the people's face in each image and preserve, the bashful figure of people that is called personnel to be identified, and by the bashful figure of a plurality of people carry out yardstick, towards the normalized with illumination, the bashful figure of people after normalized is as original sample sequence to be identified
Figure 872108DEST_PATH_IMAGE003
, wherein J is the number of times of personnel's man face image acquiring to be identified;
Step 3, recognition of face: personnel to be identified enter identified region and carry out after data acquisition, according to personnel identity information inquiry original licensed sample sequence to be identified, obtain registration sample sequence corresponding to personnel to be identified
Figure 2013104771259100002DEST_PATH_IMAGE004
; Calculate registration sample sequence
Figure 295262DEST_PATH_IMAGE005
orthogonal subspaces with sample sequence to be identified
Figure 91048DEST_PATH_IMAGE003
orthogonal subspaces
Figure 202224DEST_PATH_IMAGE007
, calculate two sub spaces with similarity measure, according to the similarity measure of two sub spaces, judge that whether people's face to be identified consistent with registrant's face.
Further, said method also comprises: when carrying out authentication, while confirming whether people to be identified is he or she, registrant's face of people's face to be identified and appointment carries out subspace and compares, subspace similarity measure compares with the threshold value of setting in advance, thereby judges.
Further, said method also comprises: when carrying out identification, while confirming that whom people to be identified is, the registrant's face in people's face to be identified and database carries out subspace one by one to be compared, or finds out the registrant face classification the most similar to it by sorter.
Further, the method for the similarity measure between above-mentioned judgement two sub spaces is the main included angle cosine between judgement two sub spaces.
Further, the detailed process of the main included angle cosine between above-mentioned judgement two sub spaces is: the angle of setting between two sub spaces is , the angle between two sub spaces
Figure 24796DEST_PATH_IMAGE009
be less than the angle threshold value of setting
Figure 2013104771259100002DEST_PATH_IMAGE010
time, judge that personnel to be identified are consistent with accredited personnel, identify successfully; Otherwise judge that personnel to be identified and accredited personnel are inconsistent, recognition failures; Wherein
Figure 159236DEST_PATH_IMAGE011
.
 
