WO2022105015A1 - Face recognition method and system, terminal, and storage medium - Google Patents

Face recognition method and system, terminal, and storage medium Download PDF

Info

Publication number
WO2022105015A1
WO2022105015A1 PCT/CN2020/139671 CN2020139671W WO2022105015A1 WO 2022105015 A1 WO2022105015 A1 WO 2022105015A1 CN 2020139671 W CN2020139671 W CN 2020139671W WO 2022105015 A1 WO2022105015 A1 WO 2022105015A1
Authority
WO
WIPO (PCT)
Prior art keywords
layer
face recognition
neural network
input
face
Prior art date
Application number
PCT/CN2020/139671
Other languages
French (fr)
Chinese (zh)
Inventor
钱静
彭树宏
Original Assignee
中国科学院深圳先进技术研究院
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 中国科学院深圳先进技术研究院 filed Critical 中国科学院深圳先进技术研究院
Publication of WO2022105015A1 publication Critical patent/WO2022105015A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • 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
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/30—Noise filtering

Definitions

  • the present application belongs to the technical field of face recognition, and in particular relates to a face recognition method, system, terminal and storage medium.
  • face recognition has entered a period of rapid development at home and abroad.
  • the rapid development of face recognition is mainly due to its ability to rapidly drive the progress of related disciplines. Since face recognition is a very complex and multi-faceted combination of technologies, it usually involves the most classic image pattern processing, computer vision, computer Graphics, scientific understanding, physiology, psychology, AI, mathematical logic calculation and other disciplines are combined and intersected to form a new field, which can be applied in real life and combined with AI, can lead society to a higher level. in living conditions.
  • face recognition technology has considerable application potential.
  • face recognition has a wide range of applications, such as mobile phone face unlocking, door lock face recognition, public security using face recognition to solve crimes, and "swiping your face” for meals, etc. It also brings great convenience to people's lives. Therefore, the research on face recognition has very practical significance.
  • the main face recognition methods include geometric feature face recognition method, discriminant analysis method (Fisher), template matching method, eigenface method (Eige naface), independent principal component analysis (LCA), Hidden Marker method. Kov method (HMM), support vector machine method (SVM), singular value decomposition method (SVD), elastic graph matching method and neural network method, etc., but due to the distortion of human expressions, emotions, emotions and griefs in the process of face recognition Due to the influence of factors such as changes, the above methods have many restrictive conditions and cannot perform face recognition well.
  • the present application provides a face recognition method, system, terminal and storage medium, aiming to solve one of the above technical problems in the prior art at least to a certain extent.
  • a face recognition method comprising:
  • the feature matrix is input into the trained BP neural network, and the face recognition result is output through the BP neural network;
  • the BP neural network includes an input layer, a hidden layer and an output layer, and the input data is from all the input layers.
  • the neuron of the output layer is entered, the calculation is performed in the hidden layer, and the calculation result is input into each neuron of the output layer for calculation, and the face recognition result is obtained.
  • the technical solutions adopted in the embodiments of the present application further include: performing face capture and taking screenshots on the dynamic video, and obtaining static face pictures, further including:
  • the static face picture is preprocessed by grayscale and median filtering.
  • the preprocessing of the static face picture by using grayscale and median filtering includes:
  • Grayscale processing is performed on the static face picture, and the static face picture is converted into a grayscale matrix
  • the technical solutions adopted in the embodiments of the present application further include: the BP neural network adopts the tansig function as the transfer function between the input layer and the hidden layer, adopts the purelin linear function as the transfer function between the hidden layer and the output layer, and adopts The Sigmoid function is used as the activation function from the input layer to the hidden layer, and the Purelin linear function is used as the activation function from the hidden layer to the output layer.
  • the number of nodes in the output layer is the number of categories of faces
  • the number of nodes in the hidden layer is:
  • n represents the size of the input neuron
  • m represents the size of the output neuron
  • a represents a constant within 10.
  • the vectors of each layer in the BP neural network are respectively:
  • O t represents the t-th output layer, where t is a natural number
  • y j represents the j-th hidden layer, where j is a natural number
  • w jt is the t-th output layer O to the j-th hidden layer y
  • the weights of i, j represent the i-th neuron of the output layer and the j-th neuron of the hidden layer, respectively;
  • x i represents the ith input layer, where i is a natural number.
  • v ij is the weight from the ith input layer x to the jth hidden layer y, i represents the ith neuron in the input layer, and j represents the jth neuron in the hidden layer.
  • the inputting the feature matrix into the trained BP neural network further includes:
  • an error range is set.
  • the BP neural network When training the BP neural network, an error range is set.
  • the BP neural network When the error of the BP neural network does not reach the error range, the BP neural network returns the output result in the output layer through the back propagation algorithm. Perform cyclic calculation on the hidden layer and the input layer, and correct the weights of each layer during the cyclic calculation, so that the network error gradually decreases until the error of the BP neural network reaches the error range.
  • a face recognition method comprising:
  • Face picture acquisition module used to capture and take screenshots of dynamic videos, and obtain static face pictures
  • Face feature extraction module used to extract feature values from the static face picture to generate a feature matrix
  • Face recognition module used to input the feature matrix into the trained BP neural network, and output the face recognition result through the BP neural network;
  • the BP neural network includes an input layer, a hidden layer and an output layer, and the input data Enter from all neurons of the input layer, perform calculation in the hidden layer, input the calculation result to each neuron of the output layer for calculation, and obtain a face recognition result.
  • a terminal includes a processor and a memory coupled to the processor, wherein,
  • the memory stores program instructions for implementing the face recognition method
  • the processor is configured to execute the program instructions stored in the memory to control face recognition.
  • a storage medium storing program instructions executable by a processor, where the program instructions are used to execute the face recognition method.
  • the beneficial effects of the embodiments of the present application are: the face recognition method of the embodiments of the present application performs grayscale and latitude reduction processing on the face image, and then extracts the face features from the face image to generate Feature matrix, and use BP neural network to use back propagation algorithm for face recognition, which can greatly improve the efficiency and accuracy of face recognition.
  • FIG. 1 is a flowchart of a face recognition method according to an embodiment of the present application.
  • FIG. 2 is a schematic diagram of a grayscale matrix after grayscale conversion is performed on a face picture according to an embodiment of the present application
  • FIG. 3 is a schematic diagram of a small matrix after cutting a grayscale matrix according to an embodiment of the present application
  • FIG. 4 is a schematic diagram of a code for building a BP neural network according to an embodiment of the application.
  • Fig. 5 is the schematic diagram of the BP neural network variable of the embodiment of the application.
  • FIG. 6 is a schematic structural diagram of a BP neural network according to an embodiment of the application.
  • FIG. 7 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 8 according to an embodiment of the present application;
  • FIG. 8 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 16 according to an embodiment of the present application;
  • FIG. 9 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 24 according to an embodiment of the present application.
  • FIG. 10 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 32 according to an embodiment of the present application;
  • 11 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 48 according to an embodiment of the present application;
  • FIG. 12 is a schematic structural diagram of a face recognition system according to an embodiment of the present application.
  • FIG. 13 is a schematic structural diagram of a terminal according to an embodiment of the present application.
  • FIG. 14 is a schematic structural diagram of a storage medium according to an embodiment of the present application.
  • first”, “second” and “third” in the present invention are only used for description purposes, and should not be understood as indicating or implying relative importance or implying the number of indicated technical features. Thus, a feature defined as “first”, “second”, “third” may expressly or implicitly include at least one of that feature.
  • "a plurality of” means at least two, such as two, three, etc., unless otherwise expressly and specifically defined. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship between various components under a certain posture (as shown in the accompanying drawings).
  • FIG. 1 is a flowchart of a face recognition method according to an embodiment of the present application.
  • the face recognition method of the embodiment of the present application includes the following steps:
  • S10 Capture and take screenshots of the faces in the dynamic video, and obtain a certain number of static face pictures with different expressions and looks;
  • S20 Preprocess all static face images by grayscale, median filtering and other methods
  • the purpose of preprocessing is to reduce the interference and noise of the static face image, and to enhance the contrast of the face target and the image background.
  • the preprocessing process specifically includes:
  • S21 first perform grayscale processing on the static face image, and convert the static face image into a grayscale matrix
  • the embodiment of the present invention adopts the grayscale technology to preprocess the static face picture, converts the static face picture into a grayscale matrix, and directly converts the input static face picture into a grayscale matrix by using the imread function.
  • the matrix is a process of converting a picture into an abstract.
  • the processing principle is to convert the picture into a digital information, and use the value of the matrix to represent the eigenvalue of each pixel.
  • the converted grayscale matrix is shown in Figure 2.
  • the latitude reduction process is to reduce the latitude of a picture with a high latitude, and process and extract some unique features of a low latitude as a feature mark of the picture to represent the picture.
  • the purpose of latitude reduction is to reduce the complexity of the data, which is conducive to more obvious representation of picture information.
  • this scheme adopts svd (Singular Value Decompositionm, singular value decomposition) algorithm to extract eigenvalues, and the formula is as follows:
  • A is an X*Y matrix
  • U be an N*N square matrix
  • a and U are both orthogonal vectors, and the left side of U is a left singular vector
  • ⁇ is an X*Y matrix, the diagonal
  • the arity on is 0, and its name is called singular value
  • V T is an N*N matrix
  • an orthogonal vector is also called a right singular vector; the formula is as follows:
  • the column vector (left singular vector) of U is the eigenvector of AAT; at the same time, the column vector (right singular vector) of V is the eigenvector of A T A; the singular value of M ( the non-zero diagonal elements of ⁇ ) is AA T or the square root of the non-zero eigenvalues of A T A.
  • the embodiment of the present application can reduce the value of the data stream by performing feature value extraction on the static face picture, which brings convenience to subsequent work and improves the processing speed.
  • S50 Input the training set into the BP neural network (Back Propagation, forward propagation neural network) for classifier training, and output the face recognition result;
  • BP neural network Back Propagation, forward propagation neural network
  • the BP neural network In this step, after the BP neural network is constructed, input the corresponding number of neurons and the number of iterative training times, and then connect the transfer functions of each layer to train the classifier.
  • the construction code of the BP neural network is shown in Figure 4. After running the code, variables such as the number of neurons, the two transfer functions between layers, the training function, the training target, the number of training iterations and the learning rate need to be manually input. After construction, you can see the net variable in the variable window, as shown in Figure 5.
  • the structure of the BP neural network in the embodiment of the present application is shown in FIG. 6 , which includes an input layer, a hidden layer ranging from one to N layers, and an output layer.
  • the BP neural network algorithm is a back-propagation algorithm, and its calculation process is as follows: the tansig function is used as the transfer function between the input layer and the hidden layer, the purelin linear function is used as the transfer function between the hidden layer and the output layer, and the Sigmoid function is used as the input.
  • the activation function from the layer to the hidden layer adopts the Purelin linear function as the activation function from the hidden layer to the output layer; first, the neurons in each layer are initialized, and the input samples first enter from all the neurons in the input layer, and then in the hidden layer.
  • the processing is performed layer by layer in the containing layer, and finally the data is sent to each neuron in the output layer for calculation, and the learning result is output after the neuron is calculated.
  • the number of nodes in the output layer is the number of categories of faces, and the number of nodes in the hidden layer is:
  • n represents the size of the input neuron
  • m represents the size of the output neuron
  • a represents a constant within 10.
  • the vectors of each layer in the BP neural network are different.
  • the vector settings of each layer in the embodiment of this application are: the input layer is x, the hidden layer is Y, the output layer is O, and an expected vector value is set. is d, and for the output layer, there are:
  • the output layer is O
  • O t represents the t-th output layer, where t is a natural number.
  • the hidden layer is Y
  • yj represents the jth hidden layer, where j is a natural number.
  • w jt is the weight from the t-th output layer O to the j-th hidden layer y
  • i, j represent the ith neuron of the output layer and the jth neuron of the hidden layer, respectively.
  • the input layer is x
  • x i represents the ith input layer, where i is a natural number.
  • v ij is the weight from the ith input layer x to the jth hidden layer y
  • i represents the ith neuron in the input layer
  • j represents the jth neuron in the hidden layer.
  • an error range is also set.
  • the BP neural network when training the BP neural network, an error range is also set.
  • the BP neural network returns the output result in the output layer through a back propagation process.
  • the hidden layer and the input layer are cyclically calculated, and the weights of each layer are corrected in the cyclical calculation process, so that the network error gradually decreases until the output result reaches the preset error range.
  • the network error is expressed by E:
  • E is the square of the difference between the output layer O and the actual value d. According to the least squares method, the smaller the value of E, the closer the output layer O is to the true value d . If the E value does not reach the set error range, adjust the i, j values of each layer in the BP neural network to reduce the size of the E value, until the E value reaches the set error range, the optimal network parameters can be obtained.
  • the recognition rate and total recognition rate of the BP neural network in the training set and the test set are counted, and the total recognition rate of the BP neural network is calculated under the condition that the feature dimensions are 8, 16, 32, and 48 respectively.
  • the recognition rate is counted to verify the face recognition accuracy of the BP neural network under different feature dimensions, as follows:
  • FIG 7 it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 8. Under the condition that the feature dimension is 8, the neurons are added in multiples, and the training times are gradually increased to test the total recognition rate. The test results show that the recognition rate of face recognition is basically under the condition of low feature dimension. Below 80%, the overall recognition rate is low, so a lower feature dimension is not used.
  • FIG 8 it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 16. Under the condition that the feature dimension is 16, when the number of neurons is less than 120, it can be found that the total recognition rate can exceed 80%, but when there are too many neurons, the overall recognition rate will decrease, indicating that the neuron in face recognition The more the number of neurons, the better. Too many neurons will reduce the recognition rate, and the training time of the neural network will be too long.
  • FIG. 9 it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 24.
  • the recognition rate decreases with the increase of neurons, and the recognition rate can be as high as 91.5% when the number of training times is 4000 among 120 neurons.
  • FIG. 10 it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 32. Under the condition that the feature dimension is 32, when the number of neurons is 120, the overall recognition rate of the network exceeds 90%, and the highest recognition rate can reach 93.5% when the number of training is 6000 times.
  • FIG 11 it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 48.
  • the recognition rate can reach more than 80%, which is lower than the recognition rate when the feature dimension is 32.
  • the recognition rate will be lost. A lot of time, so don't take too high feature dimensionality.
  • the embodiment of the present application preferably adopts the feature dimension of 32, the number of neurons to be 120, and the highest recognition rate (up to 93.5%) when the number of training times is 6000. .
  • Face recognition after performing grayscale and latitude reduction processing on the face image, face feature extraction is performed on the face image, a feature matrix is generated, and a back propagation algorithm is used by using a BP neural network. Face recognition can greatly improve the efficiency and accuracy of face recognition.
  • FIG. 12 is a schematic structural diagram of a face recognition system according to an embodiment of the present application.
  • the face recognition system 40 of the embodiment of the present application includes:
  • Face picture acquisition module 41 used to capture and take screenshots of dynamic videos, and obtain static face pictures
  • Face feature extraction module 42 used to extract feature values from the static face picture to generate a feature matrix
  • Face recognition module 43 for inputting the feature matrix into the trained BP neural network, and outputting the face recognition result through the BP neural network; wherein, the BP neural network includes an input layer, a hidden layer and an output layer , the input data is entered from all the neurons in the input layer, the calculation is performed in the hidden layer, and the calculation result is input into each neuron of the output layer for calculation to obtain a face recognition result.
  • the BP neural network includes an input layer, a hidden layer and an output layer , the input data is entered from all the neurons in the input layer, the calculation is performed in the hidden layer, and the calculation result is input into each neuron of the output layer for calculation to obtain a face recognition result.
  • FIG. 13 is a schematic structural diagram of a terminal according to an embodiment of the present application.
  • the terminal 50 includes a processor 51 and a memory 52 coupled to the processor 51 .
  • the memory 52 stores program instructions for implementing the above-mentioned face recognition method.
  • the processor 51 is configured to execute program instructions stored in the memory 52 to control face recognition.
  • the processor 51 may also be referred to as a CPU (Central Processing Unit, central processing unit).
  • the processor 51 may be an integrated circuit chip with signal processing capability.
  • the processor 51 may also be a general purpose processor, digital signal processor (DSP), application specific integrated circuit (ASIC), off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component .
  • DSP digital signal processor
  • ASIC application specific integrated circuit
  • FPGA off-the-shelf programmable gate array
  • a general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • FIG. 14 is a schematic structural diagram of a storage medium according to an embodiment of the present application.
  • the storage medium of this embodiment of the present application stores a program file 61 capable of implementing all the above methods, wherein the program file 61 may be stored in the above-mentioned storage medium in the form of a software product, and includes several instructions to make a computer device (which may It is a personal computer, a server, or a network device, etc.) or a processor that executes all or part of the steps of the methods of the various embodiments of the present invention.
  • a computer device which may It is a personal computer, a server, or a network device, etc.
  • a processor that executes all or part of the steps of the methods of the various embodiments of the present invention.
  • 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 disk and other media that can store program codes , or terminal devices such as computers, servers, mobile phones, and tablets.

Landscapes

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

Abstract

The present application relates to a face recognition method and system, a terminal, and a storage medium. The method comprises: performing face capture on a dynamic video and performing screenshot capture to obtain static face images; extracting feature values from the static face images and generating a feature matrix; inputting the feature matrix into a trained backpropagation (BP) neural network, and outputting a face recognition result by means of the BP neural network, the BP neural network comprising an input layer, a hidden layer, and an output layer, input data entering from all neurons of the input layer, calculation being performed in the hidden layer, and the calculation result being input into each neuron of the output layer for calculation, to obtain a face recognition result. According to the present application, by extracting face features from face images, generating a feature matrix, and performing face recognition by a BP neural network by means of a BP algorithm, the face recognition efficiency and recognition precision can be greatly improved.

Description

一种人脸识别方法、***、终端以及存储介质A face recognition method, system, terminal and storage medium 技术领域technical field
本申请属于人脸识别技术领域，特别涉及一种人脸识别方法、***、终端以及存储介质。The present application belongs to the technical field of face recognition, and in particular relates to a face recognition method, system, terminal and storage medium.
背景技术Background technique
早在上世纪60年代，经历了半个世纪之久的发展，到如今，人脸识别已进入国内外的高速发展期。人脸识别能够快速发展主要在于其能够快速的带动相关学科的进步，由于人脸识别是一个非常复杂且涉及多方面技术的结合体，通常会涉及到最经典的图像模式处理、计算机视觉、计算机图形学、科学方面的认识、生理学、心理学、AI、数学逻辑计算等多种学科的结合交叉，搭成一个全新的领域，应用于实际生活中，结合AI，可以引领社会进入更高层次的生活条件中。同时，人脸识别技术具有相当大的应用潜力，目前，人脸识别的应用领域非常广泛，例如手机人脸解锁、门锁人脸识别、公安运用人脸识别破案以及吃饭“刷脸”等，给人们的生活也带来了极大的便利。因此，人脸识别的研究具有非常实用的意义。As early as the 1960s, after half a century of development, face recognition has entered a period of rapid development at home and abroad. The rapid development of face recognition is mainly due to its ability to rapidly drive the progress of related disciplines. Since face recognition is a very complex and multi-faceted combination of technologies, it usually involves the most classic image pattern processing, computer vision, computer Graphics, scientific understanding, physiology, psychology, AI, mathematical logic calculation and other disciplines are combined and intersected to form a new field, which can be applied in real life and combined with AI, can lead society to a higher level. in living conditions. At the same time, face recognition technology has considerable application potential. At present, face recognition has a wide range of applications, such as mobile phone face unlocking, door lock face recognition, public security using face recognition to solve crimes, and "swiping your face" for meals, etc. It also brings great convenience to people's lives. Therefore, the research on face recognition has very practical significance.
现有技术中，主要的人脸识别方法包括几何特征的人脸识别方法、判别分析法(Fisher)、模板匹配法、特征脸法(Eige naface)、独立主元分析法(LCA)、隐马尔可夫方法(HMM)、支持向量机法(SVM)、奇异值分解法(SVD)、弹性图匹配方法及神经网络方法等，但由于人脸识别过程中会受到人的表情扭曲、喜怒哀乐变化等因素的影响，上述方法都存在较多的限制性条件，不能很好的进行人脸识别。In the prior art, the main face recognition methods include geometric feature face recognition method, discriminant analysis method (Fisher), template matching method, eigenface method (Eige naface), independent principal component analysis (LCA), Hidden Marker method. Kov method (HMM), support vector machine method (SVM), singular value decomposition method (SVD), elastic graph matching method and neural network method, etc., but due to the distortion of human expressions, emotions, emotions and sorrows in the process of face recognition Due to the influence of factors such as changes, the above methods have many restrictive conditions and cannot perform face recognition well.
发明内容SUMMARY OF THE INVENTION
本申请提供了一种人脸识别方法、***、终端以及存储介质，旨在至少在一定程度上解决现有技术中的上述技术问题之一。The present application provides a face recognition method, system, terminal and storage medium, aiming to solve one of the above technical problems in the prior art at least to a certain extent.
为了解决上述问题，本申请提供了如下技术方案：In order to solve the above problems, the application provides the following technical solutions:
一种人脸识别方法，包括：A face recognition method, comprising:
对动态视频进行人脸捕捉并截图，获取静态人脸图片；Capture and take screenshots of dynamic videos to obtain static face pictures;
对所述静态人脸图片进行特征值提取，生成特征矩阵；Extracting feature values from the static face picture to generate a feature matrix;
将所述特征矩阵输入训练好的BP神经网络，通过所述BP神经网络输出人脸识别结果；所述BP神经网络包括输入层、隐含层以及输出层，输入数据从所述输入层的所有的神经元进入，在所述隐含层中进行计算，将计算结果输入所述输出层的每个神经元进行计算，得到人脸识别结果。The feature matrix is input into the trained BP neural network, and the face recognition result is output through the BP neural network; the BP neural network includes an input layer, a hidden layer and an output layer, and the input data is from all the input layers. The neuron of the output layer is entered, the calculation is performed in the hidden layer, and the calculation result is input into each neuron of the output layer for calculation, and the face recognition result is obtained.
本申请实施例采取的技术方案还包括：所述对动态视频进行人脸捕捉并截图，获取静态人脸图片还包括：The technical solutions adopted in the embodiments of the present application further include: performing face capture and taking screenshots on the dynamic video, and obtaining static face pictures, further including:
采用灰度化及中值滤波对所述静态人脸图片进行预处理。The static face picture is preprocessed by grayscale and median filtering.
本申请实施例采取的技术方案还包括：所述采用灰度化及中值滤波对所述静态人脸图片进行预处理包括：The technical solutions adopted in the embodiments of the present application further include: the preprocessing of the static face picture by using grayscale and median filtering includes:
对所述静态人脸图片进行灰度处理，将所述静态人脸图片转换成灰度矩阵；Grayscale processing is performed on the static face picture, and the static face picture is converted into a grayscale matrix;
将所述灰度矩阵切割为预定数量的小矩阵；cutting the grayscale matrix into a predetermined number of small matrices;
对所述切割后的小矩阵进行降纬处理。Weft reduction processing is performed on the cut small matrix.
本申请实施例采取的技术方案还包括：所述BP神经网络采用tansig函数作为所述输入层与隐含层的传递函数，采用purelin线性函数作为所述隐含层与输出层 的传递函数，采用Sigmoid函数作为所述输入层到隐含层的激活函数，采用Purelin线性函数作为所述隐含层到输出层的激活函数。The technical solutions adopted in the embodiments of the present application further include: the BP neural network adopts the tansig function as the transfer function between the input layer and the hidden layer, adopts the purelin linear function as the transfer function between the hidden layer and the output layer, and adopts The Sigmoid function is used as the activation function from the input layer to the hidden layer, and the Purelin linear function is used as the activation function from the hidden layer to the output layer.
本申请实施例采取的技术方案还包括：所述输出层节点数为人脸的类别数，所述隐含层节点数为：The technical solutions adopted in the embodiments of the present application further include: the number of nodes in the output layer is the number of categories of faces, and the number of nodes in the hidden layer is:
Figure PCTCN2020139671-appb-000001
Figure PCTCN2020139671-appb-000001
上式中，n代表输入神经元的大小，m代表输出神经元的大小，a代表10以内的常数。In the above formula, n represents the size of the input neuron, m represents the size of the output neuron, and a represents a constant within 10.
本申请实施例采取的技术方案还包括：所述BP神经网路中每一层的向量分别为：The technical solutions adopted in the embodiments of the present application further include: the vectors of each layer in the BP neural network are respectively:
设置输入层为x，隐含层为Y，输出层为O，以及一个预期向量值设为d，对于输出层，有：Set the input layer to x, the hidden layer to Y, the output layer to O, and an expected vector value to d. For the output layer, we have:
Figure PCTCN2020139671-appb-000002
Figure PCTCN2020139671-appb-000002
上式中，O t表示第t个输出层，其中t为自然数；y j表示第j个隐含层，其中j为自然数；w jt为第t个输出层O到第j个隐含层y的权重，i，j分别表示输出层的第i个神经元和隐含层的第j个神经元； In the above formula, O t represents the t-th output layer, where t is a natural number; y j represents the j-th hidden layer, where j is a natural number; w jt is the t-th output layer O to the j-th hidden layer y The weights of i, j represent the i-th neuron of the output layer and the j-th neuron of the hidden layer, respectively;
对于隐含层，有：For the hidden layer, there are:
Figure PCTCN2020139671-appb-000003
Figure PCTCN2020139671-appb-000003
上式中，x i表示第i个输入层，其中i为自然数。v ij为第i个输入层x到第j个隐含层y的权重，i表示输入层的第i个神经元，j表示隐含层的第j个神经元。 In the above formula, x i represents the ith input layer, where i is a natural number. v ij is the weight from the ith input layer x to the jth hidden layer y, i represents the ith neuron in the input layer, and j represents the jth neuron in the hidden layer.
本申请实施例采取的技术方案还包括：所述将所述特征矩阵输入训练好的BP神经网络还包括：The technical solutions adopted in the embodiments of the present application further include: the inputting the feature matrix into the trained BP neural network further includes:
在训练所述BP神经网络时，设置一个误差范围，当所述BP神经网络的误差没有达到所述误差范围时，所述BP神经网络通过逆向传播算法将所述输出结果在输 出层当中逆向返回给所述隐含层和输入层进行循环计算，并在循环计算过程中修正各层的权值，使得网络误差逐渐下降，直到所述BP神经网络的误差达到所述误差范围。When training the BP neural network, an error range is set. When the error of the BP neural network does not reach the error range, the BP neural network returns the output result in the output layer through the back propagation algorithm. Perform cyclic calculation on the hidden layer and the input layer, and correct the weights of each layer during the cyclic calculation, so that the network error gradually decreases until the error of the BP neural network reaches the error range.
本申请实施例采取的另一技术方案为：一种人脸识别方法，包括：Another technical solution adopted in the embodiment of the present application is: a face recognition method, comprising:
人脸图片获取模块：用于对动态视频进行人脸捕捉并截图，获取静态人脸图片；Face picture acquisition module: used to capture and take screenshots of dynamic videos, and obtain static face pictures;
人脸特征提取模块：用于对所述静态人脸图片进行特征值提取，生成特征矩阵；Face feature extraction module: used to extract feature values from the static face picture to generate a feature matrix;
人脸识别模块：用于将所述特征矩阵输入训练好的BP神经网络，通过所述BP神经网络输出人脸识别结果；所述BP神经网络包括输入层、隐含层以及输出层，输入数据从所述输入层的所有的神经元进入，在所述隐含层中进行计算，将计算结果输入所述输出层的每个神经元进行计算，得到人脸识别结果。Face recognition module: used to input the feature matrix into the trained BP neural network, and output the face recognition result through the BP neural network; the BP neural network includes an input layer, a hidden layer and an output layer, and the input data Enter from all neurons of the input layer, perform calculation in the hidden layer, input the calculation result to each neuron of the output layer for calculation, and obtain a face recognition result.
本申请实施例采取的又一技术方案为：一种终端，所述终端包括处理器、与所述处理器耦接的存储器，其中，Another technical solution adopted by the embodiments of the present application is: a terminal, the terminal includes a processor and a memory coupled to the processor, wherein,
所述存储器存储有用于实现所述人脸识别方法的程序指令；The memory stores program instructions for implementing the face recognition method;
所述处理器用于执行所述存储器存储的所述程序指令以控制人脸识别。The processor is configured to execute the program instructions stored in the memory to control face recognition.
本申请实施例采取的又一技术方案为：一种存储介质，存储有处理器可运行的程序指令，所述程序指令用于执行所述人脸识别方法。Another technical solution adopted by the embodiments of the present application is: a storage medium storing program instructions executable by a processor, where the program instructions are used to execute the face recognition method.
相对于现有技术，本申请实施例产生的有益效果在于：本申请实施例的人脸识别方法通过对人脸图像进行灰度以及降纬处理后，对人脸图片进行人脸特征提取，生成特征矩阵，并利用BP神经网络采用逆向传播算法进行人脸识别，能极大的提高人脸识别效率以及识别精度。Compared with the prior art, the beneficial effects of the embodiments of the present application are: the face recognition method of the embodiments of the present application performs grayscale and latitude reduction processing on the face image, and then extracts the face features from the face image to generate Feature matrix, and use BP neural network to use back propagation algorithm for face recognition, which can greatly improve the efficiency and accuracy of face recognition.
附图说明Description of drawings
图1是本申请实施例的人脸识别方法的流程图；1 is a flowchart of a face recognition method according to an embodiment of the present application;
图2为本申请实施例对人脸图片进行灰度转换后的灰度矩阵示意图；2 is a schematic diagram of a grayscale matrix after grayscale conversion is performed on a face picture according to an embodiment of the present application;
图3为本申请实施例对灰度矩阵进行切割后的小矩阵示意图；3 is a schematic diagram of a small matrix after cutting a grayscale matrix according to an embodiment of the present application;
图4为本申请实施例的BP神经网络搭建代码示意图；4 is a schematic diagram of a code for building a BP neural network according to an embodiment of the application;
图5为本申请实施例的BP神经网络变量示意图；Fig. 5 is the schematic diagram of the BP neural network variable of the embodiment of the application;
图6为本申请实施例的BP神经网络结构示意图；6 is a schematic structural diagram of a BP neural network according to an embodiment of the application;
图7为本申请实施例在特征维数为8时BP神经网络的识别率示意图；7 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 8 according to an embodiment of the present application;
图8为本申请实施例在特征维数为16时BP神经网络的识别率示意图；8 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 16 according to an embodiment of the present application;
图9为本申请实施例在特征维数为24时BP神经网络的识别率示意图；9 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 24 according to an embodiment of the present application;
图10为本申请实施例在特征维数为32时BP神经网络的识别率示意图；10 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 32 according to an embodiment of the present application;
图11为本申请实施例在特征维数为48时BP神经网络的识别率示意图；11 is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 48 according to an embodiment of the present application;
图12是本申请实施例的人脸识别***的结构示意图；12 is a schematic structural diagram of a face recognition system according to an embodiment of the present application;
图13为本申请实施例的终端结构示意图；FIG. 13 is a schematic structural diagram of a terminal according to an embodiment of the present application;
图14为本申请实施例的存储介质的结构示意图。FIG. 14 is a schematic structural diagram of a storage medium according to an embodiment of the present application.
具体实施方式Detailed ways
下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅是本发明的一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.
本发明中的术语“第一”、“第二”、“第三”仅用于描述目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第 一”、“第二”、“第三”的特征可以明示或者隐含地包括至少一个该特征。本发明的描述中，“多个”的含义是至少两个，例如两个，三个等，除非另有明确具体的限定。本发明实施例中所有方向性指示(诸如上、下、左、右、前、后……)仅用于解释在某一特定姿态(如附图所示)下各部件之间的相对位置关系、运动情况等，如果该特定姿态发生改变时，则该方向性指示也相应地随之改变。此外，术语“包括”和“具有”以及它们任何变形，意图在于覆盖不排他的包含。例如包含了一系列步骤或单元的过程、方法、***、产品或设备没有限定于已列出的步骤或单元，而是可选地还包括没有列出的步骤或单元，或可选地还包括对于这些过程、方法、产品或设备固有的其它步骤或单元。The terms "first", "second" and "third" in the present invention are only used for description purposes, and should not be understood as indicating or implying relative importance or implying the number of indicated technical features. Thus, a feature defined as "first", "second", "third" may expressly or implicitly include at least one of that feature. In the description of the present invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise expressly and specifically defined. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship between various components under a certain posture (as shown in the accompanying drawings). , motion situation, etc., if the specific posture changes, the directional indication also changes accordingly. Furthermore, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but optionally also includes unlisted steps or units, or optionally also includes For other steps or units inherent to these processes, methods, products or devices.
在本文中提及“实施例”意味着，结合实施例描述的特定特征、结构或特性可以包含在本发明的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例，也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是，本文所描述的实施例可以与其它实施例相结合。Reference herein to an "embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present invention. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment, nor a separate or alternative embodiment that is mutually exclusive of other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.
请参阅图1，是本申请实施例的人脸识别方法的流程图。本申请实施例的人脸识别方法包括以下步骤：Please refer to FIG. 1 , which is a flowchart of a face recognition method according to an embodiment of the present application. The face recognition method of the embodiment of the present application includes the following steps:
S10：对动态视频中的人脸进行捕捉并截图，得到一定数量的不同表情、神色的静态人脸图片；S10: Capture and take screenshots of the faces in the dynamic video, and obtain a certain number of static face pictures with different expressions and looks;
S20：采用灰度化、中值滤波等方法对所有静态人脸图片进行预处理；S20: Preprocess all static face images by grayscale, median filtering and other methods;
本步骤中，预处理的目的是为了降低静态人脸图片的干扰和噪声，增强人脸目标以及图像背景的对比度。预处理过程具体包括：In this step, the purpose of preprocessing is to reduce the interference and noise of the static face image, and to enhance the contrast of the face target and the image background. The preprocessing process specifically includes:
S21：首先对静态人脸图片进行灰度处理，将静态人脸图片转换成灰度矩阵；S21: first perform grayscale processing on the static face image, and convert the static face image into a grayscale matrix;
其中，本发明实施例采用灰度化技术对静态人脸图片进行预处理，将静态人脸图片转换成灰度矩阵，利用imread函数直接将输入的静态人脸图片转换成灰度 矩阵，灰度矩阵是一个将图片转换成抽象的过程，其处理原理是将图片转化成一个数字信息，并用矩阵的数值表示各个像素的特征值，转换后的灰度矩阵如图2所示。Among them, the embodiment of the present invention adopts the grayscale technology to preprocess the static face picture, converts the static face picture into a grayscale matrix, and directly converts the input static face picture into a grayscale matrix by using the imread function. The matrix is a process of converting a picture into an abstract. The processing principle is to convert the picture into a digital information, and use the value of the matrix to represent the eigenvalue of each pixel. The converted grayscale matrix is shown in Figure 2.
S22：将灰度矩阵进行切割，生成一定数量(具体数量可根据实际场景进行设定)的小矩阵；S22: Cut the grayscale matrix to generate a certain number of small matrices (the specific number can be set according to the actual scene);
其中，切割后的小矩阵如图3所示。Among them, the small matrix after cutting is shown in Figure 3.
S23：对分割后的小矩阵进行降纬处理；S23: Perform latitude reduction processing on the divided small matrix;
其中，降纬处理即为纬度高的图片进行纬度降低，将低纬度的一些独特特征进行处理及提取，作为图片的特征标记，以用表示图片。降纬的目的是为了减少数据的复杂程度，有利于更加明显的表示图片信息。Among them, the latitude reduction process is to reduce the latitude of a picture with a high latitude, and process and extract some unique features of a low latitude as a feature mark of the picture to represent the picture. The purpose of latitude reduction is to reduce the complexity of the data, which is conducive to more obvious representation of picture information.
S30：对预处理的静态人脸图片进行特征值(包括特殊或特别点)提取，生成由一定数量的特征值图片组成的特征矩阵；S30: Extract feature values (including special or special points) of the preprocessed static face picture, and generate a feature matrix composed of a certain number of feature value pictures;
本步骤中，本方案采用svd(Singular Value Decompositionm,奇异值分解)算法进行特征值提取，公式如下：In this step, this scheme adopts svd (Singular Value Decompositionm, singular value decomposition) algorithm to extract eigenvalues, and the formula is as follows:
A＝U∑V T  (1) A=U∑V T (1)
表示为：设A为一个X*Y的矩阵，U为N*N的方阵，A和U都是正交向量，U左侧为左奇异向量；Σ为X*Y的矩阵，对角线上的元数都为0，其名字称为奇异值；V T为N*N的矩阵，正交向量又名右奇异向量；得公式如下： Represented as: let A be an X*Y matrix, U be an N*N square matrix, A and U are both orthogonal vectors, and the left side of U is a left singular vector; Σ is an X*Y matrix, the diagonal The arity on is 0, and its name is called singular value; V T is an N*N matrix, and an orthogonal vector is also called a right singular vector; the formula is as follows:
AA T＝U∑V TV∑ TU T＝U(∑∑ T)U T AA T =U∑V T V∑ T U T =U(∑∑ T )U T
A TA＝V∑ TU TU∑V T＝V(∑ T∑)V T  (2) A T A=V∑ T U T U∑V T =V(∑ T ∑)V T (2)
所以得：U的列向量(左奇异向量)是AA T的特征向量；同时，V的列向量(右奇异向量)是A TA的特征向量；M的奇异值(Σ的非零对角元素)则是AA T或者A TA的非零特征值的平方根。 So we get: the column vector (left singular vector) of U is the eigenvector of AAT; at the same time, the column vector (right singular vector) of V is the eigenvector of A T A; the singular value of M ( the non-zero diagonal elements of Σ ) is AA T or the square root of the non-zero eigenvalues of A T A.
基于上述，本申请实施例通过对静态人脸图片进行特征值提取可以减少数据 流的值，给后续的各项工作带来了便利，并提高处理速度。Based on the above, the embodiment of the present application can reduce the value of the data stream by performing feature value extraction on the static face picture, which brings convenience to subsequent work and improves the processing speed.
S40：将特征矩阵划分为训练集和测试集，并对训练集和测试集进行归一化处理；S40: Divide the feature matrix into a training set and a test set, and normalize the training set and the test set;
本步骤中，特征矩阵划分方式为：运行程序时，在matlab命令窗口，输入“[pn,pnewn,t,num_train,num_test]＝train_test(feature,num_train)；”，其中num_train和num_test分别是每个人用于训练和测试的图片数(num_train必须是1-10)，可以在运行程序时进行设定；运行程序后，可以在Matlab的变量空间中看到一个名为pn的变量，该变量是经过归一化处理后的训练集，pnewn是归一化后的测试集。In this step, the feature matrix is divided as follows: when running the program, in the matlab command window, enter "[pn,pnewn,t,num_train,num_test]=train_test(feature,num_train);", where num_train and num_test are each person The number of pictures used for training and testing (num_train must be 1-10), which can be set when running the program; after running the program, you can see a variable named pn in the variable space of Matlab, which is a The normalized training set, and pnewn is the normalized test set.
S50：将训练集输入BP神经网络(Back Propagation，前向传播神经网络)进行分类器训练，并输出人脸识别结果；S50: Input the training set into the BP neural network (Back Propagation, forward propagation neural network) for classifier training, and output the face recognition result;
本步骤中，当BP神经网络构建完成后，输入相应的神经元个数以及迭代训练次数，然后将各层的传递函数连接好即可进行分类器的训练。BP神经网络的搭建代码如图4所示，在运行该代码之后，需要手动输入神经元个数、层之间的两个传递函数、训练函数、训练目标、训练迭代次数和学习速率等变量，构建好之后可以在变量窗口看到net变量，具体如图5所示。In this step, after the BP neural network is constructed, input the corresponding number of neurons and the number of iterative training times, and then connect the transfer functions of each layer to train the classifier. The construction code of the BP neural network is shown in Figure 4. After running the code, variables such as the number of neurons, the two transfer functions between layers, the training function, the training target, the number of training iterations and the learning rate need to be manually input. After construction, you can see the net variable in the variable window, as shown in Figure 5.
本申请实施例的BP神经网络结构如图6所示，其包括一层输入层、一到N层的隐含层以及一层输出层。BP神经网路算法为逆向传播算法，其计算过程具体为：采用tansig函数作为输入层与隐含层的传递函数，采用purelin线性函数作为隐含层与输出层的传递函数，采用Sigmoid函数作为输入层到隐含层的激活函数，采用Purelin线性函数作为隐含层到输出层的激活函数；首先对各层的神经元进行初始化，输入样本先从输入层的所有的神经元进入，然后在隐含层当中进行一层一层的处理，最后数据被运送到输出层的每个神经元进行计算，当神经元计算完之后输出学习结果。其中输出层的节点数为人脸的类别数，隐含层节点数为：The structure of the BP neural network in the embodiment of the present application is shown in FIG. 6 , which includes an input layer, a hidden layer ranging from one to N layers, and an output layer. The BP neural network algorithm is a back-propagation algorithm, and its calculation process is as follows: the tansig function is used as the transfer function between the input layer and the hidden layer, the purelin linear function is used as the transfer function between the hidden layer and the output layer, and the Sigmoid function is used as the input. The activation function from the layer to the hidden layer adopts the Purelin linear function as the activation function from the hidden layer to the output layer; first, the neurons in each layer are initialized, and the input samples first enter from all the neurons in the input layer, and then in the hidden layer. The processing is performed layer by layer in the containing layer, and finally the data is sent to each neuron in the output layer for calculation, and the learning result is output after the neuron is calculated. The number of nodes in the output layer is the number of categories of faces, and the number of nodes in the hidden layer is:
Figure PCTCN2020139671-appb-000004
Figure PCTCN2020139671-appb-000004
公式(3)中，n代表输入神经元的大小，m代表输出神经元的大小，a代表10以内的常数。In formula (3), n represents the size of the input neuron, m represents the size of the output neuron, and a represents a constant within 10.
BP神经网路中每一层的向量都不一样，本申请实施例中每一层的向量设置分别为：输入层为x，隐含层为Y，输出层为O，以及一个预期向量值设为d，对于输出层，有：The vectors of each layer in the BP neural network are different. The vector settings of each layer in the embodiment of this application are: the input layer is x, the hidden layer is Y, the output layer is O, and an expected vector value is set. is d, and for the output layer, there are:
Figure PCTCN2020139671-appb-000005
Figure PCTCN2020139671-appb-000005
公式(4)中，输出层为O，O t表示第t个输出层，其中t为自然数。隐含层为Y，y j表示第j个隐含层，其中j为自然数。w jt为第t个输出层O到第j个隐含层y的权重，i，j分别表示输出层的第i个神经元和隐含层的第j个神经元。 In formula (4), the output layer is O, and O t represents the t-th output layer, where t is a natural number. The hidden layer is Y, and yj represents the jth hidden layer, where j is a natural number. w jt is the weight from the t-th output layer O to the j-th hidden layer y, i, j represent the ith neuron of the output layer and the jth neuron of the hidden layer, respectively.
对于隐含层，有：For the hidden layer, there are:
Figure PCTCN2020139671-appb-000006
Figure PCTCN2020139671-appb-000006
公式(5)中，输入层为x，x i表示第i个输入层，其中i为自然数。v ij为第i个输入层x到第j个隐含层y的权重，i表示输入层的第i个神经元，j表示隐含层的第j个神经元。 In formula (5), the input layer is x, and x i represents the ith input layer, where i is a natural number. v ij is the weight from the ith input layer x to the jth hidden layer y, i represents the ith neuron in the input layer, and j represents the jth neuron in the hidden layer.
本申请实施例中，在训练BP神经网络时，还设置有一个误差范围，当BP算法的输出结果没有达到该误差范围时，BP神经网络通过一个逆向传播过程将输出结果在输出层当中逆向返回给隐含层和输入层进行循环计算，并在循环计算过程中修正各层的权值，使得网络误差逐渐下降，直到输出结果达到预设的误差范围。具体的，网络误差用E表达：In the embodiment of the present application, when training the BP neural network, an error range is also set. When the output result of the BP algorithm does not reach the error range, the BP neural network returns the output result in the output layer through a back propagation process. The hidden layer and the input layer are cyclically calculated, and the weights of each layer are corrected in the cyclical calculation process, so that the network error gradually decreases until the output result reaches the preset error range. Specifically, the network error is expressed by E:
Figure PCTCN2020139671-appb-000007
Figure PCTCN2020139671-appb-000007
公式(6)中，E是输出层O与实际值d的差值的平方，
Figure PCTCN2020139671-appb-000008
为第t个实际值d t减去第t个输出层O t的平方累积求和，根据最小二乘法，E值越小,输出层O越接 近真实值d。如果E值没有达到设定的误差范围，则调整BP神经网络中各层的i，j数值，以降低E值的大小，直到E值达到设定的误差范围，即可得到最优网络参数。 In formula (6), E is the square of the difference between the output layer O and the actual value d,
Figure PCTCN2020139671-appb-000008
According to the least squares method, the smaller the value of E, the closer the output layer O is to the true value d . If the E value does not reach the set error range, adjust the i, j values of each layer in the BP neural network to reduce the size of the E value, until the E value reaches the set error range, the optimal network parameters can be obtained.
S50：将测试集输入训练好的BP神经网络进行模型性能评估；S50: Input the test set into the trained BP neural network for model performance evaluation;
进一步地，本申请实施例通过统计BP神经网络在训练集和测试集下的识别率以及总识别率，并分别在特征维数为8、16、32、48的条件下对BP神经网络的总识别率进行统计，以对BP神经网络在不同特征维数下的人脸识别精度进行验证，具体如下：Further, in the embodiment of the present application, the recognition rate and total recognition rate of the BP neural network in the training set and the test set are counted, and the total recognition rate of the BP neural network is calculated under the condition that the feature dimensions are 8, 16, 32, and 48 respectively. The recognition rate is counted to verify the face recognition accuracy of the BP neural network under different feature dimensions, as follows:
(1)特征维数为8(1) The feature dimension is 8
如图7所示，为特征维数为8时BP神经网络的识别率示意图。在特征维数为8的条件下，以倍数的形式增加神经元，逐步增加训练次数对总识别率进行测试，测试结果表明在特征维数比较低的条件下，人脸识别的识别率基本在80％以下，总体识别率偏低，因此不采用较低的特征维数。As shown in Figure 7, it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 8. Under the condition that the feature dimension is 8, the neurons are added in multiples, and the training times are gradually increased to test the total recognition rate. The test results show that the recognition rate of face recognition is basically under the condition of low feature dimension. Below 80%, the overall recognition rate is low, so a lower feature dimension is not used.
(2)特征维数为16(2) The feature dimension is 16
如图8所示，为特征维数为16时BP神经网络的识别率示意图。在特征维数为16的条件下，当神经元低于120个时可以发现总识别率可以超过80％，但是当神经元过多时，总体识别率反而会降低，表明人脸识别中神经元的个数并不是越多越好，神经元个数太多会导致识别率降低，且训练神经网络的时间也会过长。As shown in Figure 8, it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 16. Under the condition that the feature dimension is 16, when the number of neurons is less than 120, it can be found that the total recognition rate can exceed 80%, but when there are too many neurons, the overall recognition rate will decrease, indicating that the neuron in face recognition The more the number of neurons, the better. Too many neurons will reduce the recognition rate, and the training time of the neural network will be too long.
(3)特征维数24(3) Feature dimension 24
如图9所示，为特征维数为24时BP神经网络的识别率示意图。在特征维数为24时，识别率随着神经元的增加而降低，在120个神经元当中训练次数为4000次时识别率可高达91.5％。As shown in Figure 9, it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 24. When the feature dimension is 24, the recognition rate decreases with the increase of neurons, and the recognition rate can be as high as 91.5% when the number of training times is 4000 among 120 neurons.
(4)特征维数32(4) Feature dimension 32
如图10所示，为特征维数为32时BP神经网络的识别率示意图。在特征维数为32的条件下，当神经元个数为120个时，网络总体识别率超过90％以上， 其中当训练次数为6000次时最高识别率可达93.5％。As shown in Figure 10, it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 32. Under the condition that the feature dimension is 32, when the number of neurons is 120, the overall recognition rate of the network exceeds 90%, and the highest recognition rate can reach 93.5% when the number of training is 6000 times.
(5)特征维数48(5) Feature dimension 48
如图11所示，为特征维数为48时BP神经网络的识别率示意图。在特征维数为48的条件下，识别率可达80％以上，比特征维数为32时的识别率有所降低，而且由于特征维数较大，在后期的训练识别过程中，会损耗很多时间，因此不采用太高的特征维数。As shown in Figure 11, it is a schematic diagram of the recognition rate of the BP neural network when the feature dimension is 48. Under the condition that the feature dimension is 48, the recognition rate can reach more than 80%, which is lower than the recognition rate when the feature dimension is 32. Moreover, due to the large feature dimension, in the later training and recognition process, the recognition rate will be lost. A lot of time, so don't take too high feature dimensionality.
基于上述精度验证分析表明，在特征维数逐步增加的过程中，人脸识别的准确率呈现开口向下的抛物线形态，另外，神经元的个数也会影响人脸识别的准确率，太多的神经元会降低识别率，且导致训练时间过长；因此本申请实施例优选采用特征维数为32、神经元个数为120，并且训练次数为6000次时识别率最高(达93.5％)。Based on the above precision verification analysis, it is shown that in the process of gradually increasing the feature dimension, the accuracy of face recognition presents a parabolic shape with an opening downward. In addition, the number of neurons will also affect the accuracy of face recognition, too many The number of neurons will reduce the recognition rate and cause the training time to be too long; therefore, the embodiment of the present application preferably adopts the feature dimension of 32, the number of neurons to be 120, and the highest recognition rate (up to 93.5%) when the number of training times is 6000. .
基于上述，本申请实施例的人脸识别方法通过对人脸图像进行灰度以及降纬处理后，对人脸图片进行人脸特征提取，生成特征矩阵，并利用BP神经网络采用逆向传播算法进行人脸识别，能极大的提高人脸识别效率以及识别精度。Based on the above, in the face recognition method of the embodiment of the present application, after performing grayscale and latitude reduction processing on the face image, face feature extraction is performed on the face image, a feature matrix is generated, and a back propagation algorithm is used by using a BP neural network. Face recognition can greatly improve the efficiency and accuracy of face recognition.
请参阅图12，是本申请实施例的人脸识别***的结构示意图。本申请实施例的人脸识别***40包括：Please refer to FIG. 12 , which is a schematic structural diagram of a face recognition system according to an embodiment of the present application. The face recognition system 40 of the embodiment of the present application includes:
人脸图片获取模块41：用于对动态视频进行人脸捕捉并截图，获取静态人脸图片；Face picture acquisition module 41: used to capture and take screenshots of dynamic videos, and obtain static face pictures;
人脸特征提取模块42：用于对所述静态人脸图片进行特征值提取，生成特征矩阵；Face feature extraction module 42: used to extract feature values from the static face picture to generate a feature matrix;
人脸识别模块43：用于将所述特征矩阵输入训练好的BP神经网络，通过所述BP神经网络输出人脸识别结果；其中，所述BP神经网络包括输入层、隐含层以及输出层，输入数据从所述输入层的所有的神经元进入，在所述隐含层中进行计算，将计算结果输入所述输出层的每个神经元进行计算，得到人脸识别结果。Face recognition module 43: for inputting the feature matrix into the trained BP neural network, and outputting the face recognition result through the BP neural network; wherein, the BP neural network includes an input layer, a hidden layer and an output layer , the input data is entered from all the neurons in the input layer, the calculation is performed in the hidden layer, and the calculation result is input into each neuron of the output layer for calculation to obtain a face recognition result.
请参阅图13，为本申请实施例的终端结构示意图。该终端50包括处理器51、与处理器51耦接的存储器52。Please refer to FIG. 13 , which is a schematic structural diagram of a terminal according to an embodiment of the present application. The terminal 50 includes a processor 51 and a memory 52 coupled to the processor 51 .
存储器52存储有用于实现上述人脸识别方法的程序指令。The memory 52 stores program instructions for implementing the above-mentioned face recognition method.
处理器51用于执行存储器52存储的程序指令以控制人脸识别。The processor 51 is configured to execute program instructions stored in the memory 52 to control face recognition.
其中，处理器51还可以称为CPU(Central Processing Unit，中央处理单元)。处理器51可能是一种集成电路芯片，具有信号的处理能力。处理器51还可以是通用处理器、数字信号处理器(DSP)、专用集成电路(ASIC)、现成可编程门阵列(FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。The processor 51 may also be referred to as a CPU (Central Processing Unit, central processing unit). The processor 51 may be an integrated circuit chip with signal processing capability. The processor 51 may also be a general purpose processor, digital signal processor (DSP), application specific integrated circuit (ASIC), off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component . A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
请参阅图14，为本申请实施例的存储介质的结构示意图。本申请实施例的存储介质存储有能够实现上述所有方法的程序文件61，其中，该程序文件61可以以软件产品的形式存储在上述存储介质中，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)或处理器(processor)执行本发明各个实施方式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，或者是计算机、服务器、手机、平板等终端设备。Please refer to FIG. 14 , which is a schematic structural diagram of a storage medium according to an embodiment of the present application. The storage medium of this embodiment of the present application stores a program file 61 capable of implementing all the above methods, wherein the program file 61 may be stored in the above-mentioned storage medium in the form of a software product, and includes several instructions to make a computer device (which may It is a personal computer, a server, or a network device, etc.) or a processor that executes all or part of the steps of the methods of the various embodiments of the present invention. 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 disk and other media that can store program codes , or terminal devices such as computers, servers, mobile phones, and tablets.
对所公开的实施例的上述说明，使本领域专业技术人员能够实现或使用本申请。对这些实施例的多种修改对本领域的专业技术人员来说将是显而易见的，本申请中所定义的一般原理可以在不脱离本申请的精神或范围的情况下，在其它实施例中实现。因此，本申请将不会被限制于本申请所示的这些实施例，而是要符合与本申请所公开的原理和新颖特点相一致的最宽的范围。The above description of the disclosed embodiments enables any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined in this application may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

  1. 一种人脸识别方法，其特征在于，包括：A face recognition method, comprising:
    对动态视频进行人脸捕捉并截图，获取静态人脸图片；Capture and take screenshots of dynamic videos to obtain static face pictures;
    对所述静态人脸图片进行特征值提取，生成特征矩阵；Extracting feature values from the static face picture to generate a feature matrix;
    将所述特征矩阵输入训练好的BP神经网络，通过所述BP神经网络输出人脸识别结果；所述BP神经网络包括输入层、隐含层以及输出层，输入数据从所述输入层的所有的神经元进入，在所述隐含层中进行计算，将计算结果输入所述输出层的每个神经元进行计算，得到人脸识别结果。The feature matrix is input into the trained BP neural network, and the face recognition result is output through the BP neural network; the BP neural network includes an input layer, a hidden layer and an output layer, and the input data is from all the input layers. The neuron of the output layer is entered, the calculation is performed in the hidden layer, and the calculation result is input into each neuron of the output layer for calculation, and the face recognition result is obtained.
  2. 根据权利要求1所述的人脸识别方法，其特征在于，所述对动态视频进行人脸捕捉并截图，获取静态人脸图片还包括：The face recognition method according to claim 1, characterized in that, performing face capture and screenshots on the dynamic video, and obtaining a static face picture further comprises:
    采用灰度化及中值滤波对所述静态人脸图片进行预处理。The static face picture is preprocessed by grayscale and median filtering.
  3. 根据权利要求2所述的人脸识别方法，其特征在于，所述采用灰度化及中值滤波对所述静态人脸图片进行预处理包括：The face recognition method according to claim 2, wherein the preprocessing of the static face picture by grayscale and median filtering comprises:
    对所述静态人脸图片进行灰度处理，将所述静态人脸图片转换成灰度矩阵；Grayscale processing is performed on the static face picture, and the static face picture is converted into a grayscale matrix;
    将所述灰度矩阵切割为预定数量的小矩阵；cutting the grayscale matrix into a predetermined number of small matrices;
    对所述切割后的小矩阵进行降纬处理。Weft reduction processing is performed on the cut small matrix.
  4. 根据权利要求1所述的人脸识别方法，其特征在于：face recognition method according to claim 1, is characterized in that:
    所述BP神经网络采用tansig函数作为所述输入层与隐含层的传递函数，采用purelin线性函数作为所述隐含层与输出层的传递函数，采用Sigmoid函数作为所述输入层到隐含层的激活函数，采用Purelin线性函数作为所述隐含层到输出层的激活函数。The BP neural network adopts the tansig function as the transfer function between the input layer and the hidden layer, the purelin linear function as the transfer function between the hidden layer and the output layer, and the sigmoid function as the input layer to the hidden layer. The activation function uses Purelin linear function as the activation function from the hidden layer to the output layer.
  5. 根据权利要求4所述的人脸识别方法，其特征在于，所述输出层节点数为人脸的类别数，所述隐含层节点数为：The face recognition method according to claim 4, wherein the number of nodes in the output layer is the number of categories of faces, and the number of nodes in the hidden layer is:
    Figure PCTCN2020139671-appb-100001
    Figure PCTCN2020139671-appb-100001
    上式中，n代表输入神经元的大小，m代表输出神经元的大小，a代表10以内的常数。In the above formula, n represents the size of the input neuron, m represents the size of the output neuron, and a represents a constant within 10.
  6. 根据权利要求5所述的人脸识别方法，其特征在于，所述BP神经网路中每一层的向量分别为：The face recognition method according to claim 5, wherein the vectors of each layer in the BP neural network are respectively:
    设置输入层为x，隐含层为Y，输出层为O，以及一个预期向量值设为d，对于输出层，有：Set the input layer to x, the hidden layer to Y, the output layer to O, and an expected vector value to d. For the output layer, we have:
    Figure PCTCN2020139671-appb-100002
    Figure PCTCN2020139671-appb-100002
    上式中，O t表示第t个输出层，其中t为自然数；y j表示第j个隐含层，其中j为自然数；w jt为第t个输出层O到第j个隐含层y的权重，i，j分别表示输出层的第i个神经元和隐含层的第j个神经元； In the above formula, O t represents the t-th output layer, where t is a natural number; y j represents the j-th hidden layer, where j is a natural number; w jt is the t-th output layer O to the j-th hidden layer y The weights of i, j represent the i-th neuron of the output layer and the j-th neuron of the hidden layer, respectively;
    对于隐含层，有：For the hidden layer, there are:
    Figure PCTCN2020139671-appb-100003
    Figure PCTCN2020139671-appb-100003
    上式中，x i表示第i个输入层，其中i为自然数，v ij为第i个输入层x到第j个隐含层y的权重，i表示输入层的第i个神经元，j表示隐含层的第j个神经元。 In the above formula, x i represents the ith input layer, where i is a natural number, v ij is the weight from the ith input layer x to the jth hidden layer y, i represents the ith neuron of the input layer, j represents the jth neuron of the hidden layer.
  7. 根据权利要求1至6任一项所述的人脸识别方法，其特征在于，所述将所述特征矩阵输入训练好的BP神经网络还包括：The face recognition method according to any one of claims 1 to 6, wherein the inputting the feature matrix into the trained BP neural network further comprises:
    在训练所述BP神经网络时，设置一个误差范围，当所述BP神经网络的误差没有达到所述误差范围时，所述BP神经网络通过逆向传播算法将所述输出结果在输出层当中逆向返回给所述隐含层和输入层进行循环计算，并在循环计算过程中修正各层的权值，使得网络误差逐渐下降，直到所述BP神经网络的误差达到所述误差范围。When training the BP neural network, an error range is set. When the error of the BP neural network does not reach the error range, the BP neural network returns the output result in the output layer through the back propagation algorithm. Perform cyclic calculation on the hidden layer and the input layer, and correct the weights of each layer during the cyclic calculation, so that the network error gradually decreases until the error of the BP neural network reaches the error range.
  8. 一种人脸识别***，其特征在于，包括：A face recognition system, comprising:
    人脸图片获取模块：用于对动态视频进行人脸捕捉并截图，获取静态人脸图片；Face picture acquisition module: used to capture and take screenshots of dynamic videos, and obtain static face pictures;
    人脸特征提取模块：用于对所述静态人脸图片进行特征值提取，生成特征矩阵；Face feature extraction module: used to extract feature values from the static face picture to generate a feature matrix;
    人脸识别模块：用于将所述特征矩阵输入训练好的BP神经网络，通过所述BP神经网络输出人脸识别结果；所述BP神经网络包括输入层、隐含层以及输出层，输入数据从所述输入层的所有的神经元进入，在所述隐含层中进行计算，将计算结果输入所述输出层的每个神经元进行计算，得到人脸识别结果。Face recognition module: used to input the feature matrix into the trained BP neural network, and output the face recognition result through the BP neural network; the BP neural network includes an input layer, a hidden layer and an output layer, and the input data Enter from all neurons of the input layer, perform calculation in the hidden layer, input the calculation result to each neuron of the output layer for calculation, and obtain a face recognition result.
  9. 一种终端，其特征在于，所述终端包括处理器、与所述处理器耦接的存储器，其中，A terminal, characterized in that the terminal includes a processor and a memory coupled to the processor, wherein,
    所述存储器存储有用于实现权利要求1-7任一项所述的人脸识别方法的程序指令；Described memory stores the program instruction that is used to realize the method for face recognition according to any one of claim 1-7;
    所述处理器用于执行所述存储器存储的所述程序指令以控制人脸识别。The processor is configured to execute the program instructions stored in the memory to control face recognition.
  10. 一种存储介质，其特征在于，存储有处理器可运行的程序指令，所述程序指令用于执行权利要求1至7任一项所述人脸识别方法。A storage medium, characterized in that it stores program instructions executable by a processor, and the program instructions are used to execute the face recognition method according to any one of claims 1 to 7.
PCT/CN2020/139671 2020-11-17 2020-12-25 Face recognition method and system, terminal, and storage medium WO2022105015A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011284648.8 2020-11-17
CN202011284648.8A CN112257672A (en) 2020-11-17 2020-11-17 Face recognition method, system, terminal and storage medium

Publications (1)

Publication Number Publication Date
WO2022105015A1 true WO2022105015A1 (en) 2022-05-27

Family

ID=74265877

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/139671 WO2022105015A1 (en) 2020-11-17 2020-12-25 Face recognition method and system, terminal, and storage medium

Country Status (2)

Country Link
CN (1) CN112257672A (en)
WO (1) WO2022105015A1 (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103679139A (en) * 2013-11-26 2014-03-26 闻泰通讯股份有限公司 Face recognition method based on particle swarm optimization BP network
CN105139003A (en) * 2015-09-17 2015-12-09 桂林远望智能通信科技有限公司 Dynamic face identification system and method
US20180211102A1 (en) * 2017-01-25 2018-07-26 Imam Abdulrahman Bin Faisal University Facial expression recognition
CN109543637A (en) * 2018-11-29 2019-03-29 中国科学院长春光学精密机械与物理研究所 A kind of face identification method, device, equipment and readable storage medium storing program for executing
CN111652021A (en) * 2019-04-30 2020-09-11 上海铼锶信息技术有限公司 Face recognition method and system based on BP neural network

Family Cites Families (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100729273B1 (en) * 2005-02-04 2007-06-15 오병주 A method of face recognition using pca and back-propagation algorithms
CN103631761B (en) * 2012-08-29 2018-02-27 睿励科学仪器（上海）有限公司 Parallel processing architecture carries out matrix operation and for the method for strict ripple coupling analysis
CN105095962B (en) * 2015-07-27 2017-07-28 中国汽车工程研究院股份有限公司 A kind of material dynamic mechanical performance prediction method based on BP artificial neural networks
CN106056059B (en) * 2016-05-20 2019-02-12 合肥工业大学 The face identification method of multi-direction SLGS feature description and performance cloud Weighted Fusion
CN106443453A (en) * 2016-07-04 2017-02-22 陈逸涵 Lithium battery SOC estimation method based on BP neural network
CN107424146A (en) * 2017-06-28 2017-12-01 北京理工大学 A kind of infrared polarization method for objectively evaluating image quality and system
CN107527018A (en) * 2017-07-26 2017-12-29 湖州师范学院 Momentum method for detecting human face based on BP neural network
CN109902546B (en) * 2018-05-28 2020-11-06 华为技术有限公司 Face recognition method, face recognition device and computer readable medium
CN108875639A (en) * 2018-06-20 2018-11-23 甘肃万维信息技术有限责任公司 A kind of optimization and recognition methods based on genetic algorithm recognition of face
CN109919099A (en) * 2019-03-11 2019-06-21 重庆科技学院 A kind of user experience evaluation method and system based on Expression Recognition
CN110110673B (en) * 2019-05-10 2020-11-27 杭州电子科技大学 Face recognition method based on bidirectional 2DPCA and cascade forward neural network
CN110472693B (en) * 2019-08-22 2021-11-19 华东交通大学 Image processing and classifying method and system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103679139A (en) * 2013-11-26 2014-03-26 闻泰通讯股份有限公司 Face recognition method based on particle swarm optimization BP network
CN105139003A (en) * 2015-09-17 2015-12-09 桂林远望智能通信科技有限公司 Dynamic face identification system and method
US20180211102A1 (en) * 2017-01-25 2018-07-26 Imam Abdulrahman Bin Faisal University Facial expression recognition
CN109543637A (en) * 2018-11-29 2019-03-29 中国科学院长春光学精密机械与物理研究所 A kind of face identification method, device, equipment and readable storage medium storing program for executing
CN111652021A (en) * 2019-04-30 2020-09-11 上海铼锶信息技术有限公司 Face recognition method and system based on BP neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
GONG CHENG-QING: "Algorithm of Face Recognition Based on BP Neural Network", COMPUTER KNOWLEDGE AND TECHNOLOGY, CN, vol. 13, no. 16, 1 June 2017 (2017-06-01), CN , pages 170 - 171,174, XP055932940, ISSN: 1009-3044, DOI: 10.14004/j.cnki.ckt.2017.1995 *

Also Published As

Publication number Publication date
CN112257672A (en) 2021-01-22

Similar Documents

Publication Publication Date Title
Liu et al. Hard negative generation for identity-disentangled facial expression recognition
CN112784764B (en) Expression recognition method and system based on local and global attention mechanism
Tang et al. Extreme learning machine for multilayer perceptron
Jia et al. Fisher non-negative matrix factorization for learning local features
WO2021134277A1 (en) Emotion recognition method, intelligent device, and computer-readable storage medium
Laraba et al. 3D skeleton‐based action recognition by representing motion capture sequences as 2D‐RGB images
WO2019169942A1 (en) Anti-angle and occlusion interference fast face recognition method
WO2019200749A1 (en) Facial recognition method, apparatus, computing device and storage medium
MageshKumar et al. Gabor features and LDA based face recognition with ANN classifier
Arora et al. AutoFER: PCA and PSO based automatic facial emotion recognition
Barros et al. Real-time gesture recognition using a humanoid robot with a deep neural architecture
CN112395979B (en) Image-based health state identification method, device, equipment and storage medium
WO2022052530A1 (en) Method and apparatus for training face correction model, electronic device, and storage medium
Yang et al. Facial expression recognition based on dual-feature fusion and improved random forest classifier
Oveisi et al. Tree-structured feature extraction using mutual information
CN108875767A (en) Method, apparatus, system and the computer storage medium of image recognition
Huang et al. Adaptive deep supervised autoencoder based image reconstruction for face recognition
CN112906520A (en) Gesture coding-based action recognition method and device
Jadhav et al. HDL-PI: hybrid DeepLearning technique for person identification using multimodal finger print, iris and face biometric features
Xu et al. Feature-level fusion method based on KFDA for multimodal recognition fusing ear and profile face
Guo et al. Facial expression recognition: a review
WO2022105015A1 (en) Face recognition method and system, terminal, and storage medium
Li et al. Feature extraction based on deep‐convolutional neural network for face recognition
Li et al. Memory-augmented autoencoder based continuous authentication on smartphones with conditional transformer gans
Tefas et al. Variants of dynamic link architecture based on mathematical morphology for frontal face authentication

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: 20962306

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20962306

Country of ref document: EP

Kind code of ref document: A1