WO2023000864A1 - 一种人脸识别方法及系统 - Google Patents

一种人脸识别方法及系统 Download PDF

Info

Publication number
WO2023000864A1
WO2023000864A1 PCT/CN2022/098308 CN2022098308W WO2023000864A1 WO 2023000864 A1 WO2023000864 A1 WO 2023000864A1 CN 2022098308 W CN2022098308 W CN 2022098308W WO 2023000864 A1 WO2023000864 A1 WO 2023000864A1
Authority
WO
WIPO (PCT)
Prior art keywords
face
image
spectral
spectrum
face recognition
Prior art date
Application number
PCT/CN2022/098308
Other languages
English (en)
French (fr)
Inventor
黄翊东
崔开宇
张巍
冯雪
刘仿
Original Assignee
清华大学
北京与光科技有限公司
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 清华大学, 北京与光科技有限公司 filed Critical 清华大学
Publication of WO2023000864A1 publication Critical patent/WO2023000864A1/zh

Links

Images

Classifications

    • 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
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06N3/08—Learning methods
    • 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
    • G06V10/60—Extraction of image or video features relating to illumination properties, e.g. using a reflectance or lighting model
    • 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
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • 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
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • 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
    • 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

Definitions

  • the present application relates to the field of spectral imaging technology, in particular to a face recognition method and system.
  • Face recognition is a biometric identification technology based on human facial feature information. By collecting images or videos containing human faces for analysis, it can automatically infer its attributes such as identity, expression, age and gender. As a kind of biometric identification technology, face recognition has the characteristics of non-mandatory, non-contact, simple operation and good concealment, so it is widely used in the fields of security, management supervision and multimedia entertainment.
  • the present application provides a face recognition method and system.
  • This application provides a face recognition method, including:
  • the human face spectral image to be identified is input into the trained human face spectral image feature extraction model, and the target feature vector of the human face spectral image is obtained, wherein the trained human face spectral image feature extraction model It is obtained by training a machine learning model on a sample face spectral image marked with the identity information label of the face and the authenticity category spectral information label;
  • the target feature vector is recognized, and the face recognition and living body recognition results are obtained.
  • the present application provides a face recognition method, and the acquisition of the face spectral image to be recognized includes:
  • the face recognition spectrum imaging chip includes a light modulation layer, an image sensor layer, and a signal processing circuit layer, and the light modulation layer, the image sensor layer, and the signal processing circuit layer are connected vertically from top to bottom, in:
  • the light modulation layer is used to receive the light signal reflected by the face to be recognized and perform light modulation;
  • the image sensor layer is used to convert the optical signal reflected by the face to be recognized after light modulation into an electrical signal, and the electrical signal includes spatial information of the human face image and skin spectral information;
  • the signal processing circuit layer is used to process the spatial information of the face image and the skin spectrum information output by the image sensor layer to obtain a face recognition result.
  • the face spectral imaging chip further includes a lens group, wherein the lens group is located on the upper surface of the light modulation layer and is connected to the light modulation layer for Focusing and imaging the light signal reflected by the face to obtain the light signal reflected by the face to be recognized.
  • the light modulation layer includes at least one light modulation unit, and the light modulation unit includes a plurality of micro-nano structure arrays, and each micro-nano structure array is arranged according to different preset rules. , uniformly distributed and arranged through holes are provided, and the shapes of the through holes of each micro-nano structure array are different.
  • a plurality of photosensitive pixel units are distributed on the upper surface of the image sensor layer, and each micro-nano structure array corresponds to at least one photosensitive pixel unit.
  • described trained face spectrum image feature extraction model is obtained through following steps of training:
  • the sample training set is constructed.
  • the authenticity category spectral information label is the spectrum of the living human face, simulated human face and non-human face category information label;
  • the sample training set is input into the machine learning model for training to obtain a trained face image feature extraction model, and the machine learning model is a convolutional neural network.
  • the described sample training set is input into the machine learning model for training to obtain a trained facial spectral image feature extraction model, including:
  • the convolution kernel in the convolutional neural network is trained through the sample training set, and if the preset training conditions are met, a trained face spectrum image feature extraction model is obtained, wherein the volume The kernel is used to detect the contours of the corner points of the face and the spectral characteristics of the skin.
  • the application also provides a face recognition system, including:
  • Extract feature vector module for inputting the face spectrum image to be identified into the trained face spectrum image feature extraction model, to obtain the target feature vector of the face spectrum image, wherein the trained The facial spectral image feature extraction model is obtained by training the machine learning model with the sample facial spectral images marked with human face information labels and authenticity category spectral information labels;
  • the recognition module is configured to recognize the target feature vector according to the similarity evaluation standard, and obtain a face recognition result.
  • the system also includes:
  • a sample training set module which is used to construct a sample training set according to the sample human face spectral image marked with a human face information label and a true and false category spectral information label, the human face information label is the identity information label to which the face belongs, and the authenticity
  • the category spectral information label is the spectral information label of living human face, simulated human face and non-human face category;
  • the training feature extraction model module is used to input the sample training set into the machine learning model for training to obtain a trained face image feature extraction model, and the machine learning model is a convolutional neural network.
  • the present application also provides a non-transitory computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of any one of the face recognition methods described above are implemented.
  • a face recognition method and system provided by the present application by obtaining a face spectral image, inputting the face spectral image into a feature extraction model of the face spectral image, and obtaining the target feature vector of the spatial characteristics and spectral reflection characteristics of the face; Using the similarity evaluation standard, the target eigenvector is compared with the eigenvector in the database.
  • the spectral characteristics of the skin are used to realize face recognition and living body recognition at the same time, which makes up for the shortcomings of traditional face detection.
  • Security loopholes improve the accuracy of face recognition results and enhance the security of face recognition systems.
  • Fig. 1 is a schematic flow chart of the face recognition method provided by the present application.
  • FIG. 2 is a schematic structural diagram of a face recognition spectral imaging chip provided by the present application.
  • FIG. 3 is a schematic structural diagram of the micro-nano structure array of the light modulation layer in the face recognition spectral imaging chip provided by the present application;
  • FIG. 4 is a schematic structural diagram of the face recognition system provided by the present application.
  • FIG. 5 is a schematic structural diagram of an electronic device provided by the present application.
  • Ordinary RGB cameras can only capture images with three channels of RGB.
  • the liveness detection algorithm based on ordinary RGB cameras only uses image information of three channels of RGB.
  • it needs to be identified through more complex algorithms such as video sequence analysis.
  • the real-time performance is poor and the reliability is poor, and the face recognition rate for highly simulated faces is also low.
  • Fig. 1 is a schematic flow chart of the face recognition method provided by the present application. As shown in Fig. 1, the present application provides a face recognition method, including:
  • Step 101 obtaining the spectrum image of the human face to be identified
  • Step 102 input the face spectrum image to be recognized into the trained face spectrum image feature extraction model, and obtain the target feature vector of the face spectrum image, wherein the trained face spectrum image
  • the feature extraction model is obtained by training the machine learning model from the sample face spectral images marked with the identity information label of the human face and the authenticity category spectral information label;
  • Step 103 Identify the target feature vector according to the similarity evaluation standard, and obtain face recognition and living body recognition results.
  • the spectral image of the human face to be recognized can be obtained by a computational spectral device, or by a hyperspectral device.
  • the identity information tag to which the face belongs is a tag containing the spatial characteristics of faces of different IDs.
  • the machine learning model is trained at the same time, and the feature extraction model of the face spectral image is obtained.
  • the facial spectral image feature extraction model is a machine learning model.
  • the facial spectral image is converted into a high-level
  • the target feature vector in the three-dimensional space the obtained target feature vector includes the spatial characteristics of the face and the spectral reflectance characteristics.
  • the target feature vector may be a feature vector of a living human face, a feature vector of a simulated human face, or a feature vector of a non-human face.
  • the facial spectral image feature extraction model can be a convolutional neural network, or any machine learning model, and the machine learning model can also be a Support Vector Machine (SVM for short), or a perception model. machine. This application is illustrated with the training process of convolutional neural network.
  • SVM Support Vector Machine
  • the similarity evaluation standard can be the Euclidean distance between the feature vectors, the angle between the feature vectors in the high-dimensional space, or the algorithm used to measure the similarity between vectors in the high-dimensional space .
  • the sample feature vector with the highest similarity to the target feature vector to be recognized in the existing face image database is selected as the candidate recognition result.
  • the sample feature vector in the database includes the face spatial characteristics and spectral reflection characteristics. If the similarity reaches the set threshold, the recognition is considered successful, otherwise it is regarded as a failure.
  • the target feature vectors are compared and recognized, so as to obtain the results of face recognition and living body recognition.
  • the face recognition method provided by this application obtains the face spectrum image, inputs the face spectrum image into the feature extraction model of the face spectrum image, and obtains the target feature vector of the face spatial characteristics and spectral reflection characteristics; uses the similarity evaluation Compared with the traditional face recognition method, the target eigenvector is compared with the eigenvector in the database, and the spectral characteristics of the skin are used to realize face recognition and living body recognition at the same time, which makes up for the security loopholes of traditional face detection and improves This improves the accuracy of the face recognition results and improves the security of the face recognition system.
  • the face recognition method provided in this application can be realized by a face recognition spectrum imaging chip, specifically, it can be realized by a signal processing circuit layer in the face recognition spectrum imaging chip.
  • the face recognition spectrum chip Through the face recognition spectrum chip, the face spectrum image to be recognized is obtained.
  • the face recognition spectrum imaging chip uses the face recognition spectrum imaging chip to shoot the face to be recognized, and the captured face spectrum image has hundreds of channels to obtain the face spectrum image to be recognized.
  • the face spectrum image includes face image space information and skin spectrum information; the face to be recognized may be a live face, a face image, a face in a video, or an object.
  • FIG. 2 is a schematic structural diagram of the face recognition spectral imaging chip provided by the present application.
  • the present application provides a face recognition spectral imaging chip, including a light modulation layer 2021, an image sensor layer 2022 and a signal processing circuit Layer 2023, light modulation layer 2021, image sensor layer 2022 and signal processing circuit layer 2023 are connected vertically from top to bottom, wherein:
  • the light modulation layer 2021 is used to receive the light signal reflected by the face 203 to be recognized and perform light modulation;
  • the image sensor layer 2022 is used to convert the optical signal reflected by the face to be recognized 203 after light modulation into an electrical signal, and the electrical signal includes the spatial information of the face image and the spectral information of the face after light modulation;
  • the signal processing circuit layer 2023 is configured to process the spatial information of the face image output by the image sensor layer 2022 and the spectral information of the face after light modulation, and obtain a face recognition result.
  • the face spectrum imaging chip also includes a lens group 201, wherein the lens group 201 is located on the upper surface of the light modulation layer 2021 and is connected to the light modulation layer 2021 for focusing the light signal reflected by the face 203 Imaging to obtain the light signal reflected by the face to be recognized.
  • the lens group lens in the face recognition spectrum imaging chip faces the face to be recognized.
  • a series of lens groups 201 are arranged on one side of the internal structure 202 of the chip, as shown in FIG.
  • the internal structure 202 of the face recognition spectral imaging chip the light reflected by the face passes through the lens group 201 to obtain the light signal reflected by the face after focusing and imaging, and the light signal reflected by the face after focusing and imaging is used as the light reflected by the face to be recognized signal;
  • the light modulation layer 2021 is provided with several light modulation units, each light modulation unit contains a plurality of micro-nano structure arrays, and performs light modulation on the received optical signal reflected by the face to be recognized after imaging; in the image sensor
  • the upper surface of the layer 2022 is provided with a plurality of photosensitive pixel units, and the surface of the photosensitive pixel unit area is directly prepared with a micro-nano structure array that has different modulation effects on light of different wavelengths, so that the image sensor layer 2022 can reflect the face to be recognized after light modulation.
  • the optical signal is converted into an electrical signal, and the electrical signal includes the spatial information of the human face image and the spectral information of the skin after light modulation; Data analysis and processing to determine whether the target to be recognized is a living human face, a simulated human face or a non-human face.
  • the light modulation layer 2021 is prepared directly on the image sensor layer 2022, for example, the light modulation layer is pasted, bonded, bonded, deposited on the image sensor layer 2022, the image sensor layer 2022 and the signal processing circuit layer 2023 are connected by electrical contact.
  • the types of micro-nanostructure arrays are different, and the modulation methods of different micro-nanostructure arrays are different.
  • the modulation methods include but are not limited to scattering, absorption, transmission, reflection, interference, excimer and resonance enhancement.
  • the micro-nano structure array includes but is not limited to one-dimensional photonic crystals, two-dimensional photonic crystals, surface plasmons, metamaterials, and metasurfaces.
  • Specific materials can include silicon, germanium, germanium silicon materials, silicon compounds, germanium compounds and III-V group materials, etc., and can also be metals, wherein silicon compounds include but not limited to silicon nitride, silicon dioxide and carbide silicon etc.
  • the light modulation layer 2021 grows one or more layers of materials directly on the image sensor layer 2022, and then prepares a micro-nano structure by etching, for example, etching after deposition; it can also be formed on the image sensor layer 2022 Micro-nanostructures were prepared by direct etching.
  • the image sensor layer may be a CIS wafer or a CCD image sensor.
  • the face recognition spectrum imaging chip is used to shoot the face to be recognized, and the face spectrum image with up to hundreds of channels is captured, and the information contained in the face spectrum image is much higher than the image captured by an ordinary RGB camera.
  • the spectral information of the face is obtained through the spectral imaging chip, and then the data is processed through the signal processing circuit layer, which can conveniently perform live detection to identify the disguised face.
  • the optical modulation layer and the image sensor are monolithically integrated, without discrete components, and no additional collimation components are required.
  • the preparation of the spectral imaging chip can be completed by one-time tape-out of the CMOS process, which is conducive to improving the stability of the device. It greatly promotes the miniaturization and weight reduction of imaging spectrometers, and reduces the cost of face recognition equipment.
  • the light modulation layer includes at least one light modulation unit, and the light modulation unit includes a plurality of micro-nano structure arrays, and each micro-nano structure array is provided with uniform The through holes are distributed and arranged, and the shapes of the through holes of each micro-nano structure array are different.
  • Figure 3 is a schematic structural diagram of the micro-nano structure array of the light modulation layer in the face recognition spectral imaging chip provided by the application.
  • the light modulation layer is engraved with several light modulation units, and each unit contains a variety of
  • the micro-nano structure on the light modulation layer can be a hole penetrating the plate, or a micro-nano structure with a certain depth.
  • the light modulation effect can be changed by changing the structural size parameter and/or structural shape of the micro-nano structural unit in the micro-nano structure array.
  • the unit geometry can include but not limited to circle, cross, regular polygon, rectangle and any combination thereof. This modulation can also be changed by changing the parameters of the micro-nano structure.
  • the change of the structural parameters can include but not limited to the period, radius, side length, duty cycle, thickness and other parameters of the micro-nano structure and any combination thereof.
  • the light modulation layer is made of silicon and silicon compounds with a thickness of 300 nm, and there are 1000 light modulation units in total, each light modulation unit has an overall size of 400 ⁇ m 2 , and each light modulation unit contains 25 micro-nano structure arrays.
  • Each micro-nano structure can be arranged according to different preset arrangement rules, and each micro-nano structure can be arranged periodically with the same shape, and the duty ratio is between 10% and 90%.
  • Each micro-nano structure array is any one of one-dimensional photonic crystals, two-dimensional photonic crystals, surface plasmons, metamaterials and metasurfaces.
  • a plurality of photosensitive pixel units are distributed on the upper surface of the image sensor layer, and each micro-nano structure array corresponds to at least one photosensitive pixel unit.
  • a plurality of photosensitive pixel units are arranged on the upper surface of the image sensor layer 2022, and the surface of the photosensitive pixel unit area is directly prepared with micro-nano structure arrays that have different modulation effects on light of different wavelengths, and each micro-nano structure array is connected with one or more Each photosensitive pixel unit corresponds to the vertical direction, and the light signal reflected by the face to be recognized after light modulation can be converted into an electrical signal through the image sensor layer 2022 .
  • the trained face image feature extraction model is obtained through the following steps of training:
  • the sample training set is constructed, and the authenticity information label is the spectral information label of the live human face, simulated human face and non-human face category ;
  • the sample training set is input into the machine learning model for training to obtain a trained face image feature extraction model, and the machine learning model is a convolutional neural network.
  • the sample face spectral images in the training sample set are tagged, and the tag contains two parts of information: one part is the identity information tag to which the face belongs, which is used for face recognition; the other part is the authenticity category spectral information tag , which is the spectral information of a living human face, a simulated human face or a non-human face, and is used for living body recognition.
  • the two parts of label information are used to train the same facial spectral image feature extraction model at the same time.
  • sample face spectrum image training set marked with the face spectrum image information label is input into the convolutional neural network for training, and the trained face spectrum image feature extraction model is obtained for extracting the face spectrum to be recognized The target feature vector for the image.
  • the simulated human face can be a 3D or image
  • the non-human face can be an animal or an object.
  • the category marked by the entire label can be ⁇ non-face, fake face, face 1, face 2, face 3... ⁇ , if there are n face feature information of different identities in the training sample set, then There should be a total of n+2 label categories, and the label information includes the identity information of the face and the spectral reflection characteristic information of the authenticity category.
  • the described sample training set is input into the machine learning model for training to obtain a trained facial spectral image feature extraction model, including:
  • the convolution kernel in the convolutional neural network is trained through the sample training set, and if the preset training conditions are met, a trained face spectrum image feature extraction model is obtained, wherein the volume The kernel is used to detect the contours of the corner points of the face and the spectral characteristics of the skin.
  • a large number of spectral reflectance characteristics of real faces and objects are collected as a sample training set, and the sample training set is input into the convolutional neural network to automatically train the convolution kernel to obtain the convolutional neural network.
  • the loss function value of the network if it is judged that the loss function value satisfies the training convergence condition, a trained facial spectral image feature extraction model is obtained.
  • the processes of face recognition and living body recognition can be completed independently.
  • the face space image to be recognized is input into the trained face image feature extraction model to obtain the target feature vector of the face space image, wherein the trained face image feature extraction model is made by marking the human face
  • the sample face space image of the information label is obtained by training the convolutional neural network.
  • the target feature vector includes the spatial characteristics of the face and is used to identify the identity of the face.
  • the target eigenvector is compared with the eigenvector in the database (which can only contain the spatial characteristics of the face) for recognition. If the similarity between the two reaches the set threshold, it is regarded as a successful face recognition, otherwise it is regarded as a human face. Face recognition failed.
  • the target feature vector is input into the trained classifier to obtain the living body recognition result.
  • the trained classifier is obtained by automatically training the deep neural network through the sample feature vector marked with the spectral information label of the true and false categories.
  • the classifier can also be obtained by training machine learning algorithms such as support vector machines (Support Vector Machines, SVM for short) and perceptrons.
  • support vector machines Small Vector Machines, SVM for short
  • perceptrons perceptrons
  • construct the face image database match the feature vector of the target feature vector and the face image database according to the similarity evaluation standard, and according to the matching result, obtain the spectral reflectance characteristic curve corresponding to the target feature vector; Judging by the spectral reflectance characteristic curve, if the characteristic peaks of the spectral reflectance characteristic curve are two minimum points at the 545nm and 575nm bands, then it is judged that the spectral image of the human face to be identified is a living human face image; or the spectral reflectance If the characteristic peak of the characteristic curve is the maximum point at the 850nm wavelength band, it is determined that the face spectrum image to be recognized is a live face image.
  • a convolution kernel with the same filter characteristic as "W" is designed to perform matching filtering on the human face and skin spectrum, and a judgment is made by setting a threshold to obtain the living body recognition result.
  • the skin spectrum of a human face is mainly reflected in the spectral characteristics of human skin.
  • the hemoglobin in the skin absorbs the light of 545nm and 575nm, so that the reflection curve of the skin is in the shape of "W" in the visible light band.
  • the reflectance of human skin reaches its maximum at around 850nm, then decreases rapidly with the increase of wavelength length, and then increases slightly at around 1450nm.
  • living body identification can be performed based on at least one characteristic peak of hemoglobin, such as the characteristic absorption peaks near the wavelengths of 545nm and 575nm; and/or in the near infrared band, it can also be performed through the extreme point at the 850nm band Liveness detection.
  • the original grayscale image output by the image sensor can also be used for face recognition, and the The original image is processed; the processed original grayscale image is then subjected to face recognition, and then, according to the face recognition spectrum imaging chip, the face spectrum in the image after face recognition is restored (some parts of the face spectrum Key points are restored), that is, to obtain the spectral information of the key points of the face, and finally through the calculation and analysis of the signal processing circuit layer in the face spectrum imaging chip to identify whether it is a live face.
  • three channels of RGB can be extracted from the multi-channel image captured by the spectral imaging chip, so the face recognition algorithm applicable to ordinary RGB cameras is also applicable to the spectral image of human face.
  • image analysis technology design a corner detector and a contour detector, abstract the face image into a feature vector in a high-dimensional space, and compare and identify it with the feature vector in the existing face image database according to the similarity evaluation standard .
  • the similarity evaluation standard can be the Euclidean distance between the feature vectors, the angle between the feature vectors in the high-dimensional space, or the algorithm used to measure the similarity between vectors in the high-dimensional space.
  • the face image of the successful preliminary face recognition is obtained, and then the spectral image information of the key points of the face in the face image is obtained by shooting with the spectral imaging chip.
  • the spectral image contains RGB information and has up to hundreds of channels
  • a 2D convolution kernel or a 3D convolution kernel can be constructed to extract the feature vector of the spectral image from the spectral image data cube.
  • the convolution kernel can be set in the signal processing circuit layer of the face recognition spectral imaging chip.
  • the detection of the face corner profile and skin spectral characteristics is constructed. For example, according to the "W" type feature of the skin spectrum between 500nm and 600nm, and then use the same "W" type convolution kernel to perform matching filtering on the face skin spectrum to obtain the filtered skin Spectral reflectance characteristic curve. Utilizing the spectral reflectance characteristics of human skin, by detecting the spectrum of the characteristic peak of the skin spectral reflectance characteristic curve, it is judged whether the photographed face object to be recognized is a living human face.
  • the face recognition algorithm and the living body detection algorithm are combined, and the spectral reflection characteristics of the face are obtained through the spectral imaging chip, which greatly enriches the face information and makes up for the shortcomings of traditional imaging technology. It has a strong anti-interference ability due to factors such as shooting angle, facial expression, hairstyle and makeup, and has a good recognition effect on camouflage, occlusion, masks and printed photos, which greatly improves the accuracy of face recognition results.
  • Fig. 4 is a schematic structural diagram of the face recognition system provided by the present application.
  • the present application provides a face recognition system, including a spectral image acquisition module 401, a feature vector extraction module 402 and a recognition module 403, wherein
  • the spectral image acquisition module 401 is used to obtain the spectral image of the human face to be identified;
  • the feature vector extraction module 402 is used to input the spectral image of the human face to be identified into the trained human facial spectral image feature extraction model, and obtains the spectral image of the human face to be identified
  • the target feature vector of the human face spectral image wherein the trained human face spectral image feature extraction model is a sample human face spectral image marked with the identity information label to which the face belongs and the authenticity category spectral information label, for machine learning
  • the model is obtained through training;
  • the identification module 403 is used to identify the target feature vector according to the similarity evaluation standard, and obtain the results of face recognition and living body recognition.
  • the lens group, light modulation layer and image sensor layer in the face recognition spectral imaging chip can be regarded as the spectral image acquisition module 401, which is used to obtain the face to be recognized through the lens group, light modulation layer and image sensor layer
  • the face spectrum image, the face spectrum image includes the face image spatial information and the face spectrum information
  • the feature vector extraction module 402 and the identification module 403 can be arranged at the signal processing circuit layer, for obtaining the target feature vector of the face spectrum image , and according to the similarity evaluation standard, match the target feature vector with the feature vector of the face image database, obtain the feature vector with the highest similarity in the face image database as the candidate recognition result, if the similarity meets the set threshold, then Obtain face recognition and living body recognition results.
  • a face recognition system obtaineds a face spectral image and inputs the face spectral image into a face spectral image feature extraction model to obtain the target feature vector of the face spatial characteristics and spectral reflection characteristics;
  • the target feature vector is compared with the feature vector in the database, and the spectral characteristics of the skin are used to realize face recognition and living body recognition at the same time, which makes up for the security loopholes of traditional face detection. , which improves the accuracy of face recognition results and improves the security of the face recognition system.
  • the system also includes building a sample training set module and a training feature extraction model module, wherein the building sample training set module is used to mark the identity information label to which the face belongs and the authenticity category spectral information label.
  • Sample human face spectral image, build sample training set, described true and false category spectral information label is the spectral information label of living human face, simulated human face and non-human face category;
  • Training feature extraction model module is used for described sample training set input into the machine learning model for training to obtain a trained face image feature extraction model, the machine learning model is a convolutional neural network.
  • the training feature extraction model module also includes a training feature extraction model unit, which is used to use the training set of samples to analyze the convolutional neural network based on a deep learning algorithm.
  • the convolution kernel is trained, and if the preset training conditions are met, a trained face image feature extraction model is obtained, wherein the convolution kernel is used to detect the contours of the corner points of the face and the spectral characteristics of the skin.
  • Fig. 5 is a schematic structural diagram of an electronic device provided by the present application.
  • the electronic device may include: a processor (processor) 501, a communication interface (Communications Interface) 502, a memory (memory) 503 and a communication bus 504 , where the processor 501 , the communication interface 502 , and the memory 503 communicate with each other through the communication bus 504 .
  • processor processor
  • Communication interface Communication Interface
  • memory memory
  • the processor 501 can call the logical instructions in the memory 503 to execute the face recognition method, the method comprising: acquiring a face spectrum image to be recognized; inputting the face spectrum image to be recognized into a trained face spectrum image In the image feature extraction model, the target feature vector of the face spectrum image is obtained, wherein the trained face spectrum image feature extraction model is a sample that is marked with the identity information label to which the face belongs and the authenticity category spectral information label
  • the human face spectrum image is obtained by training a machine learning model; the target feature vector is recognized according to the similarity evaluation standard, and the face recognition and living body recognition results are obtained.
  • the above logic instructions in the memory 503 may be implemented in the form of software function units and when sold or used as an independent product, may be stored in a computer-readable storage medium.
  • the technical solution of the present application is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. .
  • 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, and when the program instructions are executed by a computer
  • the computer can execute the face recognition method provided by the above-mentioned methods, and the method includes: obtaining a face spectrum image to be recognized; inputting the face spectrum image to be recognized into the trained face spectrum image feature
  • the target feature vector of the human face spectral image is obtained, wherein the trained human face spectral image feature extraction model is a sample human face marked with the identity information label to which the face belongs and the authenticity category spectral information label
  • the spectral image is obtained by training the machine learning model; according to the similarity evaluation standard, the target feature vector is recognized, and the face recognition and living body recognition results are obtained.
  • the present application also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it is implemented to perform the face recognition methods provided above, the method comprising: Obtain the face spectrum image to be recognized; input the face spectrum image to be recognized into the trained face spectrum image feature extraction model, and obtain the target feature vector of the face spectrum image, wherein the training A good face spectral image feature extraction model is obtained by training a machine learning model with a sample face spectral image marked with the identity information label of the face and the authenticity category spectral information label; according to the similarity evaluation standard, the The target feature vector is used for recognition, and the results of face recognition and living body recognition are obtained.
  • the device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in One place, or it can be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. It can be understood and implemented by those skilled in the art without any creative efforts.
  • each implementation can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware.
  • the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic discs, optical discs, etc., including several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) execute the methods described in various embodiments or some parts of the embodiments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Image Analysis (AREA)

Abstract

本申请提供一种人脸识别方法及系统，该人脸识别方法包括：获取待识别的人脸光谱图像；将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别和活体识别结果。本申请弥补了传统人脸检测的安全漏洞，提高了人脸识别结果的准确性和人脸识别系统的安全性。

Description

一种人脸识别方法及系统
相关申请的交叉引用
本申请要求于2021年07月19日提交的申请号为2021108139296，名称为“一种人脸识别方法及系统”的中国专利申请的优先权，其通过引用方式全部并入本文。
技术领域
本申请涉及光谱成像技术领域，尤其涉及一种人脸识别方法及系统。
背景技术
人脸识别是基于人的脸部特征信息进行身份识别的一种生物识别技术，通过采集含有人脸的图像或视频进行分析，自动推断其身份、表情、年龄和性别等属性。作为生物特征识别技术的一种，人脸识别具有非强制性、非接触性、操作简单及隐蔽性好等特点，因此在安防、管理监督和多媒体娱乐等领域应用广泛。
目前大多数人脸识别技术都是基于灰度图像或者彩色RGB图像，可以获得的图像信息有限，光照条件及拍摄视角等的变化对识别结果有着直接的影响，人脸表情、发型、化妆和眼镜等因素的改变都会可能导致识别准确度的下降，更有不法分子通过伪装、遮挡、面具以及打印照片的方式对识别结果造成严重干扰。基于传统成像系统的人脸识别技术，由于仅仅利用了观测对象的空间几何特征，对于各种条件变化引起的不确定性非常敏感，系统鲁棒性较差，在复杂环境下识别性能急剧下降，即使引入3D人脸信息，也无法解决硅胶面具的伪装问题，无法实现活体人脸的识别。
因此，现在亟需一种人脸识别方法及系统来解决上述问题。
发明内容
针对现有技术存在的问题，本申请提供一种人脸识别方法及系统。
本申请提供一种人脸识别方法，包括：
获取待识别的人脸光谱图像；
将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提 取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；
根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别和活体识别结果。
本申请提供一种人脸识别方法，所述获取待识别的人脸光谱图像，包括：
通过人脸识别光谱成像芯片，获取待识别的人脸光谱图像；
所述人脸识别光谱成像芯片包括光调制层、图像传感器层和信号处理电路层，所述光调制层、所述图像传感器层和所述信号处理电路层沿垂直方向从上至下依次连接，其中：
所述光调制层，用于接收待识别人脸反射的光信号，并进行光调制；
所述图像传感器层，用于将光调制后的待识别人脸反射的光信号转换为电信号，所述电信号包括人脸图像空间信息和皮肤光谱信息；
所述信号处理电路层，用于对所述图像传感器层输出的人脸图像空间信息和皮肤光谱信息进行处理，获取人脸识别结果。
根据本申请提供的一种人脸识别方法，所述人脸光谱成像芯片还包括透镜组，其中，所述透镜组，位于所述光调制层的上表面，与所述光调制层连接，用于对人脸反射的光信号进行聚焦成像，得到待识别人脸反射的光信号。
根据本申请提供的一种人脸识别方法，所述光调制层包括至少一个光调制单元，所述光调制单元包括多个微纳结构阵列，每个微纳结构阵列中按照不同预设排列规则，设置有均匀分布排列的通孔，且每个微纳结构阵列的通孔形状不同。
根据本申请提供的一种人脸识别方法，所述图像传感器层的上表面分布有多个感光像素单元，每个微纳结构阵列对应至少一个感光像素单元。
根据本申请提供的一种人脸识别方法，所述训练好的人脸光谱图像特征提取模型通过以下步骤训练得到：
根据标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，构建样本训练集，所述真伪类别光谱信息标签为活体人脸、 仿真人脸和非人脸类别的光谱信息标签；
将所述样本训练集输入到所述机器学习模型中进行训练，得到训练好的人脸图像特征提取模型，所述机器学习模型为卷积神经网络。
根据本申请提供的一种人脸识别方法，所述将所述样本训练集输入到所述机器学习模型中进行训练，得到训练好的人脸光谱图像特征提取模型，包括：
基于深度学习算法，通过所述样本训练集对所述卷积神经网络中的卷积核进行训练，若满足预设训练条件，得到训练好的人脸光谱图像特征提取模型，其中，所述卷积核用于检测人脸角点轮廓和皮肤光谱特性。
本申请还提供了一种人脸识别系统，包括：
获取光谱图像模块，获取待识别的人脸光谱图像；
提取特征向量模块，用于将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；
识别模块，用于根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别结果。
根据本申请提供的一种人脸识别系统，所述系统还包括：
构建样本训练集模块，用于根据标记有人脸信息标签和真伪类别光谱信息标签的样本人脸光谱图像，构建样本训练集，所述人脸信息标签为人脸所属身份信息标签，所述真伪类别光谱信息标签为活体人脸、仿真人脸和非人脸类别的光谱信息标签；
训练特征提取模型模块，用于将所述样本训练集输入到所述机器学习模型中进行训练，得到训练好的人脸图像特征提取模型，所述机器学习模型为卷积神经网络。
本申请还提供一种非暂态计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现如上述任一种所述人脸识别方法的步骤。
本申请提供的一种人脸识别方法及系统，通过获取人脸光谱图像，将 人脸光谱图像输入到人脸光谱图像特征提取模型中，得到人脸空间特性和光谱反射特性的目标特征向量；利用相似度评价标准，将目标特征向量与数据库中的特征向量进行对比识别，与传统人脸识别方法相比，利用皮肤光谱特性，同时实现人脸识别和活体识别，弥补了传统人脸检测的安全漏洞，提高了人脸识别结果的准确性，提升了人脸识别系统的安全性。
附图说明
为了更清楚地说明本申请或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作一简单地介绍，显而易见地，下面描述中的附图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为本申请提供的人脸识别方法的流程示意图；
图2为本申请提供的人脸识别光谱成像芯片的结构示意图；
图3为本申请提供的人脸识别光谱成像芯片中光调制层的微纳结构阵列的结构示意图；
图4为本申请提供的人脸识别系统的结构示意图；
图5为本申请提供的一种电子设备的结构示意图。
具体实施方式
为使本申请的目的、技术方案和优点更加清楚，下面将结合本申请中的附图，对本申请中的技术方案进行清楚、完整地描述，显然，所描述的实施例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。
普通的RGB相机只能拍摄得到RGB三个通道的图像，基于普通RGB相机的活体检测算法使用的仅有RGB三个通道的图像信息，一般需要通过视频序列分析等较为复杂的算法来进行识别，实时性差且可靠性差，对高仿真的人脸识别率也较低。
图1为本申请提供的人脸识别方法的流程示意图，如图1所示，本申请提供了一种人脸识别方法，包括：
步骤101，获取待识别的人脸光谱图像；
步骤102，将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；
步骤103，根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别和活体识别结果。
在本申请中，步骤101中，待识别的人脸光谱图像可以通过计算光谱装置获取，也可以通过高光谱装置获取。
进一步地，步骤102中，人脸所属身份信息标签为包含不同身份ID的人脸空间特性的标签。根据标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，同时对机器学习模型进行训练，得到人脸光谱图像特征提取模型。将待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，人脸光谱图像特征提取模型为机器学习模型，通过对机器学习模型进行训练，将人脸光谱图像转化为高维空间中的目标特征向量，得到的目标特征向量包括人脸空间特性和光谱反射特性。需要说明的是，目标特征向量可以是活体人脸的特征向量，可以是仿真人脸的特征向量，也可以是非人脸的特征向量。
需要说明的是，人脸光谱图像特征提取模型可以是卷积神经网络，也可以是任意的机器学习模型，机器学习模型还可以是支持向量机(Support Vector Machine，简称SVM)，也可以是感知机。本申请是以卷积神经网络的训练过程进行说明。
进一步地，步骤103中，相似度评价标准可以是特征向量之间的欧式距离，可以是特征向量在高维空间中的夹角，也可以是用于衡量高维空间中向量间相似度的算法。选定已有人脸图像数据库中与待识别目标特征向量相似度最高的样本特征向量作为候选识别结果，该数据库中的样本特征向量包含人脸空间特性以及光谱反射特性。若相似度达到设定的阈值，则视为识别成功，否则视为识别失败。通过上述相似度评价标准，对目标特征向量进行对比识别，从而得到人脸识别和活体识别结果。
本申请提供的人脸识别方法，通过获取人脸光谱图像，将人脸光谱图 像输入到人脸光谱图像特征提取模型中，得到人脸空间特性和光谱反射特性的目标特征向量；利用相似度评价标准，将目标特征向量与数据库中的特征向量进行对比识别，与传统人脸识别方法相比，利用皮肤光谱特性，同时实现人脸识别和活体识别，弥补了传统人脸检测的安全漏洞，提高了人脸识别结果的准确性，提升了人脸识别系统的安全性。
优选地，本申请提供的人脸识别方法可通过人脸识别光谱成像芯片实现，具体可通过人脸识别光谱成像芯片中的信号处理电路层实现。通过人脸识别光谱芯片，获取待识别的人脸光谱图像。利用人脸识别光谱成像芯片对待识别人脸进行拍摄，拍摄得到的人脸光谱图像高达数百个通道，获取待识别的人脸光谱图像。其中，人脸光谱图像包括人脸图像空间信息和皮肤光谱信息；待识别的人脸可以是活体人脸，可以是人脸图像，也可以是视频中的人脸，也可以是物体。
图2为本申请提供的人脸识别光谱成像芯片的结构示意图，如图2所示，本申请提供了一种人脸识别光谱成像芯片，包括光调制层2021、图像传感器层2022和信号处理电路层2023，光调制层2021、图像传感器层2022和信号处理电路层2023沿垂直方向从上至下依次连接，其中：
光调制层2021，用于接收待识别人脸203反射的光信号，并进行光调制；
图像传感器层2022，用于将光调制后的待识别人脸203反射的光信号转换为电信号，电信号包括人脸图像空间信息和光调制后的人脸光谱信息；
信号处理电路层2023，用于对图像传感器层2022输出的人脸图像空间信息和光调制后的人脸光谱信息进行处理，获取人脸识别结果。
可选地，人脸光谱成像芯片还包括透镜组201，其中，透镜组201位于所述光调制层2021的上表面，与光调制层2021连接，用于对人脸203反射的光信号进行聚焦成像，得到待识别人脸反射的光信号。
在本申请中，采用人脸识别光谱成像芯片对待识别人脸203进行拍摄时，人脸识别光谱成像芯片中的透镜组镜头朝向待识别人脸。在人脸识别光谱成像芯片中，一系列的透镜组201设置在芯片内部结构202的一侧，可参考图2所示，光调制层2021、图像传感器层2022和信号处理电路层2023组成了人脸识别光谱成像芯片的内部结构202，人脸反射的光线经过 透镜组201得到聚焦成像后的人脸反射的光信号，将聚焦成像后的人脸反射的光信号作为待识别人脸反射的光信号；光调制层2021上设置有若干个光调制单元，每个光调制单元包含多个微纳结构阵列，对接收到的成像后的待识别人脸反射的光信号进行光调制；在图像传感器层2022上表面设置多个感光像素单元，感光像素单元区域表面直接制备对不同波长的光有不同调制作用的微纳结构阵列，从而通过图像传感器层2022可以将光调制后的待识别人脸反射的光信号转换为电信号，电信号包括人脸图像空间信息和光调制后的皮肤光谱信息；信号处理电路层2023对图像传感器层2022输出的人脸图像空间信息和光调制后的人脸光谱信息进行数据分析处理，判断待识别目标是活体人脸、仿真人脸还是非人脸。
进一步地，光调制层2021是直接在图像传感器层2022上制备的，例如光调制层为贴附、粘接、键合、沉积于所述图像传感器层2022，图像传感器层2022和信号处理电路层2023之间通过电接触进行连接。
可选地，微纳结构阵列的结构阵列种类不同，且不同微纳结构阵列的调制方式不同，调制方式包括但不限于散射、吸收、透射、反射、干涉、激元和谐振增强。
可选地，微纳结构阵列包括但不限于一维光子晶体、二维光子晶体、表面等离子激元、超材料和超表面等。具体材料可包括硅、锗、锗硅材料、硅的化合物、锗的化合物和III-V族材料等，也可以为金属，其中，硅的化合物包括但不限于氮化硅、二氧化硅和碳化硅等。
可选地，光调制层2021是在图像传感器层2022上直接生长一层或多层材料，再通过刻蚀制备出微纳结构，例如沉积后，再刻蚀；也可以在图像传感器层2022上直接刻蚀制备出微纳结构。
可选地，图像传感器层可以是CIS晶圆，也可以是CCD图像传感器。
进一步地，采用人脸识别光谱成像芯片对待识别人脸进行拍摄，拍摄得到高达数百个通道的人脸光谱图像，人脸光谱图像所含的信息远高于普通RGB相机拍摄得到的图像。除了识别人脸所属的身份外，通过光谱成像芯片获取到人脸光谱信息，然后再经过信号处理电路层进行数据处理，可以方便地进行活体检测以识别伪装的人脸。
在本申请中，将光调制层与图像传感器单片集成，无分立元件，不需 要外加准直元件，可以通过CMOS工艺一次流片完成对光谱成像芯片的制备，有利于提高器件的稳定性，极大促进成像光谱仪的小型化和轻量化，降低人脸识别设备的成本。
在上述实施例的基础上，所述光调制层包括至少一个光调制单元，所述光调制单元包括多个微纳结构阵列，每个微纳结构阵列中按照不同预设排列规则，设置有均匀分布排列的通孔，且每个微纳结构阵列的通孔形状不同。
图3为本申请提供的人脸识别光谱成像芯片中光调制层的微纳结构阵列的结构示意图，如图3所示，光调制层上刻有若干个光调制单元，每个单元包含多种微纳结构阵列，光调制层上的微纳结构可以为贯穿平板的孔，也可为具有一定深度的微纳结构。可以通过改变微纳结构阵列中微纳结构单元的结构尺寸参数和/或结构形状来改变光调制作用，单元几何形状可包括但不限于圆、十字、正多边形、矩形以及它们的任意组合。还可通过改变微纳结构的参数来改变这种调制作用，结构参数的改变可包括但不限于微纳结构周期、半径、边长、占空比、厚度等参数以及它们的任意组合。
可选地，光调制层为厚度300nm的硅及硅的化合物，共有1000个光调制单元，每个光调制单元整体的尺寸为400μm 2，每个光调制单元包含25个微纳结构阵列。每个微纳结构可以按照不同的预设排列规则排列，每个微纳结构内可以是同一形状的周期排布，占空比为10％～90％之间。每个微纳结构阵列为一维光子晶体、二维光子晶体、表面等离子激元、超材料和超表面等中的任意一种。
在上述实施例的基础上，所述图像传感器层的上表面分布有多个感光像素单元，每个微纳结构阵列对应至少一个感光像素单元。
在本申请中，图像传感器层2022上表面设置多个感光像素单元，感光像素单元区域表面直接制备对不同波长的光有不同调制作用的微纳结构阵列，每个微纳结构阵列与一个或多个感光像素单元在垂直方向上相对应，通过图像传感器层2022可以将光调制后的待识别人脸反射的光信号转换为电信号。
在上述实施例的基础上，所述训练好的人脸图像特征提取模型通过以下步骤训练得到：
根据标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，构建样本训练集，所述真伪信息标签为活体人脸、仿真人脸和非人脸类别的光谱信息标签；
将所述样本训练集输入到所述机器学习模型中进行训练，得到训练好的人脸图像特征提取模型，所述机器学习模型为卷积神经网络。
在本申请中，对训练样本集中的样本人脸光谱图像标注标签，标注的标签包含两部分信息：一部分是人脸所属的身份信息标签，用于人脸识别；一部分是真伪类别光谱信息标签，即为活体人脸、仿真人脸或者非人脸的光谱信息，用于活体识别。两部分标签信息同时用于训练同一个人脸光谱图像特征提取模型。
进一步地，将标记有人脸光谱图像信息标签的样本人脸光谱图像训练集输入卷积神经网络中进行训练，获取训练好的人脸光谱图像特征提取模型，以用于提取待识别的人脸光谱图像的目标特征向量。
可选地，仿真人脸可以是3D或者图像，非人脸可以是动物或者物体。
可选地，整个标签标注的类别可以是{非人脸、伪装人脸、人脸1、人脸2、人脸3…}，若训练样本集中共有n个不同身份的人脸特征信息，则标注的标签类别应共有n+2个，标签标注的信息包含人脸的身份信息与真伪类别的光谱反射特性信息。
在上述实施例的基础上，所述将所述样本训练集输入到所述机器学习模型中进行训练，得到训练好的人脸光谱图像特征提取模型，包括：
基于深度学习算法，通过所述样本训练集对所述卷积神经网络中的卷积核进行训练，若满足预设训练条件，得到训练好的人脸光谱图像特征提取模型，其中，所述卷积核用于检测人脸角点轮廓和皮肤光谱特性。
在本申请中，基于深度学习算法，通过大量采集真实人脸和物体的光谱反射特性作为样本训练集，将样本训练集输入卷积神经网络中对卷积核进行自动训练，获取该卷积神经网络的损失函数值，若判断获知损失函数值满足训练收敛条件，得到训练好的人脸光谱图像特征提取模型。
在一实施例中，可以将人脸识别和活体识别过程分别独立完成。具体地，将待识别的人脸空间图像输入到训练好的人脸图像特征提取模型中，获取人脸空间图像的目标特征向量，其中，训练好的人脸图像特征提取模 型是由标记有人脸信息标签的样本人脸空间图像，对卷积神经网络进行训练得到的。该目标特征向量包括人脸空间特性，用于识别人脸所属身份。根据相似度评价标准，将目标特征向量与数据库中的特征向量(可以只包含人脸空间特性)进行对比识别，若两者相似度达到设定的阈值，则视为人脸识别成功，否则视为人脸识别失败。
进一步地，将目标特征向量输入到训练好的分类器中，得到活体识别结果。其中，训练好的分类器是通过标记有真伪类别光谱信息标签的样本特征向量，对深度神经网络进行自动训练得到的。
可选地，分类器还可以通过支持向量机(Support Vector Machines,简称SVM)、感知机等机器学习算法训练得到。
可以理解的是，活体与非活体的光谱反射特性存在着明显的差异，由此可以通过训练分类器来寻找符合活体和非活体对应的光谱反射特性的差异。
在一实施例中，构建人脸图像数据库，根据相似度评价标准，对目标特征向量和人脸图像数据库的特征向量进行匹配，并根据匹配结果，获取目标特征向量对应的光谱反射特性曲线；对光谱反射特性曲线进行判断，若光谱反射特性曲线的特征峰为545nm、575nm波段处的两个极小值点，则判断获知所述待识别的人脸光谱图像为活体人脸图像；或者光谱反射特性曲线的特征峰为850nm波段处的极大值点，则判断获知待识别的人脸光谱图像为活体人脸图像。
可选地，利用人脸皮肤光谱反射特性，设计滤波特性同样为“W型”的卷积核对人脸皮肤光谱进行匹配滤波，通过设定阈值进行判决，获取活体识别结果。
需要说明的是，人脸的皮肤光谱主要体现为人皮肤的光谱特性。皮肤中血红蛋白对545nm和575nm的光有吸收作用，使得皮肤的反射曲线在可见光波段内呈“W”形。人皮肤反射率在850nm左右达到最大值，然后随着波长长度的增加迅速变小，在1450nm左右又会小幅度增大。因此在可见光波段内，可以依据至少一个血红蛋白的特征峰，如545nm和575nm波长附近的特征吸收峰来进行活体识别；和/或在近红外波段内，也可通过850nm波段处的极值点进行活体检测。
在一实施例中，由于光谱成像芯片采用的是计算成像的方式，而不是直接成像，因此也可以利用图像传感器输出的原始灰度图像进行人脸识别，通过均衡、降噪等预处理方式对原始图像进行处理；再对处理后的原始灰度图像进行人脸识别，然后，根据人脸识别光谱成像芯片对人脸识别后的图像中的人脸光谱进行恢复(对人脸光谱中的一些关键点进行恢复)，即获取人脸关键点光谱信息，最后通过人脸光谱成像芯片中信号处理电路层的运算分析来辨别是否为活体人脸。
在另一实施例中，从光谱成像芯片拍摄得到的多通道图像可提取出RGB三个通道，因此适用于普通RGB相机的人脸识别算法同样适用于人脸光谱图像。利用图像分析技术，设计角点检测器以及轮廓检测器，将人脸图像抽象为高维空间中的特征向量，并根据相似度评价标准与已有的人脸图像数据库中的特征向量进行对比识别。其中，相似度评价标准可以是特征向量之间的欧式距离，可以是特征向量在高维空间中的夹角，也可以是用于衡量高维空间中向量间相似度的算法。选定已有人脸图像数据库中与待识别目标特征向量相似度最高的样本特征向量作为候选识别结果。若相似度达到设定的阈值，则视为初步的人脸识别成功，否则视为初步的人脸识别失败。
进一步地，根据上述相似度评价标准得到初步人脸识别成功的人脸图像，再经过光谱成像芯片拍摄得到人脸图像中人脸关键点的光谱图像信息。由于光谱图像包含RGB信息，并且存在高达上百个通道，因此可构建2D卷积核或者3D卷积核，从光谱图像数据立方中提取光谱图像的特征向量。可选地，该卷积核可设置在人脸识别光谱成像芯片的信号处理电路层中。根据人脸光谱图像在空间维度上的角点轮廓等信息，以及在光谱维度上的血红蛋白吸收特性为已有的先验知识，通过匹配滤波的原理，构建检测人脸角点轮廓和皮肤光谱特性的卷积核，例如，根据皮肤光谱在500nm至600nm之间的“W”型特征，然后利用滤波特性同样为“W”型的卷积核对人脸皮肤光谱进行匹配滤波，获取滤波后的皮肤光谱反射特性曲线。利用人皮肤的光谱反射特性，通过检测皮肤光谱反射特性曲线在特征峰处的光谱来判断拍摄的待识别人脸目标是否为活体人脸。
在本申请中，将人脸识别算法和活体检测算法结合在一起，通过光谱 成像芯片拍摄得到人脸的光谱反射特性，极大丰富了人脸信息，弥补了传统成像技术的缺点，对光照条件、拍摄视角、人脸表情、发型和化妆等因素的影响具有较强抗干扰能力，对伪装、遮挡、面具和打印照片等方式具有良好识别作用，大大提高了人脸识别结果的准确性。
图4为本申请提供的人脸识别系统的结构示意图，如图4所示，本申请提供了一种人脸识别系统，包括获取光谱图像模块401、提取特征向量模块402和识别模块403，其中，获取光谱图像模块401用于获取待识别的人脸光谱图像；提取特征向量模块402用于将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；识别模块403用于根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别和活体识别结果。
可选地，人脸识别光谱成像芯片中的透镜组、光调制层和图像传感器层可以视为获取光谱图像模块401，通过透镜组、光调制层和图像传感器层，用于获取待识别人脸的人脸光谱图像，人脸光谱图像包括人脸图像空间信息和人脸光谱信息；提取特征向量模块402和识别模块403可以设置于信号处理电路层，用于获取人脸光谱图像的目标特征向量，并根据相似度评价标准，对目标特征向量和人脸图像数据库的特征向量进行匹配，获取人脸图像数据库中满足相似度最高的特征向量作为候选识别结果，若相似度满足设定阈值，则得到人脸识别和活体识别结果。
本申请提供的一种人脸识别系统，通过获取人脸光谱图像，将人脸光谱图像输入到人脸光谱图像特征提取模型中，得到人脸空间特性和光谱反射特性的目标特征向量；利用相似度评价标准，将目标特征向量与数据库中的特征向量进行对比识别，与传统人脸识别方法相比，利用皮肤光谱特性，同时实现人脸识别和活体识别，弥补了传统人脸检测的安全漏洞，提高了人脸识别结果的准确性，提升了人脸识别系统的安全性。
在上述实施例的基础上，所述系统还包括构建样本训练集模块和训练特征提取模型模块，其中，构建样本训练集模块用于根据标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，构建样本训 练集，所述真伪类别光谱信息标签为活体人脸、仿真人脸和非人脸类别的光谱信息标签；训练特征提取模型模块用于将所述样本训练集输入到所述机器学习模型中进行训练，得到训练好的人脸图像特征提取模型，所述机器学习模型为卷积神经网络。
在上述实施例的基础上，所述训练特征提取模型模块还包括训练特征提取模型单元，训练特征提取模型单元用于基于深度学习算法，通过所述样本训练集对所述卷积神经网络中的卷积核进行训练，若满足预设训练条件，得到训练好的人脸图像特征提取模型，其中，所述卷积核用于检测人脸角点轮廓和皮肤光谱特性。
本申请提供的系统是用于执行上述各方法实施例的，具体流程和详细内容请参照上述实施例，此处不再赘述。
图5为本申请提供的一种电子设备的结构示意图，如图5所示，该电子设备可以包括：处理器(processor)501、通信接口(Communications Interface)502、存储器(memory)503和通信总线504，其中，处理器501，通信接口502，存储器503通过通信总线504完成相互间的通信。处理器501可以调用存储器503中的逻辑指令，以执行人脸识别方法，该方法包括：获取待识别的人脸光谱图像；将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别和活体识别结果。
此外，上述的存储器503中的逻辑指令可以通过软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM， Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
另一方面，本申请还提供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，计算机能够执行上述各方法所提供的人脸识别方法，该方法包括：获取待识别的人脸光谱图像；将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别和活体识别结果。
又一方面，本申请还提供一种非暂态计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现以执行上述各提供的人脸识别方法，该方法包括：获取待识别的人脸光谱图像；将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别和活体识别结果。
以上所描述的装置实施例仅仅是示意性的，其中所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。本领域普通技术人员在不付出创造性的劳动的情况下，即可以理解并实施。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到各实施方式可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件。基于这样的理解，上述技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品可以存储在计算机可读存储介质中，如ROM/RAM、磁碟、光盘等，包括若干指令用 以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行各个实施例或者实施例的某些部分所述的方法。
最后应说明的是：以上实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的范围。

Claims (10)

  1. 一种人脸识别方法，包括：
    获取待识别的人脸光谱图像；
    将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；
    根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别和活体识别结果。
  2. 根据权利要求1所述的人脸识别方法，其中，所述获取待识别的人脸光谱图像，包括：
    通过人脸识别光谱成像芯片，获取待识别的人脸光谱图像；
    所述人脸识别光谱成像芯片包括光调制层、图像传感器层和信号处理电路层，所述光调制层、所述图像传感器层和所述信号处理电路层沿垂直方向从上至下依次连接，其中：
    所述光调制层，用于接收待识别人脸反射的光信号，并进行光调制；
    所述图像传感器层，用于将光调制后的待识别人脸反射的光信号转换为电信号，所述电信号包括人脸图像空间信息和光调制后的皮肤光谱信息；
    所述信号处理电路层，用于对所述图像传感器层输出的人脸图像空间信息和光调制后的皮肤光谱信息进行处理，获取人脸识别结果。
  3. 根据权利要求2所述的人脸识别方法，其中，所述人脸光谱成像芯片还包括透镜组，其中，所述透镜组，位于所述光调制层的上表面，与所述光调制层连接，用于对人脸反射的光信号进行聚焦成像，得到待识别人脸反射的光信号。
  4. 根据权利要求2所述的人脸识别方法，其中，所述光调制层包括至少一个光调制单元，所述光调制单元包括多个微纳结构阵列，每个微纳结构阵列中按照不同预设排列规则，设置有均匀分布的通孔，且每个微纳结构阵列的通孔形状不同。
  5. 根据权利要求2所述的人脸识别方法，其中，所述图像传感器层的上表面分布有多个感光像素单元，每个微纳结构阵列对应至少一个感光像 素单元。
  6. 根据权利要求1所述的人脸识别方法，其中，所述训练好的人脸光谱图像特征提取模型通过以下步骤训练得到：
    根据标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，构建样本训练集，所述真伪类别光谱信息标签为活体人脸、仿真人脸和非人脸类别的光谱信息标签；
    将所述样本训练集输入到所述机器学习模型中进行训练，得到训练好的人脸图像特征提取模型，所述机器学习模型为卷积神经网络。
  7. 根据权利要求6所述的人脸识别方法，其中，所述将所述样本训练集输入到所述机器学习模型中进行训练，得到训练好的人脸光谱图像特征提取模型，包括：
    基于深度学习算法，通过所述样本训练集对所述卷积神经网络中的卷积核进行训练，若满足预设训练条件，得到训练好的人脸光谱图像特征提取模型，其中，所述卷积核用于检测人脸角点轮廓和皮肤光谱特性。
  8. 一种人脸识别系统，包括：
    获取光谱图像模块，获取待识别的人脸光谱图像；
    提取特征向量模块，用于将所述待识别的人脸光谱图像输入到训练好的人脸光谱图像特征提取模型中，获取所述人脸光谱图像的目标特征向量，其中，所述训练好的人脸光谱图像特征提取模型是由标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，对机器学习模型进行训练得到的；
    识别模块，用于根据相似度评价标准，对所述目标特征向量进行识别，获取人脸识别和活体识别结果。
  9. 根据权利要求8所述的人脸识别系统，所述系统还包括：
    构建样本训练集模块，用于根据标记有人脸所属身份信息标签和真伪类别光谱信息标签的样本人脸光谱图像，构建样本训练集，所述真伪类别光谱信息标签为活体人脸、仿真人脸和非人脸类别的光谱信息标签；
    训练特征提取模型模块，用于将所述样本训练集输入到所述机器学习模型中进行训练，得到训练好的人脸图像特征提取模型，所述机器学习模型为卷积神经网络。
  10. 一种非暂态计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现如权利要求1至7任一项所述人脸识别方法的步骤。
PCT/CN2022/098308 2021-07-19 2022-06-13 一种人脸识别方法及系统 WO2023000864A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110813929.6A CN115641649A (zh) 2021-07-19 2021-07-19 一种人脸识别方法及系统
CN202110813929.6 2021-07-19

Publications (1)

Publication Number Publication Date
WO2023000864A1 true WO2023000864A1 (zh) 2023-01-26

Family

ID=84940099

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/098308 WO2023000864A1 (zh) 2021-07-19 2022-06-13 一种人脸识别方法及系统

Country Status (2)

Country Link
CN (1) CN115641649A (zh)
WO (1) WO2023000864A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116311553A (zh) * 2023-05-17 2023-06-23 武汉利楚商务服务有限公司 应用于半遮挡图像下的人脸活体检测方法及装置
CN116797889A (zh) * 2023-08-24 2023-09-22 青岛美迪康数字工程有限公司 医学影像识别模型的更新方法、装置和计算机设备
CN117273747A (zh) * 2023-09-28 2023-12-22 广州佳新智能科技有限公司 基于人脸图像识别的支付方法、装置、存储介质和设备
CN117688538A (zh) * 2023-12-13 2024-03-12 上海深感数字科技有限公司 一种基于数字身份安全防范的互动教育管理方法及系统
CN117854138A (zh) * 2024-03-07 2024-04-09 深圳航天信息有限公司 基于大数据的信息采集分析方法、装置、设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105956515A (zh) * 2016-04-20 2016-09-21 西安电子科技大学 基于极成像的立体-高光谱人脸识别方法
US20180285668A1 (en) * 2015-10-30 2018-10-04 Microsoft Technology Licensing, Llc Spoofed face detection
US20190095701A1 (en) * 2017-09-27 2019-03-28 Lenovo (Beijing) Co., Ltd. Living-body detection method, device and storage medium
CN110502989A (zh) * 2019-07-16 2019-11-26 山东师范大学 一种小样本高光谱人脸识别方法及系统
CN112330331A (zh) * 2020-11-19 2021-02-05 平安普惠企业管理有限公司 基于人脸识别的身份验证方法、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
US20180285668A1 (en) * 2015-10-30 2018-10-04 Microsoft Technology Licensing, Llc Spoofed face detection
CN105956515A (zh) * 2016-04-20 2016-09-21 西安电子科技大学 基于极成像的立体-高光谱人脸识别方法
US20190095701A1 (en) * 2017-09-27 2019-03-28 Lenovo (Beijing) Co., Ltd. Living-body detection method, device and storage medium
CN110502989A (zh) * 2019-07-16 2019-11-26 山东师范大学 一种小样本高光谱人脸识别方法及系统
CN112330331A (zh) * 2020-11-19 2021-02-05 平安普惠企业管理有限公司 基于人脸识别的身份验证方法、装置、设备及存储介质

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116311553A (zh) * 2023-05-17 2023-06-23 武汉利楚商务服务有限公司 应用于半遮挡图像下的人脸活体检测方法及装置
CN116311553B (zh) * 2023-05-17 2023-08-15 武汉利楚商务服务有限公司 应用于半遮挡图像下的人脸活体检测方法及装置
CN116797889A (zh) * 2023-08-24 2023-09-22 青岛美迪康数字工程有限公司 医学影像识别模型的更新方法、装置和计算机设备
CN116797889B (zh) * 2023-08-24 2023-12-08 青岛美迪康数字工程有限公司 医学影像识别模型的更新方法、装置和计算机设备
CN117273747A (zh) * 2023-09-28 2023-12-22 广州佳新智能科技有限公司 基于人脸图像识别的支付方法、装置、存储介质和设备
CN117273747B (zh) * 2023-09-28 2024-04-19 广州佳新智能科技有限公司 基于人脸图像识别的支付方法、装置、存储介质和设备
CN117688538A (zh) * 2023-12-13 2024-03-12 上海深感数字科技有限公司 一种基于数字身份安全防范的互动教育管理方法及系统
CN117688538B (zh) * 2023-12-13 2024-06-07 上海深感数字科技有限公司 一种基于数字身份安全防范的互动教育管理方法及系统
CN117854138A (zh) * 2024-03-07 2024-04-09 深圳航天信息有限公司 基于大数据的信息采集分析方法、装置、设备及存储介质
CN117854138B (zh) * 2024-03-07 2024-05-10 深圳航天信息有限公司 基于大数据的信息采集分析方法、装置、设备及存储介质

Also Published As

Publication number Publication date
CN115641649A (zh) 2023-01-24

Similar Documents

Publication Publication Date Title
WO2023000864A1 (zh) 一种人脸识别方法及系统
US11288504B2 (en) Iris liveness detection for mobile devices
Bhattacharjee et al. Spoofing deep face recognition with custom silicone masks
Liu et al. Transferring deep representation for NIR-VIS heterogeneous face recognition
Omara et al. A novel geometric feature extraction method for ear recognition
Sankaran et al. On smartphone camera based fingerphoto authentication
KR100852629B1 (ko) 다초점 영상 시퀀스를 이용한 홍채인식 시스템 및 방법
Engelsma et al. Raspireader: Open source fingerprint reader
Wang et al. Review of ear biometrics
CN113614731A (zh) 使用软生物特征的认证验证
Du Biometrics: from fiction to practice
Maček et al. Multimodal biometric authentication in IoT: Single camera case study
Jayavadivel et al. Investigation on automated surveillance monitoring for human identification and recognition using face and iris biometric
Anthony et al. A review of face anti-spoofing methods for face recognition systems
Wasnik et al. Presentation attack detection for smartphone based fingerphoto recognition using second order local structures
Sharma et al. A survey on face presentation attack detection mechanisms: hitherto and future perspectives
Song et al. Face liveness detection based on joint analysis of rgb and near-infrared image of faces
Akhloufi et al. State of the art in infrared face recognition
Song et al. Face liveliness detection based on texture and color features
Proença Unconstrained iris recognition in visible wavelengths
Das et al. Face liveness detection based on frequency and micro-texture analysis
Galdi et al. Combining hardwaremetry and biometry for human authentication via smartphones
Kisku et al. Multithread face recognition in cloud
Majeed et al. A novel method to enhance color spatial feature extraction using evolutionary time-frequency decomposition for presentation-attack detection
Li et al. Iris recognition on mobile devices using near-infrared images

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE