CN102622588B - Dual-certification face anti-counterfeit method and device - Google Patents

Dual-certification face anti-counterfeit method and device Download PDF

Info

Publication number
CN102622588B
CN102622588B CN2012100594547A CN201210059454A CN102622588B CN 102622588 B CN102622588 B CN 102622588B CN 2012100594547 A CN2012100594547 A CN 2012100594547A CN 201210059454 A CN201210059454 A CN 201210059454A CN 102622588 B CN102622588 B CN 102622588B
Authority
CN
China
Prior art keywords
face
target
image
step
real
Prior art date
Application number
CN2012100594547A
Other languages
Chinese (zh)
Other versions
CN102622588A (en
Inventor
李子青
张志炜
雷震
易东
Original Assignee
无锡中科奥森科技有限公司
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 无锡中科奥森科技有限公司 filed Critical 无锡中科奥森科技有限公司
Priority to CN2012100594547A priority Critical patent/CN102622588B/en
Publication of CN102622588A publication Critical patent/CN102622588A/en
Application granted granted Critical
Publication of CN102622588B publication Critical patent/CN102622588B/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/00221Acquiring or recognising human faces, facial parts, facial sketches, facial expressions
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/00885Biometric patterns not provided for under G06K9/00006, G06K9/00154, G06K9/00335, G06K9/00362, G06K9/00597; Biometric specific functions not specific to the kind of biometric
    • G06K9/00899Spoof detection
    • G06K9/00906Detection of body part being alive
    • GPHYSICS
    • G06COMPUTING; CALCULATING; COUNTING
    • G06KRECOGNITION OF DATA; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K9/00Methods or arrangements for reading or recognising printed or written characters or for recognising patterns, e.g. fingerprints
    • G06K9/78Combination of image acquisition and recognition functions

Abstract

本发明涉及一种双验证人脸防伪方法及装置,该方法包括:步骤1,对采集的目标人脸进行活体检测,判断目标人脸是否具有生物活性,如果目标人脸被认定具有活体特性,则转入步骤2;步骤2,如果是在人脸识别应用中,则计算采集到的目标人脸与识别结果对应的人脸之间的相似度,若大于某一阈值,则认为该目标人脸是真实有效的人脸;如果是在人脸验证应用中,则计算采集到的目标人脸与目标人脸所声称的身份对应的人脸之间的相似度,若大于某一阈值,则认为该目标人脸是真实有效的人脸。 The present invention relates to a dual authentication face security method and apparatus, the method comprising: Step 1, the target person acquired a face of a living body is detected, it determines a target face is biologically active, if the target face is found having a biological characteristic, the process proceeds to step 2; step 2, if it is in face recognition applications, the degree of similarity between the acquired target face and the corresponding face recognition results is calculated, if more than a certain threshold value, the target person is considered face is real and effective face; If the similarity between the face authentication application, the acquisition target to the target face and the face claimed identity is calculated corresponding to the human face, if more than a certain threshold value, the target is considered to face a real and effective human face. 利用本发明的方法,通过活体检测与身份验证的结合,提供准确、可靠的人脸防伪检测结果。 Using the method of the present invention, by binding in vivo detection and authentication, to provide accurate and reliable security face detection result.

Description

双验证人脸防伪方法及装置技术领域 Double Face authentication and anti-counterfeiting method TECHNICAL FIELD

[0001] 本发明涉及一种人脸防伪方法及装置,尤其是一种双验证人脸防伪方法及装置,属于图像处理与模式识别的技术领域。 [0001] The present invention relates to a method and device security face, in particular a double-face authentication security method and a device, belonging to the technical field of image processing and pattern recognition.

背景技术 Background technique

[0002] 人脸防伪技术关系到人脸识别认证授权系统的安全性,如果没有人脸防伪功能,人脸识别认证授权系统易受到虚假人脸的攻击,进而可能引发严重的安全问题。 [0002] anti-counterfeiting technology related to face recognition security authentication and authorization system, if no face security features, face recognition authentication and authorization systems are vulnerable to false face attack, which may lead to a serious security problem. 例如,攻击者可以通过某种手段获取某一特定目标(即指定人)的人脸图像并制成照片、视频、或面具等,呈现在系统面前,以期获得非法权限。 For example, an attacker can get by some means to a specific target (ie designee) and made a face image such as photos, videos, or mask, presented in front of the system, in order to gain unauthorized access. 因此,人脸防伪技术受到越来越多的关注。 Therefore, face counterfeiting technology gets more and more attention. 目前国际上现有的人脸防伪技术,主要基于人机交互策略:系统发出特定指令,要求用户作出眨眼、发音等特定行为,进而判断输入人脸的活性。 Currently existing international human face anti-counterfeiting technology, mainly based on human-computer interaction strategy: the system to issue a specific directive to require the user to make specific behavior blink, pronunciation and so on, and then determine the activity of the input face. 根据常见的动作可以划分为以下三种方式:第一种是基于眨眼的活体检测,公开该技术的文献有:l)Gang Pan, Lin Sun,Zhaohui Wu and Shilong La0.Eyeblink-based Ant1-Spoofing in Face Recognitionfrom a Generic Webcamera, International Conference on Computer Vision,2007,2)K.Kollreider, ·H.Fronthaler and J.Bigun.Verifying Liveness by Multiple Expertsin Face Biometrics, IEEE Conference on Computer Vision and Pattern RecognitionWorkshop,2008,3)专利号为ZL200710178088.6,发明名称为“一种基于人脸生理性运动的活体检测方法及系统”的专利文献。 The common operation may be divided into the following three ways: The first is based on in vivo detection of blinking, the technique disclosed in the literature have: l) Gang Pan, Lin Sun, Zhaohui Wu and Shilong La0.Eyeblink-based Ant1-Spoofing in Face Recognitionfrom a Generic Webcamera, International Conference on Computer Vision, 2007,2) K.Kollreider, · H.Fronthaler and J.Bigun.Verifying Liveness by Multiple Expertsin Face Biometrics, IEEE Conference on Computer Vision and Pattern RecognitionWorkshop, 2008,3) Patent No. ZL200710178088.6, entitled "based on physiological movement of a living body face detection method and system" in Patent Document. 第二种是基于摇头的活体检测,公开该技术的相关文献包括:1)K.Kollreider,H.Fronthaler and J.Bigun.Evaluating Liveness by FaceImages and the Structure Tensor, IEEE Workshop on Automatic IdentificationAdvanced Technologies,2005,2)Wei Bao, Hong Li,Nan Li and Wei Jiang.A LivenessDetection Method for Face Recognition Based on Optical Flow Field,InternationalConference on Image Analysis and Signal Processing,2009。 The second is based on the shake detection of the living body, the technology disclosed in the literature include: 1) K.Kollreider, H.Fronthaler and J.Bigun.Evaluating Liveness by FaceImages and the Structure Tensor, IEEE Workshop on Automatic IdentificationAdvanced Technologies, 2005, 2) Wei Bao, Hong Li, Nan Li and Wei Jiang.A LivenessDetection Method for Face Recognition Based on Optical Flow Field, InternationalConference on Image Analysis and Signal Processing, 2009. 第三种是基于语音及嘴部动作的活体检测,公开该技术的相关文献有:G.Chetty and M.Wagner.LivenessVerification in Audio-Video Speaker Authentication.1n IOth Australian Int.Conference on Speech Science and Technology,2004。 The third is based on the in vivo detection of voice and mouth movements, the technology disclosed in the literature have: G.Chetty and M.Wagner.LivenessVerification in Audio-Video Speaker Authentication.1n IOth Australian Int.Conference on Speech Science and Technology, 2004.

[0003] 这种基于人机交互的方法由于要求使用者表现特定行为,因此用户负担较重、用户体验不佳、所需时间较长。 [0003] This approach based on human-computer interaction performance by requiring specific user behavior, so a heavier burden on the user, the user experience is poor, required a long time.

[0004] 另外,有的研究者从多光谱的角度入手,通过分析皮肤在不同光谱下的反射率进行活体检测,相关文献有:I) 1annis Pavlidis, Peter Symosek, The Imaging Issue inan Automatic Face/Disguise Detection System, IEEE workshop on Computer VisionBeyond the Visible Spectrum !Methods and Applications,2000。2)Youngshin Kim,Jaekeun Na,Seongbeak Yoon, and Juneho Y1.Masked fake face detection usingradiance measurements, J.0pt.Soc.Am, vol.26, n0.4, April 2009。 [0004] Further, some researchers start from multispectral angle detected by the analysis performed in vivo reflectance spectra of the skin in different, literature has: I) 1annis Pavlidis, Peter Symosek, The Imaging Issue inan Automatic Face / Disguise Detection System, IEEE workshop on Computer VisionBeyond the Visible Spectrum! Methods and Applications, 2000.2) Youngshin Kim, Jaekeun Na, Seongbeak Yoon, and Juneho Y1.Masked fake face detection usingradiance measurements, J.0pt.Soc.Am, vol. 26, n0.4, April 2009. 但该种方法目前还很粗糙,精度上也并不理想,还有很大的改进空间。 But the moment it is still very rough, the precision is not satisfactory, there is still much room for improvement.

[0005] 以上所述的人脸防伪方法,亦可以成为活体检测技术,因为他们都只判断目标人脸是否具有生物活性。 Face anti-counterfeiting method [0005] described above, can also be detected in vivo techniques, because they target only determines whether the face having biological activity. 然而,实际应用中,有可能出现真实人员去仿冒攻击指定人的情况,此时目标人脸确实为真实人脸,但是仍然属于攻击人脸识别系统的行为。 However, practical applications, there are real possibilities to phishing attacks nominee who appears, at this time the target person face to face is indeed real people, but still belongs to the attacks on face recognition system. 因此人脸防伪技术不应仅仅包含活体检测。 Thus anti-counterfeiting technology face detection should not only contain living. 并且上述方法普遍存在用户负担重、人机交互时间长、准确度不高等缺点,因此开发准确、快速、适用范围广的人脸防伪方法势在必行。 And the above method common user burden, human-computer interaction for a long time, the accuracy is not high shortcomings, so the development of accurate, fast, wide application of face anti-counterfeiting method is imperative.

发明内容 SUMMARY

[0006] 本发明的目的是克服现有技术中存在的不足,提供一种双验证人脸防伪方法及装置,其提高识别精确度,方便,安全可靠。 [0006] The object of the present invention is to overcome the disadvantages of the prior art, there is provided a dual face authentication security method and device, which improves recognition accuracy, convenient, safe and reliable.

[0007] 按照本发明提供的技术方案,一种双验证人脸防伪方法,所述方法包括: [0007] The technical solution according to the present invention provides a double-face authentication security, the method comprising:

[0008] 步骤1,对采集的目标人脸进行活体检测,判断目标人脸是否具有生物活性,如果目标人脸被认定具有活体特性,则转入步骤2 ; [0008] Step 1, the target acquisition human face detection performed in vivo, it is determined whether the target face having biological activity, if the target face is found having a biological characteristic, the process proceeds to step 2;

[0009] 步骤2,如果是在人脸识别应用中,则计算采集到的目标人脸与识别结果对应的人脸之间的相似度,若大于某一阈值,则认为该目标人脸是真实有效的人脸; [0009] Step 2, if it is in face recognition applications, the degree of similarity between the acquired target face and the corresponding face recognition results is calculated, if more than a certain threshold, the target face is considered to be true effective human face;

[0010] 如果是在人脸验证应用中,则计算采集到的目标人脸与目标人脸所声称的身份对应的人脸之间的相似度,若大于某一阈值,则认为该目标人脸是真实有效的人脸, Similarity between the [0010] If the face authentication application, the target is calculated collected target face and the face corresponding to the claimed identity of the human face, if more than a certain threshold, the target face is considered is a real and effective human face,

[0011] 其中步骤I与指定人无关,步骤2与指定人有关,当目标人脸同时通过步骤I和步骤2的验证之后,才能被认定为是真实有效的人脸,否则被认定为是虚假人脸。 [0011] wherein the step I nothing to do with the nominee, and the nominee about after step 2, when the target face and to verify step by step I and 2, in order to be recognized as a real and effective human face, otherwise identified as false human face.

[0012] 如果所述双验证人脸防伪方法是基于可见光,则步骤I进一步包括: [0012] If the verification of the dual anti-counterfeiting method is based on the face visible, I further comprising the step of:

[0013] 步骤101,对目标人脸进行活体检测,首先采集大量真实、虚假人脸样本,对目标人脸提取各种纹理特征,训练活体检测纹理分类器,若目标人脸被活体检测纹理分类器认定为真实人脸,则进入步骤2,否则认定为虚假人脸; [0013] In step 101, the target human faces in vivo detection, first collected a large number of true and false face samples, the target face extracted texture features, training vivo detection of texture classification, if the target face is life detecting texture classification is recognized as a real human face, go to step 2, otherwise identified as false face;

[0014] 步骤102,通过人机交互确定目标人脸的有效性,其中系统发出指令,要求用户做出一定的动作,然后系统不断检测目标人脸是否做出相应动作,若在一定时间内检测到上述动作的发生,则判断目标人脸为真实人脸,否则为虚假人脸; [0014] Step 102, determining the validity of the target computer interaction face, wherein the system instructs the user to make certain actions required, then the system continuously made as to whether the detection target face corresponding actions, detected during a certain time the occurrence of the actions described above, it is determined that the target person face to face real people, who otherwise face is false;

[0015] 只有目标人脸同时通过步骤101和102,才被认为通过步骤I的活体检测。 [0015] Only by the target face while steps 101 and 102, was considered by the step of I in vivo detection.

[0016] 步骤2进一步包括: [0016] Step 2 further comprises:

[0017] 步骤201,首先采集大量真实人脸图像,对每张人脸图像提取其纹理特征; [0017] Step 201, collecting a large number of real first face images extracted texture features of each face image;

[0018] 步骤202,然后将采集的所有人脸图像的特征向量两两相减,根据两图像是否属于同一个人,将相减后的特征向量分为类内、类间两类,利用机器学习算法训练一个两类分类器,由此训练得到的分类器可以判断输入的两个特征向量是否属于同一个人; [0018] Step 202, and the image feature vectors of all faces of the two acquired two subtraction, depending on whether two images belong to the same person, the feature vectors after the subtraction is divided into categories, between the two categories, using a machine learning training a classifier algorithm types, whereby the trained classifier can be determined whether the two input feature vectors belong to the same person;

[0019] 步骤203,如果是在人脸识别应用中,若目标人脸图像与识别结果对应的人脸图像,被步骤202中的分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸; [0019] Step 203, the face recognition application, if the target image and the face recognition result corresponding to the face image, has been identified in step 202 is classified as belonging to the same person, the target face is considered true and valid, otherwise, false face for the people;

[0020] 如果是在人脸验证应用中,则目标人脸图像与所声称的指定人身份对应的人脸图像,被步骤202中的分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸。 [0020] If it is in the face verification applications, the target face image with the claimed identity of the nominee corresponding face image, was identified in step 202 is classified as belonging to the same person, then that person faces a real and effective target otherwise, people face is false.

[0021] 如果所述双验证人脸防伪方法是基于多模态,则步骤I进一步包括: [0021] If the security of the dual face verification method is based multimodal, I further comprising the step of:

[0022] 步骤101,粗略判断目标人脸的生物活性,其中按照下面的方式中的一种或多种进行判断:通过热红外判断目标人脸的温度,判断是否接近37度;通过3D图像判断人脸的深度信息,判断面部是否为3D物体;通过超声波反射分析目标人脸的超声波反射率,判断皮肤的超声波反射率是否与真实人脸相似;通过多光谱成像分析目标人脸在不同光谱下的反射率,判断皮肤的多光谱反射率是否与真实人脸相似,如果通过上述一种或多种方式判断目标人脸的信息指标与真实人脸相似,则进入步骤102 ; [0022] Step 101, determining a rough biological activity of the target face, wherein the following manner according to one or more of the judgment: Analyzing the target temperature by thermal infrared face, determining whether proximity of 37 degrees; the 3D image is determined by depth information of the face, it is determined whether the face of a 3D object; ultrasonic reflectivity by ultrasonic reflection analysis target face, determined skin ultrasound reflection rate is similar to the real face; lower by multi-spectral imaging analysis of the target face at different spectral the reflectance of the skin is determined whether the multispectral reflectance similar to the real human face, if the face is determined by said target one or more ways information indicator similar to a real human face, the process proceeds to step 102;

[0023] 步骤102,精确判断目标人脸的生物活性,将采集到的多光谱人脸图像,利用互商图像算法进行准确的活体判断, [0023] Step 102, to accurately determine the biological activity of the target face, the collected multispectral face image, accurate in vivo determination algorithm using the cross quotient image,

[0024] 只有目标人脸同时通过步骤101和102,才被认为通过步骤I的活体检测。 [0024] Only by the target face while steps 101 and 102, was considered by the step of I in vivo detection.

[0025] 互商图像算法包括如下步骤: [0025] Mutual image supplier algorithm comprises the steps of:

[0026] 步骤1021,采集大量真人人脸和虚假人脸在不同距离下的多光谱成像构成训练数据集,对于同一个人的任意两张不同光谱下的图像进行像素级的相除,组成互商图像组,假设任意选定两个光谱X1, λ2,同一个人脸在两个光谱下的图像为'和',其互商图像定义如下: [0026] Step 1021, collecting a large number of faces and false real face constituting training data set in the multi-spectral imaging at different distances, for dividing pixel-level image of the same person at any two different spectral composition mutual supplier image group is assumed arbitrarily selected two spectra X1, λ2, human faces in the image of the same for the two spectra 'and' cross quotient image which is defined as follows:

Figure CN102622588BD00071

[0028] 其中,P表示人脸的反射率,K代表光源在人脸表面处的强度,ζ代表人脸距离光源的距离,U,y)代表人脸图像上的坐标; [0028] where, P represents the reflectance of the human face, K represents the distance of the intensity at the surface of the face, ζ is representative of a face from the light source, U, y coordinates on the representative face image light);

[0029] 步骤1022,对于所有的互商图像,在多个尺度上划分为多个重叠或不重叠的小块,提取每个小块的特征向量,将所有小块的特征向量进行组合,作为全局的特征向量; [0029] Step 1022, for all mutual quotient image is divided in a number of scales into a plurality of pieces or may not overlap, a feature vector is extracted for each tile, the feature vectors of all the pieces are combined, as the global feature vectors;

[0030] 步骤1023,基于统计学习方法,在训练数据集上训练分类器,用于区分真实、虚假人脸。 [0030] Step 1023, based on statistical learning methods, training a classifier on the training data set is used to distinguish between true and false faces.

[0031] 步骤2进一步包括: [0031] Step 2 further comprises:

[0032] 步骤201,采集大量真实人脸的多模态图像,对每张图像提取其纹理特征; Multimodal image [0032] Step 201, collecting a large number of real face, extracted texture features of each image;

[0033] 步骤202,将图像的特征向量两两相减,根据两图像是否属于同一个人,将相减后的特征向量分为类内、类间两类,利用机器学习算法训练一个两类分类器,训练得到的分类器能够判断输入的两个特征向量是否属于同一个人; [0033] Step 202, the image feature vectors of pairwise subtracting, depending on whether two images belong to the same person, the feature vectors after the subtraction into the within-class, of two categories, using a machine learning algorithm to train a binary classification , a trained classifier to obtain it can be determined whether the two input feature vectors belonging to the same person;

[0034] 步骤203,如果是在人脸识别应用中,若目标人脸图像与识别结果对应的人脸图像,被步骤202中的分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸; [0034] Step 203, the face recognition application, if the target image and the face recognition result corresponding to the face image, has been identified in step 202 is classified as belonging to the same person, the target face is considered true and valid, otherwise, false face for the people;

[0035] 如果是在人脸验证应用中,若目标人脸图像与所声称的指定人身份对应的人脸图像,被步骤202中的分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸。 [0035] If it is in the face verification applications, if the target face image with the claimed identity of the nominee corresponding face image, was identified in step 202 is classified as belonging to the same person, then that person faces a real and effective target otherwise, people face is false.

[0036] 每种不同的成像类型被称为一个模态,成像类型包括可见光成像,近红外成像,近紫外成像,热红外成像或超声波成像。 [0036] Each different type is called an imaging modality, including the type of imaging visible light imaging, near infrared imaging, imaging the near ultraviolet, infrared thermal imaging or ultrasound imaging.

[0037] 一种双验证人脸防伪装置,该装置包括: [0037] A double-face authentication security device, the apparatus comprising:

[0038] 感应单元,用于使用近红外、超声波、射频方式或可见光摄像头中的一种或多种,通过实时监控的方式,感应人脸的存在; [0038] The sensing unit for using near infrared, ultrasonic, radio frequency or visible light camera way of one or more of, by way of real-time monitoring, sensing the presence of a human face;

[0039] 多模态发生源,包含多个光谱下的主动光源、用于3D成像所需的3D结构光或者超声波发生器中的一种或多种; [0039] Multimodal generating source, comprising a plurality of active light source in the spectrum, for one or more desired structured light 3D imaging or 3D ultrasound generator;

[0040]多模态数据采集设备,用于采集人脸的多光谱成像,人体本身所发出的热红外光成像,人脸的3D图像或超声波成像中的一种或多种; [0040] Multi-modal data acquisition apparatus for acquiring multi-spectral imaging of the human face, the body's own thermal imaging infrared light emitted, ultrasound imaging or 3D image of the face of one or more;

[0041]多模态人脸检测单元,用于检测多模态图像中的人脸位置,并将检测到的人脸图像发送到多模态双验证人脸防伪单元; [0041] Multimodal face detection means for detecting a position of the multi-modal face image, and transmits the detected human face image to the multi-modal security double face authentication unit;

[0042] 多模态双验证人脸防伪单元,用于验证目标人脸是否为真实有效的人脸; [0042] The multi-modal security double face verification means for verifying whether the target face as the face of the real and effective person;

[0043] 显示单元,用于显示人脸防伪结果, [0043] a display unit for displaying the results of the security face,

[0044] 其中,多模态双验证人脸防伪单元进一步包括:多模态人脸活体检测单元,用于对目标人脸进行活体检测;多模态人脸验证单元,用于对目标人脸进行身份验证。 [0044] wherein the multimodal dual verify face security unit further comprises: a multimodal face living body detecting unit for the target human faces in vivo testing; multimodal face verification unit for the target face authentication.

[0045] 所述多模态人脸活体检测单元对目标人脸进行活体检测时,首先,粗略判断目标人脸的生物活性,其中按照下面的方式中的一种或多种进行判断:通过热红外判断目标人脸的温度,判断是否接近37度;通过3D图像判断人脸的深度信息,判断面部是否为3D物体;通过超声波反射分析目标人脸的超声波反射率,判断皮肤的超声波反射率是否与真实人脸相似;通过多光谱成像分析目标人脸在不同光谱下的反射率,判断皮肤的多光谱反射率是否与真实人脸相似,如果通过上述一种或多种方式判断目标人脸的信息指标与真实人脸相似,则继续精确判断目标人脸的生物活性,将采集到的多光谱人脸图像,利用互商图像算法进行准确的活体判断。 [0045] The multimodal human face when the target living body detecting unit detects human faces in vivo, first, rough determination of the biological activity of the target face, wherein the following manner according to one or more of the determination: by heat infrared judgment target face temperature, determines whether proximity of 37 degrees; the depth information of the 3D image is determined face, determines whether the face of a 3D object; ultrasonic reflectivity by ultrasonic reflection analysis target face, determined skin ultrasound reflection rate is similar to the real face; multi-spectral imaging analysis by the target face at different spectral reflectance of the skin is determined whether the multispectral reflectance similar to the real face, if the target person is determined by the above one or more ways of the face information indicator similar to the real face, continues to accurately determine the biological activity of the target face, the collected multispectral face image, the image supplier algorithm using the cross accurate determination in vivo.

[0046] 互商图像算法包括如下步骤: [0046] Mutual image supplier algorithm comprises the steps of:

[0047] 采集大量真人人脸和虚假人脸在不同距离下的多光谱成像构成训练数据集,对于同一个人的任意两张不同光谱下的图像进行像素级的相除,组成互商图像组,假设任意选定两个光谱λρ λ 2,同一个人脸在两个光谱下的图像为\和\,其互商图像定义如下: [0047] acquiring a plurality of faces and false real face constituting training data set in the multi-spectral imaging at different distances, for dividing pixel-level image of the same person at any two different spectral composition mutual commercially image group, Suppose two arbitrarily selected spectral λρ λ 2, at the face image of the same individual spectra for the two \ and \, mutual quotient image which is defined as follows:

[0048] [0048]

Figure CN102622588BD00081

[0049] 其中,p表示人脸的反射率,κ代表光源在人脸表面处的强度,ζ代表人脸距离光源的距离,(X,y)代表人脸图像上的坐标; [0049] where, P represents the reflectance of the face, the strength of the surface of the human face, the face distance [zeta] from the light source of the representative, the representative coordinates of the representative face image κ light (X, y);

[0050] 对于所有的互商图像,在多个尺度上划分为多个重叠或不重叠的小块,提取每个小块的特征向量,将所有小块的特征向量进行组合,作为全局的特征向量; [0050] For all mutual quotient image, divided in a number of scales into a plurality of pieces or may not overlap, a feature vector is extracted for each tile, the feature vectors of all the pieces are combined, as a global feature vector;

[0051] 基于统计学习方法,在训练数据集上训练分类器,用于区分真实、虚假人脸。 [0051] based on statistical learning methods, training a classifier on the training data set is used to distinguish between true and false faces.

[0052] 多模态人脸验证单元对目标人脸进行身份验证时,首先采集大量真实人脸的多模态图像,对每张图像提取其纹理特征;其次,将图像的特征向量两两相减,根据两图像是否属于同一个人,将相减后的特征向量分为类内、类间两类,利用机器学习算法训练一个两类分类器,训练得到的分类器能够判断输入的两个特征向量是否属于同一个人;如果是在人脸识别应用中,若目标人脸图像与识别结果对应的人脸图像,被上述两类分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸;如果是在人脸验证应用中,若目标人脸图像与所声称的指定人身份对应的人脸图像,被上述两类分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸。 Multimodality images [0052] Multimodal face authentication unit authenticates the target face, collecting a large number of real first face, extracted texture features of each image; secondly, the two two-phase image feature vector Save, depending on whether two images belong to the same person, the feature vector subtracted divided into categories, between the two categories, using a machine learning algorithm to train a classifier types, the trained classifier can be determined that the two input feature vector belong to the same person; if it is in face recognition applications, if the target face image and the recognition result corresponding face image, these two types of classifiers have been identified as belonging to the same person, then that target face a real and effective, otherwise people face is false; if it is in the face verification applications, if the target face image with the claimed identity of the nominee corresponding face image, these two types of classifiers have been identified as belonging to the same person, then that person face real target effective, otherwise false face.

[0053] 每种不同的成像类型被称为一个模态,成像类型包括可见光成像,近红外成像,近紫外成像,热红外成像或超声波成像。 [0053] Each different type is called an imaging modality, including the type of imaging visible light imaging, near infrared imaging, imaging the near ultraviolet, infrared thermal imaging or ultrasound imaging.

[0054] 本发明的优点:通过活体检测与身份验证的结合,提供准确、可靠的人脸防伪检测结果。 [0054] The advantages of the invention: by binding in vivo detection and authentication, to provide accurate and reliable security face detection result. 附图说明 BRIEF DESCRIPTION

[0055] 图1为本发明提出的在可见光下的双验证人脸防伪方法流程图; In the double face authentication security method in a flowchart of visible light [0055] FIG. 1 of the present invention provides;

[0056] 图2为本发明提出的在多模态下的双验证人脸放伪方法流程图; [0056] FIG. 2 of the present invention provides a double face verification in multimodal discharge flowchart pseudoephedrine;

[0057] 图3为本发明提出的在多模态下的双验证人脸防伪装置结构框图; [0057] In FIG 3 a block diagram of a multi-modal verification double face of the security device of the present invention proposed structure;

[0058] 图4为本发明提出的在多模态下的双验证人脸防伪装置的工作流程图; [0058] FIG. 4 of the present invention proposes a dual verifier in multimodal face security device operation flowchart;

[0059] 图5为本发明提出的在多模态下的双验证人脸防伪装置的光源覆盖范围示意图; [0059] FIG. 5 of the present invention proposes a dual verifier in multimodal schematic face coverage source security device;

[0060] 图6为本发明提出的双验证人脸防伪装置一实例中人脸区域图像灰度均值与人脸距采集装置距离之间的关系示意图; Double face authentication security device [0060] FIG. 6 is an example of the present invention provides a human face area and the face image from the gray value relationship between the distance schematic acquisition means;

[0061] 图7为在一定光谱范围内黑人和白人的人脸反射率曲线示意图; [0061] FIG. 7 is a face black and white reflectance curves within a certain spectral range a schematic view;

[0062] 图8为在一定光谱范围内几种常见造假人脸的反射率曲线示意图; [0062] FIG. 8 is a reflectance of several common false face schematic curve within a spectral range;

[0063] 图9为本发明提出的双验证人脸防伪装置一实例中多模态采集装置的面板示意图; Dual panel face authentication security device [0063] FIG. 9 of the present invention provides a multimodal example schematic acquisition device;

[0064] 图10为三种不同光谱的人脸成像示意图,从左到右依次为:可见光、850nm近红外光和400nm紫光; [0064] FIG. 10 is a person's face three different spectral imaging schematic, left to right: visible light, near-infrared light 850nm 400nm purple;

[0065] 图11为人脸热红外成像示意图; [0065] FIG 11 a schematic view of a human face thermal infrared imaging;

[0066] 图12为人脸3D成像示意图; [0066] FIG 12 a schematic view of a 3D imaging a human face;

[0067] 图13为超声波在人脸上的反射波示意图。 [0067] FIG. 13 is a schematic view of an ultrasonic wave reflected in the person's face.

具体实施方式 Detailed ways

[0068] 下面结合具体附图和实施例对本发明作进一步说明。 [0068] The following specific embodiments in conjunction with the drawings and embodiments of the present invention will be further described.

[0069] 本发明提出的人脸防伪方法的基本原理是基于双验证的思想。 [0069] The basic principle of the face anti-counterfeiting method proposed by the present invention is based on the idea of ​​a double authentication. 所谓双验证人脸防伪,包括如下两个步骤:步骤1,对输入人脸进行活体、非活体的判断,此步骤与人的身份无关,即指定人无关;步骤2,对输入人脸图像进行人脸身份验证,只有当输入人脸图像与所对应的身份相匹配时,才认定为真实有效的人脸,该步骤与指定人相关。 The so-called dual verify face security, comprising the following two steps: Step 1, the input face judge the living body, a non-living body, this step regardless of the identity of the person, i.e. designee independent; Step 2, the input face image face authentication, only when the input face image matches the corresponding identity, real and effective only identified as a human face, the steps associated with the designated person. 只有同时通过以上两步判断的输入人脸才被认为是有效、真实的人脸。 Only the same time by more than two steps to determine the input face is treated as valid, real people face.

[0070] 步骤I中使用的是人脸活体检测技术,即对人脸进行活体、非活体判断,鉴别是否为真人活体人脸;步骤2实际上是对人脸进行指定人、非指定人的验证。 Using I is [0070] Step a human face in vivo detection, i.e. human face in vivo, non-living body determination, to identify whether the face is a real living person; step 2 is actually a human face nominee, non-authorized person verification. 其中在步骤2中,如果是做人脸识别应用,则识别结果即为目标人脸所对应的身份,只有两者的相似度大于一定阈值,才通过该步验证,阈值可由管理人员根据实际需求自行设定;如果是做人脸验证应用,则目标人脸所对应的身份为目标人脸所声称的身份,输入人脸图像与对应身份人脸图像之间的相似度须大于一定阈值,才认为通过人脸身份认证。 Wherein in step 2, if it is done recognition application, the target is the human face recognition result corresponding identity, only similarity between the two is greater than the predetermined threshold value, only by the verification step, the threshold value may be based on actual demand management of their own set; if it is a man face verification application, the target face corresponding to the identity of the target face claimed identity, similarity between the input face image is the face image corresponding to the identity must be greater than a certain threshold value, it is considered that face authentication. 步骤2是对指定人与非指定人图像的分类。 Step 2 is a classification of the designated person and the non-image designated person. 通过融合步骤I和步骤2的信息,达到可靠的防伪的目的。 Information fusion step I and step 2, to achieve a reliable security purposes.

[0071] 本发明的方法之所以同时采用上述步骤I和步骤2,是因为在实际应用中,潜在的虚假人脸类型无法预估,单纯的活体检测无法一直保持高准确率。 [0071] The reason why the method of the present invention while using the above-described step I and step 2, because in practice, the potential false face type can not be estimated, a simple detection of the living body can not keep high accuracy. 而在另一方面,即使虚假人脸通过了活体检测,也有理由相信该虚假人脸与所仿冒的指定人人脸之间存在一定的差异,因此可以通过提取、鉴别该差异,进一步增强人脸防伪的精度。 On the other hand, even if false face by the in vivo detection, there is reason to believe that there are some differences between the false face to the specified face all the counterfeit, and therefore can be extracted to identify the differences, to further enhance the face security accuracy. 因此提出输入的人脸图像需要同时通过人脸活体检测和人脸身份验证,才能认定为真实有效的人脸。 Therefore, the input face image presented by the need to live face detection and face authentication in order to identify real and effective human face.

[0072] 传统的人脸防伪研究还停留在人脸活体检测上,而忽略了对输入人脸的验证。 [0072] The traditional face security research still remain on face detection in vivo, while ignoring the validation of input face. 事实上,步骤I的人脸活体检测,与指定人无关;而步骤2的人脸身份验证则是针对指定人的身份验证。 In fact, the steps I live face detection, regardless of the nominee; and step 2 of face authentication is verified against the identity of the nominee. 本发明的人脸防伪技术,结合人脸身份验证与活体检测,可以有效提高人脸防伪的可靠性。 Face security techniques of the present invention, in combination with the face authentication in vivo detection, and improve the reliability of the security face.

[0073] 在本发明的双验证人脸防伪方法中,进一步提出了在可见光下的具体应用形式和在多模态下的具体应用形式。 [0073] In the double-face authentication security method of the present invention, it is further proposed in the form of specific applications and specific applications in the form of visible light in a multi-modal.

[0074] 可见光下的双验证人脸防伪方法适用于传统的可见光人脸识别、人脸验证系统,无需额外硬件即可完成人脸防伪任务。 [0074] double face security verification method under visible light for conventional visible recognition, face verification system, no additional hardware to complete the face of security tasks. 虚假人脸图像可以看作是真实人脸图像在经过某种后处理之后得到的图像,因此相比真实图像其图像质量将有一定损失。 False face image can be seen as a real image of the face image obtained after processing after a certain, as compared to the true image of its image quality will have some losses. 通过对捕捉到的目标人脸提取多种类型的纹理信息,可以充分挖掘目标人脸在皮肤纹理细节上的表观特征,进而根据预先设定好的评估标准进行进一步的分类。 By extracting a plurality of types of texture information of the captured target face, you can fully exploit the apparent facial characteristics of the target on the skin texture detail, thus a good evaluation criteria further classified according to pre-set. 此外,也可以通过引入传统的人机交互过程,进一步增强其准确度。 In addition, by introducing traditional man-machine interaction, to further enhance its accuracy.

[0075] 对于本发明提出的多模态下的双验证人脸防伪方法,则进一步采用多种模态充分挖掘人脸本质特征。 [0075] For double face authentication security method proposed by the present invention in a multi-modal, using a variety of modalities is further fully exploit the essential features of a human face. 现有的人脸识别、人脸验证技术还仅停留在利用一种模态(例如可见光或近红外)获取人脸图像。 Existing face recognition, face verification techniques still remain in use only modality (e.g., visible or near infrared) acquires face image. 我们认为这种数据采集方式并不能充分挖掘人脸的皮肤特性,也不能达到较高的防伪精度,因此提出并设计了多模态下的双验证人脸防伪装置。 We believe that this manner of data acquisition can not fully exploit the characteristics of the skin of the face, they can not achieve a high security accuracy, it is proposed and designed double face authentication security device in a multi-modal. 该装置包括不同光谱下的多光谱成像装置、热红外成像装置、超声波成像装置等等,从不同的层面充分挖掘人脸皮肤的本质物理特性。 The apparatus comprises a multi-spectral imaging apparatus under different spectral, thermal infrared imaging apparatus, ultrasound imaging apparatus, etc., from different levels to fully exploit the nature of the physical characteristics of the face skin. 通过仔细分析真人人脸与典型虚假人脸在不同模态下的特性,选取合适的模态组合,为后续的人脸防伪算法提供最具鉴别力的特征。 By carefully analyzing the typical characteristics of the human face and false real face in different modes, select the appropriate combination of modes to provide the most discrimination for subsequent security algorithms face features.

[0076] 本发明提出的可见光下的双验证人脸防伪方法,可以在不依赖额外硬件的基础上,通过对皮肤纹理细节的分析和/或人脸的运动,对目标人脸的真伪进行准确判断。 [0076] bis verified face in the anti-counterfeiting method proposed by the present invention, visible light, can not rely on the basis of additional hardware on the analysis of the skin texture and motion details and / or face, the face of the authenticity of the target will be accurate judgment. 而本发明提出的基于多模态的双验证人脸防伪方法,相对于现有的人脸活体检测算法,不仅可以防御更多的攻击类型,而且具有用时少、用户体验良好、准确率高等特点。 While dual verify face anti-counterfeiting method multimodal based, with respect to the proposed present invention, a conventional face detection algorithm living body, not only against more types of attacks, but also when a less good user experience, and high accuracy features . 通过多模态获取的人脸信息可以提供更丰富的人脸信息,充分挖掘人脸的本质特征,增大真人人脸与虚假人脸的区分度,可以有效解决人脸的防伪难题。 Acquired by the multi-modal face information can provide richer information face, fully tap the essence of facial features, increase the real face discrimination and false face, can effectively solve the security problems people face.

[0077] 图1为本发明提出的双验证人脸防伪方法在可见光下的具体应用流程图。 Bis flowchart of specific application face authentication security method [0077] FIG. 1 of the present invention set forth in visible light. 参照图1,在活体检测步骤101中,使用皮肤纹理与面部运动相结合的人脸活体检测策略。 Referring to FIG 1, a living body is detected in step 101, using human facial skin texture and motion combining the face detection strategy in vivo. 在身份验证步骤102中,对目标人脸所对应的身份(人脸验证应用中为所声称的身份,人脸识别应用中为识别结果对应的身份)进行验证,若匹配相似度大于一定阈值,则认为是真实人脸,否则为虚假人脸。 In the authentication step 102, the target corresponding to the identity of the face (face authentication application for a claimed identity, face recognition applications, as the recognition result for the corresponding identity) to verify, if the matching similarity is greater than a predetermined threshold value, is considered to be a real human face, otherwise false face. 只有目标人脸同时通过了101和102两步才认定目标人脸为真实人脸。 Only the target face at the same time it finds a target face is the real face 101 and 102 by two steps.

[0078] 活体检测步骤101进一步包括步骤1011和步骤1012:步骤1011,首先对目标人脸提取各种纹理特征,例如LBP (Local Binary Pattern)、HOG (Histograms of OrientedGradients)特征等,然后通过采集真实虚假人脸样本通过机器学习算法(如支持向量机SVM)训练得到基于皮肤纹理的活体检测器。 [0078] The living body detecting step 101 further includes step 1011 and step 1012: Step 1011, first extracted texture features of the target human face, for example, LBP (Local Binary Pattern), HOG (Histograms of OrientedGradients) characteristics, etc., and then by collecting real false face samples by machine learning algorithms (such as support vector machines SVM) is trained in vivo detection based on skin texture. 如果判断是真实人脸,进入步骤1012。 If the judgment is real human face and proceeds to step 1012.

[0079]步骤1011的一个实例是,利用不同尺度的LBP描述子,例如对目标人脸图像进行滤波,然后对图像进行多尺度的划分,例如划分成1X1,3X3,5X5的小块,在每一块里面统计三种LBP描述子的直方图,把所有的直方图链接在一起作为目标人脸的纹理特征。 One example of [0079] Step 1011 is to use different scales LBP descriptor, for example, the target face image filtering multiscale image is then divided, for example, divided into small pieces 1X1,3X3,5X5 in each inside a histogram descriptors three LBP, all linked together as a texture histogram characteristics of the target face.

[0080] 然后采集大量真实、虚假人脸的图像,例如,采集50人的真实人脸图像,然后利用其人脸图像制作成不同大小的照片,然后再次采集照片图像。 [0080] and then collect a lot of true, false image of a human face, for example, the acquisition of 50 real face image, then use its facial images made into different sizes of photos, and then capture the photo image again. 取出人脸区域,按照上一步的操作抽取特征。 Remove the face region, in accordance with the feature extraction step of the operation. 然后利用SVM算法训练得到一个分类器。 SVM algorithm is then used to get a trained classifier. [0081] 在步骤1012,利用人机交互进一步检测目标人脸的生物活性。 Biological activity [0081] In step 1012, the use of interactive further detects the target face. 例如,可以通过人脸识别系统给出让用户眨眼、或摇头的指令。 For example, a user can transfer to a facial recognition system, a blink, or shake instructions. 通过检测目标人脸是否做出了相应动作,从而判断目标人脸是否为真实人脸。 By detecting whether the target face made the appropriate action to determine whether the true goal of a human face to face. 在该步骤中,可以利用运动估计或模板匹配算法进行面部运动估计。 In this step, using motion estimation or the face template matching motion estimation algorithm. 例如,若采用眨眼的形式,可以利用光流法计算目标人脸眼睛区域的运动矢量,进而判断是否发生了眨眼动作。 For example, the use of a blinking form, can be calculated using the optical flow motion vector of the target region of the eye of the face, and then determines whether the blinking action occurs. 或者模板匹配算法,预先训练好一个睁眼、闭眼的分类器,然后进行运动检测。 Or template matching algorithm, a good pre-trained eyes open, eyes closed, classifiers, and motion detection.

[0082] 一个人机交互的实例是,人脸识别系统给出指令要求用户在一定时间内,例如5秒,进行眨眼。 Examples [0082] a human-computer interaction, the face recognition system requires the user to give an instruction within a certain time, for example 5 seconds, blink. 通过训练好的人眼状态分类器,检测在该段时间内是否出现了睁眼-闭眼-睁眼的过程。 By the trained eye condition classifier to detect whether there has been eye opening in that period of time - eyes closed - open eyes of the process. 若出现,则认为是真实人脸,否则则认为是虚假人脸,进入步骤1021。 If there is, then that is a real human face, otherwise it is considered to be false face and proceeds to step 1021. 其中上面提到的人眼状态分类器,可以预先收集大量睁眼、闭眼图像,然后利用SVM分类器训练得到眼睛状态的分类器,用于上述的眨眼检测。 Wherein the above-mentioned eye condition classifier, a large number of open eyes can be collected in advance, the image with eyes closed, and then using the trained SVM classifier classifier eye state for the aforementioned blink detection.

[0083] 在身份验证步骤102中,对数据库中的人脸图像抽取特征(例如,LBP和Gabor特征),然后将采集的所有人脸图像的特征向量两两相减,根据两图像是否属于同一个人,将相减后的特征向量分为类内、类间两类,利用机器学习算法训练一个两类分类器,由此训练得到的分类器可以判断输入的两个特征向量是否属于同一个人; [0083] In the subtraction image feature vectors of all faces of the authentication step 102, the face image feature extraction (e.g., LBP and Gabor feature) in the database, then collected two by two, depending on whether two images belong to the same individual feature vectors subtracted divided into categories, between the two categories, using a machine learning algorithm to train a binary classification, thereby to obtain a trained classifier based on two input feature vectors belong to the same person;

[0084] 经过以上步骤,属于同一人的人脸特征之间的相似度应该大于不同人的人脸特征之间的相似度。 [0084] After the above steps, is the similarity between the facial features of the same person should be greater than the similarity between the facial features of different persons. 通过设定一个合理的阈值,可以用于身份验证:若在步骤102中目标人脸与其所声称的身份之间的相似度大于阈值,则认为通过了身份验证;否则失败。 By setting a reasonable threshold, it can be used for authentication: If the similarity between the target face and whose claim of step 102 is greater than the threshold value, is considered by authenticated; otherwise fail.

[0085] 图2为双验证人脸防伪方法在多模态的形式下的应用流程图。 [0085] FIG 2 is a flowchart showing the application authentication double face under the form of anti-counterfeiting method is multimodal. 该方法采用多模态作为载体,采集多模态人脸图像,利用多模态图像所提供的丰富信息和利用不同生物特征具有不同物理特性的特点,通过多模态信息融合,设计了合理、可靠的双验证人脸防伪算法。 The method uses multi-modal as carriers, collecting multimodal face image, using the wealth of information the multi-modal images is provided and with different biological characteristics having different physical properties characteristics, by multimodal information fusion, design reasonable, reliable verification of double face security algorithm.

[0086] 在多模态形式下的双验证人脸防伪方法包括活体验证步骤201与身份验证步骤202两步。 [0086] bis face authentication security method in a multi-modal forms include living body authentication step 201 and step 202-step authentication.

[0087] 在活体信息验证201中,采用由粗到精的两步策略。 [0087] In the biometric information authentication 201, the two-step strategy from coarse to fine.

[0088] 首先在第一步2011,利用所获取的多模态人脸信息,对输入人脸的活体特性进行粗略判断。 [0088] First, in a first step 2011, the use of multi-modal information acquired face, the input face of the rough characteristics of a living body is determined. 一个实例是:首先通过热红外图像进行温度检测,如果符合真实人体的温度范围(例如是否为37度),则通过3D人脸图像进行人脸深度信息的判断,如果判断输入人脸是一个三维物体,则继续利用超声波反射波测量输入人脸的超声波反射率,若反射率与真人人脸相似,则验查其多光谱的平均图像亮度是否在合理范围内,若合理,则判断为真人人脸,否则为虚假人脸。 One example is: firstly the temperature detected by the thermal infrared image, if they meet the true body temperature range (e.g., whether it is 37 degrees), is performed who determines the face depth information by 3D face image, if it is determined the input face is a three-dimensional object, then continue to use the reflected ultrasonic wave measuring the input face of the ultrasonic reflectivity, if the reflectance real face are similar, the checker average image brightness multispectral is within a reasonable range, if reasonable, it is determined that real people face, otherwise false face. 在该步骤中,可以根据特定的人脸模态动态选取作为粗略判断的人脸活体特性。 In this step, can be dynamically selected as a human face is determined roughly based on a particular characteristic of the living body face modality.

[0089] 当第一步2011认定为真人人脸之后,在第二步2012中,针对人脸的多模态成像,本发明提出基于互商图像的人脸活体检测算法,给出更为准确、精细的检测结果;若互商图像算法判断此人脸为真人人脸,则说明输入人脸具有生物活性。 [0089] When Step 2011 is identified after live human face, in a second step 2012, for a multi-modality imaging of the human face, the present invention provides an image detection algorithm mutual commercially living body based on the face, give a more accurate fine detection result; algorithm determines if this quotient image mutual face to face live, then the input face having biological activity. 如果在第二步2012判断为真人人脸,则为真人人脸,否则为虚假人脸。 If the face is a real person in a second step 2012 judgment, for the real human face, otherwise false face.

[0090] 在步骤2012中,利用互商图像算法进行精确的人脸活体检测。 [0090] In step 2012, the image supplier algorithm using the cross accurate detection of living body face. 互商图像是指任意两个光谱下的图像进行相应位置像素值做除法所得到的图像(Mutual Quotient Image,MQI)。 Mutual commercially image refers to an image at the corresponding position of any two spectral image pixel value do (Mutual Quotient Image, MQI) division obtained. 互商图像能反映拍摄人脸在两个波段反射率之间的关系,而且与人脸的形状无关。 Mutual business image shooting to show the relationship between the two bands face reflectivity, and regardless of the shape of the human face. 根据互商图像的定义,假设任意选定两个光谱λρ λ2,同一个人脸在两个光谱下的图像为' According to the definition of mutual quotient image, assuming two arbitrarily selected spectral λρ λ2, human faces in the image of the same spectrum as the two '

和\,其互商图像定义如下: And \, mutual quotient image which is defined as follows:

[0091] [0091]

Figure CN102622588BD00121

[0092] 其中,P表示人脸的反射率,κ代表光源在人脸表面处的强度,ζ代表人脸离光源之间的距离,(x,y)代表人脸图像上的坐标。 [0092] where, P represents the reflectance of the human face, [kappa] represents the light source intensity at the surface of the face, [zeta] representative distance between the light source face, the coordinates (x, y) representative face image.

[0093] 如果保证X1, λ 2两个光谱的光源发光功率一致,则在合适的距离范围内,X1, λ 2两种光源的强度之比约等为1,因此(4)式可以约等于 [0093] If the light emitting ensure consistent X1, λ 2 of the two spectral power is within an appropriate distance range, the X1, the intensity of λ 2 two sources as the ratio of about 1, so equation (4) may be approximately equal

[0094] [0094]

Figure CN102622588BD00122

[0095] 可以看出,此时互商图像反映了人脸在A1, λ 2两种光谱下的反射率之比,因此是一个可以反映人脸本质特性的特征,可以用来设计活体检测算法。 [0095] As can be seen, the image supplier at this time reflects the mutual ratio of the reflectance of the face at A1, λ 2 two spectral, and thus can reflect the characteristics of a human face essential characteristics, it can be used to design detection algorithms in vivo .

[0096] 在公式(5)的推导中,假设在合适的距离范围内,A1, λ 2两种光源的强度之比约等为I。 [0096] In deriving Equation (5), it is assumed within a suitable distance range, A1, λ 2 of the intensity ratio of about two light sources and the like is I. 通过合理设计光源,可以在实际中满足这一假设。 By rational design the light source can satisfy this assumption in practice. 例如,本发明采集了480nm和850nm两种光源在发光功率一致的情况下,在距离光源40cm到90cm之间,同一个人的人脸图像灰度均值的变化情况,如图6所示,可以看出,两种光源的强度之比约等为I的假设是合理的。 For example, the present invention is collected at 480nm and two light sources emitting the same power, the distance between the light source 40cm to 90cm, the same person's face image gray value changes, as shown in FIG. 6, see 850nm the ratio of the intensity of two light sources and the like of about I assumption is reasonable.

[0097] 基于多模态的人脸活体检测算法中,在获取了任意两个光谱的互商图像之后,可以设计合理特征,以便进行活体检测。 [0097] The face detection algorithm based on multi-modal living body, after acquiring the spectra of any two mutually quotient image, rational design can be characterized, for in vivo testing. 特征向量提取可以采用多种方法,如:强度直方图、Gabor滤波器等、似然比(Likelihood Ratio)等。 Feature vector extraction may use various methods, such as: intensity histogram, Gabor filters, likelihood ratio (Likelihood Ratio) the like. 在选定特征类型之后,可以对互商图像进行分块,并做多尺度的处理,得到不同尺度上、不同位置的人脸互商图像特征向量,然后大量采集真、假人脸样本,利用Boosting算法进行活体检测分类器的训练。 After the selected feature type, for interoperability may be commercially image block, multi-scale processing and do give different scales, different people mutual positions of face image feature vectors commercially, then a large number of real acquisition, dummy face samples, using Boosting vivo detection algorithm is trained classifier.

[0098] 基于多模态的人脸活体检测算法中,应充分考虑真实、造假人脸的反射率差异,进行光源选择。 [0098] face detection algorithm based on multi-modal living body, should be considered true, false difference in reflectance face, light source selection. 图7例示了多光谱下黑人和白人的人脸反射率曲线。 7 illustrates a case of black and white multispectral reflectance curve face. 图8例示了多光谱下几种常见造假人脸的反射率曲线,包括两种不同的硅胶和照片。 8 illustrates a multi-spectral reflectance false face some common curve, comprising two different silica and photos. 依据这两幅曲线,可以为多模态人脸活体检测中的光谱选择提供依据。 This is based on two curves, may provide the basis multimodal face the living body is detected in the spectrum selected.

[0099] 基于互商图像的人脸活体检测算法的具体流程如下: [0099] Based on the specific process mutual commercially person living body image face detection algorithm is as follows:

[0100] (I)、采集大量真人人脸和造假人脸在不同距离下的反射强度数据构成训练数据集,对于同一个人的任意两张不同光谱下的图像进行MQI计算。 [0100] (I), collecting a large number of real reflection intensity data false faces and at different distances from the face constituting the training data set, the image is calculated for MQI same person at any two different spectra.

[0101] (2)、在所有的MQI图像上,在多个尺度上划分为多个小块(重叠或不重叠),提取每个小块的特征向量,将所有小块的特征向量进行组合,作为全局的特征向量。 [0101] (2), on all MQI images into a plurality of small blocks on a plurality of scales (or no overlap), extracting a feature vector of each small block, the feature vectors of all the pieces are combined as the global feature vector.

[0102] (3)、基于统计学习方法,在训练数据集上训练分类器,如:SVM(支持向量机)、LDA (线性判别分析)、Boosting等。 [0102] (3), based on the statistical learning method, the training data set on training a classifier, such as: SVM (Support Vector Machine), LDA (Linear Discriminant Analysis), and the like on Boosting.

[0103] 下面通过举例来进一步说明活体检测步骤2012的互商图像算法。 [0103] The following further illustrated by way of example the step of detecting a living body image mutual supplier algorithm 2012.

[0104] 例如,采用480nm和940nm的两种光源进行成像,获得的人脸图像分别为148(|,I9400然后规定I48tl为参考图像,计算这两个波段下的互商图像为MQI94tl,48Q(x,y) = I940 (x, y)/I480 (x^ y)。本发明在此仅以举例的方式给出了两种波段的情况,也可根据实际情况选择任意多种波段的光源。 [0104] For example, 480nm and 940nm using two light source image, a face image obtained respectively 148 (|, I9400 and I48tl predetermined as a reference image, the image supplier at the cross of these two bands was calculated MQI94tl, 48Q ( x, y) = i940 (x, y) / I480 (x ^ y). in the present invention, given by way of example only the case where the two kinds of bands, the light source may be any of a variety selected bands according to the actual situation.

[0105] 128X128的MQI图像经过预处理后进行多尺度处理,分为5个尺度,其大小分别是128X128像素、64X64像素、32X32像素、16X16像素、8X8像素。 MQI image [0105] After the pretreatment 128X128 multiscale processing, divided into five scales, which are the size of 128X128 pixels, pixels 64X64, 32X32 pixels, 16X16 pixels, 8X8 pixels. 基于在训练集上通过统计学习得到的概率模型,对于互商图像上的每一点,可以计算其属于活体和非活体的似然;?(^,>^|0),;^,>^|0),其中G代表图像来自活体,g代表图像来自非活体,(x,y)为图像坐标。 Probability model on the training set obtained by statistical learning based, for each point can be calculated on mutual business image that belongs to the likelihood of living and non-living body; (^,> ^ | 0),;? ^,> ^ | 0), wherein G represents the image from the living body, g representative image from non-living body, (x, y) is the image coordinates. 将这两个量相除,可以得到互商图像的局部似然比: Dividing these two quantities can be obtained commercially partial cross-likelihood ratio of the image:

[0106] [0106]

Figure CN102622588BD00131

[0107] 对于上述多分辨率的互商图像,所有的局部似然比可以构成一个活体特征向量,其维度为21824。 [0107] For the above-described multi-resolution image mutual commercially, all the local log likelihood ratio may constitute a living body feature vector which dimension 21824.

[0108] 为了使特征更具有区分度和具有更高的运算效率,活体特征提取算法利用Boosting进行特征选择,从原始的高维度特征中挑选最具鉴别力的3000维特征。 [0108] In order to make the feature more discrimination and higher computational efficiency, dimensional feature 3000 live feature extraction algorithm using Boosting feature selecting, choosing the most discrimination from the original high-dimensional features.

[0109] 然后采集大量真、假人脸样本,组建训练数据库,按照上述Boosting挑选后的特征标号进行特征抽取,并利用支持向量机器(Support Vector Machine, SVM)方法学习得到一个两类分类器,用于对输入的特征向量进行活体、非活体的判断。 [0109] and then collect a large true, dummy face samples, the training set database, after feature extraction in accordance with the characteristic numerals Boosting selection, and two types of learning classifier to obtain a support vector machine (Support Vector Machine, SVM) method, for input feature vector is determined in vivo, non-living body.

[0110] 在身份验证步骤202中,需要对输入人脸与所其所对应的身份进行相似度验证。 [0110] In the authentication step 202, it is necessary to verify the similarity of the input face and its corresponding identity. 具体的验证算法与可见光下的验证方法102类似,不同之处在于输入的特征为多模态图像上所有特征的总和。 Specific authentication algorithm and authentication method is similar to the visible light 102, except that the sum of all the input characteristic features on the multi-modal images.

[0111] 只有当活体信息验证和身份验证两步都认定输入人脸为真人人脸,输入人脸才算通过人脸防伪判断。 [0111] Only when the biometric information, and authentication input face both steps identified as real face, the input face considered counterfeit judgment by a human face.

[0112] 下面通过举例来进一步说明身份验证步骤202中的人脸验证算法,其中以人脸验证应用为例。 [0112] The following further illustrated by way of example in the person authentication step 202 face authentication algorithm, which face authentication application as an example.

[0113] 假设每个人都有N张不同模态的图像,首先对每张人脸图像进行LBP特征和Gabor特征抽取,组成该张图像的特征向量fk,k = 1:N。 [0113] Suppose there are N number of different modalities of each image, each of the first face images on LBP and Gabor feature extraction, feature vector fk composition of the images, k = 1: N.

[0114] 然后将属于同一个多模态图像组合内的每张图像的特征向量串接成组成统一的特征向量F= Lf1 ;...;fN],则F为每一个人的多模态特征向量。 [0114] Then belong to more than one feature vector for each image in the series combination of modality images into a unified composition wherein F = Lf1 vector; ...; fN], F is for everyone multimodal Feature vector.

[0115] 在人脸验证分类器的训练过程中,正样本为属于同一个人的多光谱特征向量F之差,负样本为不属于同一个人的多光谱特征向量F之差。 [0115] In the face authentication process of training the classifier, the positive difference between the samples belonging to the same vector F of the individual multispectral characteristics, the difference between the negative samples do not belong to the same person vector F of multispectral characteristics. 利用Boosting算法进行特征挑选,得到一个特征子集。 Boosting algorithm using feature selection, to obtain a feature subset.

[0116] 对训练数据集中的每个人的多模态图像,按照Boosting选择出的样本进行特征抽取,并利用LDA算法进行判别分析。 [0116] The multimodal image each training data set, in accordance with the feature extraction Boosting selected samples using LDA discriminant analysis algorithm.

[0117] 经过以上步骤,属于同一人的人脸特征之间的相似度应该大于不同人之间的人脸特征相似度。 [0117] After the above steps, the similarity between the facial features of the same person should be larger than face feature similarity between different persons. 若在步骤202中目标人脸与其所声称的身份之间的相似度大于阈值,则认为通过了身份验证;否则失败。 If the similarity between the target face and whose claim is greater than the threshold value in step 202, it is considered by the authentication; otherwise fail.

[0118] 本发明还提出了一种双验证人脸防伪装置。 [0118] The present invention further provides a double security device to verify the face. 图3为本发明基于多模态的双验证人脸防伪装置的结构框图。 FIG 3 is a block diagram showing the structure based on bis face authentication security device of the invention is multimodal. 图4为本发明的基于多模态的双验证人脸防伪装置的工作流程图。 Double face authentication operation flowchart security device based on the multimodal FIG. 4 of the present invention. [0119] 在本发明的基于多模态的双验证人脸防伪装置中,其中的多模态包括多光谱、3D、超声波等模态中的一种或多种。 [0119] bis face authentication security device based on the multimodal, wherein multi-modal includes one or more multi-spectral, 3D, ultrasound and other modality of the present invention. 由于人脸皮肤在不同的光谱下具有不同的反射率,因此本发明引入多光谱人脸成像系统,用于采集、分析人脸在不同光谱下的成像,充分挖掘人脸的本质特性,从而为后续的人脸防伪提供丰富的人脸特征。 Since facial skin having different reflectivities at different spectra, the present invention is thus introduced into the face multispectral imaging system for collecting, analyzing the human face in different spectral imaging, to fully exploit the essential characteristics of a human face, so as to follow-up face security provides a wealth of facial features. 光谱的选取可包括近红外光、中红外光、远红外(热红外)、近紫外光等等,以尽量反映人脸的不同反射特性。 The selected spectrum may include a near-infrared light, mid-infrared, far-infrared (IR heat), near ultraviolet and the like, to try to reflect the different characteristics of reflection face. 特别的,热红外图像指人体自身热量所散发出的红外光成像,与个人的体质、生物组织特性有关,具显著个体差异性,适合用作人脸防伪的依据。 In particular, thermal infrared imaging infrared light image refers to the body's own heat exudes, and individual physical and biological characteristics of the organization, with significant individual differences, as a basis for the security of the human face. 以上光源除热红外线外,都需要多光谱采集系统提供主动光源。 Infrared light source other than heat, the need to provide active multispectral light collection system.

[0120] 本发明同时引入3D人脸图像,以及超声波成像,与多光谱图像一起,共同构成了多模态的人脸图像获取系统。 [0120] The present invention also introducing 3D facial image, and an ultrasound imaging, and multi-spectral images together to form a multi-modal face image acquisition system. 通过3D图像获取的人脸部位的深度信息,是人脸防伪的重要依据,可以抵御常见虚假人脸的攻击,例如照片、视频等。 Face depth information acquired by the site of a 3D image, is an important basis for the security of the human face, false face can resist common attacks, such as photos, videos and so on. 超声波成像的方法,通过测量人脸皮肤对于超声波的反射率,可以提供另外一种人脸皮肤的物理特性度量手段,进一步辅助人脸活体检测的需求。 The method of ultrasonic imaging, by measuring the ultrasonic facial skin reflectance may be provided another face measure physical properties of the skin means further assist in the living body needs face detection.

[0121] 结合图3和图4,本发明的基于多模态的双验证人脸防伪装置包括感应单元301,多模态发生源302、多模态数据采集设备303,多模态人脸检测单元304、多模态双验证人脸防伪单元305 (包括多模态人脸活体检测单元3051,多模态人脸身份验证单元3052),控制单元306以及显示单元307。 [0121] in conjunction with FIGS. 3 and 4, based on multi-modal dual verify face security device includes a sensing unit 301, the multi-modal generating source 302, a multi-modal data acquisition device 303, multimodal face detection according to the present invention. unit 304, a multi-modal security double face authentication unit 305 (detecting unit 3051 include multi-state face the living body, multimodal face authentication unit 3052), the control unit 306 and a display unit 307.

[0122] 感应单元301,用于使用近红外、超声波、或射频方式进行生物特征感应,或者使用可见光摄像头进行实时监控。 [0122] The sensing unit 301, for the near-infrared, ultrasound, or radio frequency biometric sensing mode, or using the real-time monitoring the visible light camera. 该单元用以在特定感应区域内感应人脸的存在,若感应到有人脸,则向控制单元306发出物体存在的信号。 The unit is used in the presence of a specific sensing region of the sensing face, if the face was sensed, the control unit 306 signals the presence of the object. 事实上,感应单元301并不能判断感应到的是人脸,只要有物体出现在感应区内,就认为是感应到了人脸。 In fact, the sensing unit 301 can not determine senses of a human face, as long as the object appears in the sensing area, is considered to be sensitive to the human face. 感应单元301可以使用近红夕卜、超声波、或射频等方式进行人脸感应,也可以简单的使用可见光摄像头进行实时监控。 The sensing unit 301 may use nearly evening Bu red, ultrasonic, or radio frequency sensing face manner, can simply use the real-time monitoring the visible light camera. 特定感应区域的大小和位置优选地设定为可以捕获整个人脸。 The size and position is preferably set to a specific sensing region can capture the entire face.

[0123] 感应单元301感应到人脸的存在,具体执行以下操作:步骤1.如果当前未检测到人脸存在,则继续循环检测;如果检测到人脸的存在,则转入步骤2 ;步骤2,等待一定时间,然后再次检测人脸,如果人脸依旧存在,则认为是有效人脸,并发送信号给控制单元306 ;如果人脸不再存在,则认为是无效人脸,转入步骤I重新开始检测。 [0123] The sensing unit 301 sensing the presence of a human face, in particular the following: 1. If the current step is not the detected face is present, the loop continues detected; if there is a human face is detected, the process proceeds to Step 2; Step 2, wait for a certain time, and then again detect the face, if the face is still present, it is considered to be effective face, and sends a signal to the control unit 306; if the face is no longer present, it is considered invalid face, proceeds to step I re-start the test.

[0124] 在一实例中,感应单元301为可见光摄像头,其以监控的方式进行人脸感应。 [0124] In one example, the sensing unit 301 is a visible light camera, which was induced to monitor face manner. 可见光摄像头循环采集图像并检测是否存在人脸。 The visible light camera and an image acquisition cycle detect the presence of a human face. 如果不存在人脸,则继续采集可见光图像进行人脸检测;如果存在人脸,则等待0.5秒钟再次采集图像并检测人脸。 If there is no human face, continue to collect visible light image of the face detection; if a face is present, it waits for 0.5 seconds to capture images and detect face again. 如果此时人脸还存在,则说明有稳定、有效的人脸出现,然后发送信号给控制单元306,开始相应的图像采集工作;如果等待之后人脸消失,说明此人脸很有可能不是进行多模态图像采集的人脸,认为是噪声而不予理会。 At this time, if there is a human face, then there is a stable, effective face appears, and then send a signal to the control unit 306, starts the corresponding image acquisition working; if after waiting the face disappears, indicating that this is not likely to be a human face multimodal image acquisition face, as noise and ignored. 继续采集可见光图像并检测是否存在人脸。 Continue to collect visible light image and detect the presence of a human face.

[0125] 多模态发生源302可以包括(但不限于)如下一种或多种设备:多个光谱下的主动光源(提供多光谱成像所需的光照),用于3D成像所需的3D结构光,超声波发生器(用以发射超声波)。 Source 302 [0125] Multimodal may occur include (but are not limited to) one or more of the following equipment: active light source (providing the desired multi-spectral imaging light) at a plurality of spectrum required for 3D imaging 3D structured light, ultrasound generator (to transmit ultrasonic). 在多光谱光源中,光谱组合可以包括可见光(此时不需要提供可见光光源),但必须包含一个或一个以上的非可见光光源的组合,光源光谱范围可以为近红外波段(740nm-4000nm),或近紫外波段(360_400nm)。 In the multi-spectral light source, the spectral composition may include a visible light (visible light source need not be provided at this time), but must contain one or more combinations of non-visible light source, the light source may be a spectral range of near-infrared (740nm-4000nm), or near ultraviolet (360_400nm). 也可以包括热红外成像,此时热红外线由人体发出,不必再架设额外光源。 It may also include a thermal infrared imaging, infrared heat emitted by the body at this time, do not have to set up the additional light source. 但光谱组合不应该包括对人体有害的光线,例如中紫外光(290-320nm波长)或近紫外光(200nm-290nm波长)。 However, the spectral composition should not include harmful rays, for example ultraviolet light (wavelength 290-320 nm) or near ultraviolet light (200nm-290nm wavelength). 3D结构光可以根据实际需求配置,例如线激光或3DNIR结构光。 3D structured light may be arranged according to the actual needs, for example, a line laser or structured light 3DNIR. 超声波发生器的频率根据实际需求进行设定,例如,可以设为50kHz ο Frequency of the ultrasonic generator is set according to actual needs, for example, it may be set to 50kHz ο

[0126] 对于其中的多光谱光源,光源发出的光应符合两个原则:1、在合适距离范围内,在多模态发生源302正前方平面中,一定面积内应保持光强大致均匀。 [0126] For multi-spectral light source therein, light from the source should meet two principles: 1, within an appropriate distance range, the multimodal front plane generating source 302, the light should be kept within a certain area substantially even stronger. 如图5所示,在采集设备正前方一定距离(d)处,一定面积内(图中所示圆形)光强应保持均匀。 As shown in FIG. 5, at (d) collecting a certain distance in front of the device, within a certain area (shown in FIG circle) should be maintained uniform light intensity. 2、发光强度应保持在合理范围内,使得成像装置既能清晰的采集到人脸图像,又不至于光强太大而引起用户的不舒适。 2, the luminous intensity should be kept within reasonable limits, such that both the image forming apparatus to clearly capture a facial image, but not so much light intensity caused by the user's discomfort.

[0127] 多模态数据采集设备303,用于采集主动光源照射在人脸上然后反射的多光谱光线,另外也用于采集人体本身所发出的热红外光,人脸的3D图像,以及人脸的超声波成像。 [0127] Multi-modal data acquisition device 303 for acquiring multi-spectral light irradiating active light source and reflected on a human face, and the other thermal infrared light emitted by the human body itself is also used for collection, a 3D image of the face, as well as human ultrasound imaging of the face. 该采集设备包括但不局限于如下一个或多个设备单元:响应各个光源光线的摄像头、响应各个光谱光线的接收管或光敏二极管、热红外感应摄像头或感应器、3D图像采集设备、超声波成像设备或接收器。 The collection device including, but not limited to, the following one or more equipment units: in response camera respective light sources of light, in response to each of the spectral light receiving tube or a photodiode, a thermal infrared sensor cameras or sensors, 3D image capture device, an ultrasonic imaging apparatus or receiver.

[0128]多模态数据采集单元303首先包括对应于302中各个光谱的成像设备用以采集人脸反射的多光谱光线,包括成像设备以及相应滤片,此外还包括热红外、3D、超声波成像设备或感应器。 [0128] multimodal data acquisition unit 303 comprises a first 302 corresponding to the respective spectral imaging apparatus to collect multispectral light reflected from the face, and a corresponding image forming apparatus comprising a filter and further comprising a thermal infrared, 3D, ultrasound imaging devices or sensors. 多光谱成像设备优选良好响应多光谱光源光线的摄像头,此时返回数据类型为图像。 Multi-spectral imaging apparatus preferably good response multispectral light source camera, then return data type as an image. 如果条件有限,也可以使用其他的接收设备,例如响应多光谱光线的接收管、光敏二极管等,此时返回数据类型为反射强度标量。 If the condition is limited, you may be used to receive other devices, for example in response to a multi-spectral light receiving tube, photodiode, etc., when return data type is the reflection intensity scalar. 多光谱光源中的一种光源可以对应一个摄像头,也可以利用单一摄像头响应多个波段的多光谱光源。 A light multi-spectral light source may correspond to a camera, the camera can also use a single light source in response to a plurality of multi-spectral band. 摄像头应在所响应的光谱处有较高的灵敏度。 Camera should have a high sensitivity at the spectral response. 对于超声波成像设备,应与302中的超声波发生器保持频率一致;若条件不允许,也可以选用超声波接收器。 For ultrasound imaging equipment, should be consistent with the frequency of the ultrasonic wave generator 302; if not possible, can also use the ultrasonic receiver. 对于热红外,优选热红外摄像头,也可以选用可以感应温度的感应器。 For the thermal infrared, preferably infrared thermal camera, may be chosen to sense temperature sensor. 对于3D摄像头而言,则采集到的是反映人脸深度信息的图像。 For 3D camera, it is a reflection of the collected facial image depth information.

[0129] 在多模态数据采集单元303的多光谱成像设备中,需要配备对应波段的滤片,用以消除环境光以及其他波段光线对本波段的干扰。 [0129] In the multi-modal data acquisition unit 303 multi-spectral imaging apparatus, it is necessary with the corresponding band filter for eliminating ambient light interference as well as other bands of the light band. 滤片应放置在相应波段的成像设备前面,并紧贴摄像头镜头或接收设备,以防止杂光进入。 Filter should be placed in front of the band corresponding image forming apparatus, and close to the camera lens or the receiving device, in order to prevent stray light from entering.

[0130] 感应单元多模态人脸检测单元304,用于对多模态图像成像设备采集的人脸图像进行预处理,然后对经过预处理的人脸图像进行检测,当所有人脸图像都被检测到人脸和眼睛的情况下认为检测到的是人脸。 [0130] The sensing unit multimodal face detection unit 304, preprocessing for multimodal image forming apparatus acquired human face image, then preprocessed detected face image when the face image are all the case has been detected face and eyes think detected a human face.

[0131] 多模态的双验证人脸防伪单元305,包括多模态人脸活体检测3051与多模态人脸验证3052两个子单元。 [0131] Dual multi-modal security face verification unit 305, including a living multimodal face detection 3051 Multimodal 3052 face authentication two subunits. 其中在多模态人脸活体检测3051中,采用上文中提到的由粗到精的两步策略设计合适的多模态人脸活体分类器;多模态人脸身份验证单元3052,从多模态人脸图像中提取能确定目标身份的信息进行人脸身份验证。 Wherein the multimodal face vivo detection 3051, the use of coarse-to-fine two-step strategy design of suitable multi-modal face biological classifier mentioned hereinabove; multimodal face authentication unit 3052, from the multi- modal face image to extract information can determine the identity of the target for face authentication. 其中,多模态人脸活体检测单元3051和多模态人脸身份验证单元3052,共同组成了本发明的基于多模态的双验证人脸防伪算法的实现单元305。 Wherein the multimodal face living body detecting unit 3051 and a multi-modal face authentication unit 3052, a common component unit based on achieving double security validation face multimodal algorithm 305 of the present invention.

[0132] 控制单元306,用于控制各个单元的工作状态、单元之间的信息通信等工作;显示单元307,用于在输出介质上显示中间结果,方便用户查询。 [0132] The control unit 306 for controlling the operation state of each unit, communication operation information between the units; a display unit 307 for displaying intermediate results on an output medium, user query.

[0133] 控制单元306用以实现多模态发生源302的工作状态以及多模态数据采集单元303的控制。 [0133] The control unit 306 controls to achieve multimodal generating source 302 and the multimodal operation state data acquisition unit 303. 可以用单片机控制,也可以采用PC机连接控制。 MCU can be controlled, PC connected to the control unit may be employed.

[0134] 参照图3和图4,控制单元306的控制方式为:在接收到感应单元301发送的人脸存在信号之后,首先给出控制信号,打开光谱I的光源,然后等待一定的时间给予摄像头曝光,然后采集对应于光谱I的摄像头的图像信号,然后关闭光谱I的光源。 [0134] Referring to FIG. 3 and FIG. 4, the control unit 306 to control: after receiving the presence signal human face sensing unit 301 sends the first control signal is given to open the source spectrum I, and waits for a given time exposing the camera is then acquired spectrum I corresponding to the image signal of the camera, and then close the source spectrum I. 然后给出信号,打开光谱2的光源,等待一定的曝光时间,容纳和采集对应于光谱2的摄像头的图像信号,然后关闭光谱2的光源,依次类推,直到所有光谱的图像数据采集完毕。 Then the signal is given, spectra of the light source 2 is open, waiting for a certain exposure time, receiving and collecting the spectral image signal corresponding to the camera 2, and then close the spectrum of the light source 2, and so on, until all spectra image data acquisition is completed. 如果某一个光谱下没有使用摄像头,而是使用了其他的接收设备,如接收管、光敏二极管等,则读取相应的接收强度数值。 If not using a camera at a certain spectrum, but the use of other receiving devices, such as a receiver tube, photodiode, etc., corresponding to the reception intensity value is read. 然后控制热红外摄像头进行图像采集。 Control then thermal infrared camera for image acquisition. 在热红外之后,控制3D摄像机进行3D人脸图像采集。 After the thermal infrared, control of the 3D camera 3D face image acquisition. 然后控制超声波发射超声波,并用超声波成像设备进行成像。 Control then ultrasonic wave transmitting ultrasonic waves, and ultrasound imaging using the imaging apparatus.

[0135] 一个实例为:控制单元306由上位机PC端软件组成。 [0135] An example is: the control unit 306 by the host computer PC terminal software. 控制单元306在接收到感应单元301发送的信号之后,首先给出光源I的开启命令,等待50ms然后给出对应光源I的摄像头(或接收管)的采集命令,由摄像头(或接收管)采集数据。 The control unit 306 after receiving the signal transmitted from the sensing unit 301 is first given on command source I, and waits 50ms and then to give the corresponding source I camera (or receiver tube) acquisition order, by the camera (or the receiver tube) collected data. 然后令光源I熄灭,给出光源2的开启命令,等待50ms,令光源2的摄像头(或接收管)进行数据采集。 I then turned off so that the light source, the light source turn-on command is given 2, latency 50ms, the camera 2 so that the light source (or receiver tube) for data acquisition. 依次类推,直至所有光源的摄像头都采集到数据为止。 And so on, until all light sources of the camera until the data are collected. 然后采集热红外以及3D图像时,此时不需要等待可以直接进行采集。 When the infrared heat is then collected and 3D images, this time can be collected directly without waiting. 然后开启超声波发射器,并通过超声波成像设备对回波进行接收和成像。 Then open the ultrasonic transmitters, and by receiving and imaging echo ultrasound imaging apparatus. 然后控制单元306会将各摄像头采集到的图像数据送入多模态人脸检测单元304。 Then the control unit 306 will be captured by the camera to the respective image data into multi-modal face detection unit 304.

[0136] 显示单元307用以显示由多模态数据采集单元303采集的人脸图像,并给出各种中间结果或反馈信息,方便人机交互。 [0136] The display unit 307 to display a face image by a multimodal data acquisition unit 303, and gives various intermediate results or feedback information to facilitate human interaction.

[0137] 值得注意的是,若上述的某种模态,没有相应的图像数据采集设备,也可以用其他的非图像式感应仪器代替。 [0137] It is noted that, if some of the above-described mode, there is no corresponding image data acquisition device can also be used in place of other non-image-based sensing instruments.

[0138]图9以举例的方式给出了多模态发生源和多模态数据采集单元的示意图。 [0138] FIG 9 is given by way of example a schematic diagram of multi-modal and multi-modal generating source data acquisition unit. 其中,多模态图像采集装置面板804起到装置框架的作用。 Wherein the multimodal image pickup apparatus 804 functions as a panel frame of the device. 面板分为上下两个部分,上半部分为多模态发生源901和多模态数据采集单元902,下半部分为显示单元905,由一块IXD屏幕组成,在两部分之间为一个可见光摄像头903,作为感应单元使用。 The panel is divided into two parts, the upper half of a multi-modal generating source 901 and a multi-modal data acquisition unit 902, the lower half of the display unit 905 by a IXD screens, between two portions of a visible light camera 903, is used as a sensing unit. 在上半部分中,三个多模态发射源分别为SOOnm多光谱光源、3D结构光源和超声波发射源。 In the upper part, three multimodal emission sources are SOOnm multispectral light source, a light source and a 3D ultrasound emission source structure. 三种发射源交叉排列,并组成矩形,这样可以保证每个发射源在装置前方一定范围内都可以形成均匀分布。 Three kinds of staggered radiation source and the composition of the rectangle, so that each emitting source can ensure a uniform distribution may be formed in the front of the apparatus a certain range. 在发射源源中央为四个成像设备(或接收设备),包括多光谱成像设备(摄像头前方都覆盖相应波段的滤片,以防止可见光或其他光谱光线的干扰)、热红外摄像头(用于采集热红外图像)、3D以及超声波成像设备。 The emission center of the stream of four imaging device (or reception device), comprising a multi-spectral imaging apparatus (a camera in front of all cover the respective band filter to prevent interference visible light or other spectral light), thermal infrared cameras (for acquiring heat infrared images), 3D imaging apparatus and ultrasound. 进行测试时,人脸应正面面对该采集装置。 When tested, people should face the front of the face of the collection device. 控制单元并不包含在多光谱采集装置的面版上,而是独立成一个部分(可以是单片机,也可以是上位机软件),与多光谱采集装置面板通过控制信号线相连接。 The control unit is not included on the face plate multispectral acquisition device, but as a separate part (may be a microcontroller, or may be PC software), connected to the panel multispectral acquisition device via the control signal line.

[0139] 多模态人脸检测单元、多模态双验证人脸防伪单元均为上位机的应用程序,在接收到采集到的多模态人脸图像后,分别送至以上两个单元,并给出相应的结果。 [0139] Multimodal face detection unit multimodal double face authentication unit are security application of host computer, after receiving the acquired face image multimodal, respectively, to the two or more units, and the corresponding results.

[0140] 上述基于多模态的双验证人脸防伪装置的工作流程如图4所示。 [0140] The multimodal-based workflow double face authentication security device is shown in FIG. 参照图4,首先由感应单元401感应人脸的存在;如果不存在人脸,则继续循环检测,而事实上,感应单元401并不能判断检测到的是人脸,只是在感应到有物体存在的时候,即认为是感应到人脸;如果存在人脸,则发出命令给控制单元402,由控制单元402发出控制命令,指导多模态发生源403开启、关闭,以及多模态数据采集单元404采集数据;然后进入多模态人脸检测单元405进行人脸检测,如果有的图像中没检测到人脸,则发信号给显示单元407。 Referring to Figure 4, first by the sensing unit 401 sensing the presence of a human face; if there is no human face detection loop continues, in fact, the sensing unit 401 can not determine that the detected human face, but in the presence of an object sensed when it that is sensitive to the human face; if there is a human face, issuing commands to the control unit 402, issued by the control unit 402 control command guidance multimodal generating source 403 is turned on, off, and multi-modal data collection unit 404 data collection; multimodal then enters the face detection unit 405 detects the face, and if the image is not a face is detected, a signal is sent to the display unit 407. 输出检测失败的信息,并返回感应单元401,重新进行图像采集;如果所有模态图像都检测到人脸,则进入多模态双验证人脸防伪单元406,并发信号给显示单元407,以便输出人脸检测信息或显示捕获的某张人脸图像;进入多模态双验证人脸防伪单元406之后进行人脸活体检测判断4061以及人脸身份验证4062,如果为造假人脸则通过显示单元407给出相应活体检测失败的信息,并返回感应单元401,进行新一轮的图像采集;如果为真人人脸也由显示单元407给出,然后等待一段时间,返回感应单元401,开始新一轮的人脸检测。 Failure information output detection, and returns the sensing unit 401, re-image capture; if all modality image are face is detected, the process proceeds multimodal dual verify face security unit 406, and transmits the signal to the display unit 407 so as to output a Zhang face image face detection information, or display the captured; enter multimodal dual verify face security unit face vivo detection judging 4061 and face authentication 4062 after 406, the display unit if false face through 407 to give the corresponding living body information detection fails, and returns the sensing unit 401, a new image acquisition; given a real face 407 is also a display unit, and then waits for a period of time, the sensing unit 401 returns to begin a new round of face detection.

[0141] 在一实例中,多模态人脸检测单元405为上位机PC端的应用程序,用于对多模态数据采集装置403采集到的每张图像调用相应的人脸检测分类器进行人脸检测。 [0141] In one example, multimodal face detection unit 405 is a PC application PC terminal, means 403 for each image acquired call the corresponding face of the classifier for multimodal data acquisition al face detection. 如果全部图像都检测到人脸,输送某张人脸图像给显示单元407用于显示(例如,选用可见光下的人脸图像),并将检测到的所有光谱下的人脸图像输入至多模态双验证人脸防伪单元406。 If all the images have a face is detected, a sheet conveying face image to the display unit 407 to display (e.g., the face image selection visible light), and the input face image detected in all spectra at most modalities double security verification unit 406 face. 若没有全部检测到人脸,则输送检测失败的结果给显示单元407显示,并返回感应单元401,重新开始图像感应。 If not all the detected face, the failure detection result is supplied to the display unit 407 display unit 401 and returns to sensing, image sensing is restarted.

[0142] 最后,本发明须指出,利用本发明提出的双验证人脸防伪方法及其装置,用户可以根据自己的需要来适用于不同的生物模态,例如人脸、虹膜等。 [0142] Finally, the present invention is to be noted that the use of a double face authentication security method and apparatus proposed by the present invention, a user may be applied to various biological modalities according to their needs, for example, face, iris and the like. 并且可以根据实际情况自由选择模态组合,例如,可以单独选用不同的光谱组合,也可以结合热红外光、3D图像或超声波成像联合使用。 And a combination of modalities can be freely selected according to the actual situation, for example, can be selected individually different spectral composition, may be combined with thermal infrared light, an image or 3D ultrasound imaging in combination.

[0143] 以上所述的具体实施例,对本发明的目的、技术方案和有益效果进行了进一步详细说明,所应理解的是,以上所述仅为本发明的具体实施例而已,并不用于限制本发明,凡在本发明的精神和原则之内,所做的任何修改、等同替换、改进等,均应包含在本发明的保护范围之内。 Specific Example [0143] above, the objectives, technical solutions, and beneficial effects of the present invention will be further described in detail, it should be understood that the above descriptions are merely embodiments of the present invention, but not intended to limit the present invention, within the spirit and principle of the present invention, any modifications, equivalent replacements, improvements, etc., should be included within the scope of the present invention.

Claims (10)

1.一种双验证人脸防伪方法,其特征是,所述方法包括: 步骤I,对采集的目标人脸进行活体检测,判断目标人脸是否具有生物活性,如果目标人脸被认定具有活体特性,则转入步骤2 ; 步骤2,如果是在人脸识别应用中,则计算采集到的目标人脸与识别结果对应的人脸之间的相似度,若大于某一阈值,则认为该目标人脸是真实有效的人脸; 如果是在人脸验证应用中,则计算采集到的目标人脸与目标人脸所声称的身份对应的人脸之间的相似度,若大于某一阈值,则认为该目标人脸是真实有效的人脸, 其中步骤I与指定人无关,步骤2与指定人有关,当目标人脸同时通过步骤I和步骤2的验证之后,才能被认定为是真实有效的人脸,否则被认定为是虚假人脸; 如果所述双验证人脸防伪方法是基于多模态,则步骤I进一步包括: 步骤101,粗略判断目标人脸的生物 A double-face authentication security method, wherein the method comprises: a step I, the acquisition of the target person's face for detection in vivo, it is determined whether the target face having biological activity, if the target has a face is recognized in vivo characteristic, the process proceeds to step 2; step 2, if it is in face recognition applications, the degree of similarity between the acquired target face and the corresponding face recognition results is calculated, if more than a certain threshold value, it is considered that target face is the face of a real and effective; If the similarity between the face authentication application, the acquisition target to the target face and the face claimed identity is calculated corresponding to the human face, if more than a certain threshold after that, it is considered that the target face is real and effective human face, wherein the step I nothing to do with the nominee, and the nominee about the step 2, when the target face and to verify step by step I and 2, in order to be recognized as the true effective face, false otherwise been identified as a human face; and if the double-face authentication method is based on the security multimodal, further comprising the step I: 101, determines the target face roughly biological step 性,其中按照下面的方式中的一种或多种进行判断:通过热红外判断目标人脸的温度,判断是否接近37度;通过3D图像判断人脸的深度信息,判断面部是否为3D物体;通过超声波反射分析目标人脸的超声波反射率,判断皮肤的超声波反射率是否与真实人脸相似;通过多光谱成像分析目标人脸在不同光谱下的反射率,判断皮肤的多光谱反射率是否与真实人脸相似,如果通过上述一种或多种方式判断目标人脸的信息指标与真实人脸相似,则进入步骤102 ; 步骤102,精确判断目标人脸的生物活性,将采集到的多光谱人脸图像,利用互商图像算法进行准确的活体判断; 只有目标人脸同时通过步骤101和102,才被认为通过步骤I的活体检测。 Of which the following manner for determining one or more of: determining a target face by thermal infrared temperature is close to 37 degrees is determined; determined by the depth information of a 3D image of a human face, the face is determined whether a 3D object; ultrasonic ultrasound reflectivity of the analysis target face, determined skin ultrasound reflection rate is similar to the real face; the reflectivity of the target face at different spectra by a multispectral imaging analysis, determine whether the skin multispectral reflectance similar to a real human face, if the face is determined by said target one or more ways information indicator similar to a real human face, the process proceeds to step 102; step 102, to accurately determine the biological activity of the target face, the collected multispectral face image, accurate determination of the living body image using the cross supplier algorithm; while only the target face at step 101 and 102, was considered by the step of I in vivo detection.
2.根据权利要求1所述的双验证人脸防伪方法,其特征是,如果所述双验证人脸防伪方法是基于可见光,则步骤I进一步包括: 步骤101,对目标人脸进行活体检测,首先采集大量真实、虚假人脸样本,对目标人脸提取各种纹理特征,训练活体检测纹理分类器,若目标人脸被活体检测纹理分类器认定为真实人脸,则进入步骤2,否则认定为虚假人脸; 步骤102,通过人机交互确定目标人脸的有效性,其中系统发出指令,要求用户做出一定的动作,然后系统不断检测目标人脸是否做出相应动作,若在一定时间内检测到上述动作的发生,则判断目标人脸为真实人脸,否则为虚假人脸; 只有目标人脸同时通过步骤101和102,才被认为通过步骤I的活体检测。 2. Verify double face of the anti-counterfeiting method as claimed in claim 1, characterized in that, if the double-face security authentication method is based on the visible light, I further comprising the step of: step 101, the target for in vivo detection face, first collected a large number of true and false face samples, the target face extracted texture features, training vivo detection of texture classification, if the target face has been identified in vivo detection of texture classification is as real human face, go to step 2, otherwise identified human face is false; step 102, it is determined by the validity of the target interactive face, wherein the system instructs the user to make certain actions required, then the system continuously made as to whether the detection target face corresponding action, if at a certain time detected within the above-described operation occurs, it is determined that the target face to face transactions, otherwise false face; while only the target face at step 101 and 102, was considered by the step of I in vivo detection.
3.根据权利要求2所述的双验证人脸防伪方法,其特征是,步骤2进一步包括: 步骤201,首先采集大量真实人脸图像,对每张人脸图像提取其纹理特征; 步骤202,然后将采集的所有人脸图像的特征向量两两相减,根据两图像是否属于同一个人,将相减后的特征向量分为类内、类间两类,利用机器学习算法训练一个两类分类器,由此训练得到的分类器可以判断输入的两个特征向量是否属于同一个人; 步骤203,如果是在人脸识别应用中,若目标人脸图像与识别结果对应的人脸图像,被步骤202中的分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸; 如果是在人脸验证应用中,则目标人脸图像与所声称的指定人身份对应的人脸图像,被步骤202中的分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸。 3. Verify double face of the anti-counterfeiting method as claimed in claim 2, wherein step 2 further comprising: a step 201, a large number of real acquired first facial image, texture features extracted face image of each person; step 202, then everyone will feature vectors of face images acquired two two subtraction, depending on whether the two images belong to the same person, the feature vector after the subtraction divided into categories, between two categories, using a machine learning algorithm to train a binary classification , thereby to obtain a trained classifier based on two input feature vectors belong to the same person; step 203, if it is in face recognition applications, if the target image and the face recognition result corresponding to the face image, the step of 202 classifier identified as belonging to the same person, then that target face a real and effective, otherwise false face; if it is in the face verification applications, the target face image with the claimed identity of the nominee corresponding face the image was identified in step 202 is classified as belonging to the same person, then that target face a real and effective, otherwise false face.
4.根据权利要求1所述的双验证人脸防伪方法,其特征是,互商图像算法包括如下步骤:步骤1021,采集大量真人人脸和虚假人脸在不同距离下的多光谱成像构成训练数据集,对于同一个人的任意两张不同光谱下的图像进行像素级的相除,组成互商图像组,假设任意选定两个光谱入” λ2,同一个人脸在两个光谱下的图像为&和/4,其互商图像定义如下: The double-face authentication of the anti-counterfeiting method as claimed in claim 1, characterized in that the mutual image supplier algorithm comprises the following steps: Step 1021, a large number of collecting live training constituting false faces and face multispectral imaging at different distances data set, for the different spectral images of the same person at any stage of the two divided pixel composition mutual commercially image group, the spectrum is assumed arbitrarily selected two "λ2, the same person in the face image in the two spectra and & / 4, which cross quotient image defined as follows:
Figure CN102622588BC00031
其中,P表示人脸的反射率,K代表光源在人脸表面处的强度,ζ代表人脸距离光源的距离,(X,y)代表人脸图像上的坐标; 步骤1022,对于所有的互商图像,在多个尺度上划分为多个重叠或不重叠的小块,提取每个小块的特征向量,将所有小块的特征向量进行组合,作为全局的特征向量; 步骤1023,基于统计学习方法,在训练数据集上训练分类器,用于区分真实、虚假人脸。 Wherein, P represents the reflectance of the human face, K represents the distance of the intensity at the surface of the face, the face from the light source [zeta] representative of the coordinates of the representative face image source (X, y); step 1022, for all mutual quotient image, divided over a plurality of scales into a plurality of pieces or may not overlap, a feature vector is extracted for each tile, the feature vectors of all the pieces are combined, as the global feature vectors; step 1023, based on statistical learning, training in the training data set classifier used to distinguish between true and false faces.
5.根据权利要求1所述的双验证人脸防伪方法,其特征是,步骤2进一步包括: 步骤201,采集大量真实人脸的多模态图像,对每张图像提取其纹理特征; 步骤202,将图像的特征向量两两相减,根据两图像是否属于同一个人,将相减后的特征向量分为类内、类间两类,利用机器学习算法训练一个两类分类器,训练得到的分类器能够判断输入的两个特征向量是否属于同一个人; 步骤203,如果是在人脸识别应用中,若目标人脸图像与识别结果对应的人脸图像,被步骤202中的分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸; 如果是在人脸验证应用中,若目标人脸图像与所声称的指定人身份对应的人脸图像,被步骤202中的分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸。 The double-face authentication of the anti-counterfeiting method as claimed in claim 1, wherein step 2 further comprises: Step 201, collecting a large number of real multimodal image face, extracted texture features of each image; step 202 the image feature vector pairwise subtracting, depending on whether two images belong to the same person, the feature vectors after the subtraction is divided into classes, between two categories, using a machine learning algorithm to train a classifier two training obtained two sorter capable of determining whether the input feature vectors belonging to the same person; step 203, if it is in face recognition applications, if the target face image and face image corresponding to the recognition result, was identified in step 202 is classified as belong to the same person, then that target face a real and effective, otherwise false face for the people; if it is in the face verification applications, if the target face image with the claimed identity of the nominee corresponding face image, is in step 202 classifier identified as belonging to the same person, then that target face a real and effective, otherwise false face.
6.根据权利要求1所述的双验证人脸防伪方法,其特征是,每种不同的成像类型被称为一个模态,成像类型包括可见光成像,近红外成像,近紫外成像,热红外成像或超声波成像。 The double-face authentication of the anti-counterfeiting method as claimed in claim 1, characterized in that each different type is called an imaging modality, including the type of imaging visible light imaging, near infrared imaging, imaging the near ultraviolet, infrared thermal imaging or ultrasound imaging.
7.—种双验证人脸防伪装置,该装置包括: 感应单元,用于使用近红外、超声波、射频方式或可见光摄像头中的一种或多种,通过实时监控的方式,感应人脸的存在; 多模态发生源,包含多个光谱下的主动光源、用于3D成像所需的3D结构光或者超声波发生器中的一种或多种; 多模态数据采集设备,用于采集人脸的多光谱成像,人体本身所发出的热红外光成像,人脸的3D图像或超声波成像中的一种或多种; 多模态人脸检测单元,用于检测多模态图像中的人脸位置,并将检测到的人脸图像发送到多模态双验证人脸防伪单元; 多模态双验证人脸防伪单元,用于验证目标人脸是否为真实有效的人脸; 显示单元,用于显示人脸防伪结果, 其中,多模态双验证人脸防伪单元进一步包括:多模态人脸活体检测单元,用于对目标人脸进行活体检测;多模态人脸验证 7.- Double kinds of face authentication security device, the apparatus comprising: sensing means for using one or more near-infrared, ultrasonic, radio frequency or visible light camera mode is, by way of real-time monitoring, sensing the presence of a human face ; multimodal generating source, comprising a plurality of active light source in the spectrum, for one or more desired imaging 3D 3D structured light or an ultrasonic generator; multimodal data acquisition device for acquiring facial the multi-spectral imaging, thermal imaging infrared light emitted by the human body itself, ultrasound imaging or 3D image face of one or more; multimodal face detection unit for detecting a multi-modal images in face position and the detected face image to the multi-modal security double face authentication unit; multimodal double face security verification unit configured to verify whether a human face as a target real and effective face; a display unit, with security results on the display face, wherein the multimodal double face authentication security unit further comprises: a multimodal vivo face detection unit, a target for in vivo detection of human faces; multimodal face verification 元,用于对目标人脸进行身份验证; 其特征是,所述多模态人脸活体检测单元对目标人脸进行活体检测时,首先,粗略判断目标人脸的生物活性,其中按照下面的方式中的一种或多种进行判断:通过热红外判断目标人脸的温度,判断是否接近37度;通过3D图像判断人脸的深度信息,判断面部是否为3D物体;通过超声波反射分析目标人脸的超声波反射率,判断皮肤的超声波反射率是否与真实人脸相似;通过多光谱成像分析目标人脸在不同光谱下的反射率,判断皮肤的多光谱反射率是否与真实人脸相似,如果通过上述一种或多种方式判断目标人脸的信息指标与真实人脸相似,则继续精确判断目标人脸的生物活性,将采集到的多光谱人脸图像,利用互商图像算法进行准确的活体判断。 Element, for the target face authentication; wherein said multimodal person when the face living body detecting unit target human faces in vivo detection, first, a rough determination of biological activity of the target face, which in the following embodiment one or more of the determination: analyzing the target face by thermal infrared temperature is close to 37 degrees is determined; determining depth information of the 3D image of a human face, the face is determined whether a 3D object; target person by ultrasonic reflection analysis the ultrasonic reflectance of the face, determining skin ultrasound reflection rate is similar to the real face; the reflectivity of the target face at different spectra by a multispectral imaging analysis, determine whether the skin multispectral reflectance similar to the real face, if Analyzing the target face of one or more ways by the above information indicator similar to the real face, continues to accurately determine the biological activity of the target face, the collected multispectral face image, accurate image using the cross commercially algorithm Analyzing the living body.
8.根据权利要求7所述的双验证人脸防伪装置,其特征是,互商图像算法包括如下步骤:采集大量真人人脸和虚假人脸在不同距离下的多光谱成像构成训练数据集,对于同一个人的任意两张不同光谱下的图像进行像素级的相除,组成互商图像组,假设任意选定两个光谱X1, λ 2,同一个人脸在两个光谱下的图像为/Λ和及,其互商图像定义如下: 8. Double face authentication security device according to claim 7, characterized in that the mutual image supplier algorithm comprises the steps of: acquiring a plurality of faces and false real face constituting training data set in the multi-spectral imaging at different distances, carried out for the image of the same person at any two different spectra divided by the pixel level, image group composed of mutual providers, assuming two arbitrarily selected spectral X1, λ 2, individual images of the same face in the two spectra is / Λ and and that mutual quotient image is defined as follows:
Figure CN102622588BC00041
其中,P表示人脸的反射率,K代表光源在人脸表面处的强度,ζ代表人脸距离光源的距离,(X,y)代表人脸图像上的坐标; 对于所有的互商图像,在多个尺度上划分为多个重叠或不重叠的小块,提取每个小块的特征向量,将所有小块的特征向量进行组合,作为全局的特征向量; 基于统计学习方法,在训练数据集上训练分类器,用于区分真实、虚假人脸。 Wherein, P represents the reflectance of the human face, K represents the intensity of light at the surface of the human face, the face distance from the light source [zeta] representative of, (X, y) coordinate on the representative face image; for all mutual quotient image, divided in a number of scales into a plurality of pieces or may not overlap, a feature vector is extracted for each tile, the feature vectors of all the pieces are combined, as the global feature vectors; method based on statistical learning, training data classifier on the training set, used to distinguish between true and false faces.
9.根据权利要求7所述的双验证人脸防伪装置,其特征是,多模态人脸验证单元对目标人脸进行身份验证时,首先采集大量真实人脸的多模态图像,对每张图像提取其纹理特征;其次,将图像的特征向量两两相减,根据两图像是否属于同一个人,将相减后的特征向量分为类内、类间两类,利用·机器学习算法训练一个两类分类器,训练得到的分类器能够判断输入的两个特征向量是否属于同一个人;如果是在人脸识别应用中,若目标人脸图像与识别结果对应的人脸图像,被上述两类分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸;如果是在人脸验证应用中,若目标人脸图像与所声称的指定人身份对应的人脸图像,被上述两类分类器认定为属于同一人,则认为目标人脸真实有效,否则为虚假人脸。 9. A multi-modal images bis face authentication security device according to claim 7, wherein the multimodal face authentication unit authenticates the target face, collecting a large number of real first face, each images extracted texture features; secondly, the image feature vectors of pairwise subtracting, depending on whether two images belong to the same person, the feature vector subtracted divided into categories, between the two categories, using a machine learning algorithm trained · is a binary classification, the trained classifier can be determined whether the two input feature vectors belonging to the same person; if it is in face recognition applications, if the target image and the face recognition result corresponding to the face images by the above two class classifier identified as belonging to the same person, then that target face a real and effective, otherwise false face for the people; if it is in the face verification applications, if the target face image with the claimed identity of the nominee corresponding face image, these two types of classifiers have been identified as belonging to the same person, then that target face a real and effective, otherwise false face.
10.根据权利要求7-9任一项所述的双验证人脸防伪装置,其特征是:每种不同的成像类型被称为一个模态,成像类型包括可见光成像,近红外成像,近紫外成像,热红外成像或超声波成像。 10. The double face authentication security device according to any of claims 7-9, characterized in that: each of the different types is called an imaging modality, including the type of imaging visible imaging, infrared imaging near, near-ultraviolet imaging, ultrasound imaging or thermal infrared imaging.
CN2012100594547A 2012-03-08 2012-03-08 Dual-certification face anti-counterfeit method and device CN102622588B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2012100594547A CN102622588B (en) 2012-03-08 2012-03-08 Dual-certification face anti-counterfeit method and device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN2012100594547A CN102622588B (en) 2012-03-08 2012-03-08 Dual-certification face anti-counterfeit method and device
PCT/CN2013/000228 WO2013131407A1 (en) 2012-03-08 2013-03-05 Double verification face anti-counterfeiting method and device

Publications (2)

Publication Number Publication Date
CN102622588A CN102622588A (en) 2012-08-01
CN102622588B true CN102622588B (en) 2013-10-09

Family

ID=46562498

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2012100594547A CN102622588B (en) 2012-03-08 2012-03-08 Dual-certification face anti-counterfeit method and device

Country Status (2)

Country Link
CN (1) CN102622588B (en)
WO (1) WO2013131407A1 (en)

Families Citing this family (55)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102622588B (en) * 2012-03-08 2013-10-09 无锡中科奥森科技有限公司 Dual-certification face anti-counterfeit method and device
US9032510B2 (en) * 2012-09-11 2015-05-12 Sony Corporation Gesture- and expression-based authentication
CN103036680A (en) * 2012-12-10 2013-04-10 中国科学院计算机网络信息中心 Realm name certification system and method based on biological feature recognition
CN103198294B (en) * 2013-02-22 2016-05-11 广州市朗辰电子科技有限公司 A method of identifying the type of video or image viewpoint
CN104683302A (en) * 2013-11-29 2015-06-03 国际商业机器公司 Authentication method, authentication device, terminal equipment, authentication server and system
CN103634120A (en) * 2013-12-18 2014-03-12 上海市数字证书认证中心有限公司 Method and system for real-name authentication based on face recognition
CN103793690B (en) * 2014-01-27 2017-08-18 天津科技大学 The body of living organisms and Application of a method for detecting blood flow based on the detected skin
WO2015136675A1 (en) * 2014-03-13 2015-09-17 日本電気株式会社 Detecting device, detecting method, and recording medium
CN103886301B (en) * 2014-03-28 2017-01-18 北京中科奥森数据科技有限公司 One human face detection method in vivo
CN103984941B (en) * 2014-06-10 2017-04-12 深圳市赛为智能股份有限公司 Attendance recognition method and apparatus
US9251427B1 (en) 2014-08-12 2016-02-02 Microsoft Technology Licensing, Llc False face representation identification
US9633269B2 (en) * 2014-09-05 2017-04-25 Qualcomm Incorporated Image-based liveness detection for ultrasonic fingerprints
CN105472296B (en) * 2014-09-09 2019-02-05 联想(北京)有限公司 Real-time method of calibration and device
CN105868677A (en) * 2015-01-19 2016-08-17 阿里巴巴集团控股有限公司 Live human face detection method and device
CN105447441B (en) * 2015-03-19 2019-03-29 北京眼神智能科技有限公司 Face authentication method and device
CN104794386A (en) * 2015-04-08 2015-07-22 天脉聚源(北京)传媒科技有限公司 Data processing method and device based on face recognition
US10275672B2 (en) 2015-04-29 2019-04-30 Beijing Kuangshi Technology Co., Ltd. Method and apparatus for authenticating liveness face, and computer program product thereof
US20180173979A1 (en) * 2015-06-29 2018-06-21 Beijing Kuangshi Technology Co., Ltd. Living body detection method, living body detection system, and computer program product
CN106557726A (en) * 2015-09-25 2017-04-05 北京市商汤科技开发有限公司 Face identity identification system and method with silence detection in vivo
CN106341380B (en) * 2015-10-15 2018-02-16 收付宝科技有限公司 A method for remote user authentication, apparatus and system
CN105426815A (en) * 2015-10-29 2016-03-23 北京汉王智远科技有限公司 Living body detection method and device
US20180285668A1 (en) * 2015-10-30 2018-10-04 Microsoft Technology Licensing, Llc Spoofed face detection
CN105488486B (en) * 2015-12-07 2018-10-30 清华大学 Face recognition method and apparatus for preventing attacks photo
CN105512632B (en) 2015-12-09 2019-04-05 北京旷视科技有限公司 Biopsy method and device
CN105447483B (en) * 2015-12-31 2019-03-22 徐州旷视数据科技有限公司 Biopsy method and device
CN107182218A (en) * 2015-12-31 2017-09-19 深圳先进技术研究院 Authentication method and apparatus
US9619723B1 (en) * 2016-02-17 2017-04-11 Hong Kong Applied Science and Technology Research Institute Company Limited Method and system of identification and authentication using facial expression
WO2017151747A1 (en) * 2016-03-02 2017-09-08 EyeVerify Inc. Spoof detection using proximity sensors
CN105893828A (en) * 2016-05-05 2016-08-24 南京甄视智能科技有限公司 Human face verification driving testing system and method based on mobile terminal
CN106096519A (en) * 2016-06-01 2016-11-09 腾讯科技(深圳)有限公司 Living body identification method and apparatus
CN106127103B (en) * 2016-06-12 2019-06-25 广州广电运通金融电子股份有限公司 A kind of offline identity authentication method and device
CN106203320A (en) * 2016-07-06 2016-12-07 惠州Tcl移动通信有限公司 Human face recognition optimization method and system based on mobile terminal
CN106407914A (en) * 2016-08-31 2017-02-15 北京旷视科技有限公司 Method for detecting human faces, device and remote teller machine system
US20180068160A1 (en) 2016-09-07 2018-03-08 Mei-Yen Lee Optical imaging system with variable light field for biometrics application
CN106372615A (en) * 2016-09-19 2017-02-01 厦门中控生物识别信息技术有限公司 Face anti-counterfeiting identification method and apparatus
CN106599772A (en) * 2016-10-31 2017-04-26 北京旷视科技有限公司 Living body authentication method, identity authentication method and device
CN106682578A (en) * 2016-11-21 2017-05-17 北京交通大学 Human face recognition method based on blink detection
CN106845916A (en) * 2016-11-30 2017-06-13 浙江水马环保科技有限公司 Intelligent APP attendance management and state monitoring method based on water purifier
CN106845917A (en) * 2016-11-30 2017-06-13 浙江水马环保科技有限公司 Intelligent APP attendance management system based on water purifiers
CN106651302A (en) * 2016-11-30 2017-05-10 浙江水马环保科技有限公司 Intelligent PC attendance management and state monitoring method through water purifier
CN106845915A (en) * 2016-11-30 2017-06-13 浙江水马环保科技有限公司 Water purifier intelligent PC attendance management system
CN106778559A (en) * 2016-12-01 2017-05-31 北京旷视科技有限公司 Living body detection method and apparatus
CN106845345A (en) * 2016-12-15 2017-06-13 重庆凯泽科技股份有限公司 Living body detection method and device
CN106682607A (en) * 2016-12-23 2017-05-17 山东师范大学 Offline face recognition system and offline face recognition method based on low-power-consumption embedded and infrared triggering
CN106650666A (en) * 2016-12-26 2017-05-10 北京旷视科技有限公司 Method and device for detection in vivo
CN108229325A (en) * 2017-03-16 2018-06-29 北京市商汤科技开发有限公司 Face detection method and system, electronic equipment, program and medium
CN106980836A (en) * 2017-03-28 2017-07-25 北京小米移动软件有限公司 Identity verification method and device
CN107273794A (en) * 2017-04-28 2017-10-20 北京建筑大学 Method and apparatus for identifying living body in human face identification process
US10303932B2 (en) * 2017-07-05 2019-05-28 Midea Group Co., Ltd. Face recognition in a residential environment
CN107454335A (en) * 2017-08-31 2017-12-08 广东欧珀移动通信有限公司 Image processing method and apparatus thereof, computer readable storage medium and mobile terminal
CN107563329A (en) * 2017-09-01 2018-01-09 广东欧珀移动通信有限公司 Image processing method and device, computer-readable storage medium and mobile terminal
CN107808115A (en) * 2017-09-27 2018-03-16 联想(北京)有限公司 Living body detection method, device and storage medium
CN107844764A (en) * 2017-10-31 2018-03-27 广东欧珀移动通信有限公司 Image processing method and device, electronic equipment, and computer readable storage medium
CN108564033A (en) * 2018-04-12 2018-09-21 Oppo广东移动通信有限公司 Security verification method and device based on structured light and terminal equipment
CN108596061A (en) * 2018-04-12 2018-09-28 Oppo广东移动通信有限公司 Face recognition method and device, mobile terminal and storage medium

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101702198B (en) * 2009-11-19 2011-11-23 浙江大学 Identification method for video and living body faces based on background comparison
CN101770613A (en) * 2010-01-19 2010-07-07 北京智慧眼科技发展有限公司 Social insurance identity authentication method based on face recognition and living body detection
US8675926B2 (en) * 2010-06-08 2014-03-18 Microsoft Corporation Distinguishing live faces from flat surfaces
CN101964056B (en) * 2010-10-26 2012-06-27 徐勇 Bimodal face authentication method with living body detection function and system
CN102622588B (en) * 2012-03-08 2013-10-09 无锡中科奥森科技有限公司 Dual-certification face anti-counterfeit method and device

Also Published As

Publication number Publication date
CN102622588A (en) 2012-08-01
WO2013131407A1 (en) 2013-09-12

Similar Documents

Publication Publication Date Title
Wang et al. Infrared imaging of hand vein patterns for biometric purposes
US7881524B2 (en) Information processing apparatus and information processing method
Lee A novel biometric system based on palm vein image
CN102339382B (en) Multispectral imaging biometrics
KR101415287B1 (en) Method, computer-readable storage device and computing device for liveness detercion
Bowyer et al. Image understanding for iris biometrics: A survey
CN101536384B (en) Spatial-spectral fingerprint spoof detection
KR101356358B1 (en) Computer-implemented method and apparatus for biometric authentication based on images of an eye
Zhang et al. Face liveness detection by learning multispectral reflectance distributions.
CN103440479B (en) A living human face detection method and system
US9311535B2 (en) Texture features for biometric authentication
US9836647B2 (en) Iris biometric recognition module and access control assembly
Sun et al. Improving iris recognition accuracy via cascaded classifiers
US8391590B2 (en) System and method for three-dimensional biometric data feature detection and recognition
KR101393717B1 (en) Facial recognition technology
US20080212849A1 (en) Method and Apparatus For Facial Image Acquisition and Recognition
US8483450B1 (en) Quality metrics for biometric authentication
CN101404060B (en) Human face recognition method based on visible light and near-infrared Gabor information amalgamation
CN101030244B (en) Automatic identity discriminating method based on human-body physiological image sequencing estimating characteristic
CN101669824B (en) Biometrics-based device for detecting indentity of people and identification
CN100453040C (en) Identity recognition instrument based on characteristics of subcutaneous vein of dorsum of hand and recognition method
CN100458831C (en) Human face model training module and method, human face real-time certification system and method
CN101485570B (en) Vein authentication device and vein authentication method
US20050238208A1 (en) Handheld biometric computer for 2D/3D image capture
CN2828935Y (en) Image obtaining recognition device for human face recognition

Legal Events

Date Code Title Description
C06 Publication
C10 Entry into substantive examination
C53 Correction of patent for invention or patent application
COR Change of bibliographic data

Free format text: CORRECT: APPLICANT; FROM: WUXI DIGIT AUTHENMETRIC TECHNOLOGY CO. LTD. TO: AUTHENMETRIC CO. LTD.

C14 Grant of patent or utility model
C41 Transfer of patent application or patent right or utility model
C41 Transfer of patent application or patent right or utility model