WO2020151489A1 - 基于面部识别的活体检测的方法、电子设备和存储介质 - Google Patents

基于面部识别的活体检测的方法、电子设备和存储介质 Download PDF

Info

Publication number
WO2020151489A1
WO2020151489A1 PCT/CN2020/070712 CN2020070712W WO2020151489A1 WO 2020151489 A1 WO2020151489 A1 WO 2020151489A1 CN 2020070712 W CN2020070712 W CN 2020070712W WO 2020151489 A1 WO2020151489 A1 WO 2020151489A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
visible light
infrared
measured
light image
Prior art date
Application number
PCT/CN2020/070712
Other languages
English (en)
French (fr)
Inventor
王升国
赵先林
申川
Original Assignee
杭州海康威视数字技术股份有限公司
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 杭州海康威视数字技术股份有限公司 filed Critical 杭州海康威视数字技术股份有限公司
Priority to EP20744256.7A priority Critical patent/EP3916627A4/en
Priority to US17/425,242 priority patent/US11830230B2/en
Publication of WO2020151489A1 publication Critical patent/WO2020151489A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/10—Image acquisition
    • G06V10/12—Details of acquisition arrangements; Constructional details thereof
    • G06V10/14—Optical characteristics of the device performing the acquisition or on the illumination arrangements
    • G06V10/143—Sensing or illuminating at different wavelengths
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/54—Extraction of image or video features relating to texture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation
    • G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/40—Spoof detection, e.g. liveness detection
    • G06V40/45—Detection of the body part being alive

Definitions

  • This application relates to the technical field of face recognition, and in particular to a method, electronic equipment and storage medium for living body detection based on face recognition.
  • biometric identification has been applied in the security industry.
  • biometric recognition includes face recognition, fingerprint recognition, and iris recognition.
  • face recognition As an example, face recognition technology is becoming more and more mature, and the recognition accuracy rate of face recognition in a specific scene is as high as 95% or more, and sometimes it can even directly distinguish the faces of twins.
  • the accuracy of face recognition is getting higher and higher, faces in photos and videos in real scenes can be mistaken for real faces, which brings opportunities for criminals and huge losses to legitimate users. unnecessary trouble.
  • the main attack methods facing face recognition include: (1) Photo attack methods that print out high-definition and realistic photos, dig out important areas of human faces, and replace real faces with the important areas of human faces.
  • the photos include black-and-white photos and color-printed photos.
  • Important areas of the face can be areas where the nose, eyes, mouth, etc. are located; (2) Obtain a pre-recorded video of a real face, and use the face in the video to replace the real face.
  • the above-mentioned video can be a real face video obtained from a social networking site or a real face video recorded by a camera in a public place; (3) A high-precision three-dimensional (Three Dimensional, 3D) printer is used to create a realistic Face model, and the above-mentioned face model instead of the real face model attack method, etc. Therefore, there is a need to propose a new technical solution that can further perform living body detection based on the results of facial recognition.
  • the purpose of the embodiments of the present application is to provide a method, electronic device and storage medium for living body detection based on facial recognition to enable living body detection.
  • an embodiment of the present application provides a method for living body detection based on facial recognition, including: acquiring an infrared image to be measured and a visible light image to be measured respectively; performing edge detection and texture feature extraction on the infrared image to be measured; Perform feature extraction on the visible light image to be measured through a convolutional neural network; based on the edge detection result of the infrared image to be measured, the result of the texture feature extraction, and the visible light image to be measured through the convolutional neural network Based on the result of feature extraction, it is determined whether the infrared image to be tested and the visible light image to be tested pass the living body detection.
  • an embodiment of the present application provides a device for living body detection based on facial recognition, including: an acquisition module for acquiring an infrared image to be measured and a visible light image to be measured; The infrared image performs edge detection and texture feature extraction; the processing module is also used to perform feature extraction on the visible light image to be measured through a convolutional neural network; the discrimination module is used to perform edge detection based on the infrared image to be measured Determine whether the infrared image to be tested and the visible light image to be tested pass the in vivo detection result from the result of, the result of the texture feature extraction and the result of feature extraction of the visible light image to be measured through a convolutional neural network.
  • an embodiment of the present application provides an electronic device, including: a memory, a processor, and computer-executable instructions stored on the memory and running on the processor, and the computer-executable instructions are The processor implements the steps of the method described in the first aspect when executed.
  • an embodiment of the present application provides a computer-readable storage medium, where the computer-readable storage medium is used to store computer-executable instructions.
  • the computer-executable instructions are executed by a processor, the above-mentioned first aspect is implemented The steps of the method described.
  • embodiments of the present application provide a system for living body detection based on facial recognition, including: an image capture device for capturing infrared images and visible light images; electronic equipment including: a processor; and a computer arranged to store A memory for executable instructions, when the executable instructions are executed, the processor is used to perform the following operations: respectively acquiring an infrared image to be tested and a visible light image to be tested; performing edge detection and texture feature extraction on the infrared image to be tested Feature extraction is performed on the visible light image to be measured through a convolutional neural network; based on the result of edge detection of the infrared image to be measured, the result of the texture feature extraction, and a convolutional neural network for the visible light to be measured As a result of image feature extraction, it is determined whether the infrared image to be tested and the visible light image to be tested pass the living body detection.
  • the infrared image to be measured and the visible light image to be measured are obtained separately, edge detection and texture feature extraction are performed on the infrared image to be measured, and the visible light image to be measured is characterized by a convolutional neural network.
  • Extraction based on the result of edge detection of the infrared image to be measured, the result of the above-mentioned texture feature extraction and the result of feature extraction of the visible light image to be measured through a convolutional neural network, to determine whether the infrared image to be measured and the visible light image to be measured pass the living body detection,
  • This process can combine the advantages of the three technologies of edge detection, texture feature extraction and convolutional neural network, and can effectively perform live detection.
  • the infrared image to be measured and the visible light image to be measured include the image area of the human face, it can efficiently determine whether the human face in the image is a living human face, thereby improving the accuracy of the judgment.
  • FIG. 1 shows a schematic flowchart of a method for living body detection based on facial recognition provided by an embodiment of the present application
  • FIG. 2 shows another schematic flow chart of the method for living body detection based on facial recognition provided by an embodiment of the present application
  • FIG. 3 shows another schematic flow chart of the method for living body detection based on facial recognition provided by an embodiment of the present application
  • FIG. 4 shows a schematic structural diagram of an apparatus for living body detection based on facial recognition provided by an embodiment of the present application
  • FIG. 5 shows another schematic structural diagram of a device for living body detection based on facial recognition provided by an embodiment of the present application
  • FIG. 6 is a schematic diagram of the hardware structure of an electronic device that executes the method for living body detection based on facial recognition provided by an embodiment of the present application.
  • FIG. 1 shows a schematic flowchart of a method for living body detection based on facial recognition provided by an embodiment of the present application.
  • the method may be executed by an electronic device.
  • the above-mentioned electronic device may be a terminal device or a server device.
  • the above method can be executed by software or hardware installed on the terminal device or the server device.
  • the aforementioned server equipment includes, but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.
  • the method includes the following steps S10-S40.
  • the above-mentioned infrared image to be measured and the visible light image to be measured may be infrared images and visible light images respectively collected by an image collection device that performs image collection for the same scene.
  • the infrared image to be measured and the visible light image to be measured may include the image area of the same face.
  • the infrared image to be measured may include image areas of multiple human faces, and the visible light image to be measured also includes image areas of the multiple human faces.
  • the above-mentioned scene may be the scene where the access control device is located.
  • the above-mentioned image acquisition device may be a device that can collect not only infrared images but also visible light images, such as a binocular camera.
  • the infrared image to be measured and the visible light image to be measured may be images collected by the image collecting device at the same time stamp.
  • the above-mentioned image acquisition device can also collect multiple infrared images and multiple visible light images, and then select the infrared image and the visible light image in the image area of the same face from the multiple infrared images and multiple visible light images, as the waiting Measure the infrared image and the visible light image to be measured.
  • S20 Perform edge detection and texture feature extraction on the infrared image to be measured.
  • edge detection In image processing and computer vision, you can perform edge detection on an image to detect edge information in the image. For example, edge detection is performed on the image to detect pixels with obvious brightness changes in the image.
  • the texture feature is a visual feature that reflects the homogeneity of the image.
  • One of the attack methods faced by face recognition is the tablet electronic product attack method. This attack method refers to: the tablet electronic product pretends to be real by displaying non-real faces in photos or playing non-real faces in videos. human face. Due to the high frequency interference of flat panel electronic products, a large amount of moiré may be generated when displaying photos or videos, which changes the characteristics of the images presented by the panel electronic products. In this case, when performing face recognition, it is possible to determine whether there is moiré in the image by analyzing the image characteristics, and then quickly distinguish whether the image presents a real face or an unreal face.
  • the corresponding texture feature can be extracted from the intrinsic relationship between imaging pixels in the infrared image to be measured.
  • S30 Perform feature extraction on the visible light image to be measured through a convolutional neural network.
  • Convolutional neural network is a network model that is widely used in deep learning.
  • the network model has a multi-layer structure. Each layer performs feature extraction on the input data of the layer. These extracted features are in the form of two-dimensional images. Continue to be entered into the next layer.
  • the result of edge detection of the infrared image to be measured can be called the first result
  • the result of texture feature extraction is called the second result
  • the result of feature extraction of the visible light image to be measured through a convolutional neural network is called The third result.
  • the above-mentioned first result, second result, and third result are all features of the image
  • the above-mentioned first result, second result, and third result can be feature-fused, and then based on the feature The result of the fusion determines whether the infrared image to be tested and the visible light image to be tested pass the living body detection.
  • a weighted calculation may be performed on the first result, the second result, and the third result, and the result of the weighted calculation may be used as the result of the feature fusion.
  • the infrared image to be measured and the visible light image to be measured pass the living body detection; the infrared image to be measured and the visible light image to be measured fail to pass the living body detection based on the second result; the infrared image to be measured and the visible light image to be measured are judged based on the third result.
  • the visible light image is measured by living body detection. According to statistics, the number of results that passed the living body detection is: 2, which means that the number of results that have not passed the living body detection is: 1, then the final judgment result is: the infrared image to be measured and the visible light image to be measured have not passed the living body detection.
  • the infrared image to be measured and the visible light image to be measured include the image area of the same face, the infrared image to be measured and the visible light image to be measured are indicated by living body detection: the infrared image to be measured and the visible light image to be measured are collected for the real face The image; the infrared image to be tested and the visible light image to be tested failed to pass the live detection means: the infrared image to be tested and the visible light image to be tested are images collected for real faces, for example, images collected for photos, images collected for videos, etc. .
  • the infrared image to be measured and the visible light image to be measured are obtained separately, and the edge detection and texture feature extraction are performed on the infrared image to be measured, and the convolutional neural network is used to treat Feature extraction is performed on the visible light image.
  • this process can combine the advantages of the three technologies of edge detection, texture feature extraction and convolutional neural network to effectively perform live body detection.
  • the infrared image to be measured and the visible light image to be measured include the image area of the human face, it can efficiently determine whether the human face in the image is a living human face, thereby improving the accuracy of the judgment.
  • the infrared image and the visible light image can also be collected by the image acquisition device, and then the face detection algorithm is used to detect the infrared image and the visible light image. Locate faces separately in.
  • the infrared image to be measured can be obtained from the infrared image and the visible light image according to the results of locating the human face in the infrared image and the visible light image. And the visible light image to be measured.
  • the aforementioned image acquisition device may acquire infrared images according to a preset infrared image acquisition frequency, and acquire visible light images according to a predetermined visible light image acquisition frequency.
  • the above-mentioned infrared image collection frequency and the visible light image collection frequency may be the same or different, which is not limited in the embodiment of the present application.
  • the face detection algorithm when a face detection algorithm is used to locate a human face in an infrared image and a visible light image respectively, the face detection algorithm can be used to detect the face area in the infrared image and the visible light image, that is, in Face positioning in infrared image and visible light image.
  • the position of the face feature point in the infrared image and the face feature point position in the visible light image can also be determined on the basis of the detected face area.
  • the obtained face localization result may include the information of the face in the image, the position of the facial feature point, etc.
  • the information of the above-mentioned area may be the coordinates of two diagonal vertices of the rectangular area, etc.
  • the above-mentioned facial feature point positions may include the positions of the feature points used to describe the contour of the human face in the image, the positions of the feature points used to describe the human eyes in the image, the positions of the feature points used to describe the human mouth in the image, and so on.
  • the infrared image and the visible light image including the image area of the same human face can be selected from the infrared image and the visible light image as the infrared image to be measured and the visible light image to be measured, respectively.
  • the infrared image and the visible light image that match the information of the region of the human face in the image and match the positions of the facial feature points may be determined as the infrared image to be measured and the visible light image to be measured.
  • the interpupillary distance of the human eye can be calculated according to the position of the facial feature points, and then when the ratio between the interpupillary distances is greater than the second preset threshold , It is considered that the positions of the facial feature points match.
  • the deflection angle and interpupillary distance of the face in the infrared image according to the position of the facial feature point in the infrared image, and obtain the visible light image according to the position of the face feature point in the visible light image
  • the deflection angle and interpupillary distance of the face in the middle according to the obtained deflection angle and interpupillary distance, select the infrared image to be measured and the visible light image to be measured from the infrared image and the visible image.
  • the deflection angle and interpupillary distance of the face can be used to indicate the posture of the face.
  • the deflection angle and interpupillary distance of the face in the infrared image represent the posture of the face
  • the deflection angle and interpupillary distance between the face in the visible light image When the postures of the human faces shown are consistent, it can be considered that the infrared image and the visible light image include the image area of the same human face, which can be used as the infrared image to be measured and the visible light image to be measured, respectively.
  • the angle difference between the deflection angles is less than the preset difference, and the ratio between the interpupillary distances is greater than the third preset threshold, it can be considered that the postures of the faces are consistent.
  • FIG. 2 shows another schematic flowchart of a method for living body detection based on facial recognition provided by an embodiment of the present application.
  • the method may be executed by an electronic device.
  • the above-mentioned electronic device may be a terminal device or a server device.
  • the above method can be executed by software or hardware installed on the terminal device or the server device.
  • the aforementioned server equipment includes, but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.
  • the method includes the following steps S11-S40.
  • S11 The infrared image and the visible light image are collected by the image acquisition device, and the human face is located in the infrared image and the visible light image through the face detection algorithm.
  • the image acquisition device may include a binocular camera or the like.
  • this step includes: detecting a human face area in an infrared image and a visible light image through a face detection algorithm.
  • the number of infrared faces in the infrared image and the position of the face feature points can be determined on the basis of the detected face area, and the visible light face in the visible light image can be determined The number and location of facial feature points.
  • the above process realizes the positioning of the human face in the infrared image and the visible light image respectively.
  • the aforementioned infrared human face refers to the image area where the human face is located in the infrared image.
  • the aforementioned visible light human face refers to the image area where the human face is located in the visible light image.
  • the number of infrared human faces in the infrared image and the number of visible human faces in the visible light image can be used to roughly determine whether the infrared image and the visible light image contain the image area of the same human face.
  • the probability of the image area containing the same face in the infrared image and the visible light image is lower. Conversely, if the number of infrared faces is the same as the number of visible light faces, then The probability of the infrared image and the image area containing the same face in the light image is higher.
  • S12 Obtain the deflection angle and the interpupillary distance of the face according to the position of the facial feature point in the infrared image and the position of the facial feature point in the visible light image.
  • the deflection angle and interpupillary distance of the face in the infrared image are obtained according to the position of the facial feature points in the infrared image, and the position of the face in the visible light image is obtained according to the position of the facial feature points in the visible light image. Deflection angle and interpupillary distance.
  • the distance between the two eyes of a person can be calculated according to the positions of the feature points related to the eyes of the human face, and then the interpupillary distance of the face can be determined according to the aforementioned distance.
  • S13 According to the obtained deflection angle and interpupillary distance, select the infrared image to be measured and the visible light image to be measured from the infrared image and the visible light image collected by the image collecting device.
  • the deflection angle and interpupillary distance of the face can reflect the posture of the face, in application scenarios such as face recognition and face detection, the posture of the face represents the face facing the image acquisition device, and the collected image quality is higher. , It can get better results when performing face recognition and face detection for such images. For this reason, after the infrared image and visible light image are collected by the image acquisition device, the above infrared image and visible light image can be filtered according to the deflection angle and interpupillary distance of the face, and the posture of the face is filtered to indicate that the face is not facing the image acquisition device. Image. For example, infrared images and visible light images with a deflection angle greater than a preset angle and a pupil distance less than a preset distance are filtered out.
  • the image quality can be detected according to the average brightness value of each pixel in the image. Specifically, for each infrared image collected by the image acquisition device, the average brightness value of each pixel can be calculated, and for each visible light image collected by the image acquisition device, the average brightness value of each pixel can be calculated. When the average brightness value is less than the first preset brightness value, the image is dark and the image quality is poor. When the average brightness value is greater than the second preset brightness value, the image is too bright, may be overexposed, and the image quality is also poor . In this way, based on the above situation, the poor quality images in the above infrared image and visible light image can be filtered out.
  • the foregoing first preset brightness value and second preset brightness value may be set according to specific application scenarios, which are not limited in the embodiment of the present application.
  • Average pixel value Average pixel value, interpupillary distance, deflection angle, etc.
  • S13 selects the infrared image to be measured and the visible light image to be measured, it is equivalent to obtaining the infrared image to be measured and the visible light image to be measured in this step. In this case, after executing the above S13, It is equivalent to finishing this step.
  • selecting the infrared image to be measured and the visible light image to be measured in S13 can be understood as merely selecting the image, and the infrared image to be measured and the visible light image to be measured are not acquired or read.
  • the infrared image to be measured and the visible light image to be measured can be acquired according to the selection result of S13.
  • S20 Perform edge detection and texture feature extraction on the infrared image to be measured.
  • the edge detection of the infrared image to be measured includes: filtering the noise in the infrared image to be measured through Gaussian transformation; and performing edge detection on the infrared image to be measured after filtering the noise through the Sobel operator , Obtain the edge detection result; count the histogram information of the edge detection result for the number of edge pixels in different directions, and filter the noise in the edge detection result according to the histogram information obtained by statistics.
  • the edge information of the image area of each face can be obtained.
  • the above edge information can be used as the feature of the image area of the face, which is called the face feature .
  • the infrared image to be tested is Gaussian transformation
  • the high-frequency information in the infrared image to be tested can be filtered out, and the noise in the image is often expressed as high-frequency information. Therefore, the infrared image to be tested can be filtered out after Gaussian transformation is performed. Noise in the infrared image to be measured.
  • the edge information of the image content can be detected to obtain the edge image, which is called the edge detection result here. For example, detecting the edge information of the face in the image.
  • the above-mentioned different directions may include a horizontal direction and a vertical direction.
  • the edge detection result is an edge image
  • the number of edge pixels contained in each pixel row in the edge image can be counted .
  • histogram statistical information and/or can also count the number of edge pixels contained in each pixel column in the edge image as histogram statistical information.
  • live detection can be performed based on the texture of a static image.
  • the above-mentioned living body is a real face
  • real face detection when it is performed based on the static image texture, it can be based on the local binary pattern (English: Local Binary Pattern, abbreviation: LBP), Gabor wavelet, gradient direction histogram (English : Histogram of Oriented Gradients, abbreviation: HOG) etc. to realize real face detection.
  • LBP Local Binary Pattern
  • HOG gradient direction histogram
  • live detection can be performed based on dynamic texture.
  • real face recognition can be performed by learning the structure and dynamic information of the real face microtexture, and using LBP to expand the feature operator in the airspace.
  • extracting the texture feature of the infrared image to be tested includes: extracting the texture feature of the infrared image to be tested through a dynamic local ternary pattern (English: Dynamic Local Ternary Pattern, abbreviation: DLTP).
  • DLTP Dynamic Local Ternary Pattern
  • the above-mentioned DLTP is evolved from the Local Ternary Pattern (LTP).
  • LTP evolved from the local binary pattern (LBP).
  • the pixel value of the current pixel is g c
  • the gray values of P adjacent pixels that are centered on the current pixel and adjacent to the current pixel are g 1 , g 2 ,..., g P .
  • the pixel values of each adjacent pixel after the binarization process are weighted and summed to obtain the local three-value mode value of the current pixel which is
  • the x c and y c are the horizontal and vertical coordinates of the current pixel in the image.
  • s(g i -g c ) represents the pixel value after binarization processing is performed on the i-th adjacent pixel.
  • the value of the above ⁇ is more difficult to set.
  • the above-mentioned ⁇ can be determined by Weber’s law, and the expression of Weber’s law is:
  • the above x and y represent the horizontal and vertical coordinates of the pixel in the image.
  • S30 Perform feature extraction on the visible light image to be measured through a convolutional neural network.
  • Convolutional neural network is a network model that is widely used in deep learning.
  • the network model has a multi-layer structure. Each layer performs feature extraction on the input data of the layer. These extracted features are in the form of two-dimensional images. Continue to be entered into the next layer.
  • the size of each original image in the real face database can be used as the size of the input image of the aforementioned convolutional neural network, so that the aforementioned convolutional neural network can be
  • the network performs feature extraction on input images of one size, thereby reducing the amount of calculation that multi-scale input images bring to the convolutional neural network.
  • the original images in the aforementioned real face database can be used as training samples to train the aforementioned convolutional neural network, so that the aforementioned convolutional neural network learns the original images in the real face database.
  • Features of real faces in the image can be used as training samples to train the aforementioned convolutional neural network, so that the aforementioned convolutional neural network learns the original images in the real face database.
  • the result of edge detection of the above-mentioned infrared image to be measured, the result of the above-mentioned texture feature extraction, and the result of feature extraction of the visible light image to be measured through a convolutional neural network can be feature fused, and then the feature fusion As a result, it is detected whether the infrared image to be measured and the visible light image to be measured pass the living body detection.
  • the fully connected layer may contain multiple nodes, and each node is used to obtain the result of edge detection, texture feature extraction, and feature extraction of the visible light image to be measured through a convolutional neural network.
  • the fully connected layer can integrate the features corresponding to the above three results.
  • the output results of the above-mentioned fully connected layer may be classified by a classifier such as softmax to obtain whether the above-mentioned infrared image and the visible light image to be measured pass the living body detection.
  • the softmax classifier can judge the input information of the softmax classifier by setting a threshold, that is, judge the output result of the above-mentioned fully connected layer, and when the input information of the softmax classifier is greater than the preset threshold, the judgment
  • the infrared image to be measured and the visible light image to be measured are images for real human faces and pass live detection; otherwise, it is determined that the infrared image to be measured and the visible light image to be measured are images for non-real human faces and fail the live detection.
  • the infrared image to be measured and the visible light image to be measured can be obtained separately, edge detection and texture feature extraction are performed on the infrared image to be measured, and the visible light image to be measured is performed through the convolutional neural network.
  • the infrared image to be measured and the visible light image to be measured include the image area of the human face, it can efficiently determine whether the human face in the image is a living human face, thereby improving the accuracy of the judgment.
  • FIG. 3 shows another schematic flowchart of a method for living body detection based on facial recognition provided by an embodiment of the present application.
  • the method may be executed by an electronic device.
  • the above-mentioned electronic device may be a terminal device or a server device.
  • the above method can be executed by software or hardware installed on the terminal device or the server device.
  • the aforementioned server equipment includes, but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.
  • the method may include the following steps S11-S40.
  • S11 The infrared image and the visible light image are collected by the image acquisition device, and the human face is located in the infrared image and the visible light image through the face detection algorithm.
  • this step includes: detecting a human face area in an infrared image and a visible light image through a face detection algorithm.
  • the number of infrared faces in the infrared image and the position of the face feature points can be determined on the basis of the detected face area, and the visible light face in the visible light image can be determined The number and location of facial feature points.
  • the above process realizes the positioning of the human face in the infrared image and the visible light image respectively.
  • S12 Obtain the deflection angle and the interpupillary distance of the face according to the position of the facial feature point in the infrared image and the position of the facial feature point in the visible light image.
  • the deflection angle and interpupillary distance of the face in the infrared image are obtained according to the position of the facial feature points in the infrared image, and the position of the face in the visible light image is obtained according to the position of the facial feature points in the visible light image. Deflection angle and interpupillary distance.
  • the distance between the two eyes of a person can be calculated according to the positions of the feature points related to the eyes of the human face, and then the interpupillary distance of the face can be determined according to the aforementioned distance.
  • S13 According to the obtained deflection angle and interpupillary distance, select the infrared image to be measured and the visible light image to be measured from the infrared image and the visible light image collected by the image collecting device.
  • S14 Perform gray-scale pixel processing on the infrared image to be tested to obtain an infrared gray-scale image.
  • the infrared image to be tested may be grayscale transformed to obtain the grayscale image corresponding to the infrared image to be tested, and the grayscale image is used as the infrared grayscale image
  • S15 Perform normalization processing on the visible light image to be measured to obtain a normalized visible light image.
  • Normalizing an image refers to the process of performing a series of standard processing transformations on an image to transform it into a fixed standard image.
  • the standard image is called a normalized image.
  • the aforementioned four-channel image includes: red, green, and blue RGB channel images and infrared gray-scale channel images.
  • the above-mentioned RGB channel image is an image corresponding to the three RGB channels respectively.
  • the aforementioned infrared gray-scale channel image is: the aforementioned infrared gray-scale image corresponding to the fourth channel.
  • S20 Perform edge detection and texture feature extraction on the infrared image to be measured.
  • edge detection may be performed on the infrared gray channel image
  • texture feature extraction may be performed on the above-mentioned RGB channel image
  • performing edge detection on the infrared image to be measured includes: filtering the noise in the infrared image to be measured through Gaussian transformation; and performing the Sobel operator on the infrared image to be measured after filtering the noise.
  • Perform edge detection to obtain edge detection results; count the histogram information of the edge detection results for the number of edge pixels in different directions, and filter out the noise in the edge detection results according to the statistical histogram information.
  • the edge information of the image area of each face can be obtained.
  • the above edge information can be used as the feature of the image area of the face, which is called the face feature .
  • live detection can be performed based on the texture of a static image.
  • real face detection when the above-mentioned living body is a real face, when real face detection is performed based on static image texture, real face detection can be realized based on LBP, Gabor wavelet, HOG, etc.
  • live detection can be performed based on dynamic texture.
  • real face recognition can be performed by learning the structure and dynamic information of the real face microtexture, and using LBP to expand the feature operator in the airspace.
  • extracting the texture feature of the infrared image to be tested includes: extracting the texture feature of the infrared image to be tested through a dynamic local ternary pattern (English: Dynamic Local Ternary Pattern, abbreviation: DLTP).
  • DLTP Dynamic Local Ternary Pattern
  • the above-mentioned DLTP is evolved from the Local Ternary Pattern (LTP).
  • LTP evolved from the local binary pattern (LBP).
  • the pixel value of the current pixel is g c
  • the gray values of P adjacent pixels that are centered on the current pixel and adjacent to the current pixel are g 1 , g 2 ,..., g P .
  • the pixel values of each adjacent pixel after the binarization process are weighted and summed to obtain the local three-value mode value of the current pixel which is
  • the x c and y c are the horizontal and vertical coordinates of the current pixel in the image.
  • s(g i -g c ) represents the pixel value after binarization processing is performed on the i-th adjacent pixel.
  • the value of the above ⁇ is more difficult to set.
  • the above-mentioned ⁇ can be determined by Weber’s law, and the expression of Weber’s law is:
  • the above x and y represent the horizontal and vertical coordinates of the pixel in the image.
  • S30 Perform feature extraction on the visible light image to be measured through a convolutional neural network.
  • Convolutional neural network is a network model that is widely used in deep learning.
  • the network model has a multi-layer structure. Each layer performs feature extraction on the input data of the layer. These extracted features are in the form of two-dimensional images. Continue to be entered into the next layer.
  • the above-mentioned four-channel image can be extracted through a convolutional neural network.
  • the infrared image to be measured and the visible light image to be measured can be obtained separately, edge detection and texture feature extraction are performed on the infrared image to be measured, and the visible light image to be measured is passed through the convolutional neural network.
  • the infrared image to be measured and the visible light image to be measured include the image area of the human face, it can efficiently determine whether the human face in the image is a living human face, thereby improving the accuracy of the judgment.
  • FIG. 4 shows a schematic structural diagram of a device for living body detection based on facial recognition provided by an embodiment of the present application.
  • the device 100 includes: an acquisition module 110, a processing module 120, and a discrimination module 130.
  • the obtaining module 110 is used to obtain the infrared image to be measured and the visible light image to be measured respectively.
  • the processing module 120 is configured to perform edge detection and texture feature extraction on the infrared image to be tested.
  • the processing module 120 is further configured to perform feature extraction on the visible light image to be measured through a convolutional neural network.
  • the discrimination module 130 is configured to determine the result of edge detection of the infrared image to be measured, the result of the texture feature extraction, and the result of feature extraction of the visible light image to be measured through a convolutional neural network. Check whether the infrared image and the visible light image to be measured pass the living body detection.
  • the acquisition module 110 is specifically configured to collect infrared images and visible light images by using an image acquisition device; use a face detection algorithm to locate human faces in the infrared image and the visible light image respectively; According to the results of locating the human face in the infrared image and the visible light image, respectively, the infrared image to be measured and the visible light image to be measured are obtained from the infrared image and the visible light image.
  • the acquisition module 110 is specifically configured to detect the face area in the infrared image and the visible light image through the face detection algorithm; determine the face in the infrared image Feature point location; determining the face feature point location in the visible light image.
  • the acquisition module 110 is specifically configured to obtain the deflection angle and interpupillary distance of the face in the infrared image according to the position of the facial feature points in the infrared image, and according to the Obtain the deflection angle and interpupillary distance of the face in the visible light image; according to the obtained deflection angle and interpupillary distance, from the infrared image and the visible light image, select the to-be-measured Infrared image and visible light image to be measured.
  • the acquisition module 110 is specifically configured to collect infrared images and visible light images by using an image acquisition device; use a face detection algorithm to locate human faces in the infrared image and the visible light image respectively; According to the results of locating the human face in the infrared image and the visible light image, respectively, the infrared image to be measured and the visible light image to be measured are obtained from the infrared image and the visible light image.
  • the acquisition module 110 is specifically configured to detect the face area in the infrared image and the visible light image through the face detection algorithm; determine the face in the infrared image Feature point location; determining the face feature point location in the visible light image.
  • the processing module 120 is further configured to perform gray-scale pixel processing on the infrared image to be tested to obtain an infrared gray-scale image; perform normalization processing on the visible light image to be tested to obtain a normalized A visible light image; the normalized visible light image and the infrared gray-scale image are fused into a four-channel image, wherein the four-channel image includes: a red, green, and blue RGB channel image and an infrared gray channel image, the The RGB channel image is: the image corresponding to the three RGB channels, and the infrared gray channel image is the infrared gray image corresponding to the fourth channel.
  • the processing module 120 is specifically configured to perform edge detection on the infrared gray channel image, and perform texture feature extraction on the RGB channel image; and perform convolutional neural network on the four-channel Image feature extraction.
  • the processing module 120 is specifically configured to filter the noise in the infrared image to be measured through Gaussian transformation; to edge the infrared image to be measured after the noise is filtered through the Sobel operator. Detect to obtain an edge detection result; count histogram information of the edge detection result for the number of edge pixels in different directions, and filter out noise in the edge detection result according to the histogram information obtained by statistics.
  • the processing module 120 is specifically configured to extract the texture feature of the infrared image to be measured through a dynamic local three-value pattern.
  • FIG. 5 shows a schematic structural diagram of an apparatus for living body detection based on facial recognition provided by an embodiment of the present application.
  • the apparatus 100 includes: an acquisition module 110, a processing module 120, a discrimination module 130, and a screening module 140.
  • the screening module 140 is configured to obtain the deflection angle and interpupillary distance of the face in the infrared image according to the position of the facial feature points in the infrared image, and obtain the facial feature point positions in the visible light image
  • the deflection angle and interpupillary distance of the human face in the visible light image; according to the obtained deflection angle and interpupillary distance, the infrared image to be measured and the visible light image to be measured are selected from the infrared image and the visible light image.
  • the obtaining module 110 is used to obtain the infrared image to be measured and the visible light image to be measured respectively.
  • the processing module 120 is configured to perform edge detection and texture feature extraction on the infrared image to be tested.
  • the processing module 120 is further configured to perform feature extraction on the visible light image to be measured through a convolutional neural network.
  • the discrimination module 130 is configured to determine the result of edge detection of the infrared image to be measured, the result of extracting texture features, and the result of feature extraction of the visible light image to be measured through a convolutional neural network. Check whether the infrared image and the visible light image to be measured pass the living body detection.
  • the device 100 provided in the embodiment of the present application can execute the methods described in the foregoing method embodiments, and realize the functions and beneficial effects of the methods described in the foregoing method embodiments, and will not be repeated here.
  • FIG. 6 shows a schematic diagram of the hardware structure of an electronic device that implements a method for living body detection based on facial recognition provided by an embodiment of the present application.
  • the electronic device may have relatively large differences due to different configurations or performances.
  • the memory 702 may store one or more storage application programs or data.
  • the memory 702 may be short-term storage or persistent storage.
  • the application program stored in the memory 702 may include one or more modules (not shown in the figure), and each module may include a series of computer-executable instructions in the electronic device.
  • the processor 701 may be configured to communicate with the memory 702, and execute a series of computer-executable instructions in the memory 702 on the electronic device.
  • the electronic device may also include one or more power supplies 703, one or more wired or wireless network interfaces 704, one or more input and output interfaces 705, one or more keyboards 706, and so on.
  • the electronic device shown in FIG. 6 may further include an image acquisition device.
  • the electronic device includes an image acquisition device that separately obtains an infrared image to be measured and a visible light image to be measured; a processor; and a memory arranged to store computer-executable instructions.
  • the processor is used to perform the following operations: respectively obtain the infrared image to be measured and the visible light image to be measured; perform edge detection and texture feature extraction on the infrared image to be measured; and perform the convolutional neural network on the visible light image to be measured Perform feature extraction; based on the result of edge detection of the infrared image to be tested, the result of texture feature extraction, and the result of feature extraction of the visible light image to be tested through a convolutional neural network, determine the infrared image to be tested Whether the image and the visible light image to be measured pass the living body detection.
  • the electronic device that executes the method for living body detection based on facial recognition provided in the embodiments of the present application can execute the methods described in the foregoing method embodiments, and realize the functions of the methods described in the foregoing method embodiments. And the beneficial effects will not be repeated here.
  • the electronic devices in the embodiments of the present application exist in various forms, including but not limited to the following devices.
  • Mobile communication equipment This type of equipment is characterized by mobile communication functions, and its main goal is to provide voice and data communications.
  • Such terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones.
  • Ultra-mobile personal computer equipment This type of equipment belongs to the category of personal computers, has calculation and processing functions, and generally also has mobile Internet features.
  • Such terminals include: PDA, MID and UMPC devices, such as iPad.
  • Portable entertainment equipment This type of equipment can display and play multimedia content.
  • Such devices include: audio and video players (such as iPod), handheld game consoles, e-books, as well as smart toys and portable car navigation devices.
  • Server A device that provides computing services.
  • the structure of a server includes a processor, hard disk, memory, system bus, etc.
  • the server is similar to a general computer architecture, but due to the need to provide highly reliable services, it is in terms of processing capacity and stability. , Reliability, security, scalability, and manageability.
  • an embodiment of the present application also provides a system for living body detection based on facial recognition, including: an image capture device for capturing infrared images and visible light images; an electronic device including: a processor; and a storage computer arranged
  • a memory for executable instructions when the executable instructions are executed, the processor is used to perform the following operations: obtain infrared images and visible light images collected by the image acquisition device, and filter out infrared images to be detected and visible light to be detected Image; edge detection and texture feature extraction of the infrared image to be tested by convolutional neural network; feature extraction of the visible light image to be tested; based on the result of edge detection of the infrared image to be tested, the texture
  • the result of feature extraction and the result of feature extraction of the visible light image to be measured through the convolutional neural network determine whether the infrared image to be measured and the visible light image to be measured pass the living body detection.
  • the embodiment of the present application also provides a system for living detection based on facial recognition that can execute the methods described in the foregoing method embodiments, and realize the functions and beneficial effects of the methods described in the foregoing method embodiments. , I won’t repeat it here.
  • an embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium is used to store computer-executable instructions, and the computer-executable instructions are executed by a processor to implement the following processes: The infrared image to be measured and the visible light image to be measured; edge detection and texture feature extraction are performed on the infrared image to be measured; feature extraction is performed on the visible light image to be measured through a convolutional neural network; based on the infrared image to be measured The result of the edge detection, the result of the texture feature extraction and the result of the feature extraction of the visible light image to be measured through a convolutional neural network determine whether the infrared image to be measured and the visible light image to be measured pass the living body detection.
  • the computer-readable storage medium includes read-only memory (Read-Only Memory, ROM for short), random access memory (Random Access Memory, RAM for short), magnetic disks, or optical disks.
  • Read-Only Memory ROM for short
  • random access memory Random Access Memory, RAM for short
  • magnetic disks or optical disks.
  • an embodiment of the present application also provides a computer program product, the computer program product includes a computer program stored on a non-transitory computer-readable storage medium, the computer program includes program instructions, when the program instructions When executed by a computer, the following processes are realized: the infrared image to be measured and the visible light image to be measured are respectively obtained; edge detection and texture feature extraction are performed on the infrared image to be measured; the visible light image to be measured is characterized by a convolutional neural network Extraction; based on the result of edge detection of the infrared image to be tested, the result of the texture feature extraction and the result of feature extraction of the visible light image to be tested through a convolutional neural network, the determination of the infrared image to be tested and Whether the visible light image to be measured passes the living body detection.

Abstract

一种基于面部识别的活体检测的方法、电子设备和存储介质，该方法包括：分别获取待测红外图像和待测可见光图像（S10），对所述待测红外图像进行边缘检测和纹理特征提取（S20），通过卷积神经网络对所述待测可见光图像进行特征提取（S30），基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测（S40）。一种基于面部识别的活体检测的方法、电子设备和存储介质能够结合边缘检测、纹理特征提取和卷积神经网络三种技术的优势，有效的进行活体检测，高效的判别图像中的人脸是否属于活体，提升了判别准确率。

Description

基于面部识别的活体检测的方法、电子设备和存储介质
本申请要求于2019年1月25日提交中国专利局、申请号为201910072693.8发明名称为“基于面部识别的活体检测的方法、电子设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及人脸识别技术领域，尤其涉及一种基于面部识别的活体检测的方法、电子设备和存储介质。
背景技术
随着人工智能(英文：Artificial Intelligence，缩写：AI)行业的迅速崛起，生物特征识别已经应用于安防行业中。例如，上述生物特征识别包括人脸识别、指纹识别和虹膜识别等。
以人脸识别为例，人脸识别技术越来越成熟，人脸识别在特定场景中识别准确率高达95％以上，甚至有时能直接区分双胞胎人脸。但是随着人脸识别的准确率越来越高，能够将现实场景内照片、视频中的人脸误认为真实人脸，给不法分子带来可乘之机，给合法用户带来巨大损失或者不必要的麻烦。
目前，人脸识别面临的主要攻击方式包括：(1)打印出高清逼真的照片、挖取人脸重要区域、并以挖取的人脸重要区域代替真实人脸的照片攻击方式，其中，上述照片包括黑白照片和彩印照片，人脸重要区域可以是鼻子、眼睛、嘴巴等所在区域；(2)获得预先录制的一段真实人脸视频、并以视频中的人脸代替真实人脸的视频攻击方式，其中，上述视频可以是从社交网站获得的一段真实人脸视频或者公共场合摄像头录制的真实人脸视频；(3)通过高精准的三维(Three Dimensional，3D)打印机制作出一张逼真的人脸模型、并以上述人脸模型代替真实人脸的模型攻击方式等。因此，有需要提出一种新的技术方案，能够基于面部识别的结果进一步进行活体检测。
公开于该背景技术部分的信息仅仅旨在增加对本发明的总体背景的理解，而不应当被视为承认或以任何形式暗示该信息已为本领域一般技术人员所公知的现有技术。
发明内容
本申请实施例的目的是提供一种基于面部识别的活体检测的方法、电子设备和存储介质，以能够进行活体检测。
为解决上述技术问题，本申请实施例是通过以下各方面实现的。
第一方面，本申请实施例提供了一种基于面部识别的活体检测的方法，包括：分别获取待测红外图像和待测可见光图像；对所述待测红外图像进行边缘检测和纹理特征提取； 通过卷积神经网络对所述待测可见光图像进行特征提取；基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
第二方面，本申请实施例提供了一种基于面部识别的活体检测的装置，包括：获取模块，用于分别获取待测红外图像和待测可见光图像；处理模块，用于对所述待测红外图像进行边缘检测和纹理特征提取；所述处理模块，还用于通过卷积神经网络对所述待测可见光图像进行特征提取；判别模块，用于基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
第三方面，本申请实施例提供了一种电子设备，包括：存储器、处理器和存储在所述存储器上并可在所述处理器上运行的计算机可执行指令，所述计算机可执行指令被所述处理器执行时实现如上述第一方面所述的方法的步骤。
第四方面，本申请实施例提供了一种计算机可读存储介质，所述计算机可读存储介质用于存储计算机可执行指令，所述计算机可执行指令被处理器执行时实现如上述第一方面所述的方法的步骤。
第五方面，本申请实施例提供了一种基于面部识别的活体检测的系统，包括：图像采集器件，用于采集红外图像和可见光图像；电子设备，包括：处理器；以及被安排成存储计算机可执行指令的存储器，所述可执行指令在被执行时使用所述处理器执行以下操作：分别获取待测红外图像和待测可见光图像；对所述待测红外图像进行边缘检测和纹理特征提取；通过卷积神经网络对所述待测可见光图像进行特征提取；基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
在应用本申请实施例提供的方案进行活体检测时，通过分别获取待测红外图像和待测可见光图像，对待测红外图像进行边缘检测和纹理特征提取，通过卷积神经网络对待测可见光图像进行特征提取，基于对待测红外图像进行边缘检测的结果、上述纹理特征提取的结果和通过卷积神经网络对待测可见光图像进行特征提取的结果，判断待测红外图像和待测可见光图像是否通过活体检测，这一过程能够结合边缘检测、纹理特征提取和卷积神经网络三种技术的优势，能够有效的进行活体检测。在待测红外图像和待测可见光图像中包括人脸的图像区域的情况下，能够高效的判别图像中的人脸是否属于活体的人脸，从而提升了判别准确率。
附图说明
为了更清楚地说明本申请实施例和现有技术的技术方案，下面对实施例和现有技术中 所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1示出本申请实施例提供的基于面部识别的活体检测的方法的一种流程示意图；
图2示出本申请实施例提供的基于面部识别的活体检测的方法的另一种流程示意图；
图3示出本申请实施例提供的基于面部识别的活体检测的方法的另一种流程示意图；
图4示出本申请实施例提供的基于面部识别的活体检测的装置的一种结构示意图；
图5示出本申请实施例提供的基于面部识别的活体检测的装置的另一种结构示意图；
图6为执行本申请实施例提供的基于面部识别的活体检测的方法的电子设备的硬件结构示意图。
具体实施方式
为使本申请的目的、技术方案、及优点更加清楚明白，以下参照附图并举实施例，对本申请进一步详细说明。显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
图1示出本申请实施例提供的基于面部识别的活体检测的方法的一种流程示意图。该方法可以由电子设备执行，例如，上述电子设备可以是终端设备或服务端设备。换言之，上述方法可以由安装在终端设备或服务端设备的软件或硬件来执行。上述服务端设备包括但不限于：单台服务器、服务器集群、云端服务器或云端服务器集群等。如图1所示，该方法包括以下步骤S10-S40。
S10：分别获取待测红外图像和待测可见光图像。
本申请的一个实施例中，上述待测红外图像和待测可见光图像可以为：针对同一场景进行图像采集的图像采集器件分别采集的红外图像和可见光图像。
具体的，上述待测红外图像和待测可见光图像中可以包括同一人脸的图像区域。另外，上述待测红外图像中可以包括多个人脸的图像区域，上述待测可见光图像中也包括上述多个人脸的图像区域。
上述场景可以是门禁设备所在场景等。上述图像采集器件可以是双目摄像头等不仅可以采集到红外图像、而且可以采集到可见光图像的器件。
本申请的另一个实施例中，上述待测红外图像和待测可见光图像可以是上述图像采集器件在同一时间戳下采集的图像。
另外，上述图像采集器件也可以采集多张红外图像和多张可见光图像，然后从上述多张红外图像和多张可见光图像中选择存在同一人脸的图像区域的红外图像和可见光图像， 分别作为待测红外图像和待测可见光图像。
S20：对待测红外图像进行边缘检测和纹理特征提取。
在图像处理和计算机视觉中，可以对图像进行边缘检测，检测出图像中的边缘信息。例如，对图像进行边缘检测，检测出图像中亮度变化明显的像素点。
纹理特征是一种反映图像中同质现象的视觉特征。人脸识别所面临的一种攻击方式为平板电子产品攻击方式，这种攻击方式是指：平板电子产品以显示照片中的非真实人脸或者播放视频中的非真实人脸的方式、冒充真实人脸。由于平板电子产品受高频干扰在显示照片或者视频时可能会产生大量摩尔纹，使得平板电子产品所呈现图像的特征发生变化。这种情况下，在进行人脸识别时，可以通过分析图像特征,判断图像中是否存在摩尔纹，进而快速的区分图像中呈现的是真实人脸还是非真实人脸。
本申请的一个实施例中，可以通过待测红外图像中成像像素之间内在关系提取相应的纹理特征。
S30：通过卷积神经网络对待测可见光图像进行特征提取。
卷积神经网络是深度学习中使用较多的一种网络模型，该网络模型具有多层结构，每个层对该层的输入数据进行特征提取，这些被提取出来的特征以二维图像的形式继续被输入下一个层。
S40：基于对待测红外图像进行边缘检测的结果、上述纹理特征提取的结果和通过卷积神经网络对待测可见光图像进行特征提取的结果，判断待测红外图像和待测可见光图像是否通过活体检测。
为便于表述，可以将对待测红外图像进行边缘检测的结果称为第一结果，将纹理特征提取的结果称为第二结果，将通过卷积神经网络对待测可见光图像进行特征提取的结果称为第三结果。
本申请的一个实施例中，由于上述第一结果、第二结果和第三结果均为图像的特征，因此，可以对上述第一结果、第二结果和第三结果进行特征融合，然后基于特征融合的结果判断待测红外图像和待测可见光图像是否通过活体检测。
例如，可以对上述第一结果、第二结果和第三结果进行加权计算，将加权计算的结果作为上述特征融合的结果。
本申请的另一个实施例中，还可以分别基于第一结果、第二结果和第三结果判断待测红外图像和待测可见光图像是否通过活体检测，然后，统计判断结果的数量，以统计数量最高的结果作为最终判断结果。
例如，基于第一结果判断待测红外图像和待测可见光图像通过活体检测；基于第二结果判断待测红外图像和待测可见光图像未通过活体检测；基于第三结果判断待测红外图像 和待测可见光图像通过活体检测。经统计表示通过活体检测的结果的数量为：2，表示未通过活体检测的结果的数量为：1，则最终判断结果为：待测红外图像和待测可见光图像未通过活体检测。
在上述待测红外图像和待测可见光图像中包括同一人脸的图像区域时，待测红外图像和待测可见光图像通过活体检测表示：待测红外图像和待测可见光图像为针对真实人脸采集的图像；待测红外图像和待测可见光图像未通过活体检测表示：待测红外图像和待测可见光图像为针对真实人脸采集的图像，例如，针对照片采集的图像、针对视频采集的图像等。
由此可见，本实施例提供的基于面部识别的活体检测的方案中，通过分别获取待测红外图像和待测可见光图像，对待测红外图像进行边缘检测和纹理特征提取，通过卷积神经网络对待测可见光图像进行特征提取，基于对待测红外图像进行边缘检测的结果、上述纹理特征提取的结果和通过卷积神经网络对待测可见光图像进行特征提取的结果，判断待测红外图像和待测可见光图像是否通过活体检测，这一过程能够结合边缘检测、纹理特征提取和卷积神经网络三种技术的优势，能够有效的进行活体检测。在待测红外图像和待测可见光图像中包括人脸的图像区域的情况下，能够高效的判别图像中的人脸是否属于活体的人脸，从而提升了判别准确率。
本申请的一个实施例中，在上述S10分别获取待测红外图像和待测可见光图像之前，还可以利用图像采集器件采集红外图像和可见光图像，然后通过人脸检测算法，在红外图像和可见光图像中分别定位人脸。
这种情况下，上述S10分别获取待测红外图像和待测可见光图像时，可以根据在红外图像和可见光图像中分别定位人脸的结果，从红外图像和可见光图像中，分别获取待测红外图像和待测可见光图像。
具体的，上述图像采集器件可以按照预先设定的红外图像采集频率采集红外图像，并按照预先定的可见光图像采集频率采集可见光图像。其中，上述红外图像采集频率和可见光图像采集频率可以相同，也可以不相同，本申请实施例并不对此进行限定。
本申请的一个实施例中，通过人脸检测算法，在红外图像和可见光图像中分别定位人脸时，可以通过人脸检测算法，检测红外图像和可见光图像中的人脸区域，也就是，在红外图像和可见光图像中进行人脸定位。除此之外，在检测出人脸区域之后，还可以在所检测出人脸区域的基础上，确定红外图像中的人脸特征点位置以及可见光图像中的人脸特征点位置。
鉴于上述情况，通过人脸检测算法在红外图像和可见光图像中分别定位人脸时，所得人脸定位结果中可以包括人脸在图像中区域的信息、人脸特征点位置等。
其中，人脸在图像中的区域为矩形区域的情况下，上述区域的信息可以是矩形区域两个对角顶点的坐标等。上述人脸特征点位置可以包括图像中用于描述人脸轮廓的特征点的位置、图像中用于描述人眼睛的特征点的位置、图像中用于描述人嘴巴的特征点的位置等等。
具体的，可以通过上述定位人脸的结果，从红外图像和可见光图像中选择包括同一人脸的图像区域的红外图像和可见光图像，分别作为待测红外图像和待测可见光图像。
本申请的一个实施例中，可以将人脸在图像中区域的信息相匹配、且人脸特征点位置相匹配的红外图像和可见光图像确定为待测红外图像和待测可见光图像。
例如，区域重合度大于第一预设阈值时，可以认为区域的信息相匹配。另外，在人脸特征点包括用于表示人眼睛的特征点的情况下，可以根据人脸特征点位置计算人眼睛的瞳距大小，然后在瞳距大小间的比例大于第二预设阈值时，认为人脸特征点位置相匹配。
本申请的另一个实施例中，还可以根据红外图像中的人脸特征点位置，获得红外图像中人脸的偏转角和瞳距，并根据可见光图像中的人脸特征点位置，获得可见光图像中人脸的偏转角和瞳距；根据所获得的偏转角和瞳距，从红外图像和可见光图像中，选择待测红外图像和待测可见光图像。
具体的，可以通过人脸的偏转角和瞳距表示人脸的姿态，当红外图像中人脸的偏转角和瞳距表示的人脸的姿态、与可见光图像中人脸的偏转角和瞳距表示的人脸的姿态一致时，可以认为红外图像和可见光图像中包括同一人脸的图像区域，可以分别作为待测红外图像和待测可见光图像。
例如，在偏转角之间的角度差小于预设差值、且瞳距间的比例大于第三预设阈值时，可以认为人脸的姿态一致。
下面再通过图2所示的具体实施例，对本申请实施例提供的活体检测方法进行详细说明。
图2示出本申请实施例提供的基于面部识别的活体检测的方法的另一种流程示意图。该方法可以由电子设备执行，例如，上述电子设备可以是终端设备或服务端设备。换言之，上述方法可以由安装在终端设备或服务端设备的软件或硬件来执行。上述服务端设备包括但不限于：单台服务器、服务器集群、云端服务器或云端服务器集群等。如图2所示，该方法包括以下步骤S11-S40。
S11：利用图像采集器件采集红外图像和可见光图像，通过人脸检测算法，在红外图像和可见光图像中分别定位人脸。
在一种可能的实现方式中，图像采集器件可以包括双目摄像头等。
本申请的一个实施例中，本步骤包括：通过人脸检测算法，检测红外图像和可见光图 像中的人脸区域。另外，在检测出上述人脸区域之后，还可以在所检测出人脸区域的基础上，确定红外图像中的红外人脸个数和人脸特征点位置，并确定可见光图像中的可见光人脸个数和人脸特征点位置。上述过程实现了在红外图像和可见光图像中分别定位人脸。
其中，上述红外人脸是指红外图像中人脸所在图像区域。上述可见光人脸是指可见光图像中人脸所在图像区域。
具体的，可以通过红外图像中红外人脸个数和可见光图像中可见光人脸个数，粗略判断红外图像和可见光图像中是否包含同一人脸的图像区域。
若红外人脸个数与可见光人脸个数不同，则红外图像和可见光图像中包含同一人脸的图像区域的概率较低，反之，若红外人脸个数与可见光人脸个数相同，则红外图像和可将光图像中包含同一人脸的图像区域的概率较高。
S12：根据红外图像中的人脸特征点位置和可见光图像中的人脸特征点位置，获得人脸的偏转角和瞳距。
具体的，本步骤中，根据红外图像中的人脸特征点位置，获得红外图像中人脸的偏转角和瞳距，并根据可见光图像中的人脸特征点位置，获得可见光图像中人脸的偏转角和瞳距。
本申请的一个实施例中，可以根据人脸特征点中与人眼睛相关的特征点的位置计算人两只眼睛之间的距离，然后根据上述距离确定人脸的瞳距。
S13：根据所获得的偏转角和瞳距，从利用图像采集器件采集的红外图像和可见光图像中，选择待测红外图像和待测可见光图像。
由于人脸的偏转角和瞳距能够反映人脸的姿态，在人脸识别、人脸检测等应用场景中，人脸的姿态表征人脸面向图像采集器件时，所采集到的图像质量较高，针对这样的图像进行人脸识别、人脸检测时能够得到较好的结果。为此，在图像采集器件采集到红外图像和可见光图像后，可以根据人脸的偏转角和瞳距对上述红外图像和可见光图像进行过滤，过滤掉人脸的姿态表征人脸未面向图像采集器件的图像。例如，过滤掉偏转角大于预设角度、且瞳距小于预设距离的红外图像和可见光图像。
所以，依据偏转角和瞳距这两个参数可以过滤掉图像采集器件所采集的红外图像和可见光图像中质量较差的人脸图像，从而提高活体检测的鲁棒性。
本申请的一个实施例中，可以根据图像中各个像素点的平均亮度值进行图像质量进行检测。具体的，可以针对图像采集器件采集的每一红外图像，计算各个像素点的平均亮度值，针对图像采集器件采集的每一可见光图像，计算各个像素点的平均亮度值。当平均亮度值小于第一预设亮度值时，说明图像偏暗，图像质量欠佳，当平均亮度值大于第二预设亮度值时，说明图像过亮，可能曝光过度，图像质量也欠佳。这样基于上述情况可以过滤 掉上述红外图像和可见光图像中质量欠佳的图像。
上述第一预设亮度值和第二预设亮度值可以根据具体应用场景设定，本申请实施例并不对此进行限定。
另外，在过滤红外图像和可见光图像中质量欠佳的图像时，还可以通过以下信息中的一种或者多种相结合的方式实现：
平均像素值、瞳距、偏转角等。
S10：分别获取待测红外图像和待测可见光图像。
一种情况下，上述S13选择待测红外图像和待测可见光图像后，也就相当于本步骤中获取了待测红外图像和待测可见光图像，这种情况下，执行完上述S13后也就相当于执行完本步骤。
另一种情况下，上述S13选择待测红外图像和待测可见光图像可以理解为仅仅是选择了图像，并没有获取或者读取待测红外图像和待测可见光图像。这种情况下，本步骤中，可以根据上述S13的选择结果获取上述待测红外图像和待测可见光图像。
S20：对待测红外图像进行边缘检测和纹理特征提取。
在一种可能的实现方式中，对待测红外图像进行边缘检测包括：通过高斯变换滤除待测红外图像中的噪声；通过索贝尔sobel算子对滤除噪声后的待测红外图像进行边缘检测，得到边缘检测结果；统计上述边缘检测结果在不同方向针对边缘像素点数量的直方图信息，并根据统计得到的直方图信息滤除边缘检测结果中的噪声。这样在待测红外图像中存在人脸的图像区域时，可以得到各个人脸的图像区域的边缘信息，这种情况下，可以将上述边缘信息作为人脸的图像区域的特征，称为人脸特征。
其中，对待测红外图像进行高斯变换时，可以滤除待测红外图像中的高频信息，而图像中的噪声往往表现为高频信息，因此，对待测红外图像进行高斯变换后，可以滤除待测红外图像中的噪声。
当然，也可以采用其他变换方式滤除待测红外图像中的噪声，本申请实施例并不对此进行限定。
通过sobel算子对滤除噪声后的待测红外图像进行边缘检测时，可以检测出图像内容的边缘信息，得到边缘图像，在此称为边缘检测结果。例如，检测出图像中人脸的边缘信息。
另外，上述不同方向可以包括水平方向和垂直方向。
具体的，统计上述边缘检测结果在不同方向针对边缘像素点数量的直方图信息时，由于上述边缘检测结果为边缘图像，因此，可以统计边缘图像中每一像素行中所包含边缘像素点的数量，作为直方图统计信息，和/或还可以统计边缘图像中每一像素列所包含边缘 像素点的数量，作为直方图统计信息。
由于沿像素行或者像素列图像中人脸的边缘信息呈现的边缘像素点数量较多，所以，在各个像素行或者像素列对应的直方图信息表示的边缘像素点数量较少时，这些像素行或者像素列中的边缘像素点不是人脸的边缘像素点的概率较高，为此可以从上述边缘图像中过滤掉这些像素点。
在一种可能的实现方式中，可以基于静态图像纹理进行活体检测。这种情况下，需从图像中提取一个或多个具有运动不变性质的特征，比如，图像中的边界线条或角点等，并依据这些特征建立活体检测模型。然后通过上述活体检测模型检测图像是否为针对活体采集的图像。
具体的，在上述活体为真实人脸时，基于静态图像纹理进行真实人脸检测时，可以基于局部二值模式(英文：Local Binary Pattern，缩写：LBP)、Gabor小波、梯度方向直方图(英文：Histogram of Oriented Gradients,缩写：HOG)等实现真实人脸检测。
在一种可能的实现方式中，可以基于动态纹理进行活体检测。这种情况下，在上述活体为真实人脸时，可以通过学习真实人脸微纹理的结构和动态信息，并利用LBP在空域进行特征算子扩展的方式进行真实人脸识别。
在一种可能的实现方式中，对待测红外图像进行纹理特征提取包括：通过动态局部三值模式(英文：Dynamic Local Ternary Pattern，缩写：DLTP)提取待测红外图像的纹理特征。
具体的，上述DLTP是由局部三值模式(Local Ternary Pattern，LTP)演变来的。LTP由局部二模式(local binary pattern，LBP)演变而来。
下面简述基于LTP得到DLTP信息的过程。
假设，当前像素点的像素值为g c，以该当前像素点为中心、且与该当前像素点相邻的P个相邻像素点的灰度值分别g 1,g 2,…,g P。
首先，以g c±τ为阈值，对上述相邻像素点进行二值化处理。
然后，根据相邻像素点的不同位置，对二值化处理后各个相邻像素点的像素值进行加权求和，得到当前像素点的局部三值模式值
Figure PCTCN2020070712-appb-000001
即
Figure PCTCN2020070712-appb-000002
Figure PCTCN2020070712-appb-000003
其中，上述x c、y c为上述当前像素点在图像中的横、纵坐标。s(g i-g c)表示对第i个相邻像素点进行二值化处理后的像素值。
局部三值模式用于提取图像的纹理特征时，上述τ的取值比较难设置。本申请的一个实施例中，上述τ可以通过韦伯定律Weber’s law确定,韦伯定律表达式为：
Figure PCTCN2020070712-appb-000004
最后，通过局部三值模式值得DLTP直方图为：
Figure PCTCN2020070712-appb-000005
Figure PCTCN2020070712-appb-000006
其中，上述x、y表示像素点在图像中的横、纵坐标。
S30：通过卷积神经网络对待测可见光图像进行特征提取。
卷积神经网络是深度学习中使用较多的一种网络模型，该网络模型具有多层结构，每个层对该层的输入数据进行特征提取，这些被提取出来的特征以二维图像的形式继续被输入下一个层。
本申请的一个实施例中，在设计上述卷积神经网络的结构时，可以利用真实人脸数据库中各原始图像的大小作为上述卷积神经网络的输入图像的大小，这样可以使得上述卷积神经网络对一种大小的输入图像进行特征提取，从而减少多尺度输入图像对卷积神经网络带来的过多计算量。
在对上述卷积神经网络进行训练时，可以以上述真实人脸数据库中的原始图像为训练样本，对上述卷积神经网络进行训练，使得上述卷积神经网络学习到真实人脸数据库中各个原始图像中真实人脸的特征。
S40：基于对待测红外图像进行边缘检测的结果、上述纹理特征提取的结果和通过卷积神经网络对待测可见光图像进行特征提取的结果，判断待测红外图像和待测可见光图像是否通过活体检测。
本申请的一个实施例中，可以对上述待测红外图像进行边缘检测的结果、上述纹理特征提取的结果和通过卷积神经网络对待测可见光图像进行特征提取的结果进行特征融合，然后依据特征融合结果检测上述待测红外图像和待测可见光图像是否通过活体检测。
具体的，进行特征融合时，可以通过网络模型的全连接层实现。
全连接层可以包含多个节点，每一个节点分别用于获得上述对待测红外图像进行边缘检测的结果、纹理特征提取的结果和通过卷积神经网络对待测可见光图像进行特征提取的结果，这样上述全连接层可以将上述三种结果对应的特征综合起来。
由于上述全连接层具有全相连的特性，一般全连接层的参数也是最多的。例如在VGG16中，第一个全连接层FC1有4096个节点，上一层池化层POOL2有7*7*512＝25088个节点，则上述FC1需要具有4096*25088个权值，这些权值需要消耗很大的内存。
本申请的一个实施例中，可以通过softmax等分类器对上述全连接层的输出结果进行分类，获得上述待测红外图像和待测可见光图像是否通过活体检测。例如，softmax分类器可以通过设置阈值的方式对上述softmax分类器的输入信息进行判断，也就是，对上述全连接层的输出结果进行判断，当softmax分类器的输入信息大于预设阈值时，判定待测红外图像和待测可见光图像为针对真实人脸的图像，通过活体检测；反之，判定待测红外图像和待测可见光图像为针对非真实人脸的图像，未通过活体检测。
由此，在应用本实施例提供的方案进行活体检测时，能够分别获取待测红外图像和待测可见光图像，对待测红外图像进行边缘检测和纹理特征提取，通过卷积神经网络对待测可见光图像进行特征提取，基于对待测红外图像进行边缘检测的结果、上述纹理特征提取的结果和通过卷积神经网络对待测可见光图像进行特征提取的结果，判断待测红外图像和待测可见光图像是否通过活体检测，这一过程能够结合边缘检测、纹理特征提取和卷积神经网络三种技术的优势，能够有效的进行活体检测。在待测红外图像和待测可见光图像中包括人脸的图像区域的情况下，能够高效的判别图像中的人脸是否属于活体的人脸，从而提升了判别准确率。
图3示出本申请实施例提供的基于面部识别的活体检测的方法的另一种流程示意图。该方法可以由电子设备执行，例如，上述电子设备可以是终端设备或服务端设备。换言之，上述方法可以由安装在终端设备或服务端设备的软件或硬件来执行。上述服务端设备包括但不限于：单台服务器、服务器集群、云端服务器或云端服务器集群等。如图3所示，该方法可以包括以下步骤S11-S40。
S11：利用图像采集器件采集红外图像和可见光图像，通过人脸检测算法，在红外图像和可见光图像中分别定位人脸。
在一种可能的实现方式中，本步骤包括：通过人脸检测算法，检测红外图像和可见光图像中的人脸区域。另外，在检测出上述人脸区域之后，还可以在所检测出人脸区域的基础上，确定红外图像中的红外人脸个数和人脸特征点位置，并确定可见光图像中的可见光人脸个数和人脸特征点位置。上述过程实现了在红外图像和可见光图像中分别定位人脸。
S12：根据红外图像中的人脸特征点位置和可见光图像中的人脸特征点位置，获得人脸的偏转角和瞳距。
具体的，本步骤中，根据红外图像中的人脸特征点位置，获得红外图像中人脸的偏转角和瞳距，并根据可见光图像中的人脸特征点位置，获得可见光图像中人脸的偏转角和瞳 距。
本申请的一个实施例中，可以根据人脸特征点中与人眼睛相关的特征点的位置计算人两只眼睛之间的距离，然后根据上述距离确定人脸的瞳距。
S13：根据所获得的偏转角和瞳距，从利用图像采集器件采集的红外图像和可见光图像中，选择待测红外图像和待测可见光图像。
依据偏转角和瞳距这两个参数可以过滤掉图像采集器件所采集的红外图像和可见光图像中质量较差的人脸图像，从而提高活体检测的鲁棒性。
S10：分别获取待测红外图像和待测可见光图像。
S14：对待测红外图像进行灰度像素处理得到红外灰度图像。
本申请的一个实施例中，可以对待测红外图像进行灰度变换，得到待测红外图像对应的灰度图像，将该灰度图像作为红外灰度图像
S15：对待测可见光图像进行归一化处理，得到归一化可见光图像。
对图像进行归一化处理是指对图像进行一系列标准的处理变换，使之变换为一固定标准图像的过程，该标准图像称作归一化图像。
S16：将归一化可见光图像和红外灰度图像融合为四通道图像。
其中，上述四通道图像包括：红绿蓝RGB通道图像和红外灰度通道图像。
上述RGB通道图像为：RGB三个通道分别对应的图像。上述红外灰度通道图像为：第四个通道对应的上述红外灰度图像。
S20：对待测红外图像进行边缘检测和纹理特征提取。
本申请的一个实施例中，可以对红外灰度通道图像进行边缘检测，对上述RGB通道图像进行纹理特征提取。
在一种可能的实现方式中，对所述待测红外图像进行边缘检测包括：通过高斯变换滤除待测红外图像中的噪声；通过索贝尔sobel算子对滤除噪声后的待测红外图像进行边缘检测，得到边缘检测结果；统计上述边缘检测结果在不同方向针对边缘像素点数量的直方图信息，并根据统计得到的直方图信息滤除边缘检测结果中的噪声。这样在待测红外图像中存在人脸的图像区域时，可以得到各个人脸的图像区域的边缘信息，这种情况下，可以将上述边缘信息作为人脸的图像区域的特征，称为人脸特征。
在一种可能的实现方式中，可以基于静态图像纹理进行活体检测。这种情况下，需从图像中提取一个或多个具有运动不变性质的特征，比如，图像中的边界线条或角点等，并依据这些特征建立活体检测模型。然后通过上述活体检测模型检测图像是否为针对活体采集的图像。
具体的，在上述活体为真实人脸时，基于静态图像纹理进行真实人脸检测时，可以基 于LBP、Gabor小波、HOG等实现真实人脸检测。
在一种可能的实现方式中，可以基于动态纹理进行活体检测。这种情况下，在上述活体为真实人脸时，可以通过学习真实人脸微纹理的结构和动态信息，并利用LBP在空域进行特征算子扩展的方式进行真实人脸识别。
在一种可能的实现方式中，对待测红外图像进行纹理特征提取包括：通过动态局部三值模式(英文：Dynamic Local Ternary Pattern，缩写：DLTP)提取待测红外图像的纹理特征。
具体的，上述DLTP是由局部三值模式(Local Ternary Pattern，LTP)演变来的。LTP由局部二模式(local binary pattern，LBP)演变而来。
下面简述基于LTP得到DLTP信息的过程。
假设，当前像素点的像素值为g c，以该当前像素点为中心、且与该当前像素点相邻的P个相邻像素点的灰度值分别g 1,g 2,…,g P。
首先，以g c±τ为阈值，对上述相邻像素点进行二值化处理。
然后，根据相邻像素点的不同位置，对二值化处理后各个相邻像素点的像素值进行加权求和，得到当前像素点的局部三值模式值
Figure PCTCN2020070712-appb-000007
即
Figure PCTCN2020070712-appb-000008
Figure PCTCN2020070712-appb-000009
其中，上述x c、y c为上述当前像素点在图像中的横、纵坐标。s(g i-g c)表示对第i个相邻像素点进行二值化处理后的像素值。
局部三值模式用于提取图像的纹理特征时，上述τ的取值比较难设置。本申请的一个实施例中，上述τ可以通过韦伯定律Weber’s law确定,韦伯定律表达式为：
Figure PCTCN2020070712-appb-000010
最后，通过局部三值模式值得DLTP直方图为：
Figure PCTCN2020070712-appb-000011
Figure PCTCN2020070712-appb-000012
其中，上述x、y表示像素点在图像中的横、纵坐标。
S30：通过卷积神经网络对待测可见光图像进行特征提取。
卷积神经网络是深度学习中使用较多的一种网络模型，该网络模型具有多层结构，每个层对该层的输入数据进行特征提取，这些被提取出来的特征以二维图像的形式继续被输入下一个层。
在一种可能的实现方式中，通过卷积神经网络对待测可见光图像进行特征提取时，可以通过卷积神经网络对上述四通道图像进行特征提取。
S40：基于对待测红外图像进行边缘检测的结果、上述纹理特征提取的结果和通过卷积神经网络对待测可见光图像进行特征提取的结果，判断待测红外图像和待测可见光图像是否通过活体检测。
由此，在应用本实施例提供的方案进行活体检测时，能够分别获取待测红外图像和待测可见光图像，对待测红外图像进行边缘检测和纹理特征提取，对待测可见光图像通过卷积神经网络进行特征提取，基于对待测红外图像进行边缘检测的结果、上述纹理特征提取的结果和通过卷积神经网络对待测可见光图像进行特征提取的结果，判断待测红外图像和待测可见光图像是否通过活体检测，这一过程能够结合边缘检测、纹理特征提取和卷积神经网络三种技术的优势，能够有效的进行活体检测。在待测红外图像和待测可见光图像中包括人脸的图像区域的情况下，能够高效的判别图像中的人脸是否属于活体的人脸，从而提升了判别准确率。
图4示出本申请实施例提供的一种基于面部识别的活体检测的装置的结构示意图，该装置100包括：获取模块110、处理模块120和判别模块130。
获取模块110用于分别获取待测红外图像和待测可见光图像。处理模块120用于对所述待测红外图像进行边缘检测和纹理特征提取。所述处理模块120还用于通过卷积神经网络对所述待测可见光图像进行特征提取。判别模块130，用于基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
在一种可能的实现方式中，所述获取模块110具体用于利用图像采集器件采集红外图像和可见光图像；通过人脸检测算法，在所述红外图像和所述可见光图像中分别定位人脸；根据在所述红外图像和所述可见光图像中分别定位人脸的结果，从所述红外图像和所述可见光图像中，分别获取待测红外图像和待测可见光图像。
在一种可能的实现方式中，所述获取模块110具体用于所述通过人脸检测算法，检测所述红外图像和所述可见光图像中的人脸区域；确定所述红外图像中的人脸特征点位置；确定所述可见光图像中的人脸特征点位置。
在一种可能的实现方式中，所述获取模块110，具体用于根据所述红外图像中的人脸 特征点位置，获得所述红外图像中人脸的偏转角和瞳距，并根据所述可见光图像中的人脸特征点位置，获得所述可见光图像中人脸的偏转角和瞳距；根据所获得的偏转角和瞳距，从所述红外图像和可见光图像中，选择所述待测红外图像和待测可见光图像。
在一种可能的实现方式中，所述获取模块110具体用于利用图像采集器件采集红外图像和可见光图像；通过人脸检测算法，在所述红外图像和所述可见光图像中分别定位人脸；根据在所述红外图像和所述可见光图像中分别定位人脸的结果，从所述红外图像和所述可见光图像中，分别获取待测红外图像和待测可见光图像。
在一种可能的实现方式中，所述获取模块110具体用于所述通过人脸检测算法，检测所述红外图像和所述可见光图像中的人脸区域；确定所述红外图像中的人脸特征点位置；确定所述可见光图像中的人脸特征点位置。
在一种可能的实现方式中，所述处理模块120还用于对所述待测红外图像进行灰度像素处理得到红外灰度图像；对所述待测可见光图像进行归一化处理，得到归一化可见光图像；将所述归一化可见光图像和所述红外灰度图像融合为四通道图像，其中，所述四通道图像包括：红绿蓝RGB通道图像和红外灰度通道图像，所述RGB通道图像为：RGB三个通道分别对应的图像，所述红外灰度通道图像为：第四个通道对应的所述红外灰度图像。
在一种可能的实现方式中，所述处理模块120具体用于对所述红外灰度通道图像进行边缘检测，对所述RGB通道图像进行纹理特征提取；通过卷积神经网络对所述四通道图像进行特征提取。
在一种可能的实现方式中，所述处理模块120具体用于通过高斯变换滤除所述待测红外图像中的噪声；通过索贝尔sobel算子对滤除噪声后的待测红外图像进行边缘检测，得到边缘检测结果；统计所述边缘检测结果在不同方向针对边缘像素点数量的直方图信息，并根据统计得到的直方图信息滤除所述边缘检测结果中的噪声。
在一种可能的实现方式中，所述处理模块120具体用于通过动态局部三值模式提取所述待测红外图像的纹理特征。
图5示出本申请实施例提供的一种基于面部识别的活体检测的装置的结构示意图，该装置100包括：获取模块110、处理模块120、判别模块130和筛选模块140。
筛选模块140用于根据所述红外图像中的人脸特征点位置，获得所述红外图像中人脸的偏转角和瞳距，并根据所述可见光图像中的人脸特征点位置，获得所述可见光图像中人脸的偏转角和瞳距；根据所获得的偏转角和瞳距，从所述红外图像和可见光图像中，选择所述待测红外图像和待测可见光图像。获取模块110用于分别获取待测红外图像和待测可见光图像。处理模块120用于对所述待测红外图像进行边缘检测和纹理特征提取。所述处理模块120还用于通过卷积神经网络对所述待测可见光图像进行特征提取。判别模块130， 用于基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
本申请实施例提供的该装置100，可执行前文方法实施例中所述的各方法，并实现前文方法实施例中所述的各方法的功能和有益效果，在此不再赘述。
图6示出执行本申请实施例提供的一种基于面部识别的活体检测的方法的电子设备的硬件结构示意图，如图6所示，该电子设备可因配置或性能不同而产生比较大的差异，可以包括一个或一个以上的处理器701和存储器702，存储器702中可以存储有一个或一个以上存储应用程序或数据。其中，存储器702可以是短暂存储或持久存储。存储在存储器702的应用程序可以包括一个或一个以上模块(图示未示出)，每个模块可以包括对该电子设备中的一系列计算机可执行指令。更进一步地，处理器701可以设置为与存储器702通信，在该电子设备上执行存储器702中的一系列计算机可执行指令。该电子设备还可以包括一个或一个以上电源703，一个或一个以上有线或无线网络接口704，一个或一个以上输入输出接口705，一个或一个以上键盘706等。
本申请的一个实施例中，上述图6所示的电子设备还可以包括图像采集器件。
在一个具体的实施例中，该电子设备包括图像采集器件，分别获取待测红外图像和待测可见光图像；处理器；以及被安排成存储计算机可执行指令的存储器，所述可执行指令在被执行时使用所述处理器执行以下操作：分别获取待测红外图像和待测可见光图像；对所述待测红外图像进行边缘检测和纹理特征提取；通过卷积神经网络对所述待测可见光图像进行特征提取；基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
由此，执行本申请实施例提供的一种基于面部识别的活体检测的方法的电子设备可执行前文方法实施例中所述的各方法，并实现前文方法实施例中所述的各方法的功能和有益效果，在此不再赘述。
本申请实施例的电子设备以多种形式存在，包括但不限于以下设备。
(1)移动通信设备:这类设备的特点是具备移动通信功能，并且以提供话音、数据通信为主要目标。这类终端包括:智能手机(例如iPhone)、多媒体手机、功能性手机，以及低端手机等。
(2)超移动个人计算机设备:这类设备属于个人计算机的范畴，有计算和处理功能，一般也具备移动上网特性。这类终端包括:PDA、MID和UMPC设备等，例如iPad。
(3)便携式娱乐设备:这类设备可以显示和播放多媒体内容。该类设备包括:音频、视频 播放器(例如iPod)，掌上游戏机，电子书，以及智能玩具和便携式车载导航设备。
(4)服务器:提供计算服务的设备，服务器的构成包括处理器、硬盘、内存、系统总线等，服务器和通用的计算机架构类似，但是由于需要提供高可靠的服务，因此在处理能力、稳定性、可靠性、安全性、可扩展性、可管理性等方面要求较高。
(5)其他具有数据交互功能的电子装置。
进一步地，本申请实施例还提供了一种基于面部识别的活体检测的系统，包括：图像采集设备，用于采集红外图像和可见光图像；电子设备，包括：处理器；以及被安排成存储计算机可执行指令的存储器，所述可执行指令在被执行时使用所述处理器执行以下操作：获取所述图像采集设备采集的红外图像和可见光图像，筛选出待检测的红外图像和待检测的可见光图像；通过卷积神经网络对所述待测红外图像进行边缘检测和纹理特征提取；对所述待测可见光图像进行特征提取；基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
由此，本申请实施例还提供的一种基于面部识别的活体检测的系统能够执行前文方法实施例中所述的各方法，并实现前文方法实施例中所述的各方法的功能和有益效果，在此不再赘述。
进一步地，本申请实施例还提供了一种计算机可读存储介质，所述计算机可读存储介质用于存储计算机可执行指令，所述计算机可执行指令被处理器执行时实现以下流程：分别获取待测红外图像和待测可见光图像；对所述待测红外图像进行边缘检测和纹理特征提取；通过卷积神经网络对所述待测可见光图像进行特征提取；基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
由此，所述计算机可执行指令被处理器执行时能够执行前文方法实施例中所述的各方法，并实现前文方法实施例中所述的各方法的功能和有益效果，在此不再赘述。
其中，所述的计算机可读存储介质包括只读存储器(Read-Only Memory，简称ROM)、随机存取存储器(Random Access Memory，简称RAM)、磁碟或者光盘等。
进一步地，本申请实施例还提供了一种计算机程序产品，所述计算机程序产品包括存储在非暂态计算机可读存储介质上的计算机程序，所述计算机程序包括程序指令，当所述程序指令被计算机执行时，实现以下流程：分别获取待测红外图像和待测可见光图像；对所述待测红外图像进行边缘检测和纹理特征提取；通过卷积神经网络对所述待测可见光图像进行特征提取；基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图 像和待测可见光图像是否通过活体检测。
由此，执行本申请实施例提供的计算机程序产品能够执行前文方法实施例中所述的各方法，并实现前文方法实施例中所述的各方法的功能和有益效果，在此不再赘述。
本说明书中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于装置、电子设备、计算机可读介质、计算机程序产品、系统实施例而言，由于其基本相似于方法实施例，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
以上所述仅为本申请的实施例而已，并不用于限制本申请。对于本领域技术人员来说，本申请可以有各种更改和变化。凡在本申请的精神和原理之内所作的任何修改、等同替换、改进等，均应包含在本申请的权利要求范围之内。

Claims (26)

  1. 一种基于面部识别的活体检测的方法，其特征在于，包括：
    分别获取待测红外图像和待测可见光图像；
    对所述待测红外图像进行边缘检测和纹理特征提取；
    通过卷积神经网络对所述待测可见光图像进行特征提取；
    基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
  2. 根据权利要求1所述的方法，其特征在于，在所述分别获取待测红外图像和待测可见光图像之前，还包括：
    利用图像采集器件采集红外图像和可见光图像；
    通过人脸检测算法，在所述红外图像和所述可见光图像中分别定位人脸；
    所述分别获取待测红外图像和待测可见光图像，包括：
    根据在所述红外图像和所述可见光图像中分别定位人脸的结果，从所述红外图像和所述可见光图像中，分别获取待测红外图像和待测可见光图像。
  3. 根据权利要求2所述的方法，其特征在于，所述通过人脸检测算法，在所述红外图像和所述可见光图像中分别定位人脸包括：
    所述通过人脸检测算法，检测所述红外图像和所述可见光图像中的人脸区域；
    确定所述红外图像中的人脸特征点位置；
    确定所述可见光图像中的人脸特征点位置。
  4. 根据权利要求3所述的方法，其特征在于，所述根据在所述红外图像和所述可见光图像中分别定位人脸的结果，从所述红外图像和所述可见光图像中，分别获取待测红外图像和待测可见光图像，包括：
    根据所述红外图像中的人脸特征点位置，获得所述红外图像中人脸的偏转角和瞳距，并根据所述可见光图像中的人脸特征点位置，获得所述可见光图像中人脸的偏转角和瞳距；
    根据所获得的偏转角和瞳距，从所述红外图像和可见光图像中，选择所述待测红外图像和待测可见光图像。
  5. 根据权利要求1所述的方法，其特征在于，在分别获取待测红外图像和待测可见光图像之后，还包括：
    对所述待测红外图像进行灰度像素处理得到红外灰度图像；
    对所述待测可见光图像进行归一化处理，得到归一化可见光图像；
    将所述归一化可见光图像和所述红外灰度图像融合为四通道图像，其中，所述四通道 图像包括：红绿蓝RGB通道图像和红外灰度通道图像，所述RGB通道图像为：RGB三个通道分别对应的图像，所述红外灰度通道图像为：第四个通道对应的所述红外灰度图像。
  6. 根据权利要求5所述的方法，其特征在于，对所述待测红外图像进行边缘检测和纹理特征提取，包括：
    对所述红外灰度通道图像进行边缘检测，对所述RGB通道图像进行纹理特征提取；
    通过卷积神经网络对所述待测可见光图像进行特征提取，包括：
    通过卷积神经网络对所述四通道图像进行特征提取。
  7. 根据权利要求1所述的方法，其特征在于，对所述待测红外图像进行边缘检测，包括：
    通过高斯变换滤除所述待测红外图像中的噪声；
    通过索贝尔sobel算子对滤除噪声后的待测红外图像进行边缘检测，得到边缘检测结果；
    统计所述边缘检测结果在不同方向针对边缘像素点数量的直方图信息，并根据统计得到的直方图信息滤除所述边缘检测结果中的噪声。
  8. 根据权利要求1所述的方法，其特征在于，对所述待测红外图像进行纹理特征提取，包括：
    通过动态局部三值模式提取所述待测红外图像的纹理特征。
  9. 一种电子设备，包括：
    图像采集器件，分别获取待测红外图像和待测可见光图像；
    处理器；以及
    被安排成存储计算机可执行指令的存储器，所述可执行指令在被执行时使用所述处理器执行以下操作：
    对所述待测红外图像进行边缘检测和纹理特征提取；
    通过卷积神经网络对所述待测可见光图像进行特征提取；
    基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
  10. 根据权利要求9所述的电子设备，其特征在于，在所述分别获取待测红外图像和待测可见光图像之前，还执行：
    利用图像采集器件采集红外图像和可见光图像；
    通过人脸检测算法，在所述红外图像和所述可见光图像中分别定位人脸；
    所述分别获取待测红外图像和待测可见光图像，包括：
    根据在所述红外图像和所述可见光图像中分别定位人脸的结果，从所述红外图像和所述可见光图像中，分别获取待测红外图像和待测可见光图像。
  11. 根据权利要求10所述的电子设备，其特征在于，所述通过人脸检测算法，在所述红外图像和所述可见光图像中分别定位人脸包括：
    所述通过人脸检测算法，检测所述红外图像和所述可见光图像中的人脸区域；
    确定所述红外图像中的人脸特征点位置；
    确定所述可见光图像中的人脸特征点位置。
  12. 根据权利要求11所述的电子设备，其特征在于，所述根据在所述红外图像和所述可见光图像中分别定位人脸的结果，从所述红外图像和所述可见光图像中，分别获取待测红外图像和待测可见光图像，包括：
    根据所述红外图像中的人脸特征点位置，获得所述红外图像中人脸的偏转角和瞳距，并根据所述可见光图像中的人脸特征点位置，获得所述可见光图像中人脸的偏转角和瞳距；
    根据所获得的偏转角和瞳距，从所述红外图像和可见光图像中，选择所述待测红外图像和待测可见光图像。
  13. 根据权利要求9所述的电子设备，其特征在于，在分别获取待测红外图像和待测可见光图像之后，还执行：
    对所述待测红外图像进行灰度像素处理得到红外灰度图像；
    对所述待测可见光图像进行归一化处理，得到归一化可见光图像；
    将所述归一化可见光图像和所述红外灰度图像融合为四通道图像，其中，所述四通道图像包括：红绿蓝RGB通道图像和红外灰度通道图像，所述RGB通道图像为：RGB三个通道分别对应的图像，所述红外灰度通道图像为：第四个通道对应的所述红外灰度图像。
  14. 根据权利要求13所述的电子设备，其特征在于，对所述待测红外图像进行边缘检测和纹理特征提取，包括：
    对所述红外灰度通道图像进行边缘检测，对所述RGB通道图像进行纹理特征提取；
    通过卷积神经网络对所述待测可见光图像进行特征提取，包括：
    通过卷积神经网络对所述四通道图像进行特征提取。
  15. 根据权利要求9所述的电子设备，其特征在于，对所述待测红外图像进行边缘检测，包括：
    通过高斯变换滤除所述待测红外图像中的噪声；
    通过索贝尔sobel算子对滤除噪声后的待测红外图像进行边缘检测，得到边缘检测结果；
    统计所述边缘检测结果在不同方向针对边缘像素点数量的直方图信息，并根据统计得 到的直方图信息滤除所述边缘检测结果中的噪声。
  16. 根据权利要求9所述的电子设备，其特征在于，对所述待测红外图像进行纹理特征提取，包括：
    通过动态局部三值模式提取所述待测红外图像的纹理特征。
  17. 一种计算机可读介质，所述计算机可读介质存储一个或多个程序，所述一个或多个程序当被包括多个应用程序的电子设备执行时，使得所述电子设备执行以下操作：
    分别获取待测红外图像和待测可见光图像；
    对所述待测红外图像进行边缘检测和纹理特征提取；
    通过卷积神经网络对所述待测可见光图像进行特征提取；
    基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
  18. 一种基于面部识别的活体检测的系统，其特征在于，包括：
    图像采集设备，用于采集红外图像和可见光图像；
    电子设备，包括：处理器；以及被安排成存储计算机可执行指令的存储器，所述可执行指令在被执行时使用所述处理器执行以下操作：
    获取所述图像采集设备采集的红外图像和可见光图像，筛选出待检测的红外图像和待检测的可见光图像；
    对所述待测红外图像进行边缘检测和纹理特征提取；
    通过卷积神经网络对所述待测可见光图像进行特征提取；
    基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
  19. 一种基于面部识别的活体检测的装置，其特征在于，包括：
    获取模块，用于分别获取待测红外图像和待测可见光图像；
    处理模块，用于对所述待测红外图像进行边缘检测和纹理特征提取；
    所述处理模块，还用于通过卷积神经网络对所述待测可见光图像进行特征提取；
    判别模块，用于基于对所述待测红外图像进行边缘检测的结果、所述纹理特征提取的结果和通过卷积神经网络对所述待测可见光图像进行特征提取的结果，判断所述待测红外图像和待测可见光图像是否通过活体检测。
  20. 根据权利要求19所述的装置，其特征在于，
    所述获取模块，具体用于利用图像采集器件采集红外图像和可见光图像；通过人脸检 测算法，在所述红外图像和所述可见光图像中分别定位人脸；根据在所述红外图像和所述可见光图像中分别定位人脸的结果，从所述红外图像和所述可见光图像中，分别获取待测红外图像和待测可见光图像。
  21. 根据权利要求20所述的装置，其特征在于，
    所述获取模块，具体用于所述通过人脸检测算法，检测所述红外图像和所述可见光图像中的人脸区域；确定所述红外图像中的人脸特征点位置；确定所述可见光图像中的人脸特征点位置。
  22. 根据权利要求21所述的装置，其特征在于，
    所述获取模块，具体用于根据所述红外图像中的人脸特征点位置，获得所述红外图像中人脸的偏转角和瞳距，并根据所述可见光图像中的人脸特征点位置，获得所述可见光图像中人脸的偏转角和瞳距；根据所获得的偏转角和瞳距，从所述红外图像和可见光图像中，选择所述待测红外图像和待测可见光图像。
  23. 根据权利要求19所述的装置，其特征在于，
    所述处理模块，还用于对所述待测红外图像进行灰度像素处理得到红外灰度图像；对所述待测可见光图像进行归一化处理，得到归一化可见光图像；将所述归一化可见光图像和所述红外灰度图像融合为四通道图像，其中，所述四通道图像包括：红绿蓝RGB通道图像和红外灰度通道图像，所述RGB通道图像为：RGB三个通道分别对应的图像，所述红外灰度通道图像为：第四个通道对应的所述红外灰度图像。
  24. 根据权利要求23所述的装置，其特征在于，
    所述处理模块，具体用于对所述红外灰度通道图像进行边缘检测，对所述RGB通道图像进行纹理特征提取；通过卷积神经网络对所述四通道图像进行特征提取。
  25. 根据权利要求19所述的装置，其特征在于，
    所述处理模块，具体用于通过高斯变换滤除所述待测红外图像中的噪声；通过索贝尔sobel算子对滤除噪声后的待测红外图像进行边缘检测，得到边缘检测结果；统计所述边缘检测结果在不同方向针对边缘像素点数量的直方图信息，并根据统计得到的直方图信息滤除所述边缘检测结果中的噪声。
  26. 根据权利要求19所述的装置，其特征在于，
    所述处理模块，具体用于通过动态局部三值模式提取所述待测红外图像的纹理特征。
PCT/CN2020/070712 2019-01-25 2020-01-07 基于面部识别的活体检测的方法、电子设备和存储介质 WO2020151489A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP20744256.7A EP3916627A4 (en) 2019-01-25 2020-01-07 LIVING BODY DETECTION METHOD BASED ON FACIAL RECOGNITION, AND ELECTRONIC DEVICE AND STORAGE MEDIA
US17/425,242 US11830230B2 (en) 2019-01-25 2020-01-07 Living body detection method based on facial recognition, and electronic device and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910072693.8 2019-01-25
CN201910072693.8A CN111488756B (zh) 2019-01-25 2019-01-25 基于面部识别的活体检测的方法、电子设备和存储介质

Publications (1)

Publication Number Publication Date
WO2020151489A1 true WO2020151489A1 (zh) 2020-07-30

Family

ID=71735432

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/070712 WO2020151489A1 (zh) 2019-01-25 2020-01-07 基于面部识别的活体检测的方法、电子设备和存储介质

Country Status (4)

Country Link
US (1) US11830230B2 (zh)
EP (1) EP3916627A4 (zh)
CN (1) CN111488756B (zh)
WO (1) WO2020151489A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111914758A (zh) * 2020-08-04 2020-11-10 成都奥快科技有限公司 一种基于卷积神经网络的人脸活体检测方法及装置
CN112116580A (zh) * 2020-09-22 2020-12-22 中用科技有限公司 用于摄像头支架的检测方法、系统及设备
CN112183422A (zh) * 2020-10-09 2021-01-05 成都奥快科技有限公司 一种基于时空特征的人脸活体检测方法、装置、电子设备及存储介质
CN112232204A (zh) * 2020-10-16 2021-01-15 中科智云科技有限公司 基于红外图像的活体检测方法
CN112329674A (zh) * 2020-11-12 2021-02-05 北京环境特性研究所 基于多纹理特征融合的结冰湖泊检测方法和装置
CN112434647A (zh) * 2020-12-09 2021-03-02 浙江光珀智能科技有限公司 一种人脸活体检测方法
CN112699826A (zh) * 2021-01-05 2021-04-23 风变科技（深圳）有限公司 人脸检测方法、装置、计算机设备和存储介质
CN113486829A (zh) * 2021-07-15 2021-10-08 京东科技控股股份有限公司 人脸活体检测方法、装置、电子设备及存储介质
CN113627300A (zh) * 2021-08-02 2021-11-09 中电福富信息科技有限公司 基于深度学习的人脸识别与活体检测方法

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112163988B (zh) * 2020-08-17 2022-12-13 中国人民解放军93114部队 红外图像的生成方法、装置、计算机设备和可读存储介质
CN112070954A (zh) * 2020-09-14 2020-12-11 上海商汤智能科技有限公司 活体识别方法、装置、设备及存储介质
CN114387324A (zh) * 2021-12-22 2022-04-22 北京的卢深视科技有限公司 深度成像方法、装置、电子设备和计算机可读存储介质
CN115798004A (zh) * 2022-11-27 2023-03-14 深圳市巨龙创视科技有限公司 基于局部区域的人脸打卡方法、装置、电子设备及介质
CN115601818B (zh) * 2022-11-29 2023-04-07 海豚乐智科技(成都)有限责任公司 一种轻量化可见光活体检测方法及装置
CN116189800B (zh) * 2023-02-23 2023-08-18 深圳大学 基于气体检测的模式识别方法、装置、设备及存储介质
CN116977154B (zh) * 2023-09-22 2024-03-19 南方电网数字电网研究院有限公司 可见光图像与红外图像融合储存方法、装置、设备和介质
CN117542127A (zh) * 2024-01-09 2024-02-09 南方科技大学 基于多光谱偏振光的皮肤检测方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107992794A (zh) * 2016-12-30 2018-05-04 腾讯科技（深圳）有限公司 一种活体检测方法、装置和存储介质
CN108280418A (zh) * 2017-12-12 2018-07-13 北京深醒科技有限公司 脸部图像的欺骗识别方法及装置
CN108764091A (zh) * 2018-05-18 2018-11-06 北京市商汤科技开发有限公司 活体检测方法及装置、电子设备和存储介质
CN108921100A (zh) * 2018-07-04 2018-11-30 武汉高德智感科技有限公司 一种基于可见光图像与红外图像融合的人脸识别方法及系统
CN108985134A (zh) * 2017-06-01 2018-12-11 重庆中科云丛科技有限公司 基于双目摄像机的人脸活体检测及刷脸交易方法及系统

Family Cites Families (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6920236B2 (en) * 2001-03-26 2005-07-19 Mikos, Ltd. Dual band biometric identification system
WO2008130906A1 (en) * 2007-04-17 2008-10-30 Mikos, Ltd. System and method for using three dimensional infrared imaging to provide psychological profiles of individuals
US9208492B2 (en) * 2013-05-13 2015-12-08 Hoyos Labs Corp. Systems and methods for biometric authentication of transactions
JP6729394B2 (ja) * 2015-01-13 2020-07-22 ソニー株式会社 画像処理装置、画像処理方法、プログラム及びシステム
US20230031545A1 (en) * 2015-08-18 2023-02-02 Mbl Limited Robotic kitchen systems and methods in an instrumented environment with electronic cooking libraries
KR102483642B1 (ko) * 2016-08-23 2023-01-02 삼성전자주식회사 라이브니스 검사 방법 및 장치
CN108229325A (zh) * 2017-03-16 2018-06-29 北京市商汤科技开发有限公司 人脸检测方法和系统、电子设备、程序和介质
CN107545241B (zh) * 2017-07-19 2022-05-27 百度在线网络技术（北京）有限公司 神经网络模型训练及活体检测方法、装置及存储介质
CN108197586B (zh) * 2017-12-12 2020-04-21 北京深醒科技有限公司 脸部识别方法和装置
CN107918773B (zh) * 2017-12-13 2021-06-04 汉王科技股份有限公司 一种人脸活体检测方法、装置及电子设备
KR102466997B1 (ko) * 2018-01-22 2022-11-14 삼성전자주식회사 라이브니스 검사 방법 및 장치
CN108339401A (zh) 2018-03-06 2018-07-31 安徽郁萃生物科技有限公司 一种节能型膜分离装置
CN108399401B (zh) * 2018-03-27 2022-05-03 百度在线网络技术（北京）有限公司 用于检测人脸图像的方法和装置
WO2019196559A1 (zh) * 2018-04-12 2019-10-17 Oppo广东移动通信有限公司 人脸识别方法、装置及移动终端、存储介质
US10956714B2 (en) * 2018-05-18 2021-03-23 Beijing Sensetime Technology Development Co., Ltd Method and apparatus for detecting living body, electronic device, and storage medium
CN108470169A (zh) * 2018-05-23 2018-08-31 国政通科技股份有限公司 人脸识别系统及方法
CN108921041A (zh) * 2018-06-06 2018-11-30 深圳神目信息技术有限公司 一种基于rgb和ir双目摄像头的活体检测方法及装置
CN109034102B (zh) * 2018-08-14 2023-06-16 腾讯科技（深圳）有限公司 人脸活体检测方法、装置、设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107992794A (zh) * 2016-12-30 2018-05-04 腾讯科技（深圳）有限公司 一种活体检测方法、装置和存储介质
CN108985134A (zh) * 2017-06-01 2018-12-11 重庆中科云丛科技有限公司 基于双目摄像机的人脸活体检测及刷脸交易方法及系统
CN108280418A (zh) * 2017-12-12 2018-07-13 北京深醒科技有限公司 脸部图像的欺骗识别方法及装置
CN108764091A (zh) * 2018-05-18 2018-11-06 北京市商汤科技开发有限公司 活体检测方法及装置、电子设备和存储介质
CN108921100A (zh) * 2018-07-04 2018-11-30 武汉高德智感科技有限公司 一种基于可见光图像与红外图像融合的人脸识别方法及系统

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111914758A (zh) * 2020-08-04 2020-11-10 成都奥快科技有限公司 一种基于卷积神经网络的人脸活体检测方法及装置
CN112116580A (zh) * 2020-09-22 2020-12-22 中用科技有限公司 用于摄像头支架的检测方法、系统及设备
CN112116580B (zh) * 2020-09-22 2023-09-05 中用科技有限公司 用于摄像头支架的检测方法、系统及设备
CN112183422A (zh) * 2020-10-09 2021-01-05 成都奥快科技有限公司 一种基于时空特征的人脸活体检测方法、装置、电子设备及存储介质
CN112232204A (zh) * 2020-10-16 2021-01-15 中科智云科技有限公司 基于红外图像的活体检测方法
CN112329674A (zh) * 2020-11-12 2021-02-05 北京环境特性研究所 基于多纹理特征融合的结冰湖泊检测方法和装置
CN112329674B (zh) * 2020-11-12 2024-03-12 北京环境特性研究所 基于多纹理特征融合的结冰湖泊检测方法和装置
CN112434647A (zh) * 2020-12-09 2021-03-02 浙江光珀智能科技有限公司 一种人脸活体检测方法
CN112699826A (zh) * 2021-01-05 2021-04-23 风变科技（深圳）有限公司 人脸检测方法、装置、计算机设备和存储介质
CN113486829A (zh) * 2021-07-15 2021-10-08 京东科技控股股份有限公司 人脸活体检测方法、装置、电子设备及存储介质
CN113486829B (zh) * 2021-07-15 2023-11-07 京东科技控股股份有限公司 人脸活体检测方法、装置、电子设备及存储介质
CN113627300A (zh) * 2021-08-02 2021-11-09 中电福富信息科技有限公司 基于深度学习的人脸识别与活体检测方法

Also Published As

Publication number Publication date
CN111488756A (zh) 2020-08-04
US20220092882A1 (en) 2022-03-24
EP3916627A4 (en) 2022-09-07
EP3916627A1 (en) 2021-12-01
CN111488756B (zh) 2023-10-03
US11830230B2 (en) 2023-11-28

Similar Documents

Publication Publication Date Title
WO2020151489A1 (zh) 基于面部识别的活体检测的方法、电子设备和存储介质
CN108334848B (zh) 一种基于生成对抗网络的微小人脸识别方法
WO2021036436A1 (zh) 一种人脸识别方法及装置
Faraji et al. Face recognition under varying illuminations using logarithmic fractal dimension-based complete eight local directional patterns
CN110852160A (zh) 以图像为基准的生物识别系统及计算机实施方法
CN109670430A (zh) 一种基于深度学习的多分类器融合的人脸活体识别方法
CN108416291B (zh) 人脸检测识别方法、装置和系统
CN105335719A (zh) 活体检测方法及装置
CN109858375A (zh) 活体人脸检测方法、终端及计算机可读存储介质
CN112215043A (zh) 一种人脸活体检测方法
CN108470178B (zh) 一种结合深度可信度评价因子的深度图显著性检测方法
CN110543848B (zh) 一种基于三维卷积神经网络的驾驶员动作识别方法及装置
WO2013075295A1 (zh) 低分辨率视频的服装识别方法及系统
CN111784658B (zh) 一种用于人脸图像的质量分析方法和系统
CN111325107A (zh) 检测模型训练方法、装置、电子设备和可读存储介质
CN114842524B (zh) 一种基于不规则显著性像素簇的人脸鉴伪方法
CN106529441A (zh) 基于模糊边界分片的深度动作图人体行为识别方法
CN115147936A (zh) 一种活体检测方法、电子设备、存储介质及程序产品
CN111767879A (zh) 一种活体检测方法
CN108446639A (zh) 低功耗增强现实设备
CN115294162B (zh) 目标识别方法、装置、设备及存储介质
CN108647600A (zh) 人脸识别方法、设备及计算机可读存储介质
CN113221606B (zh) 一种基于ims视频会议登录的人脸识别方法
CN114299569A (zh) 一种基于眼球运动的安全人脸认证方法
CN115995103A (zh) 人脸活体检测方法、装置、计算机可读存储介质及设备

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20744256

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2020744256

Country of ref document: EP

Effective date: 20210825