CN101923641B - Improved human face recognition method - Google Patents

Improved human face recognition method Download PDF

Info

Publication number
CN101923641B
CN101923641B CN2010102766799A CN201010276679A CN101923641B CN 101923641 B CN101923641 B CN 101923641B CN 2010102766799 A CN2010102766799 A CN 2010102766799A CN 201010276679 A CN201010276679 A CN 201010276679A CN 101923641 B CN101923641 B CN 101923641B
Authority
CN
China
Prior art keywords
human face
plane
image
unique point
error amount
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
Application number
CN2010102766799A
Other languages
Chinese (zh)
Other versions
CN101923641A (en
Inventor
王忠立
宋永瑞
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.)
Beijing Jiaotong University
Original Assignee
Beijing Jiaotong University
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 Beijing Jiaotong University filed Critical Beijing Jiaotong University
Priority to CN2010102766799A priority Critical patent/CN101923641B/en
Publication of CN101923641A publication Critical patent/CN101923641A/en
Application granted granted Critical
Publication of CN101923641B publication Critical patent/CN101923641B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

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
    • G06V40/172—Classification, e.g. identification
    • 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
    • G06V40/40—Spoof detection, e.g. liveness detection
    • G06V40/45—Detection of the body part being alive

Abstract

The invention discloses an improved human face recognition method, which belongs to the technical field of human face recognition. The improved human face recognition method comprises the following steps of: processing two or more acquired human face images, extracting a plurality of 2D characteristic points of the human face areas of the images, and establishing a corresponding relation of the human face 2D characteristic points of different images; judging whether the extracted human face 2D characteristic points belong to the same space plane or not according to the restriction conditions which the projections of the points of the same space plane on different images meet in a 3D space; and accordingly, judging whether the acquired images are plane scenes or not so as to determine whether the human faces are from other 2D pictures or not and prevent a human face recognition system from being deceived by other pictures. The method solves the problem that the traditional human face recognition system is easily deceived by the pictures, and improves the reliability of the human face recognition system by determining whether an input image is from a real human face or a picture.

Description

A kind of improved face identification method
Technical field
The present invention relates to 2D graphical analysis and treatment technology in the biometrics identification technology, extract minutiae from two width of cloth or several 2D images specifically, and judge whether it belongs to the method for the same space planar point.
Background technology
Face recognition technology is a kind of in the biometrics identification technology.Each country all takes much count of face recognition technology at present, and many major companies have also released the identity identifying technology based on recognition of face, have wide practical use in fields such as video monitoring, multimedia, process control, identifications.Along with this The Application of Technology increases gradually, along with and shortcoming on some recognition technologies of coming also is utilized.Traditional face recognition technology based on characteristic through the unique point of extraction human face region, and according to certain the intrinsic relation between the special unique point, is discerned.Its basis is the relation between the unique point on the image, and has ignored the space attribute of unique point.This system is easy to cheated by photo in application, and its reliability and security are received and had a strong impact on.Outstanding feature of the present invention is; In the conplane image that under different points of view, obtains in the space, should satisfy the isoplanar constraint, adopt stable operator; Like LMEDS; Calculate the plane homography matrix reliably according to character pair point, and based on this, carry out the isoplanar characteristic of unique point and differentiate.
Summary of the invention
The object of the present invention is to provide; Utilization is estimated the isoplanar attribute of two-dimensional images unique point; Whether the area of space corresponding with judging characteristic point is same plane; Reliably whether judge image from photo, but not actual living body faces, and then the technology of raising people face detecting reliability.
In order to achieve the above object; Technical solution of the present invention provides a kind of improved face identification method; It is characterized in that; Should satisfy plane homography matrix constraint according to the corresponding point of isoplanar, space o'clock on two width of cloth planes of delineation, the unique point of the human face region that extracts carried out the isoplanar constraint differentiate, to confirm whether facial image obtains from plane picture.
Described method, it comprises the following steps:
Step 1 is obtained the image of two width of cloth or several same people's faces through shooting;
Step 2 is carried out feature extraction to the image that obtains and is handled, and obtains a plurality of 2D unique points, utilizes the vision matching technology to set up the corresponding relation of unique point;
Step 3, the character pair point on arbitrary extracting two width of cloth images calculates the plane homography matrix; The error amount of the unique point that the plane homography matrix calculation procedure 2 that utilization calculates is obtained;
Step 4; Set an error higher limit; If the error amount of the unique point that step 3 calculates is smaller or equal to the error higher limit, then description of step 2 described unique points are positioned at same plane, if the error amount of unique point is greater than the error higher limit of setting then change step 5 over to;
Step 5 is determined the human face region in the image through method for detecting human face, and the unique point in the human face region is used to recomputate the plane homography matrix; And the error amount of the interior unique point of calculating human face region; If error amount is then judged unique point and belonged to same plane smaller or equal to the error amount of setting, the spatial point that human face region is corresponding is the plane; If error amount is greater than the error amount of setting, then judging the image that photographs is real facial image.
The said image of step 1 is same people's face imaging to be obtained under diverse location by single camera, or takes acquisition by two or more video cameras at synchronization, or video camera is fixed the image to the people's face acquisition time that moves
The calculating of above-mentioned plane homography matrix adopts least square method to estimate or the LMEDS algorithm.
The advantage of the more existing face recognition technology of the present invention is:
When carrying out recognition of face; Space plane attribute to the unique point that is used to discern is differentiated; Confirm whether these unique points belong to same plane; Screen picture and actual facial image with this, thereby avoid, improved security and reliability during this type systematic uses by the photograph image deception.The method that the present invention proposes is directly handled two dimensional image, scene is not set the constraint condition of priori, has more ubiquity.
Description of drawings
Fig. 1 is the face identification method process flow diagram that the present invention proposes.
Embodiment
Below in conjunction with accompanying drawing and embodiment the present invention is described further.
As shown in Figure 1, a kind of improved face identification method may further comprise the steps:
1). obtain the image of two width of cloth or several people's faces;
2). the image that obtains is carried out feature point extraction handle, obtain a plurality of 2D unique points, utilize the vision matching technology to set up the corresponding relation of unique point;
3). the character pair point on arbitrary extracting two width of cloth images, calculate plane homography matrix H;
4). calculate the error amount of all character pair points;
5). set an error higher limit, if calculate error amount in the step 4 smaller or equal to setting value, then unique point that participate in to calculate of explanation belongs to the isoplanar point, if the error amount of unique point greater than setting value change over to step next step;
6). determine the unique point of people's face surveyed area with method for detecting human face, and utilize unique point to calculate the plane homography matrix;
7). calculate the corresponding error of face characteristic provincial characteristics point;
8). set an error higher limit, whether the error that determining step 7 draws is smaller or equal to this setting value;
9). if then unique point belongs to the isoplanar point, if otherwise unique point does not belong to the isoplanar point, and be real facial image.
Principle of the present invention is: same video camera is carried out to picture to same people's face under two diverse locations; Perhaps two video cameras are simultaneously to people's face imaging, and perhaps video camera is fixing to mobile people's face acquisition time image, and principle is the same; Belong to the point in the same space plane in the scene; Coordinate position in two width of cloth images should satisfy plane homography matrix constraint, that is:
p i ′ ∝ Hp i
Here p ' i, p iBe on two width of cloth images, the image coordinate that the same space point is corresponding, H is the plane homography matrix.
In the space, three points on same straight line can uniquely not determine a plane, and space plane can be used parameter [n T,-d] TDescribe, wherein n is the normal vector of space plane, and d is that space plane is the distance of round dot to video camera.This space plane parameter can be used three unique points, calculates with linear method.
On the plane of delineation, defining point to the distance on plane is:
e i 2 ( p i , p i ′ , H ) = | | p i - H - 1 p i ′ | | 2 + | | p i ′ - Hp i | | 2 .
When detecting, at first the image that obtains is extracted the 2D unique point, and set up the corresponding relation between the characteristic on the plane.
To feature of interest point on the plane of delineation and the corresponding point in other image thereof,, calculate the right distance error value of each point according to the range formula of putting the plane.If error amount all is not more than the error higher limit of a setting, explain that then the corresponding spatial point of these unique points is the point on the same plane, conclude that thus the image that is obtained is from 2D plane facial images such as pictures.
Have greater than the error amount of setting if calculate the error amount of unique point this moment, then determine human face region characteristic of correspondence point, and utilize the unique point in the human face region to recomputate the plane homography matrix with method for detecting human face; Calculate the corresponding error of unique point in the human face region then; Set an error higher limit, whether the error of unique point is smaller or equal to this setting value in the determining step human face region.Belong to the isoplanar point if the unique point of human face region then is described, the image that is obtained is to be obtained by picture etc.; If otherwise unique point does not belong to the isoplanar point, can conclude that then the image that obtains is from real people's face.
Calculate the plane homography matrix for matched feature points, can adopt LMEDS to stablize method of estimation to improve the robustness of result of calculation.
In general; The location of unique point is noisy in the image characteristics extraction process; In addition, the correspondence of unique point is set up process, may produce wrong factors such as match point; The present invention has reduced the influence of these factors to the plane testing result through adopting the robust method for parameter estimation, makes the result more reliable.
Embodiment 1
A video camera is fixed on the support.During experiment, the people is positioned at suitable position, video camera the place ahead, because human body has light exercise, two width of cloth facial images of inscribing when video camera obtains two respectively.
Two width of cloth images that obtain are carried out feature point extraction handle, establish in first width of cloth image, the unique point of extraction is p i(i=1,2,3 ..., n), in second width of cloth image, the unique point of obtaining is p ' i(i=1,2,3 ..., m).Utilize the stereoscopic vision matching technique, set up the corresponding relation of these unique points:
p i ′ ↔ p i (i＝1，2，3，...，k)
According to the isoplanar equation of constraint,
Figure GSB00000799921400052
adopts least square method to estimate to obtain the value of plane homography matrix H.
Calculate the error amount of each character pair point:
e i 2 ( p i , p i ′ , H ) = | | p i - H - 1 p i ′ | | 2 + | | p i ′ - Hp i | | 2 (i＝1，2，3，...k)
If
e i≤ε t(i＝1，2，3，...，k)
In the formula, ε tBe the error amount of setting.Think that then these unique points are corresponding to same space plane.
All smaller or equal to setting value, then utilize human face detection tech in two width of cloth images, to detect human face region if not the error of all unique points, the unique point that belongs to human face region in the matched feature points that has obtained is screened.Only the character pair point with human face region recomputates plane homography matrix H f, and calculate the error amount of human face region character pair point, set an error higher limit ε TfIf:
e i≤ε tf (i＝1，2，3，...，m)
In the formula, m thinks then that for the corresponding matched feature points number that belongs to human face region these unique points of human face region are corresponding to same space plane.Otherwise think that the image that is obtained comes from real people's face.
Embodiment 2
Obtaining of facial image is that two video cameras have a certain degree, and takes same individual face simultaneously.To the image extract minutiae taken and set up corresponding relation, utilize the LMEDS method of good reliability to estimate to obtain the H matrix.Calculate each unique point error amount and with the error higher limit of setting relatively; If be not more than setting value then unique point belong to the isoplanar point; If greater than setting value; Then utilize human face detection tech to detect human face region, the unique point of extracting in the human face region recomputates the plane homography matrix, calculates the error amount of human face region unique point and compares with the error amount of setting; If then think these points of human face region corresponding to same plane smaller or equal to setting value, if greater than setting value then the image that obtains of decidable from real people's face.
The present invention can realize stablizing, carrying out reliably the plane attribute differentiation of unique point, can prevent to be cheated by the input picture in 2D images such as photo source in the recognition of face.

Claims (3)

1. an improved face identification method is characterized in that, comprises following key step:
Step 1 is obtained the image of two width of cloth or several same people's faces through shooting;
Step 2 is carried out feature extraction to the image that obtains and is handled, and obtains a plurality of 2D unique points, utilizes the vision matching technology to set up the corresponding relation of unique point;
Step 3, the character pair point on arbitrary extracting two width of cloth images calculates the plane homography matrix; The error amount of the unique point that the plane homography matrix calculation procedure 2 that utilization calculates is obtained;
Step 4; Set an error higher limit; If the error amount of the unique point that step 3 calculates is smaller or equal to the error higher limit, then the unique point obtained of description of step 2 is positioned at same plane, if the error amount of unique point is greater than the error higher limit of setting then change step 5 over to;
Step 5 is determined the human face region in the image through method for detecting human face, and the unique point in the human face region is used to recomputate the plane homography matrix; And the error amount of the interior unique point of calculating human face region; If error amount is then judged unique point and belonged to same plane smaller or equal to the error amount of setting, the spatial point that human face region is corresponding is the plane; If error amount is greater than the error amount of setting, then judging the image that photographs is real facial image.
2. face identification method according to claim 1; It is characterized in that: the said image of step 1 is same people's face imaging to be obtained under diverse location by single camera; Or take to obtain at synchronization, or the fixing image of video camera to people's face acquisition time of moving by two or more video cameras.
3. face identification method according to claim 1 is characterized in that: the calculating of said plane homography matrix adopts least square method to estimate or the LMEDS algorithm.
CN2010102766799A 2010-09-09 2010-09-09 Improved human face recognition method Expired - Fee Related CN101923641B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2010102766799A CN101923641B (en) 2010-09-09 2010-09-09 Improved human face recognition method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2010102766799A CN101923641B (en) 2010-09-09 2010-09-09 Improved human face recognition method

Publications (2)

Publication Number Publication Date
CN101923641A CN101923641A (en) 2010-12-22
CN101923641B true CN101923641B (en) 2012-08-08

Family

ID=43338565

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2010102766799A Expired - Fee Related CN101923641B (en) 2010-09-09 2010-09-09 Improved human face recognition method

Country Status (1)

Country Link
CN (1) CN101923641B (en)

Families Citing this family (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102194106B (en) * 2011-05-11 2013-01-16 西安理工大学 Human face recognition method used in door access system
FR2997211B1 (en) * 2012-10-18 2021-01-01 Morpho PROCESS FOR AUTHENTICATION OF AN IMAGE CAPTURE OF A THREE-DIMENSIONAL ENTITY
CN103106703A (en) * 2013-01-14 2013-05-15 张平 Anti-cheating driver training recorder
CN103051887A (en) * 2013-01-23 2013-04-17 河海大学常州校区 Eagle eye-imitated intelligent visual sensing node and work method thereof
CN103971132A (en) * 2014-05-27 2014-08-06 重庆大学 Method for face recognition by adopting two-dimensional non-negative sparse partial least squares
CN104217504A (en) * 2014-08-26 2014-12-17 杭州摩科商用设备有限公司 Identity recognition self-service terminal and corresponding certificate of house property printing terminal
CN104217503A (en) * 2014-08-26 2014-12-17 杭州摩科商用设备有限公司 Self-service terminal identity identification method and corresponding house property certificate printing method
US10198645B2 (en) * 2014-11-13 2019-02-05 Intel Corporation Preventing face-based authentication spoofing
CN111144293A (en) * 2015-09-25 2020-05-12 北京市商汤科技开发有限公司 Human face identity authentication system with interactive living body detection and method thereof
CN106557726B (en) * 2015-09-25 2020-06-09 北京市商汤科技开发有限公司 Face identity authentication system with silent type living body detection and method thereof
CN105868733A (en) * 2016-04-21 2016-08-17 腾讯科技（深圳）有限公司 Face in-vivo validation method and device
CN106897675B (en) * 2017-01-24 2021-08-17 上海交通大学 Face living body detection method combining binocular vision depth characteristic and apparent characteristic
CN107368817B (en) * 2017-07-26 2020-02-21 湖南云迪生物识别科技有限公司 Face recognition method and device
CN107437120A (en) * 2017-09-22 2017-12-05 南京多伦科技股份有限公司 The management system and management method of driving training
US10776609B2 (en) 2018-02-26 2020-09-15 Samsung Electronics Co., Ltd. Method and system for facial recognition
CN110751757A (en) * 2019-09-11 2020-02-04 河海大学 Unlocking method based on face image processing and intelligent lock
CN111339958B (en) * 2020-02-28 2023-08-29 南京鑫之派智能科技有限公司 Face living body detection method and system based on monocular vision

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101196986A (en) * 2007-12-25 2008-06-11 哈尔滨工业大学 Three-dimensional palm print identity identifier and its identification method
US7403658B2 (en) * 2005-04-15 2008-07-22 Microsoft Corporation Direct homography computation by local linearization

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2006138525A2 (en) * 2005-06-16 2006-12-28 Strider Labs System and method for recognition in 2d images using 3d class models

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7403658B2 (en) * 2005-04-15 2008-07-22 Microsoft Corporation Direct homography computation by local linearization
CN101196986A (en) * 2007-12-25 2008-06-11 哈尔滨工业大学 Three-dimensional palm print identity identifier and its identification method

Also Published As

Publication number Publication date
CN101923641A (en) 2010-12-22

Similar Documents

Publication Publication Date Title
CN101923641B (en) Improved human face recognition method
CN105243386B (en) Face living body judgment method and system
EP2426642B1 (en) Method, device and system for motion detection
CN109784130B (en) Pedestrian re-identification method, device and equipment thereof
CN110796079A (en) Multi-camera visitor identification method and system based on face depth features and human body local depth features
CN109598242B (en) Living body detection method
US8933783B2 (en) Detection of fraud for access control system of biometric type
CN106407875A (en) Target feature extraction method and apparatus
CN106033614B (en) A kind of mobile camera motion object detection method under strong parallax
CN105631515A (en) People flow counting system
US20160335523A1 (en) Method and apparatus for detecting incorrect associations between keypoints of a first image and keypoints of a second image
CN110991266A (en) Binocular face living body detection method and device
CN112926464B (en) Face living body detection method and device
Argyros et al. Binocular hand tracking and reconstruction based on 2D shape matching
Krinidis et al. A robust and real-time multi-space occupancy extraction system exploiting privacy-preserving sensors
KR100922429B1 (en) Pose robust human detection and tracking method using the stereo image
CN109767826A (en) A kind of acquisition methods and medical photography imaging system of medical treatment photographed data
Yu et al. Accurate motion detection in dynamic scenes based on ego-motion estimation and optical flow segmentation combined method
Zhang et al. Detecting photographic composites using two-view geometrical constraints
Yan et al. Exposing photo manipulation with inconsistent perspective geometry
CN110119692B (en) Planar portrait detection method based on double cameras
KR101711307B1 (en) Portable and Computer Equipment Unlock System using Depth Face Recognition
Pandey et al. A detailed analysis on feature extraction techniques of panoramic image stitching algorithm
D’Apuzzo et al. Three-dimensional human face feature extraction from multi images
CN110232353B (en) Method and device for acquiring depth position of scene personnel

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
C17 Cessation of patent right
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20120808

Termination date: 20130909