Further, above-mentioned people's face testing process is specific as follows:
First adopt Haar characteristic image to travel through in the image of 20 * 20, the pixel of white portion and deduct black region pixel and, the value obtaining is referred to as face characteristic value;
Then from the face database of registration in advance, cut out a large amount of people's face picture and background picture, as training sample; Training sample is normalized to the image of 20 * 20 sizes, in big or small like this picture, extracts Haar eigenwert; These Haar features have formed Weak Classifier, and its function expression is:
Wherein: the image of 20 * 20 sizes,
Figure 2013104771259100002DEST_PATH_IMAGE014
for Haar feature,
Figure 176663DEST_PATH_IMAGE015
for symbol designator,
Figure 2013104771259100002DEST_PATH_IMAGE016
for threshold value.
Further, said method also comprises training strong classifier, and the training process of strong classifier need to pass through
Figure 884725DEST_PATH_IMAGE017
inferior iteration, its detailed process is as follows:
(1) given training sample set , altogether
Figure 72124DEST_PATH_IMAGE002
individual sample, wherein
Figure 745813DEST_PATH_IMAGE019
with
Figure 2013104771259100002DEST_PATH_IMAGE020
correspond respectively to positive sample and negative sample;
Figure 180205DEST_PATH_IMAGE017
maximum cycle for training;
(2) initialization sample weight is
Figure 188613DEST_PATH_IMAGE021
, be the initial probability distribution of training sample;
(3) iteration training for the first time individual sample, obtains first optimum Weak Classifier;
(4) improve last round of in the weight of misjudged sample;
(5) new sample and last time are put together and carried out the training of a new round by the sample of misclassification;
(6) 4-5 step is carried out in circulation,
Figure 704968DEST_PATH_IMAGE017
after wheel, obtain
Figure 372579DEST_PATH_IMAGE017
individual optimum Weak Classifier;
(7) combination
Figure 602703DEST_PATH_IMAGE017
individual optimum Weak Classifier obtains strong classifier, and array mode is as follows
Figure 2013104771259100002DEST_PATH_IMAGE022
Further, said method also comprises face characteristic leaching process, and the registrant's face collecting in step 1 and step 2 and people's face to be identified are carried out carrying out the recognition of face of step 3 after feature extraction again.
Further, above-mentioned feature extracting method is principal component analysis (PCA).
By adopting above technical scheme, beneficial effect of the present invention is: in identification and two stages of identity registration, all gather respectively multiple image, and carry out respectively the detection of people's face, the multiframe facial image recognizing is set up respectively to subspace separately, calculate the similarity measure between two sub spaces, thereby judge that whether people's face to be identified is similar to registrant's face.The present invention adopts the method for comparing between two sub spaces, and identification accuracy and stability are all improved, and compares with the scheme of a plurality of camera lenses simultaneously, and hardware cost is lower, and system stability improves.The present invention adopts same video camera to take multiframe facial image at cognitive phase, this multiframe facial image is formed to subspace, and compare with the subspace that the multiframe facial image in when registration forms, in the situation that not increasing hardware device, greatly improved the accuracy of identification.
Accompanying drawing explanation
Fig. 1 is the face identification method process flow diagram based on multiple image of the present invention.
Embodiment
In order to make object of the present invention, technical scheme and advantage clearer, below in conjunction with Figure of description and specific embodiment, the present invention is described in further detail.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.
The face identification method process flow diagram based on multiple image of the present invention as shown in Figure 1.Its concrete steps comprise identity registration and two stages of identification; In the identity registration stage, first sequentially everyone is carried out to multiple image collection, every two field picture recognition of face, face characteristic extraction and face characteristic record; In the identification stage, equally sequentially carry out multiple image collection, every two field picture recognition of face, face characteristic extraction and face characteristic record; Then according to personnel identity information to be identified, (such as No. ID, identity) obtains the corresponding face characteristic at registration phase record, the face characteristic that the identification stage obtains is measured with the similarity that the corresponding face characteristic of registration phase record carries out two sub spaces, whether legal thereby personnel to be identified are judged in knowledge.Such as, when personnel to be identified are legal, gate is opened, and permission personnel pass through, otherwise gate is not opened, and sends alerting signal simultaneously.
Embodiment particularly:
The invention discloses a kind of face identification method based on multiple image, it specifically comprises the following steps:
Step 1, personnel identity registration: gather the image that multiframe comprises accredited personnel people's face, such as requiring accredited personnel face in the face of camera lens, conversion towards (face, upward, down, towards left, towards the right side) carry out multi collect, the image that comprises accredited personnel people's face collecting is carried out to the detection of people's face, extract the people's face in each image and preserve, the bashful figure of people that is called accredited personnel, and by the bashful figure of a plurality of accredited personnel's people carry out yardstick, towards the normalized with illumination.Yardstick normalization: refer to the facial image of different sizes is zoomed in or out to unified size; Towards normalization: refer to the facial image tilting is rotated to normal angled; Unitary of illumination: refer to the brightness that increases or reduce facial image, make facial image overall brightness consistent.Yardstick, towards the detailed process with unitary of illumination, belong to the conventional means of those skilled in the art, do not repeat them here its detailed process.The bashful figure of people after normalization is as an original licensed sample sequence
Figure 85899DEST_PATH_IMAGE001
, wherein K is
Figure 967137DEST_PATH_IMAGE002
the number of times of individual accredited personnel's man face image acquiring,
Figure 556381DEST_PATH_IMAGE002
also can be set as accredited personnel's No. ID, identity.Personnel to be identified by time, according to No. ID, its identity
Figure 601697DEST_PATH_IMAGE002
can obtain fast the image that personnel to be identified gather when registration.
Step 2, personnel's data acquisition to be identified: gather the image that multiframe comprises personnel people's face to be identified, the image that comprises personnel people's face to be identified collecting is carried out to the detection of people's face, extract the people's face in each image and preserve, the bashful figure of people that is called personnel to be identified, and by the bashful figure of a plurality of people carry out yardstick, towards the normalized with illumination.Yardstick normalization: refer to the facial image of different sizes is zoomed in or out to unified size; Towards normalization: refer to the facial image tilting is rotated to normal angled; Unitary of illumination: refer to the brightness that increases or reduce facial image, make facial image overall brightness consistent.Similarly, yardstick, towards the normalized detailed process with illumination, belong to that those skilled in the art commonly use technological means, do not repeat them here its detailed process.The bashful figure of people after normalization is as an original sample sequence to be identified
Figure 682392DEST_PATH_IMAGE003
, wherein J is the number of times of personnel's man face image acquiring to be identified.
Step 3, recognition of face: personnel to be identified enter identified region and carry out after data acquisition, according to No. ID, identity
Figure 168868DEST_PATH_IMAGE002
query sample database, obtains registration sample sequence corresponding to personnel to be identified
Figure 443861DEST_PATH_IMAGE004
; Calculate registration sample sequence
Figure 336993DEST_PATH_IMAGE005
orthogonal subspaces with sample sequence to be identified
Figure 876876DEST_PATH_IMAGE003
orthogonal subspaces
Figure 322769DEST_PATH_IMAGE007
, calculate two sub spaces
Figure 749203DEST_PATH_IMAGE008
with
Figure 751794DEST_PATH_IMAGE007
similarity measure, according to the similarity measure of two sub spaces, judge that whether people's face to be identified consistent with registrant's face.When carrying out authentication, while confirming whether people to be identified is he or she, registrant's face of people's face to be identified and appointment carries out subspace to be compared, and subspace similarity measure compares with the threshold value of setting in advance, thereby judges.When carrying out identification, while confirming that whom people to be identified is, the registrant's face in people's face to be identified and database carries out subspace one by one to be compared, or finds out the registrant face classification the most similar to it by sorter.Judge that the similarity measure between two sub spaces can adopt the main included angle cosine between judgement two sub spaces to realize, also can select the similarity measure between other subspaces to calculate, such as: Kernel Principal Angles, Symmetric Distance algorithm.The detailed process that judges the main included angle cosine between two sub spaces is: the angle of setting between two sub spaces is
Figure 960665DEST_PATH_IMAGE009
, the angle between two sub spaces
Figure 62613DEST_PATH_IMAGE009
be less than the angle threshold value of setting
Figure 491189DEST_PATH_IMAGE010
time, judge that personnel to be identified are consistent with accredited personnel, identify successfully; Otherwise judge that personnel to be identified and accredited personnel are inconsistent, recognition failures.Wherein
Figure 969575DEST_PATH_IMAGE011
, can prove , wherein
Figure 2013104771259100002DEST_PATH_IMAGE024
for matrix
Figure 574311DEST_PATH_IMAGE025
eigenvalue of maximum.The thinking of its proof is based upon in linear algebra theoretical foundation, leading role's degree of two linear subspaces can obtain by the svd of matrix, the length of proof is long, refer to paper G.H. Golub, Numerical Methods for Computing Angles Between Linear Subspaces. math.Comp., 27 (1973), pp.579 – 594.
Prior art adopts many camera lenses to gather facial image to be identified, obtain the multiframe facial image of synchronization different angles, although it is a plurality of that sample point to be identified has, but just compare with registration sample point one by one, remain comparison between points, lack related information between sample point, the distributed intelligence of the attitude that is beyond expression, light, expression.Compare with the comparison method between sample of the present invention subspace, identification accuracy and less stable, and because the introducing of a plurality of camera lenses, hardware cost is higher, and system stability reduces.The present invention gathers respectively multiple image in identification and two stages of identity registration, and carry out respectively the detection of people's face, the multiframe facial image recognizing is set up respectively to subspace separately, calculate the similarity measure between two sub spaces, thereby judge that whether people's face to be identified is similar to registrant's face.The present invention adopts the method for comparing between two sub spaces, and identification accuracy and stability are all improved, and compares with the scheme of a plurality of camera lenses simultaneously, and hardware cost is lower, and system stability improves.The present invention adopts same video camera to take multiframe facial image at cognitive phase, this multiframe facial image is formed to subspace, and compare with the subspace that the multiframe facial image in when registration forms, in the situation that not increasing hardware device, greatly improved the accuracy of identification.
The process that said extracted goes out the people's face in each image can adopt people's face detection mode conventional in prior art, such as adopting HOG algorithm, SIFT algorithm etc.; Sorter can be used SVM, neural network etc.
People's face testing process in the present invention is specific as follows:
First adopt Haar characteristic image to travel through in the image of 20 * 20, the pixel of white portion and deduct black region pixel and, the value obtaining is referred to as face characteristic value.
Then from the face database of registration in advance, cut out a large amount of people's face picture and background picture, as training sample.Training sample is normalized to the image of 20 * 20 sizes, in big or small like this picture, extracts Haar eigenwert.These Haar features have formed Weak Classifier, and its function expression is:
Figure 224604DEST_PATH_IMAGE012
Wherein:
Figure 506681DEST_PATH_IMAGE013
the image of 20 * 20 sizes,
Figure 526082DEST_PATH_IMAGE014
for Haar feature,
Figure 32149DEST_PATH_IMAGE015
for symbol designator, for threshold value.
The training process of strong classifier need to pass through inferior iteration:
(1) given training sample set
Figure 545673DEST_PATH_IMAGE018
, altogether
Figure 911058DEST_PATH_IMAGE002
individual sample, wherein
Figure 286675DEST_PATH_IMAGE019
with
Figure 972872DEST_PATH_IMAGE020
correspond respectively to positive sample and negative sample;
Figure 366813DEST_PATH_IMAGE017
maximum cycle for training;
(2) initialization sample weight is
Figure 417945DEST_PATH_IMAGE021
, be the initial probability distribution of training sample;
(3) iteration training for the first time
Figure 28662DEST_PATH_IMAGE002
individual sample, obtains first optimum Weak Classifier;
(4) improve last round of in the weight of misjudged sample;
(5) new sample and last time are put together and carried out the training of a new round by the sample of misclassification;
(6) 4-5 step is carried out in circulation,
Figure 190653DEST_PATH_IMAGE017
after wheel, obtain
Figure 704680DEST_PATH_IMAGE017
individual optimum Weak Classifier;
(7) combination
Figure 989030DEST_PATH_IMAGE017
individual optimum Weak Classifier obtains strong classifier, and array mode is as follows
Figure 73661DEST_PATH_IMAGE022
Because the training stage has been done yardstick normalization to training sample, so this strong classifier can only be checked through the people's face under same yardstick.In order further to improve the stability of algorithm, can 5 yardsticks (such as: original image is first yardstick, 4 yardsticks are below by obtaining original image size reduction to 1/2,1/4,1/8,1/16 of original size) 5 strong classifiers of lower training, between them, by cascade, form final sorter.
Further, said method also comprises face characteristic leaching process, and the registrant's face collecting in step 1 and step 2 and people's face to be identified are carried out carrying out the recognition of face of step 3 after feature extraction again.Generally, the subwindow that comprises facial image detecting can not be directly used in recognition of face, and this is that excessive dimension can cause " dimension disaster " problem because these subwindows generally comprise the pixel of 400 left and right.Secondly, because the human face expression in these subwindows differs, even also have noise, block the situation mixed and disorderly with background.Therefore processing mode is by characteristic extraction procedure preferably, reduces the dimension of data, reduces picture noise, promotes the conspicuousness of data.In numerous feature extraction algorithms, principal component analysis (PCA) (Principal Component Analysis) is because it has analytic solution, and without iteration, counting yield is high and be used widely.Therefore, the preferred mode of the present invention is to adopt principal component analytical method to extract face characteristic.
Given coefficient and parameter in the above embodiments; be to provide to those skilled in the art and realize or use of the present invention; the present invention does not limit and only gets aforementioned disclosed numerical value; in the situation that not departing from inventive concept; those skilled in the art can make various modifications or adjustment to above-described embodiment; thereby protection scope of the present invention do not limit by above-described embodiment, and it should be the maximum magnitude that meets the inventive features that claims mention.

Claims (9)

1. the face identification method based on multiple image, it specifically comprises the following steps:
Step 1, personnel identity registration: gather the image that multiframe comprises accredited personnel people's face, the image that comprises accredited personnel people's face collecting is carried out to the detection of people's face, extract the people's face in each image and preserve, be called accredited personnel's the bashful figure of people; By the bashful figure of a plurality of accredited personnel's people carry out yardstick, towards the normalized with illumination, the bashful figure of people after normalized is as original licensed sample sequence
Figure 2013104771259100001DEST_PATH_IMAGE001
, wherein K is
Figure 411148DEST_PATH_IMAGE002
the number of times of individual accredited personnel's man face image acquiring, the total quantity that N is accredited personnel;
Step 2, personnel's data acquisition to be identified: gather the image that multiframe comprises personnel people's face to be identified, the image that comprises personnel people's face to be identified collecting is carried out to the detection of people's face, extract the people's face in each image and preserve, the bashful figure of people that is called personnel to be identified, and by the bashful figure of a plurality of people carry out yardstick, towards the normalized with illumination, the bashful figure of people after normalized is as original sample sequence to be identified
Figure 2013104771259100001DEST_PATH_IMAGE003
, wherein J is the number of times of personnel's man face image acquiring to be identified;
Step 3, recognition of face: personnel to be identified enter identified region and carry out after data acquisition, according to personnel identity information inquiry original licensed sample sequence to be identified, obtain registration sample sequence corresponding to personnel to be identified
Figure 373288DEST_PATH_IMAGE004
; Calculate registration sample sequence
Figure DEST_PATH_IMAGE005
orthogonal subspaces
Figure 814764DEST_PATH_IMAGE006
with sample sequence to be identified
Figure 723290DEST_PATH_IMAGE003
orthogonal subspaces , calculate two sub spaces with
Figure 346350DEST_PATH_IMAGE007
similarity measure, according to the similarity measure of two sub spaces, judge that whether people's face to be identified consistent with registrant's face.
2. the face identification method based on multiple image as claimed in claim 1, it is characterized in that described method also comprises: when carrying out authentication, while confirming whether people to be identified is he or she, registrant's face of people's face to be identified and appointment carries out subspace and compares, subspace similarity measure compares with the threshold value of setting in advance, thereby judges.
3. the face identification method based on multiple image as claimed in claim 1, it is characterized in that described method also comprises: when carrying out identification, while confirming that whom people to be identified is, registrant's face in people's face to be identified and database carries out subspace one by one to be compared, or finds out the registrant face classification the most similar to it by sorter.
4. the face identification method based on multiple image as claimed in claim 1, is characterized in that the method for the similarity measure between described judgement two sub spaces is for judging the main included angle cosine between two sub spaces.
5. the face identification method based on multiple image as claimed in claim 4, is characterized in that the detailed process of the main included angle cosine between described judgement two sub spaces is: the angle of setting between two sub spaces is
Figure DEST_PATH_IMAGE009
, the angle between two sub spaces
Figure 591517DEST_PATH_IMAGE009
be less than the angle threshold value of setting
Figure 560741DEST_PATH_IMAGE010
time, judge that personnel to be identified are consistent with accredited personnel, identify successfully; Otherwise judge that personnel to be identified and accredited personnel are inconsistent, recognition failures; Wherein
Figure DEST_PATH_IMAGE011
.
6. the face identification method based on multiple image as claimed in claim 1, is characterized in that described people's face testing process is specific as follows:
First adopt Haar characteristic image to travel through in the image of 20 * 20, the pixel of white portion and deduct black region pixel and, the value obtaining is referred to as face characteristic value;
Then from the face database of registration in advance, cut out a large amount of people's face picture and background picture, as training sample; Training sample is normalized to the image of 20 * 20 sizes, in big or small like this picture, extracts Haar eigenwert; These Haar features have formed Weak Classifier, and its function expression is:
Figure 230233DEST_PATH_IMAGE012
Wherein:
Figure DEST_PATH_IMAGE013
the image of 20 * 20 sizes, for Haar feature,
Figure DEST_PATH_IMAGE015
for symbol designator,
Figure 497714DEST_PATH_IMAGE016
for threshold value.
7. the face identification method based on multiple image as claimed in claim 6, is characterized in that described method also comprises training strong classifier, and the training process of strong classifier need to pass through
Figure DEST_PATH_IMAGE017
inferior iteration, its detailed process is as follows:
(1) given training sample set
Figure 383762DEST_PATH_IMAGE018
, altogether
Figure 414034DEST_PATH_IMAGE002
individual sample, wherein
Figure DEST_PATH_IMAGE019
with
Figure 382603DEST_PATH_IMAGE020
correspond respectively to positive sample and negative sample; maximum cycle for training;
(2) initialization sample weight is
Figure DEST_PATH_IMAGE021
, be the initial probability distribution of training sample;
(3) iteration training for the first time
Figure 913390DEST_PATH_IMAGE002
individual sample, obtains first optimum Weak Classifier;
(4) improve last round of in the weight of misjudged sample;
(5) new sample and last time are put together and carried out the training of a new round by the sample of misclassification;
(6) 4-5 step is carried out in circulation, after wheel, obtain
Figure 245462DEST_PATH_IMAGE017
individual optimum Weak Classifier;
(7) combination
Figure 760757DEST_PATH_IMAGE017
individual optimum Weak Classifier obtains strong classifier, and array mode is as follows
Figure 808347DEST_PATH_IMAGE022
8. the face identification method based on multiple image as claimed in claim 1, it is characterized in that described method also comprises face characteristic leaching process, the registrant's face collecting in step 1 and step 2 and people's face to be identified are carried out carrying out the recognition of face of step 3 after feature extraction again.
9. the face identification method based on multiple image as claimed in claim 8, is characterized in that described feature extracting method is principal component analysis (PCA).
CN201310477125.9A 2013-10-14 2013-10-14 Face recognition method based on multi-frame images Pending CN103530648A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310477125.9A CN103530648A (en) 2013-10-14 2013-10-14 Face recognition method based on multi-frame images

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310477125.9A CN103530648A (en) 2013-10-14 2013-10-14 Face recognition method based on multi-frame images

Publications (1)

Publication Number Publication Date
CN103530648A true CN103530648A (en) 2014-01-22

Family

ID=49932644

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310477125.9A Pending CN103530648A (en) 2013-10-14 2013-10-14 Face recognition method based on multi-frame images

Country Status (1)

Country Link
CN (1) CN103530648A (en)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104091160A (en) * 2014-07-14 2014-10-08 成都万维图新信息技术有限公司 Human face detecting method
CN104794459A (en) * 2015-05-07 2015-07-22 北京丰华联合科技有限公司 Video personnel identification method
CN106331589A (en) * 2015-06-26 2017-01-11 神讯电脑(昆山)有限公司 Webcam system and remote monitoring method thereof
CN106909905A (en) * 2017-03-02 2017-06-30 中科视拓(北京)科技有限公司 A kind of multi-modal face identification method based on deep learning
CN107133575A (en) * 2017-04-13 2017-09-05 中原智慧城市设计研究院有限公司 A kind of monitor video pedestrian recognition methods again based on space-time characteristic
CN107153820A (en) * 2017-05-10 2017-09-12 电子科技大学 A kind of recognition of face and movement locus method of discrimination towards strong noise
CN107563360A (en) * 2017-09-30 2018-01-09 百度在线网络技术(北京)有限公司 Information acquisition method and device
CN109993028A (en) * 2017-12-29 2019-07-09 技嘉科技股份有限公司 Human face recognition device and method, the method for promoting image identification rate
CN111291638A (en) * 2020-01-19 2020-06-16 上海云从汇临人工智能科技有限公司 Object comparison method, system, equipment and medium
CN111325131A (en) * 2020-02-14 2020-06-23 杭州电子科技大学 Micro-expression detection method based on self-adaptive transition frame removing depth network
WO2020215229A1 (en) * 2019-04-24 2020-10-29 深圳市汇顶科技股份有限公司 Face registration method, face registration device, server and storage medium
CN112308914A (en) * 2020-03-06 2021-02-02 北京字节跳动网络技术有限公司 Method, apparatus, device and medium for processing information

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1971639A (en) * 2005-11-22 2007-05-30 上海银晨智能识别科技有限公司 Method for preventing embezzlement of bank card and system to realize the same method
CN101216884A (en) * 2007-12-29 2008-07-09 北京中星微电子有限公司 A method and system for face authentication
CN101976356A (en) * 2010-09-30 2011-02-16 惠州市华阳多媒体电子有限公司 Face identification method for surfing identification policy, and identification system thereof
CN102142083A (en) * 2011-04-18 2011-08-03 苏州市慧视通讯科技有限公司 Face recognition method based on LDA (Linear Discriminant Analysis) subspace learning
US8050465B2 (en) * 2006-08-11 2011-11-01 DigitalOptics Corporation Europe Limited Real-time face tracking in a digital image acquisition device
CN103246877A (en) * 2013-05-13 2013-08-14 北京工业大学 Image contour based novel human face recognition method

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1971639A (en) * 2005-11-22 2007-05-30 上海银晨智能识别科技有限公司 Method for preventing embezzlement of bank card and system to realize the same method
US8050465B2 (en) * 2006-08-11 2011-11-01 DigitalOptics Corporation Europe Limited Real-time face tracking in a digital image acquisition device
CN101216884A (en) * 2007-12-29 2008-07-09 北京中星微电子有限公司 A method and system for face authentication
CN101976356A (en) * 2010-09-30 2011-02-16 惠州市华阳多媒体电子有限公司 Face identification method for surfing identification policy, and identification system thereof
CN102142083A (en) * 2011-04-18 2011-08-03 苏州市慧视通讯科技有限公司 Face recognition method based on LDA (Linear Discriminant Analysis) subspace learning
CN103246877A (en) * 2013-05-13 2013-08-14 北京工业大学 Image contour based novel human face recognition method

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
吴贤星: "人脸信息多尺度几何分析与应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
方娟: "人机交互中人脸识别方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
李勇: "基于RFID与人脸识别技术的智能门禁系统研究与设计", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
李阳: "人脸特征子空间方法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *
王慧泽: "基于AdaBoost算法和Fisher线性准则的人脸识别研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104091160B (en) * 2014-07-14 2017-12-08 成都万维图新信息技术有限公司 A kind of method for detecting human face
CN104091160A (en) * 2014-07-14 2014-10-08 成都万维图新信息技术有限公司 Human face detecting method
CN104794459A (en) * 2015-05-07 2015-07-22 北京丰华联合科技有限公司 Video personnel identification method
CN106331589A (en) * 2015-06-26 2017-01-11 神讯电脑(昆山)有限公司 Webcam system and remote monitoring method thereof
CN106909905A (en) * 2017-03-02 2017-06-30 中科视拓(北京)科技有限公司 A kind of multi-modal face identification method based on deep learning
CN106909905B (en) * 2017-03-02 2020-02-14 中科视拓(北京)科技有限公司 Multi-mode face recognition method based on deep learning
CN107133575A (en) * 2017-04-13 2017-09-05 中原智慧城市设计研究院有限公司 A kind of monitor video pedestrian recognition methods again based on space-time characteristic
CN107133575B (en) * 2017-04-13 2020-06-12 中原智慧城市设计研究院有限公司 Monitoring video pedestrian re-identification method based on space-time characteristics
CN107153820A (en) * 2017-05-10 2017-09-12 电子科技大学 A kind of recognition of face and movement locus method of discrimination towards strong noise
US10740597B2 (en) 2017-09-30 2020-08-11 Baidu Online Network Technology (Beijing) Co., Ltd. Method and apparatus for acquiring information
CN107563360A (en) * 2017-09-30 2018-01-09 百度在线网络技术(北京)有限公司 Information acquisition method and device
CN109993028A (en) * 2017-12-29 2019-07-09 技嘉科技股份有限公司 Human face recognition device and method, the method for promoting image identification rate
WO2020215229A1 (en) * 2019-04-24 2020-10-29 深圳市汇顶科技股份有限公司 Face registration method, face registration device, server and storage medium
CN111291638A (en) * 2020-01-19 2020-06-16 上海云从汇临人工智能科技有限公司 Object comparison method, system, equipment and medium
CN111325131A (en) * 2020-02-14 2020-06-23 杭州电子科技大学 Micro-expression detection method based on self-adaptive transition frame removing depth network
CN111325131B (en) * 2020-02-14 2023-06-23 杭州电子科技大学 Micro-expression detection method based on self-adaptive transition frame depth network removal
CN112308914A (en) * 2020-03-06 2021-02-02 北京字节跳动网络技术有限公司 Method, apparatus, device and medium for processing information
CN112308914B (en) * 2020-03-06 2024-07-12 北京字节跳动网络技术有限公司 Method, apparatus, device and medium for processing information

Similar Documents

Publication Publication Date Title
CN103530648A (en) Face recognition method based on multi-frame images
US10672140B2 (en) Video monitoring method and video monitoring system
Shi et al. Real-time traffic light detection with adaptive background suppression filter
Ma et al. A reliable people counting system via multiple cameras
CN106934359B (en) Multi-view gait recognition method and system based on high-order tensor subspace learning
US9008365B2 (en) Systems and methods for pedestrian detection in images
CN104091147B (en) A kind of near-infrared eyes positioning and eye state identification method
CN102855496A (en) Method and system for authenticating shielded face
CN102902986A (en) Automatic gender identification system and method
KR101222100B1 (en) Apparatus for detecting frontal face
CN106650574A (en) Face identification method based on PCANet
CN111259756A (en) Pedestrian re-identification method based on local high-frequency features and mixed metric learning
Raj et al. Vacant parking lot detection system using random forest classification
Guofeng et al. Traffic sign recognition based on SVM and convolutional neural network
Halidou et al. Fast pedestrian detection based on region of interest and multi-block local binary pattern descriptors
Gou et al. mom: Mean of moments feature for person re-identification
Ou et al. Vehicle logo recognition based on a weighted spatial pyramid framework
Strueva et al. Student attendance control system with face recognition based on neural network
Rahman et al. LDA based paper currency recognition system using edge histogram descriptor
CN104318224A (en) Face recognition method and monitoring equipment
Subburaman et al. Alternative search techniques for face detection using location estimation and binary features
Desai et al. Real-time implementation of Indian license plate recognition system
KR101686246B1 (en) System, method and program product for camera-based object analysis
Satsangi et al. License plate recognition: a comparative study on thresholding, OCR and machine learning approaches
Abaza et al. Human ear detection in the thermal infrared spectrum

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20140122