CN113221830B - Super-division living body identification method, system, terminal and storage medium - Google Patents

Super-division living body identification method, system, terminal and storage medium Download PDF

Info

Publication number
CN113221830B
CN113221830B CN202110605475.3A CN202110605475A CN113221830B CN 113221830 B CN113221830 B CN 113221830B CN 202110605475 A CN202110605475 A CN 202110605475A CN 113221830 B CN113221830 B CN 113221830B
Authority
CN
China
Prior art keywords
image data
super
rgb
network
image
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110605475.3A
Other languages
Chinese (zh)
Other versions
CN113221830A (en
Inventor
徐玲玲
戴磊
刘玉宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
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 Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Priority to CN202110605475.3A priority Critical patent/CN113221830B/en
Publication of CN113221830A publication Critical patent/CN113221830A/en
Application granted granted Critical
Publication of CN113221830B publication Critical patent/CN113221830B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • G06V40/161—Detection; Localisation; Normalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/40—Spoof detection, e.g. liveness detection
    • G06V40/45—Detection of the body part being alive

Landscapes

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

Abstract

The invention discloses a super-division living body identification method, a system, a terminal and a storage medium. The method comprises the following steps: respectively acquiring first image data, second image data, third image data and fourth image data which contain different formats under different scenes; inputting the first image data into a convolution network for training to obtain a trained convolution network, wherein the output of the convolution network is the face classification result of each detection target in the first image data and the score of each classification; inputting the third image data into a convolution network for training to obtain a trained RGB (red, green and blue) super-division recognition network, wherein the output of the RGB super-division recognition network is the classification scores of live photos and non-live photos in the third image data; and respectively inputting the second image data and the fourth image data into an RGB super-division recognition network, and training the RGB super-division recognition network again to obtain a trained RGB+IR super-division living body recognition network. The invention has the capability of resisting non-living body image attack and can improve the face recognition effect.

Description

Super-division living body identification method, system, terminal and storage medium
Technical Field
The invention relates to the technical field of face recognition, in particular to a super-division living body recognition method, a system, a terminal and a storage medium.
Background
Face recognition is a biological recognition technology for carrying out identity recognition based on facial feature information of people. In the prior art, face recognition technology generally collects face images through a camera, inputs the face images into a face recognition network, extracts features of the face images through the face recognition network, and judges whether the face images and the face images stored in an archive belong to the same person. However, whether a live image (i.e., a real image) or a non-live image (i.e., an image forged by a photo) is photographed by a camera, a two-dimensional image is obtained finally, and the existing face recognition network has no resistance to attack of the non-live image, so that the authenticity of the input face image cannot be accurately distinguished. Meanwhile, the face image generally includes two formats of RGB image and IR (Infrared Radiation, infrared) image, and the face recognition effect is poor because the RGB image is greatly affected by illumination. While the IR image is slightly affected by light, more image information can be obtained under the condition of large backlight or dim light, the detail information of the IR image is insufficient, and the IR image has the defects of large loss, poor recognition effect and the like under the conditions of long distance, low resolution or blurred scenes. Based on the above-mentioned shortcomings, it is necessary to provide a face recognition network that can be applied to both RGB images and IR images to improve the face recognition effect.
Disclosure of Invention
The invention provides a super-split living body identification method, a system, a terminal and a storage medium, which aim to solve the technical problems that the existing face identification technology has no resistance to non-living body image attack, the authenticity of an input face image cannot be accurately identified, the loss is large, the identification effect is poor and the like by utilizing RGB images and IR images.
In order to solve the technical problems, the invention adopts the following technical scheme:
a method of supersplit living body identification, comprising:
respectively acquiring first image data, second image data, third image data and fourth image data which contain different formats under different scenes; the first image data is an original face image in an RGB format, the second image data is an original face image in an RGB format and an IR format which are in one-to-one correspondence, the third image data is a live photo and a non-live photo in an RGB format, and the fourth image data is a live photo and a non-live photo in an RGB format and an IR format which are in one-to-one correspondence;
inputting the first image data into a convolution network for training to obtain a trained convolution network; the output of the convolution network is the face classification result of each detection target in the first image data and the score of each classification;
locking network parameters of a convolution network trained by the first image data, and inputting the third image data into the convolution network for training to obtain a trained RGB (red, green and blue) super-resolution identification network; the output of the RGB super-resolution identification network is the classification scores of the living body and the non-living body photos in the third image data;
respectively inputting the second image data and the fourth image data into the RGB super-resolution identification network, and training the RGB super-resolution identification network again to obtain a trained RGB+IR super-resolution living body identification network;
and carrying out face recognition on the image to be recognized through the RGB+IR super-division living body recognition network.
The technical scheme adopted by the embodiment of the invention further comprises the following steps: the step of respectively acquiring the first image data, the second image data, the third image data and the fourth image data which contain different formats under different scenes further comprises the steps of:
and respectively carrying out augmentation operation on the RGB format original face images in the first image data and the second image data, and carrying out label marking on the live photos and the non-live photos in the third image data and the fourth image data.
The technical scheme adopted by the embodiment of the invention further comprises the following steps: the step of performing augmentation operation on the original face image in the RGB format in the first image data and the second image data respectively specifically includes:
at least two sets of operations are selected from the illumination, blur, noise or compression processing, respectively, to combine and augment the first and second image data.
The technical scheme adopted by the embodiment of the invention further comprises the following steps: the training process of inputting the first image data into a convolutional network for training specifically comprises the following steps:
inputting the first image data subjected to the augmentation operation into a convolution network for feature image extraction, carrying out convolution operation on the extracted feature image through a convolution layer and a full connection layer, changing the feature image into a feature vector, carrying out face classification on the feature vector according to an input image to obtain a face classification result of each detection target and a score of each classification, and obtaining a cross entropy according to the face classification result and the score of each classification to obtain a cross entropy loss value ce_loss;
performing up-sampling operation on the feature map, obtaining an up-sampling image with the same size as an original face image in the first image data, calculating the Euclidean distance between the up-sampling image and the original face image, and calculating a loss value pix_loss according to the Euclidean distance;
and adding the cross entropy loss value ce_loss and the loss value pix_loss to obtain a loss function loss of the convolution network, and reversely updating the convolution network according to the loss function loss to obtain the trained convolution network.
The technical scheme adopted by the embodiment of the invention further comprises the following steps: the cross entropy loss value ce_loss is calculated in the following manner:
in the above formula, n is the number of classifications of the first image data, and label i Label, v for the ith class of the first image data i The score obtained for the feature vector generated for the i classified images of the first image data.
The technical scheme adopted by the embodiment of the invention further comprises the following steps: the loss value pix_loss is calculated in the following manner:
in the above formula, gp is an up-sampling image, rp is a corresponding original face image, m=h×w×c, and h, w and c are the height, width and channel number of the image respectively.
The technical scheme adopted by the embodiment of the invention further comprises the following steps: the training process of inputting the third image data into the convolution network for training specifically comprises the following steps:
inputting the third image data marked by the label into a convolution network of locking network parameters to perform feature map extraction, convolution and classification operation, obtaining classification scores of live photos and non-live photos in the third image data, and calculating a live loss value live_loss according to the classification scores of the live photos and the non-live photos and the label mark to obtain a trained RGB super-division recognition network;
the living body loss value live_loss is calculated by the following steps:
in the above, l 1 Is the living body fraction, l 2 Is the fraction of non-living bodies.
The technical scheme adopted by the embodiment of the invention further comprises the following steps: the inputting the second image data and the fourth image data into the RGB super-resolution recognition network respectively, and the training the RGB super-resolution recognition network again includes:
summing the RGB face image after the augmentation operation in the second image data with the corresponding IR face image to generate new second image data;
inputting the new second image data into a trained RGB (red, green and blue) super-resolution identification network, and training the RGB super-resolution identification network to obtain a fine-tuned super-resolution identification network;
summing the RGB live photo and the non-live photo in the fourth image data marked by the label with the corresponding IR live photo and the non-live photo to generate new fourth image data;
and inputting the new fourth image data into the trimmed super-resolution identification network, and training the super-resolution identification network again to obtain a trained RGB+IR super-resolution living body identification network.
The embodiment of the invention adopts another technical scheme that: a supersplit living body identification system, comprising:
and a data acquisition module: the method comprises the steps of respectively acquiring first image data, second image data, third image data and fourth image data which contain different formats under different scenes; the first image data is an original face image in an RGB format, the second image data is an original face image in an RGB format and an IR format which are in one-to-one correspondence, the third image data is a live photo and a non-live photo in an RGB format, and the fourth image data is a live photo and a non-live photo in an RGB format and an IR format which are in one-to-one correspondence;
a first model training module: the method comprises the steps of inputting the first image data into a convolution network for training to obtain a trained convolution network, wherein the output of the convolution network is the face classification result of each detection target in the first image data and the score of each classification;
and a second model training module: the network parameters of the convolution network trained by the first image data are locked, the third image data are input into the convolution network for training, and a trained RGB super-resolution identification network is obtained; the output of the RGB super-resolution identification network is the classification scores of the living body and the non-living body photos in the third image data;
model fine tuning module: and the RGB super-resolution recognition network is used for respectively inputting the second image data and the fourth image data into the RGB super-resolution recognition network, training the RGB super-resolution recognition network again to obtain a trained RGB+IR super-resolution living body recognition network, and carrying out face recognition on the image to be recognized through the RGB+IR super-resolution living body recognition network.
The embodiment of the invention adopts the following technical scheme: a terminal comprising a processor, a memory coupled to the processor, wherein,
the memory stores program instructions for implementing the above-mentioned supersplit living body identification method;
the processor is configured to execute the program instructions stored by the memory to perform the super-split living body identification operation.
The embodiment of the invention adopts the following technical scheme: a storage medium storing program instructions executable by a processor for performing the above-described supersplit living body identification method.
According to the super-split living body identification method, system, terminal and storage medium, the RGB super-split identification network is obtained through training of the face images in the RGB format and the living photos and the non-living photos in the RGB format, the RGB super-split identification network is retrained through the original face images in the RGB format and the IR format which are in one-to-one correspondence and the living photos and the non-living photos in the RGB format and the IR format which are in one-to-one correspondence, the RGB+IR super-split living body identification network is obtained, the face identification network has the capability of resisting the attack of the non-living body images, the image quality can be improved on the premise that the system consumption is not increased, the face identification effect is further improved, the workload of training and integrating individual living models is avoided, the system structure is simpler, and the operation speed is faster. The embodiment of the invention can be applied to not only an independent RGB scene, but also a binocular camera scene of RGB+IR, and the application scene is wider.
Drawings
FIG. 1 is a flow chart of a method for identifying a supersplit living body according to a first embodiment of the present invention;
FIG. 2 is a flow chart of a method for supersplit living body identification according to a second embodiment of the present invention;
FIG. 3 is a diagram illustrating a network training process according to an embodiment of the present invention;
FIG. 4 is a schematic structural view of an ultra-split living body recognition system according to an embodiment of the present invention;
fig. 5 is a schematic diagram of a terminal structure according to an embodiment of the present invention;
fig. 6 is a schematic diagram of a storage medium structure according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The terms "first," "second," "third," and the like in this disclosure are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first", "a second", and "a third" may explicitly or implicitly include at least one such feature. In the description of the present invention, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise. All directional indications (such as up, down, left, right, front, back … …) in embodiments of the present invention are merely used to explain the relative positional relationship, movement, etc. between the components in a particular gesture (as shown in the drawings), and if the particular gesture changes, the directional indication changes accordingly. Furthermore, the terms "comprise" and "have," as well as any variations thereof, are intended to cover a non-exclusive inclusion. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those listed steps or elements but may include other steps or elements not listed or inherent to such process, method, article, or apparatus.
Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the invention. The appearances of such phrases in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments. Those of skill in the art will explicitly and implicitly appreciate that the embodiments described herein may be combined with other embodiments.
Referring to fig. 1, a flowchart of a method for identifying a super-split living body according to a first embodiment of the present invention is shown. The super-division living body identification method of the first embodiment of the present invention includes the steps of:
s10: respectively acquiring first image data, second image data, third image data and fourth image data which contain different formats under different scenes;
in this step, the scene includes, but is not limited to, an attendance scene, a person license scene, a monitoring scene, and the like. The first image data contains a plurality of RGB format original face images of the detection targets, the second image data contains a plurality of RGB format original face images and IR format original face images of the detection targets, and the number of face images in the first image data is larger than that of the second image data. The third image data contains live photos and non-live photos in RGB format of a plurality of detection targets, and the fourth image data contains live photos and non-live photos in RGB format and IR format of a plurality of detection targets in one-to-one correspondence.
S11: inputting the first image data into a convolution network for training to obtain a trained convolution network; the output of the convolution network is the face classification result of each detection target in the first image data and the score of each classification;
s12: locking network parameters of a convolution network trained by the first image data, inputting the third image data into the convolution network for training to obtain a trained RGB super-resolution (i.e. super-resolution) identification network; the output of the RGB super-division recognition network is the classification scores of the live photo and the non-live photo in the third image data;
s13: respectively inputting the second image data and the fourth image data into an RGB super-division recognition network, and training the RGB super-division recognition network again to obtain a trained RGB+IR super-division living body recognition network;
s14: and carrying out face recognition on the image to be recognized through an RGB+IR super-division living body recognition network.
Fig. 2 is a flow chart of a method for identifying a super-split living body according to a second embodiment of the invention. The super-split living body identification method of the second embodiment of the present invention includes the steps of:
s20: respectively acquiring first image data, second image data, third image data and fourth image data which contain a plurality of detection targets in different formats under different scenes;
in this step, the scene includes, but is not limited to, an attendance scene, a person license scene, a monitoring scene, and the like. The first image data contains a plurality of RGB format original face images of the detection targets, the second image data contains a plurality of RGB format original face images and IR format original face images of the detection targets, and the number of face images in the first image data is larger than that of the second image data. The third image data contains live photos and non-live photos in RGB format of a plurality of detection targets, and the fourth image data contains live photos and non-live photos in RGB format and IR format of a plurality of detection targets in one-to-one correspondence.
S21: performing augmentation operation on the RGB format original face images in the first image data and the second image data respectively, and performing label marking on the live photos and the non-live photos in the third image data and the fourth image data;
in the step, the augmentation operation comprises illumination, blurring, noise or/and compression treatment and other operations; the light processing algorithm comprises strong light, dark light, side light, point light source and the like, the fuzzy processing algorithm comprises motion fuzzy, gaussian fuzzy, median fuzzy and the like, the noise processing algorithm comprises Gaussian noise, pretzel noise and the like, the enhancement processing algorithm comprises contrast conversion, saturation conversion, gamma conversion, log conversion and the like, and the compression processing algorithm comprises picture compression, picture interpolation scaling and the like. When the first image data and the second image data are subjected to the augmentation operation, at least two groups of operations are selected from illumination, blurring, noise or compression processing respectively, and any one of the at least two groups of operations is sequentially selected to carry out combined augmentation on the image. Preferably, in the embodiment of the present invention, three sets of operations including illumination, blur and compression are selected, and a strong light, a median blur and a picture compression algorithm are selected from the three sets of operation algorithms, respectively, and then strong light, median blur and picture compression are sequentially performed on the first image data and the second image data, respectively, so as to obtain first image data and second image data after the augmentation operation.
In the embodiment of the invention, the marking of the third image data and the fourth image data is specifically: the label of the live photo is 1, the label of the non-live photo is 0, and other parameters can be adopted for labeling.
S22: inputting the first image data subjected to the augmentation operation into a convolutional network for feature map extraction, performing convolutional operation on the extracted feature map through a CONV layer (convolutional layer) and an FC layer (fully connected layer), changing the feature map into a feature vector, performing softmax face classification on the feature vector according to the input image data to obtain a face classification result of each detection target and a score of each classification, and obtaining cross entropy of the face classification result and the score of each classification to obtain a cross entropy loss value ce_loss;
in this step, the convolution network includes, but is not limited to resnet, desnet, shufflenet, etc., and the cross entropy loss value ce_loss is specifically:
in the formula (1), n is the classification number of the first image data, and label i Label (label), v for the ith class of first image data i The score obtained for the feature vector generated for the i classified images of the first image data.
S23: the extracted feature map is subjected to up-sampling operation, up-sampled images with the same size as the original face image in the first image data are obtained, euclidean distance (real distance between two points in m-dimensional space) between the up-sampled images and the original face image is calculated, and loss value pix_loss is calculated according to the Euclidean distance:
in the formula (2), gp is an up-sampling image, rp is a corresponding original face image in the first image data, m=h×w×c, and h, w, c are respectively the height, width and channel number of the image.
S24: the convolution network is reversely updated by using a loss function loss=ce_loss+pix_loss, and a trained convolution network is obtained;
s25, locking network parameters such as a feature map extraction layer, a convolution layer, a full connection layer, feature vectors and the like of the trained convolution network, inputting third image data marked by label into the convolution network to perform feature map extraction, convolution and classification operation, obtaining classification scores of live photos and non-live photos, and calculating a live loss value live_loss according to the classification scores of the live photos and the non-live photos and the marked label to obtain the trained RGB (red, green and blue) super-division identification network;
in this step, please refer to fig. 3, which is a schematic diagram of the network training process. The black frame line part is the locked network parameter part. The living body loss value live_loss is calculated by the following specific method:
in the formula (3), l 1 Is the living body fraction, l 2 Is the fraction of non-living bodies.
Based on the steps, training of the RGB super-resolution recognition network is completed, the RGB super-resolution recognition network is provided with living body output, and the feature vectors and living body classification can be directly applied to face recognition engineering.
S26: summing the RGB face image after the augmentation operation in the second image data with the corresponding IR face image to generate new second image data;
s27: inputting new second image data into the trained RGB super-resolution identification network, and training the RGB super-resolution identification network to obtain a fine-tuned super-resolution identification network;
in this step, the fine tuning training process of the RGB super-resolution identification network is the same as the convolutional network training process described in S22 to S24, and this step is not repeated. It can be appreciated that in the fine tuning training process of the RGB super-resolution identification network, only the euclidean distance between the up-sampled image and the original face image in RGB format in the second image data needs to be calculated when the loss function pix_loss is calculated.
S28: summing the RGB live photo and the non-live photo in the marked fourth image data with the corresponding IR live photo and the corresponding non-live photo to generate new fourth image data;
s29: inputting new fourth image data into the trimmed RGB super-division recognition network, training the super-division recognition network again to obtain a trained RGB+IR super-division living body recognition network, and recognizing the face of the image to be recognized through the RGB+IR super-division living body recognition network;
in this step, the fine tuning training process of the super-resolution identification network is the same as the training process of the RGB super-resolution identification network in S25, and this step is not repeated. The RGB+IR super-division living body recognition network after training can be applied to RGB and IR binocular camera scenes, and compared with single RGB recognition, the RGB+IR super-division living body recognition network has better adaptability to illumination and blurring and more accurate living body judgment.
Based on the above, the super-split living body recognition method of the embodiment of the invention obtains the RGB super-split recognition network through training the face image in the RGB format and the living photo in the RGB format and the non-living photo, retrains the RGB super-split recognition network through the original face image in the one-to-one correspondence with the RGB format and the IR format and the living photo in the one-to-one correspondence with the IR format and the non-living photo, and obtains the RGB+IR super-split living body recognition network, so that the face recognition network has the capability of resisting the attack of the non-living body image, the image quality can be improved on the premise of not increasing the system consumption, the face recognition effect is further improved, the workload of training and integrating the single living model is avoided, the system structure is simpler, and the operation speed is faster. The embodiment of the invention can be applied to not only an independent RGB scene, but also a binocular camera scene of RGB+IR, and the application scene is wider.
In an alternative embodiment, it is also possible to: and uploading the result of the super-division living body identification method to a blockchain.
Specifically, corresponding summary information is obtained based on the result of the super-split living body identification method, specifically, the summary information is obtained by hashing the result of the super-split living body identification method, for example, the summary information is obtained by using a sha256s algorithm. Uploading summary information to the blockchain can ensure its security and fair transparency to the user. The user can download the summary information from the blockchain to verify whether the result of the supersplit living identification method is tampered. The blockchain referred to in this example is a novel mode of application for computer technology such as distributed data storage, point-to-point transmission, consensus mechanisms, encryption algorithms, and the like. The Blockchain (Blockchain), which is essentially a decentralised database, is a string of data blocks that are generated by cryptographic means in association, each data block containing a batch of information of network transactions for verifying the validity of the information (anti-counterfeiting) and generating the next block. The blockchain may include a blockchain underlying platform, a platform product services layer, an application services layer, and the like.
Fig. 4 is a schematic structural diagram of an ultra-split living body recognition system according to an embodiment of the invention. The supersplit living body identification system 40 according to the embodiment of the present invention includes:
data acquisition module 41: the method comprises the steps of respectively acquiring first image data, second image data, third image data and fourth image data which contain different formats under different scenes; the first image data is an original face image in an RGB format, the second image data is an original face image in an RGB format and an IR format which are in one-to-one correspondence, the third image data is a live photo and a non-live photo in an RGB format, and the fourth image data is a live photo and a non-live photo in an RGB format and an IR format which are in one-to-one correspondence;
first model training module 42: the method comprises the steps of inputting first image data into a convolution network for training to obtain a trained convolution network; the output of the convolution network is the face classification result of each detection target in the first image data and the score of each classification;
second model training module 43: the network parameters of the convolution network trained by the first image data are locked, the third image data are input into the convolution network for training, and a trained RGB (red, green and blue) super-resolution identification network is obtained; the output of the RGB super-division recognition network is the classification scores of the live photo and the non-live photo in the third image data;
model fine tuning module 44: the method is used for respectively inputting the second image data and the fourth image data into the RGB super-division recognition network, training the RGB super-division recognition network again to obtain a trained RGB+IR super-division living body recognition network, and recognizing the face of the image to be recognized through the RGB+IR super-division living body recognition network.
Fig. 5 is a schematic diagram of a terminal structure according to an embodiment of the invention. The terminal 50 includes a processor 51, a memory 52 coupled to the processor 51.
The memory 52 stores program instructions for implementing the above-described supersplit living body identification method.
The processor 51 is configured to execute program instructions stored in the memory 52 to perform the super-split living body identification operation.
The processor 51 may also be referred to as a CPU (Central Processing Unit ). The processor 51 may be an integrated circuit chip with signal processing capabilities. Processor 51 may also be a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
Referring to fig. 6, fig. 6 is a schematic structural diagram of a storage medium according to an embodiment of the invention. The storage medium of the embodiment of the present invention stores a program file 61 capable of implementing all the methods described above, where the program file 61 may be stored in the storage medium in the form of a software product, and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor (processor) to execute all or part of the steps of the methods of the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, an optical disk, or other various media capable of storing program codes, or a terminal device such as a computer, a server, a mobile phone, a tablet, or the like.
In the several embodiments provided in the present invention, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the system embodiments described above are merely illustrative, e.g., the partitioning of elements is merely a logical functional partitioning, and there may be additional partitioning in actual implementation, e.g., multiple elements or components may be combined or integrated into another system, or some features may be omitted, or not implemented. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units. The foregoing is only the embodiments of the present invention, and therefore, the patent scope of the invention is not limited thereto, and all equivalent structures or equivalent processes using the descriptions of the present invention and the accompanying drawings, or direct or indirect application in other related technical fields, are included in the scope of the invention.

Claims (11)

1. A method of superminute living body identification, comprising:
respectively acquiring first image data, second image data, third image data and fourth image data which contain different formats under different scenes; the first image data is an original face image in an RGB format, the second image data is an original face image in an RGB format and an IR format which are in one-to-one correspondence, the third image data is a live photo and a non-live photo in an RGB format, and the fourth image data is a live photo and a non-live photo in an RGB format and an IR format which are in one-to-one correspondence;
inputting the first image data into a convolution network for training to obtain a trained convolution network; the output of the convolution network is the face classification result of each detection target in the first image data and the score of each classification;
locking network parameters of a convolution network trained by the first image data, and inputting the third image data into the convolution network for training to obtain a trained RGB (red, green and blue) super-resolution identification network; the output of the RGB super-division recognition network is the classification scores of the live photo and the non-live photo in the third image data;
respectively inputting the second image data and the fourth image data into the RGB super-resolution identification network, and training the RGB super-resolution identification network again to obtain a trained RGB+IR super-resolution living body identification network;
and carrying out face recognition on the image to be recognized through the RGB+IR super-division living body recognition network.
2. The method of claim 1, wherein the steps of obtaining the first image data, the second image data, the third image data, and the fourth image data, respectively, each having a different format in different scenes further comprise:
and respectively carrying out augmentation operation on the RGB format original face images in the first image data and the second image data, and carrying out label marking on the live photos and the non-live photos in the third image data and the fourth image data.
3. The method for super-resolution in-vivo identification as defined in claim 2, wherein the step of performing an augmentation operation on the RGB format original face image in the first image data and the second image data, respectively, comprises:
at least two sets of operations are selected from illumination, blur, noise or compression processing, respectively, to combine and augment the first and second image data.
4. The method for identifying a super-division living body according to claim 2 or 3, wherein the training process of inputting the first image data into a convolutional network for training is specifically:
inputting the first image data subjected to the augmentation operation into a convolution network for feature image extraction, carrying out convolution operation on the extracted feature image through a convolution layer and a full connection layer, changing the feature image into a feature vector, carrying out face classification on the feature vector according to an input image to obtain a face classification result of each detection target and a score of each classification, and obtaining a cross entropy according to the face classification result and the score of each classification to obtain a cross entropy loss value ce_loss;
performing up-sampling operation on the feature map, obtaining an up-sampling image with the same size as an original face image in the first image data, calculating the Euclidean distance between the up-sampling image and the original face image, and calculating a loss value pix_loss according to the Euclidean distance;
and adding the cross entropy loss value ce_loss and the loss value pix_loss to obtain a loss function loss of the convolution network, and reversely updating the convolution network according to the loss function loss to obtain the trained convolution network.
5. The method for identifying a living body according to claim 4, wherein the cross entropy loss value ce_loss is calculated by:
in the above formula, n is the number of classifications of the first image data, and label i Label, v for the ith class of the first image data i The score obtained for the feature vector generated for the i classified images of the first image data.
6. The method for identifying a living body according to claim 4, wherein the loss value pix_loss is calculated by:
in the above formula, gp is an up-sampling image, rp is a corresponding original face image, m=h×w×c, and h, w and c are the height, width and channel number of the image respectively.
7. The method for super-resolution in-vivo identification as defined in claim 4, wherein the training process of inputting the third image data into the convolutional network for training is specifically:
inputting the third image data marked by the label into a convolution network of locking network parameters to perform feature map extraction, convolution and classification operation, obtaining classification scores of live photos and non-live photos in the third image data, and calculating a live loss value live_loss according to the classification scores of the live photos and the non-live photos and the label mark to obtain a trained RGB super-division recognition network;
the living body loss value live_loss is calculated by the following steps:
in the above, l 1 Is the living body fraction, l 2 Is the fraction of non-living bodies.
8. The super-resolution living body recognition method according to claim 7, wherein inputting the second image data and the fourth image data into the RGB super-resolution recognition network, respectively, and training the RGB super-resolution recognition network again includes:
summing the RGB face image after the augmentation operation in the second image data with the corresponding IR face image to generate new second image data;
inputting the new second image data into a trained RGB (red, green and blue) super-resolution identification network, and training the RGB super-resolution identification network to obtain a fine-tuned super-resolution identification network;
summing the RGB live photo and the non-live photo in the fourth image data marked by the label with the corresponding IR live photo and the non-live photo to generate new fourth image data;
and inputting the new fourth image data into the trimmed super-resolution identification network, and training the super-resolution identification network again to obtain a trained RGB+IR super-resolution living body identification network.
9. A supersplit living body identification system, comprising:
and a data acquisition module: the method comprises the steps of respectively acquiring first image data, second image data, third image data and fourth image data which contain different formats under different scenes; the first image data is an original face image in an RGB format, the second image data is an original face image in an RGB format and an IR format which are in one-to-one correspondence, the third image data is a live photo and a non-live photo in an RGB format, and the fourth image data is a live photo and a non-live photo in an RGB format and an IR format which are in one-to-one correspondence;
a first model training module: the first image data are input into a convolution network for training, so that a trained convolution network is obtained; the output of the convolution network is the face classification result of each detection target in the first image data and the score of each classification;
and a second model training module: the network parameters of the convolution network trained by the first image data are locked, the third image data are input into the convolution network for training, and a trained RGB super-resolution identification network is obtained; the output of the RGB super-resolution identification network is the classification scores of the living body and the non-living body photos in the third image data;
model fine tuning module: and the RGB super-resolution recognition network is used for respectively inputting the second image data and the fourth image data into the RGB super-resolution recognition network, training the RGB super-resolution recognition network again to obtain a trained RGB+IR super-resolution living body recognition network, and carrying out face recognition on the image to be recognized through the RGB+IR super-resolution living body recognition network.
10. A terminal comprising a processor, a memory coupled to the processor, wherein,
the memory stores program instructions for implementing the superminute living body identification method according to any of claims 1 to 8;
the processor is configured to execute the program instructions stored by the memory to perform the superminute living body identification method.
11. A storage medium storing program instructions executable by a processor for performing the super-resolution living body identification method according to any one of claims 1 to 8.
CN202110605475.3A 2021-05-31 2021-05-31 Super-division living body identification method, system, terminal and storage medium Active CN113221830B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110605475.3A CN113221830B (en) 2021-05-31 2021-05-31 Super-division living body identification method, system, terminal and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110605475.3A CN113221830B (en) 2021-05-31 2021-05-31 Super-division living body identification method, system, terminal and storage medium

Publications (2)

Publication Number Publication Date
CN113221830A CN113221830A (en) 2021-08-06
CN113221830B true CN113221830B (en) 2023-09-01

Family

ID=77081930

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110605475.3A Active CN113221830B (en) 2021-05-31 2021-05-31 Super-division living body identification method, system, terminal and storage medium

Country Status (1)

Country Link
CN (1) CN113221830B (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109508694A (en) * 2018-12-10 2019-03-22 上海众源网络有限公司 A kind of face identification method and identification device
CN110245645A (en) * 2019-06-21 2019-09-17 北京字节跳动网络技术有限公司 Face vivo identification method, device, equipment and storage medium
CN111539311A (en) * 2020-04-21 2020-08-14 上海锘科智能科技有限公司 Living body distinguishing method, device and system based on IR and RGB double photographing
WO2021012526A1 (en) * 2019-07-22 2021-01-28 平安科技（深圳）有限公司 Face recognition model training method, face recognition method and apparatus, device, and storage medium

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109508694A (en) * 2018-12-10 2019-03-22 上海众源网络有限公司 A kind of face identification method and identification device
CN110245645A (en) * 2019-06-21 2019-09-17 北京字节跳动网络技术有限公司 Face vivo identification method, device, equipment and storage medium
WO2021012526A1 (en) * 2019-07-22 2021-01-28 平安科技（深圳）有限公司 Face recognition model training method, face recognition method and apparatus, device, and storage medium
CN111539311A (en) * 2020-04-21 2020-08-14 上海锘科智能科技有限公司 Living body distinguishing method, device and system based on IR and RGB double photographing

Also Published As

Publication number Publication date
CN113221830A (en) 2021-08-06

Similar Documents

Publication Publication Date Title
Anwar et al. Image colorization: A survey and dataset
US11830230B2 (en) Living body detection method based on facial recognition, and electronic device and storage medium
US8761446B1 (en) Object detection with false positive filtering
WO2022161286A1 (en) Image detection method, model training method, device, medium, and program product
WO2019152983A2 (en) System and apparatus for face anti-spoofing via auxiliary supervision
US11676390B2 (en) Machine-learning model, methods and systems for removal of unwanted people from photographs
CN111444744A (en) Living body detection method, living body detection device, and storage medium
CN112733802B (en) Image occlusion detection method and device, electronic equipment and storage medium
CN109271930B (en) Micro-expression recognition method, device and storage medium
CN112052831A (en) Face detection method, device and computer storage medium
CN112052830B (en) Method, device and computer storage medium for face detection
WO2022222575A1 (en) Method and system for target recognition
CN111783629A (en) Human face in-vivo detection method and device for resisting sample attack
Paschalakis et al. Real-time face detection and tracking for mobile videoconferencing
CN115862055A (en) Pedestrian re-identification method and device based on comparison learning and confrontation training
CN113642639B (en) Living body detection method, living body detection device, living body detection equipment and storage medium
CN112488072A (en) Method, system and equipment for acquiring face sample set
CN112651333A (en) Silence living body detection method and device, terminal equipment and storage medium
CN115115552B (en) Image correction model training method, image correction device and computer equipment
CN113221830B (en) Super-division living body identification method, system, terminal and storage medium
Edmunds Protection of 2D face identification systems against spoofing attacks
CN114445916A (en) Living body detection method, terminal device and storage medium
CN113723310A (en) Image identification method based on neural network and related device
CN112395929A (en) Face living body detection method based on infrared image LBP histogram characteristics
CN111126283A (en) Rapid in-vivo detection method and system for automatically filtering fuzzy human face

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant