WO2022126917A1 - 基于深度学习的人脸图像评估方法、装置、设备及介质 - Google Patents

基于深度学习的人脸图像评估方法、装置、设备及介质 Download PDF

Info

Publication number
WO2022126917A1
WO2022126917A1 PCT/CN2021/083750 CN2021083750W WO2022126917A1 WO 2022126917 A1 WO2022126917 A1 WO 2022126917A1 CN 2021083750 W CN2021083750 W CN 2021083750W WO 2022126917 A1 WO2022126917 A1 WO 2022126917A1
Authority
WO
WIPO (PCT)
Prior art keywords
face
image
evaluation model
vector
image recognition
Prior art date
Application number
PCT/CN2021/083750
Other languages
English (en)
French (fr)
Inventor
陈丹
陆进
陈斌
刘玉宇
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN202011509136.7A external-priority patent/CN112529888B/zh
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2022126917A1 publication Critical patent/WO2022126917A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20081Training; Learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20084Artificial neural networks [ANN]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30168Image quality inspection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30196Human being; Person
    • G06T2207/30201Face

Definitions

  • the present application relates to the technical field of image recognition, and in particular, to a face image evaluation method, device, device and medium based on deep learning.
  • Image recognition is an important branch in the field of deep learning.
  • the quality of face images has a direct impact on face recognition, pedestrian Reid, live detection, and even OCR detection.
  • the degree of image blur is an essential factor in evaluating image quality, so accurately evaluating the image blur degree without reference becomes the key to the problem.
  • the method for evaluating the blur degree of an image is to add a blur attenuation factor (such as a Gaussian filter).
  • a blur attenuation factor such as a Gaussian filter
  • This method is to use an algorithm to generate data to evaluate the degree of blurring of the image; the inventor found that the data used for training in this method is difficult to fully simulate the complex fuzzy state in the real scene, resulting in insufficient evaluation of the degree of blurring of the image. Precise. There is an urgent need for a method that can improve the accuracy of image blurring assessment.
  • the purpose of the embodiments of the present application is to propose a method, apparatus, device and medium for evaluating a face image based on deep learning, so as to improve the accuracy of evaluating a face image.
  • an embodiment of the present application provides a method for evaluating a face image based on deep learning, including:
  • vector extraction is performed on the image recognition area to obtain a basic vector
  • Dimension reduction processing is performed on the basic vector to obtain a target vector, and the parameters of the face evaluation model are updated according to the target vector and the labeled data to obtain a trained face evaluation model;
  • an embodiment of the present application provides a face image evaluation device based on deep learning, including:
  • vector extraction is performed on the image recognition area to obtain a basic vector
  • a technical solution adopted in the present application is to provide a computer device, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the processor executes the computer-readable instructions Implement the following steps:
  • vector extraction is performed on the image recognition area to obtain a basic vector
  • Dimension reduction processing is performed on the basic vector to obtain a target vector, and the parameters of the face evaluation model are updated according to the target vector and the labeled data to obtain a trained face evaluation model;
  • a technical solution adopted in this application is: a computer-readable storage medium, where the computer-readable instructions are executed by a processor to implement the following steps:
  • vector extraction is performed on the image recognition area to obtain a basic vector
  • Dimension reduction processing is performed on the basic vector to obtain a target vector, and the parameters of the face evaluation model are updated according to the target vector and the labeled data to obtain a trained face evaluation model;
  • Embodiments of the present application provide a deep learning-based face image evaluation method, apparatus, device, and medium.
  • the face image is divided and its gradient value is calculated, and then the vector is extracted and processed, so as to train the face evaluation model, and then output the blur degree of the face image to be evaluated, which is beneficial to improve the face image. assessment accuracy.
  • FIG. 1 is a schematic diagram of an application environment of the deep learning-based face image evaluation method provided by an embodiment of the present application
  • FIG. 2 is a flowchart of an implementation of a deep learning-based face image evaluation method provided according to an embodiment of the present application
  • FIG. 6 is another implementation flowchart of the sub-process in the deep learning-based face image evaluation method provided by the embodiment of the present application.
  • FIG. 8 is another implementation flowchart of the sub-process in the deep learning-based face image evaluation method provided by the embodiment of the present application.
  • FIG. 9 is a schematic diagram of a face image evaluation device based on deep learning provided by an embodiment of the present application.
  • FIG. 10 is a schematic diagram of a computer device provided by an embodiment of the present application.
  • the system architecture 100 may include terminal devices 101 , 102 , and 103 , a network 104 and a server 105 .
  • the network 104 is a medium used to provide a communication link between the terminal devices 101 , 102 , 103 and the server 105 .
  • the network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.
  • the user can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages and the like.
  • Various communication client applications may be installed on the terminal devices 101 , 102 and 103 , such as web browser applications, search applications, instant communication tools, and the like.
  • the terminal devices 101, 102, 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.
  • the server 105 may be a server that provides various services, such as a background server that provides support for the pages displayed on the terminal devices 101 , 102 , and 103 .
  • the deep learning-based face image evaluation method provided in the embodiments of the present application is generally executed by a server, and accordingly, the deep learning-based face image evaluation apparatus is generally configured in the server.
  • terminal devices, networks and servers in FIG. 1 are merely illustrative. There can be any number of terminal devices, networks and servers according to implementation needs.
  • FIG. 2 shows a specific implementation manner of a face image evaluation method based on deep learning.
  • the method of the present application is not limited to the flow sequence shown in FIG. 2, and the method includes the following steps:
  • S1 Acquire a face image for training, and divide the face image into a plurality of regions of the same size according to a preset number as image recognition regions.
  • the embodiment of the present application divides the face image used for training, and divides the face image into the same size according to the required number of divisions, that is, according to the preset Set the number, divide the face image into regions of the same size, and use these regions of the same size as image recognition regions for subsequent evaluation of each image recognition region.
  • the preset number is set according to the actual situation, and is not limited here. In a specific embodiment, the preset number is four.
  • S2 Perform grayscale processing on each image recognition area to obtain a grayscale image corresponding to each image recognition area.
  • the embodiments of the present application will
  • the image recognition area is processed in grayscale, and the color image is converted into a grayscale image, and then a grayscale image corresponding to each image recognition area is obtained.
  • the gradient value of the grayscale image it is possible to avoid the situation where the gradient differences in the face image are partially too clear or too blurred to cancel each other out, and in the process of selecting the image recognition area, the difference in the hair of the characters in the face image can be avoided. And the interference caused by too many backgrounds makes the face and key organs (including eyes, nose and mouth) more obvious in the face image, which makes the evaluation of the blur degree of the face image more accurate.
  • the labeled data is generated.
  • the labeled data refers to the classification accuracy of the training set used for supervised training, and is mainly used in statistical models to verify or disprove a certain research hypothesis.
  • the labeled data is used as supervision information for subsequent training of the face evaluation model, which is convenient for updating the parameters of the face evaluation model.
  • the image recognition area is input into the face evaluation model, depth feature extraction is performed on the image recognition area through the face evaluation model, and vector calculation is performed on the depth feature to obtain a column vector, and the column vector is used as a basic vector.
  • the face evaluation model is based on a deep learning network.
  • the deep learning network learns the inherent laws and representation levels of the sample data. The information obtained during the learning process is of great help to the interpretation of data such as text and images. Its ultimate goal is to enable machines to have the ability to analyze and learn like humans, and to recognize data such as text and images.
  • the vector extraction means that the face evaluation model is based on the deep learning network, and the depth feature extraction is performed on the image recognition area, and then the extracted depth feature is subjected to vector calculation.
  • the basic vector refers to the column vector obtained after the vector extraction of the image recognition area.
  • the dimensionality reduction process refers to reducing the number of channels of the basic vector, and the number of parameters has been reduced.
  • dimensionality reduction processing is performed on the basic vector to reduce the number of channels of the basic vector, so as to reduce the amount of parameters, so as to obtain the target vector, which is convenient for the subsequent update of the face evaluation model.
  • the parameters of the face evaluation model are updated according to the target vector and the labeling data, and the detailed process of obtaining the trained face evaluation model is shown in steps S51-S54, which are not repeated here to avoid repetition.
  • S6 Obtain the face image to be evaluated, input the face image to be evaluated into the trained face evaluation model, and output the evaluation result corresponding to the face image to be evaluated.
  • the face image to be evaluated is input into the trained face evaluation model, and the trained face evaluation model will divide the face image to be evaluated into the recognition area, and the divided recognition area will be blurred.
  • the fuzzy score value of each recognition area is obtained, and then the fuzzy score value is compared with the preset fuzzy threshold value to obtain the evaluation result.
  • a face image is obtained, and the face image is divided into a plurality of regions of the same size according to a preset number, which are used as image recognition regions; the image recognition regions are subjected to grayscale processing to obtain the corresponding image recognition regions for each image recognition region.
  • the grayscale image calculates gradient value of the grayscale image to obtain the corresponding gradient value of the grayscale image, and obtain the labeled data according to the gradient value; according to the face evaluation model, extract the vector of the image recognition area to obtain the basic vector; The basic vector is dimensionally reduced to obtain the target vector, and the parameters of the face evaluation model are updated according to the target vector and the labeled data to obtain a trained face evaluation model; the face image to be evaluated is obtained, and the The face image is input into the trained face evaluation model, and the evaluation result corresponding to the face image to be evaluated is output.
  • the face image is divided and its gradient value is calculated, and then the vector is extracted and processed, so as to train the face evaluation model, and then output the blur degree of the face image to be evaluated, which is beneficial to improve the face image. assessment accuracy.
  • FIG. 3 shows a specific implementation of step S4.
  • step S4 according to the face evaluation model, vector extraction is performed on the image recognition area to obtain the specific implementation process of the basic vector, which is described in detail as follows:
  • the face evaluation model is constructed based on the deep learning network, that is, the deep feature extraction is performed on the image recognition area by means of the deep learning network.
  • the depth feature extraction mainly extracts the features of key parts in the face image, such as the face contour, eyes, mouth and other parts in the face image.
  • S42 Perform a pooling process on the depth feature by means of mean pooling to obtain a column vector corresponding to the depth feature, and use the column vector as a basic vector.
  • the depth feature is essentially a kind of vector data
  • the depth feature corresponding to the image recognition area is pooled to obtain column vectors, and the column vectors corresponding to the image recognition area are stored in tensors.
  • mean-pooling refers to averaging all the values in the local receptive field.
  • the depth feature of each image recognition area is extracted, and then the depth feature is pooled by means of mean pooling to obtain the column vector corresponding to the depth feature, and the column vector is used as
  • the basic vector realizes the vector extraction of the face image, which is convenient for the subsequent update of the parameters of the face evaluation model, thereby improving the evaluation accuracy of the blur degree of the face image.
  • FIG. 4 shows a specific implementation of step S5.
  • step S5 a dimensionality reduction process is performed on the basic vector to obtain a target vector, and the parameters of the face evaluation model are updated according to the target vector and the labeled data.
  • the specific implementation process of the trained face evaluation model quantity is described in detail as follows:
  • the number of channels of the basic vector is reduced, so as to reduce the amount of subsequent parameters and the amount of calculation, and finally obtain the target vector.
  • the target vector is calculated by the sigmoid function, and the calculation result is normalized to a score value between 0 and 1, which is convenient for subsequent calculation of the loss function.
  • the sigmoid function is a common sigmoid function in biology, also known as the sigmoid growth curve.
  • the sigmoid function is often used as the activation function of a neural network to map variables between 0 and 1 due to its mono-increase and inverse-function mono-increase properties.
  • the sigmoid function calculation is performed on the target vector, and the target vector is mapped between 0 and 1, so as to facilitate the subsequent calculation of the loss function value.
  • the loss function calculation of the present application adopts the L1 loss function calculation, wherein the L1 loss function is also called minimizing the absolute error, which is to minimize the sum of the absolute values of the difference between the actual value and the predicted value. Further, in the process of calculating the loss function, the labeling data is used to supervise, so as to reduce the error in the process of calculating the loss function.
  • the gradient of the loss value is returned, and the parameters of the face evaluation model are updated.
  • the face evaluation model has a better performance, that is, when the loss value is small, the parameter update is stopped, and the trained person is obtained. face evaluation model.
  • a target vector is obtained by performing dimensionality reduction processing on a basic vector, and a sigmoid function calculation is performed on the target vector to obtain a calculation result.
  • a loss function calculation is performed on the calculation result to obtain a loss value corresponding to the image recognition area.
  • update the parameters of the face evaluation model obtain a trained face evaluation model, and realize the training of the face evaluation model, which is conducive to the evaluation of the subsequent input of the face image to be evaluated.
  • the accuracy of the evaluation of the blurring degree of the face image can be improved.
  • FIG. 5 shows a specific implementation of step S54.
  • step S54 according to the loss value corresponding to the image recognition area, the parameters of the face evaluation model are updated to obtain the trained face evaluation model.
  • the specific implementation process is described in detail as follows:
  • each region has a corresponding loss value, and the loss values of all regions of each face image are added to obtain the loss value of the entire face image.
  • the loss value that is, the target loss value.
  • the face evaluation model because in the training process of the face evaluation model, not only one face image is used, but many face images are often used, and the target loss values corresponding to different face images are different. The gradient of the target loss value is returned, and the face evaluation model is gradually updated until the face evaluation model achieves better performance.
  • the target loss value reaches the preset threshold, it indicates that the face evaluation model has performed well.
  • the updating of the parameters of the face evaluation model can be stopped to obtain a trained face evaluation model.
  • the preset value is set according to the actual situation, which is not limited here. In a specific embodiment, the default value is 0.05.
  • the target loss value is obtained by adding up the loss values corresponding to all the image recognition areas, and the target loss value is returned to the gradient according to the method of gradient return, and the parameters of the face evaluation model are updated.
  • the target loss value reaches the preset value, stop updating the parameters of the face evaluation model, obtain a trained face evaluation model, and update the parameters of the face evaluation model with the target loss value. The accuracy of the blurriness assessment of the face image.
  • FIG. 6 shows a specific implementation of step S1.
  • step S1 a face image for training is obtained, and the face image is divided into multiple regions of the same size according to a preset number, as the specific implementation process of the image recognition area, the details are as follows:
  • a face image for training is first acquired.
  • S12 Scale the face image toward the center by a preset multiple to obtain a sampling area.
  • the edge of the face image is more of the background and the hair of the character.
  • the face image will be scaled to the center by a preset multiple. , to get the sampling area.
  • the preset multiple is set according to the actual situation, which is not limited here. In a specific embodiment, the preset multiple is 0.8 times.
  • S13 Divide the sampling area into a plurality of areas of the same size according to a preset number as image recognition areas.
  • the image recognition area can be acquired.
  • FIG. 7 shows a specific implementation of step S3.
  • step S3 the gradient value of the grayscale image is calculated to obtain the gradient value corresponding to the grayscale image, and according to the gradient value, the specific implementation of the labeled data is obtained.
  • the process is detailed as follows:
  • the gradient calculation methods include: numerical method, analytical method, and back-propagation method.
  • the preset gradient calculation method is not limited. In a specific embodiment, a numerical method is used to calculate the gradient value of the grayscale image.
  • S32 Set a gradient threshold, compare the gray value with the gradient threshold, and obtain labeled data, wherein if the gradient value is greater than the gradient threshold, the labeled data is 1, and if the gradient value is less than or equal to the gradient threshold, the labeled data is 0 .
  • the labeled data is used as supervision information in the subsequent training of the face evaluation model, so the gray value is converted into labeled data according to the gradient threshold.
  • the setting of the gradient threshold is set according to the actual situation, which is not limited here.
  • the gradient value of the grayscale image is calculated, the gradient value corresponding to the grayscale image is obtained, the gradient threshold is set, the grayscale value is compared with the gradient threshold, and the labeled data is obtained. It is beneficial to the subsequent training of the face evaluation model.
  • FIG. 8 shows a specific implementation of step S6.
  • step S6 a face image to be evaluated is obtained, and the face image to be evaluated is input into the trained face evaluation model, and the output
  • the specific implementation process of the evaluation result corresponding to the face image to be evaluated is described in detail as follows:
  • S61 Acquire the face image to be evaluated, and output the score value of the image recognition area corresponding to the face image to be evaluated in the trained face evaluation model.
  • the above steps have completed the training of the face evaluation model, and the trained face evaluation model has been obtained.
  • the face image needs to be evaluated, it is only necessary to input the obtained face image to be evaluated into the trained face.
  • the trained face evaluation model divides the face image to be evaluated into regions, and scores the blur degree of each region to obtain the score value of each region.
  • the score value is compared with the preset blur threshold to obtain an evaluation result of the face image to be evaluated, for example, the evaluation result is that the image is clear.
  • the setting of the preset blur threshold is set according to the actual situation, which is not limited here.
  • the preset threshold refers to setting multiple threshold ranges, and each threshold range corresponds to an evaluation result.
  • different threshold ranges correspond to evaluation results of clear, relatively clear, relatively fuzzy, and very fuzzy.
  • the score value of the image recognition area corresponding to the face image to be evaluated is output, and the score value is compared with the preset blur threshold.
  • the comparison is performed to obtain the evaluation result corresponding to the face image to be evaluated, and the evaluation of the face image to be evaluated is realized, which is beneficial to improve the accuracy of the evaluation of the blur degree of the face image.
  • the above-mentioned face image to be evaluated can also be stored in a node of a blockchain.
  • the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM) or the like.
  • the present application provides an embodiment of a face image evaluation device based on deep learning, which is similar to the method embodiment shown in FIG. 2 .
  • the apparatus can be specifically applied to various electronic devices.
  • the deep learning-based face image evaluation device in this embodiment includes: an image recognition area acquisition module 71 , an image recognition area processing module 72 , an annotation data acquisition module 73 , a basic vector acquisition module 74 , and a face evaluation module Model training module 75 and face image evaluation module 76, wherein:
  • the image recognition area acquisition module 71 is used to acquire face images for training, and divide the face images into multiple areas of the same size as image recognition areas according to a preset number;
  • the image recognition area processing module 72 is used to perform grayscale processing on each image recognition area to obtain a grayscale map corresponding to each image recognition area;
  • An annotation data acquisition module 73 configured to calculate the gradient value of the grayscale image, obtain the gradient value corresponding to the grayscale image, and obtain the annotation data according to the gradient value;
  • the basic vector obtaining module 74 is used for performing vector extraction on the image recognition area according to the face evaluation model to obtain the basic vector;
  • the face evaluation model training module 75 is used to perform dimensionality reduction processing on the basic vector to obtain a target vector, and update the parameters of the face evaluation model according to the target vector and the labeled data to obtain a trained face evaluation model;
  • the basis vector acquisition module 74 includes:
  • the depth feature extraction unit is used to extract the depth feature of each image recognition area according to the face evaluation model
  • the pooling processing unit is used to perform pooling processing on the depth feature by means of mean pooling to obtain the column vector corresponding to the depth feature, and use the column vector as the basic vector.
  • the face evaluation model training module 75 includes:
  • the target vector acquisition unit is used to perform dimension reduction processing on the basic vector to obtain the target vector;
  • the calculation result obtaining unit is used to perform sigmoid function calculation on the target vector to obtain the calculation result;
  • the loss function calculation unit is used to perform loss function calculation on the calculation result based on the labeled data, and obtain the loss value corresponding to the image recognition area;
  • the parameter updating unit is used to update the parameters of the face evaluation model according to the loss value corresponding to the image recognition area to obtain a trained face evaluation model.
  • the parameter updating unit includes:
  • the target loss value acquisition sub-unit is used to add the loss values corresponding to all image recognition areas to obtain the target loss value
  • the target loss value return sub-unit is used for gradient return of the target loss value according to the method of gradient return to update the parameters of the face evaluation model;
  • the parameter update stop subunit is used to stop updating the parameters of the face evaluation model when the target loss value reaches the preset value, so as to obtain a trained face evaluation model.
  • the image recognition area acquisition module 71 includes:
  • a face image acquisition unit used to acquire face images for training
  • the sampling area confirmation unit is used to scale the face image to the center by a preset multiple to obtain the sampling area;
  • the image recognition area determination unit is configured to divide the sampling area into a plurality of areas of the same size according to a preset number as image recognition areas.
  • annotation data acquisition module 73 includes:
  • the gradient calculation unit is used to calculate the gradient value of the grayscale image according to the preset gradient calculation method, and obtain the gradient value corresponding to the grayscale image;
  • the labeled data determination unit is used to set the gradient threshold, compare the gray value with the gradient threshold, and obtain the labeled data, wherein, if the gradient value is greater than the gradient threshold, the labeled data is 1, and if the gradient value is less than or equal to the gradient threshold, The label data is 0.
  • the face image evaluation module 76 includes:
  • a score value obtaining unit used for obtaining the face image to be evaluated, and outputting the score value of the image recognition area corresponding to the face image to be evaluated in the trained face evaluation model
  • the evaluation result obtaining unit is configured to compare the score value with the preset blur threshold to obtain the evaluation result corresponding to the face image to be evaluated.
  • the above-mentioned face image to be evaluated can also be stored in a node of a blockchain.
  • FIG. 10 is a block diagram of the basic structure of a computer device according to this embodiment.
  • the computer device 8 includes a memory 81 , a processor 82 , and a network interface 83 that are connected to each other through a system bus. It should be pointed out that the figure only shows the computer device 8 with three components, the memory 81, the processor 82, and the network interface 83, but it should be understood that it is not required to implement all the components shown, and alternative implementations are possible. More or fewer components.
  • the computer device here is a device that can automatically perform numerical calculation and/or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, special-purpose Integrated circuit (Application Specific Integrated Circuit, ASIC), programmable gate array (Field-Programmable Gate Array, FPGA), digital processor (Digital Signal Processor, DSP), embedded equipment, etc.
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • DSP Digital Signal Processor
  • embedded equipment etc.
  • the computer equipment may be a desktop computer, a notebook computer, a palmtop computer, and a cloud server and other computing equipment.
  • Computer devices can interact with users through keyboards, mice, remote controls, touchpads, or voice-activated devices.
  • the memory 81 includes at least one type of readable storage medium, the computer-readable storage medium may be non-volatile or volatile, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (eg , SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM) ), magnetic memory, magnetic disk, optical disk, etc.
  • the memory 81 may be an internal storage unit of the computer device 8 , such as a hard disk or memory of the computer device 8 .
  • the memory 81 may also be an external storage device of the computer device 8, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, flash memory card (Flash Card), etc.
  • the memory 81 may also include both the internal storage unit of the computer device 8 and its external storage device.
  • the memory 81 is generally used to store the operating system and various application software installed on the computer device 8 , such as computer-readable instructions for the deep learning-based face image evaluation method, and the like.
  • the memory 81 can also be used to temporarily store various types of data that have been output or will be output.
  • the processor 82 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments.
  • the processor 82 is typically used to control the overall operation of the computer device 8 .
  • the processor 82 is configured to run computer-readable instructions or process data stored in the memory 81, for example, run the computer-readable instructions of the above-mentioned deep learning-based face image evaluation method, so as to realize a deep learning-based face image evaluation method.
  • image evaluation methods may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments.
  • the processor 82 is typically used to control the overall operation of the computer device 8 .
  • the processor 82 is configured to run computer-readable instructions or process data stored in the memory 81, for example, run the computer-readable instructions of the above-mentioned deep learning-based face image evaluation method, so as to realize a deep learning-based face image evaluation method.
  • image evaluation methods Various embodiments of image
  • the network interface 83 may comprise a wireless network interface or a wired network interface, and the network interface 83 is typically used to establish a communication connection between the computer device 8 and other electronic devices.
  • the present application also provides another embodiment, that is, to provide a computer-readable storage medium, where the computer-readable storage medium stores computer-readable instructions of a computer, and the computer-readable instructions of the computer can be executed by at least one processor to At least one processor is caused to execute the steps of the above-mentioned deep learning-based face image evaluation method.
  • the methods of the above embodiments can be implemented by means of software plus a necessary general hardware platform, and of course hardware can also be used, but in many cases the former is better implementation.
  • the technical solution of the present application can be embodied in the form of a software product in essence or in a part that contributes to the prior art, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, CD-ROM), including several instructions to make a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods of the various embodiments of the present application.
  • a storage medium such as ROM/RAM, magnetic disk, CD-ROM
  • the blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information to verify its Validity of information (anti-counterfeiting) and generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

Abstract

一种基于深度学习的人脸图像评估方法、装置、设备及介质,其中方法包括获取用来训练的人脸图像,并获取其图像识别区域;将图像识别区域进行灰度处理,得到灰度图,再计算灰度图的梯度值,得到灰度图对应的梯度值,根据人脸评估模型进行向量提取,将得到的基础向量进行降维处理,得到目标向量,并根据目标向量和标注数据对人脸评估模型的参数进行更新,得到训练好的人脸评估模型;将待评估的人脸图像输入到训练好的人脸评估模型中,输出评估结果。还涉及区块链技术,待评估的人脸图像存储于区块链中。该方法通过训练人脸评估模型,进而输出待评估的人脸图像模糊程度,有利于提高人脸图像的评估精准度。

Description

基于深度学习的人脸图像评估方法、装置、设备及介质
本申请要求于2020年12月18日提交中国专利局、申请号为202011509136.7,发明名称为“基于深度学习的人脸图像评估方法、装置、设备及介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本申请涉及图像识别技术领域,尤其涉及一种基于深度学习的人脸图像评估方法、装置、设备及介质。
背景技术
图像识别是深度学习领域一个重要分支,人脸图像的质量对人脸识别、行人Reid、活体检测甚至OCR检测等都有直接的影响。而图像模糊程度是评估图像质量中一个必不可少因素,因此准确评估无参考情况下图像的模糊程度就成为了问题的关键。
评估图像的模糊程度涉及深度学习技术。目前在无参考情况下,对图像的模糊程度的评估方法采用的是添加模糊衰减因子(如高斯滤波器)的方式。这种方式是通过采用生成数据的算法,评估出图像的模糊程度;发明人发现,这种方式用于训练的数据难以完全模拟真实场景中复杂的模糊状态,导致对图像的模糊程度的评估不够精准。现亟需一种能够提高对图像模糊程度评估精准度的方法。
发明内容
本申请实施例的目的在于提出一种基于深度学习的人脸图像评估方法、装置、设备及介质,以提高对人脸图像评估的精准度。
为了解决上述技术问题,本申请实施例提供一种基于深度学习的人脸图像评估方法,包括:
获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域;
将每个所述图像识别区域进行灰度处理,得到每个所述图像识别区域对应的灰度图;
计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据;
根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量;
对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型;
获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果。
为了解决上述技术问题,本申请实施例提供一种基于深度学习的人脸图像评估装置,包括:
获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域;
将每个所述图像识别区域进行灰度处理,得到每个所述图像识别区域对应的灰度图;
计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据;
根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量;
对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型;
获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果。
为解决上述技术问题,本申请采用的一个技术方案是:提供一种计算机设备,包括存储器和处理器,所述存储器中存储有计算机可读指令,所述处理器执行所述计算机可读指 令时实现如下步骤:
获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域;
将每个所述图像识别区域进行灰度处理,得到每个所述图像识别区域对应的灰度图;
计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据;
根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量;
对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型;
获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果。
为解决上述技术问题,本申请采用的一个技术方案是:一种计算机可读存储介质,所述计算机可读指令被处理器执行时实现如下步骤:
获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域;
将每个所述图像识别区域进行灰度处理,得到每个所述图像识别区域对应的灰度图;
计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据;
根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量;
对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型;
获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果。
本申请实施例提供了一种基于深度学习的人脸图像评估方法、装置、设备及介质。本申请实施例通过对人脸图像进行划分并计算其梯度值,再进行向量提取和处理,以此来训练人脸评估模型,进而输出待评估的人脸图像模糊程度,有利于提高人脸图像的评估精准度。
附图说明
为了更清楚地说明本申请中的方案,下面将对本申请实施例描述中所需要使用的附图作一个简单介绍,显而易见地,下面描述中的附图是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1是本申请实施例提供的基于深度学习的人脸图像评估方法的应用环境示意图;
图2根据本申请实施例提供的基于深度学习的人脸图像评估方法的一实现流程图;
图3是本申请实施例提供的基于深度学习的人脸图像评估方法中子流程的一实现流程图;
图4是本申请实施例提供的基于深度学习的人脸图像评估方法中子流程的又一实现流程图;
图5是本申请实施例提供的基于深度学习的人脸图像评估方法中子流程的又一实现流程图;
图6是本申请实施例提供的基于深度学习的人脸图像评估方法中子流程的又一实现流程图;
图7是本申请实施例提供的基于深度学习的人脸图像评估方法中子流程的又一实现流程图;
图8是本申请实施例提供的基于深度学习的人脸图像评估方法中子流程的又一实现流程图;
图9是本申请实施例提供的基于深度学习的人脸图像评估装置示意图;
图10是本申请实施例提供的计算机设备的示意图。
具体实施方式
除非另有定义,本文所使用的所有的技术和科学术语与属于本申请的技术领域的技术人员通常理解的含义相同;本文中在申请的说明书中所使用的术语只是为了描述具体的实施例的目的,不是旨在于限制本申请;本申请的说明书和权利要求书及上述附图说明中的术语“包括”和“具有”以及它们的任何变形,意图在于覆盖不排他的包含。本申请的说明书和权利要求书或上述附图中的术语“第一”、“第二”等是用于区别不同对象,而不是用于描述特定顺序。
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申请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。
为了使本技术领域的人员更好地理解本申请方案,下面将结合附图,对本申请实施例中的技术方案进行清楚、完整地描述。
下面结合附图和实施方式对本申请进行详细说明。
请参阅图1,系统架构100可以包括终端设备101、102、103,网络104和服务器105。网络104用以在终端设备101、102、103和服务器105之间提供通信链路的介质。网络104可以包括各种连接类型,例如有线、无线通信链路或者光纤电缆等等。
用户可以使用终端设备101、102、103通过网络104与服务器105交互,以接收或发送消息等。终端设备101、102、103上可以安装有各种通讯客户端应用,例如网页浏览器应用、搜索类应用、即时通信工具等。
终端设备101、102、103可以是具有显示屏并且支持网页浏览的各种电子设备,包括但不限于智能手机、平板电脑、膝上型便携计算机和台式计算机等等。
服务器105可以是提供各种服务的服务器,例如对终端设备101、102、103上显示的页面提供支持的后台服务器。
需要说明的是,本申请实施例所提供的基于深度学习的人脸图像评估方法一般由服务器执行,相应地,基于深度学习的人脸图像评估装置一般配置于服务器中。
应该理解,图1中的终端设备、网络和服务器的数目仅仅是示意性的。根据实现需要,可以具有任意数目的终端设备、网络和服务器。
请参阅图2,图2示出了基于深度学习的人脸图像评估方法的一种具体实施方式。
需注意的是,若有实质上相同的结果,本申请的方法并不以图2所示的流程顺序为限,该方法包括如下步骤:
S1:获取用于训练的人脸图像,并按照预设数量,将人脸图像划分为多个相同大小的区域,作为图像识别区域。
具体的,为了能够将人脸图像进行模糊程度的评估,本申请实施例将用于训练的人脸图像进行划分,并且根据需要划分的数量,将人脸图像划分为相同大小,也即按照预设数量,将人脸图像划分为相同大小的区域,并将这些相同大小的区域作为图像识别区域,以便以后续对每个图像识别区域进行评估。
需要说明的是,预设数量根据实际情况进行设定,此处不做限定。在一具体实施例中,预设数量为4个。
S2:将每个图像识别区域进行灰度处理,得到每个图像识别区域对应的灰度图。
具体的,由于光照、背景颜色等因素会对人脸图像的模糊程度的评估产生影响,故为了减少这些因素的影响,提高对人脸图像的模糊程度的评估的准确度,本申请实施例会对图像识别区域进行灰度处理,将彩色的图像转化成灰度图像,进而得到每个图像识别区域对应的灰度图。
S3:计算灰度图的梯度值,得到灰度图对应的梯度值,并根据梯度值,得到标注数据。
具体的,通过计算灰度图的梯度值,能够避免人脸图像局部过于清晰或者过于模糊等因素导致梯度差异相互抵消的情况,并且在图像识别区域选取过程中,避免人脸图像中人物头发差异以及背景过多造成的干扰,使得人脸图像中人物脸部及关键器官(包括眼睛、鼻子嘴巴)更加明显,从而使得人脸图像的模糊程度评估更加精准。
具体的,根据梯度值,生成标注数据。该标注数据指的是用于有监督训练的训练集的分类准确性,主要用于统计模型中验证或推翻某种研究假设。在本申请实施例中,标注数据作为后续人脸评估模型训练的监督信息,便于对人脸评估模型的参数的更新。
S4:根据人脸评估模型,对图像识别区域进行向量提取,得到基础向量。
具体的,将图像识别区域输入到人脸评估模型中,通过人脸评估模型对图像识别区域进行深度特征提取,再对深度特征进行向量计算,得到列向量,再将列向量作为基础向量。人脸评估模型是基于深度学习网络进行架构的,该深度学习网络是学习样本数据的内在规律和表示层次,这些学习过程中获得的信息对文字、图像等数据的解释有很大的帮助。它的最终目标是让机器能够像人一样具有分析学习能力,能够识别文字、图像等数据。
其中,向量提取是指人脸评估模型基于深度学习网络,对图像识别区域进行深度特征提取,再将提取的深度特征进行向量计算。基础向量是指对图像识别区域进行向量提取后,得到的列向量。
S5:对基础向量进行降维处理,得到目标向量,并根据目标向量和标注数据对人脸评估模型的参数进行更新,得到训练好的人脸评估模型。
具体的,降维处理是指将基础向量的通道数进行减少,已实现减少参数量。在本申请中,对基础向量进行降维处理,以减少基础向量的通道数,以减少参数量,从而得到目标向量,便于后续对人脸评估模型的更新。进一步的,根据目标向量和标注数据对人脸评估模型的参数进行更新,得到训练好的人脸评估模型的详细过程见步骤S51-S54,为避免赘述,此处不再重复。
S6:获取待评估的人脸图像,并将待评估的人脸图像输入到训练好的人脸评估模型中,输出待评估的人脸图像对应的评估结果。
具体的,将待评估的人脸图像输入到训练好的人脸评估模型中,训练好的人脸评估模型会对待评估的人脸图像进行划分识别区域,并对划分的识别区域进行模糊程度的打分,得到每个识别区域的模糊得分值,再将该模糊得分值与预设的模糊阈值进行比较,从而得到评估结果。
本实施例中,获取人脸图像,并按照预设数量,将人脸图像划分为多个相同大小的区域,作为图像识别区域;将图像识别区域进行灰度处理,得到每个图像识别区域对应的灰度图;计算灰度图的梯度值,得到灰度图对应的梯度值,并根据梯度值,得到标注数据;根据人脸评估模型,对图像识别区域进行向量提取,得到基础向量;对基础向量进行降维处理,得到目标向量,并根据目标向量和标注数据对人脸评估模型的参数进行更新,得到训练好的人脸评估模型;获取待评估的人脸图像,并将待评估的人脸图像输入到训练好的人脸评估模型中,输出待评估的人脸图像对应的评估结果。本申请实施例通过对人脸图像进行划分并计算其梯度值,再进行向量提取和处理,以此来训练人脸评估模型,进而输出待评估的人脸图像模糊程度,有利于提高人脸图像的评估精准度。
请参阅图3,图3示出了步骤S4的一种具体实施方式,步骤S4中根据人脸评估模型,对图像识别区域进行向量提取,得到基础向量的具体实现过程,详叙如下:
S41:根据人脸评估模型,提取每个图像识别区域的深度特征。
具体的,由于人脸评估模型是基于深度学习网络进行架构的,也即通过深度学习网络的方式,对图像识别区域进行深度特征提取。该深度特征提取主要提取人脸图像中关键部位的特征,例如人脸图像中的人脸脸部轮廓、眼睛、嘴巴等部位。
S42:通过均值池化的方式,对深度特征进行池化处理,得到深度特征对应的列向量,并将列向量作为基础向量。
具体的,由于深度特征本质上也是一种向量数据,通过均值池化的方式,对图像识别区域对应的深度特征进行池化处理,得到列向量,图像识别区域对应的列向量都对应存储在张量中。其中,均值池化(mean-pooling)是指对局部接受域中的所有值求均值。
在本实施中,根据人脸评估模型,提取每个图像识别区域的深度特征,再通过均值池化的方式,对深度特征进行池化处理,得到深度特征对应的列向量,并将列向量作为基础向量,实现对人脸图像进行向量提取,便于后续对人脸评估模型参数的更新,进而提高人脸图像的模糊程度的评估准确度。
请参阅图4,图4示出了步骤S5的一种具体实施方式,步骤S5中对基础向量进行降维处理,得到目标向量,并根据目标向量和标注数据对人脸评估模型的参数进行更新,得到训练好的人脸评估模型量的具体实现过程,详叙如下:
S51:对基础向量进行降维处理,得到目标向量。
具体的,通过对基础向量进行降维处理,减少基础向量的通道数,以减少后续的参数量,减少计算量,最终得到目标向量。
S52:对目标向量进行sigmoid函数计算,得到计算结果。
具体的,将目标向量进行sigmoid函数计算,将计算结果归一化到0-1之间的分数值,便于后续进行损失函数计算。
其中,sigmoid函数是一个在生物学中常见的S型函数,也称为S型生长曲线。在信息科学中,由于其单增以及反函数单增等性质,sigmoid函数常被用作神经网络的激活函数,将变量映射到0,1之间。在本申请实施例中,对目标向量进行sigmoid函数计算,将目标向量映射到0,1之间,便于后续计算损失函数值。
S53:基于标注数据,对计算结果进行损失函数计算,得到图像识别区域对应的损失值。
具体的,本申请的损失函数计算采用的是L1损失函数计算,其中,L1损失函数也被称为最小化绝对误差,就是最小化真实值和预测值之间差值的绝对值的和。进一步的,在损失函数计算过程中,通过标注数据进行监督,以减少损失函数计算过程中的误差。
S54:根据图像识别区域对应的损失值,对人脸评估模型的参数进行更新,得到训练好的人脸评估模型。
具体的,将损失值进行梯度回传,对人脸评估模型的参数进行更新,待人脸评估模型有较好的表现时,也即损失值较小时,停止对参数的更新,得到训练好的人脸评估模型。
本实施例中,通过对基础向量进行降维处理,得到目标向量,对目标向量进行sigmoid函数计算,得到计算结果,基于标注数据,对计算结果进行损失函数计算,得到图像识别区域对应的损失值,根据图像识别区域对应的损失值,对人脸评估模型的参数进行更新,得到训练好的人脸评估模型,实现对人脸评估模型的训练,有利于后续输入待评估的人脸图像的评估结果,进而实现提高人脸图像的模糊程度评估的准确度。
请参阅图5,图5示出了步骤S54的一种具体实施方式,步骤S54中根据图像识别区域对应的损失值,对人脸评估模型的参数进行更新,得到训练好的人脸评估模型的具体实现过程,详叙如下:
S541:将所有的图像识别区域对应的损失值进行相加,得到目标损失值。
具体的,由于人脸图像是经过划分成不同区域的,每个区域都有对应的损失值,将每张人脸图像所有的区域的损失值进行相加,便可以得到整张人脸图像的损失值,也即目标损失值。
S542:按照梯度回传的方式,将目标损失值进行梯度回传,对人脸评估模型的参数进行更新。
具体的,由于在对人脸评估模型的训练过程中,不单单采用一张人脸图像,往往采用的是许多张人脸图像,不同的人脸图像对应的目标损失值有所不同,将这些目标损失值进行梯度回传,逐渐对人脸评估模型进行更新,直至人脸评估模型得到较好的表现。
S543:当目标损失值达到预设值时,停止对人脸评估模型的参数的更新,得到训练好的人脸评估模型。
具体的,当目标损失值达到预设阈值时,说明人脸评估模型已经有了较好的表现,此时可以停止对人脸评估模型的参数的更新,得到训练好的人脸评估模型。
需要说明的是,预设值根据实际情况进行设定,此处不做限定。在一具体实施例中,预设值为0.05。
本实施例中,通过将所有的图像识别区域对应的损失值进行相加,得到目标损失值,按照梯度回传的方式,将目标损失值进行梯度回传,对人脸评估模型的参数进行更新,当目标损失值达到预设值时,停止对人脸评估模型的参数的更新,得到训练好的人脸评估模型,实现将目标损失值对人脸评估模型的参数进行更新,有利于提高人脸图像的模糊程度评估的准确度。
请参阅图6,图6示出了步骤S1的一种具体实施方式,步骤S1中获取用于训练的人脸图像,并按照预设数量,将人脸图像划分为多个相同大小的区域,作为图像识别区域的具体实现过程,详叙如下:
S11:获取用于训练的人脸图像。
具体的,为了对人脸评估模型进行训练,首先会获取用于训练的人脸图像。
S12:将人脸图像向中心缩放预设倍数,得到取样区域。
具体的,人脸图像的边缘更多的是人物背景和人物头发,为了减少这些人脸图像的边缘因素对后续人脸评估图像的参数更新的影响,会对人脸图像向中心缩放预设倍数,得到取样区域。
需要说明的是,预设倍数根据实际情况进行设定,此处不做限定。在一具体实施方式中,预设倍数为0.8倍。
S13:按照预设数量,将取样区域划分为多个相同大小的区域,作为图像识别区域。
具体的,由于上述步骤已经获取到取样区域,只要按照预设数量,将取样区域划分为多个相同大小的区域,即可获取到图像识别区域。
本实施例中,通过获取用于训练的人脸图像,将人脸图像向中心放缩预设倍数,得到取样区域,按照预设数量,将取样区域划分为相同大小的区域,作为图像识别区域,有利于减少其他因素对人脸图像的影响,进而有利于提高人脸图像的模糊程度评估的准确度。
请参阅图7,图7示出了步骤S3的一种具体实施方式,步骤S3中计算灰度图的梯度值,得到灰度图对应的梯度值,并根据梯度值,得到标注数据的具体实现过程,详叙如下:
S31:按照预设的梯度计算方式,计算灰度图的梯度值,得到灰度图对应的梯度值。
具体的,梯度计算方式包括:数值法,解析法,反向传播法。
需要说明的是,预设的梯度计算方式不做限定,在一具体实施例中,采用数值法计算灰度图的梯度值。
S32:设定梯度阈值,将灰度值与梯度阈值进行比较,得到标注数据,其中,若梯度值大于梯度阈值,则标注数据为1,若梯度值小于或等于梯度阈值,则标注数据为0。
具体的,标注数据用作于后续人脸评估模型训练中的监督信息,所以根据梯度阈值,将灰度值进行转化为标注数据。
需要说明的是,梯度阈值的设定根据实际情况进行设定,此处不做限定。
本实施例中,按照预设的梯度计算方式,计算灰度图的梯度值,得到灰度图对应的梯度值,设定梯度阈值,将灰度值与梯度阈值进行比较,得到标注数据,有利于后续对人脸评估模型的训练。
请参阅图8,图8示出了步骤S6的一种具体实施方式,步骤S6中获取待评估的人脸图像,并将待评估的人脸图像输入到训练好的人脸评估模型中,输出待评估的人脸图像对应的评估结果的具体实现过程,详叙如下:
S61:获取待评估的人脸图像,并通过训练好的人脸评估模型中,输出待评估的人脸 图像对应图像识别区域的得分值。
具体的,上述步骤已经对人脸评估模型训练完毕,获得了训练好的人脸评估模型,在需要对人脸图像进行评估时,只要将获取待评估的人脸图像输入到训练好的人脸评估模型中,训练好的人脸评估模型对待评估的人脸图像进行区域划分,并对每个区域进行模糊程度的打分,得到每个区域的得分值。
S62:将得分值与预设模糊阈值进行对比,得到待评估的人脸图像对应的评估结果。
具体的,将得分值与预设模糊阈值进行对比,得到待评估的人脸图像的评估结果,例如,评估结果为图像清晰。
需要说明的是,预设模糊阈值的设定根据实际情况进行设定,此处不做限定。预设阈值是指设置多个阈值范围,每个阈值范围对应一种评估结果,例如,不同阈值范围对应评估结果为清晰、较清晰、较模糊以及非常模糊等。
本实施例中,通过获取待评估的人脸图像,并通过训练好的人脸评估模型中,输出待评估的人脸图像对应图像识别区域的得分值,将得分值与预设模糊阈值进行对比,得到待评估的人脸图像对应的评估结果,实现对待评估的人脸图像进行评估,有利于提高人脸图像的模糊程度评估的准确度。
需要强调的是,为进一步保证上述待评估的人脸图像的私密和安全性,上述待评估的人脸图像还可以存储于一区块链的节点中。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机的计算机可读指令来指令相关的硬件来完成,该计算机的计算机可读指令可存储于一计算机可读取存储介质中,该计算机可读指令在执行时,可包括如上述各方法的实施例的流程。其中,前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory,ROM)等非易失性存储介质,或随机存储记忆体(Random Access Memory,RAM)等。
请参考图9,作为对上述图2所示方法的实现,本申请提供了一种基于深度学习的人脸图像评估装置的一个实施例,该装置实施例与图2所示的方法实施例相对应,该装置具体可以应用于各种电子设备中。
如图9所示,本实施例的基于深度学习的人脸图像评估装置包括:图像识别区域获取模块71、图像识别区域处理模块72、标注数据获取模块73、基础向量获取模块74、人脸评估模型训练模块75及人脸图像评估模块76,其中:
图像识别区域获取模块71,用于获取用于训练的人脸图像,并按照预设数量,将人脸图像划分为多个相同大小的区域,作为图像识别区域;
图像识别区域处理模块72,用于将每个图像识别区域进行灰度处理,得到每个图像识别区域对应的灰度图;
标注数据获取模块73,用于计算灰度图的梯度值,得到灰度图对应的梯度值,并根据梯度值,得到标注数据;
基础向量获取模块74,用于根据人脸评估模型,对图像识别区域进行向量提取,得到基础向量;
人脸评估模型训练模块75,用于对基础向量进行降维处理,得到目标向量,并根据目标向量和标注数据对人脸评估模型的参数进行更新,得到训练好的人脸评估模型;
人脸图像评估模块76,用于获取待评估的人脸图像,并将待评估的人脸图像输入到训练好的人脸评估模型中,输出待评估的人脸图像对应的评估结果。
进一步的,基础向量获取模块74包括:
深度特征提取单元,用于根据人脸评估模型,提取每个图像识别区域的深度特征;
池化处理单元,用于通过均值池化的方式,对深度特征进行池化处理,得到深度特征对应的列向量,并将列向量作为基础向量。
进一步的,人脸评估模型训练模块75包括:
目标向量获取单元,用于对基础向量进行降维处理,得到目标向量;
计算结果获取单元,用于对目标向量进行sigmoid函数计算,得到计算结果;
损失函数计算单元,用于基于标注数据,对计算结果进行损失函数计算,得到图像识别区域对应的损失值;
参数更新单元,用于根据图像识别区域对应的损失值,对人脸评估模型的参数进行更新,得到训练好的人脸评估模型。
进一步的,参数更新单元包括:
目标损失值获取子单元,用于将所有的图像识别区域对应的损失值进行相加,得到目标损失值;
目标损失值回传子单元,用于按照梯度回传的方式,将目标损失值进行梯度回传,对人脸评估模型的参数进行更新;
参数更新停止子单元,用于当目标损失值达到预设值时,停止对人脸评估模型的参数的更新,得到训练好的人脸评估模型。
进一步的,图像识别区域获取模块71包括:
人脸图像获取单元,用于获取用于训练的人脸图像;
取样区域确认单元,用于将人脸图像向中心缩放预设倍数,得到取样区域;
图像识别区域确定单元,用于按照预设数量,将取样区域划分为多个相同大小的区域,作为图像识别区域。
进一步的,标注数据获取模块73包括:
梯度计算单元,用于按照预设的梯度计算方式,计算灰度图的梯度值,得到灰度图对应的梯度值;
标注数据确定单元,用于设定梯度阈值,将灰度值与梯度阈值进行比较,得到标注数据,其中,若梯度值大于梯度阈值,则标注数据为1,若梯度值小于或等于梯度阈值,则标注数据为0。
进一步的,人脸图像评估模块76包括:
得分值获取单元,用于获取待评估的人脸图像,并通过训练好的人脸评估模型中,输出待评估的人脸图像对应图像识别区域的得分值;
评估结果获取单元,用于将得分值与预设模糊阈值进行对比,得到待评估的人脸图像对应的评估结果。
需要强调的是,为进一步保证上述待评估的人脸图像的私密和安全性,上述待评估的人脸图像还可以存储于一区块链的节点中。
为解决上述技术问题,本申请实施例还提供计算机设备。具体请参阅图10,图10为本实施例计算机设备基本结构框图。
计算机设备8包括通过系统总线相互通信连接存储器81、处理器82、网络接口83。需要指出的是,图中仅示出了具有三种组件存储器81、处理器82、网络接口83的计算机设备8,但是应理解的是,并不要求实施所有示出的组件,可以替代的实施更多或者更少的组件。其中,本技术领域技术人员可以理解,这里的计算机设备是一种能够按照事先设定或存储的指令,自动进行数值计算和/或信息处理的设备,其硬件包括但不限于微处理器、专用集成电路(Application Specific Integrated Circuit,ASIC)、可编程门阵列(Field-Programmable Gate Array,FPGA)、数字处理器(Digital Signal Processor,DSP)、嵌入式设备等。
计算机设备可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。计算机设备可以与用户通过键盘、鼠标、遥控器、触摸板或声控设备等方式进行人机交互。
存储器81至少包括一种类型的可读存储介质,所述计算机可读存储介质可以是非易失性,也可以是易失性,可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如,SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性 存储器、磁盘、光盘等。在一些实施例中,存储器81可以是计算机设备8的内部存储单元,例如该计算机设备8的硬盘或内存。在另一些实施例中,存储器81也可以是计算机设备8的外部存储设备,例如该计算机设备8上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。当然,存储器81还可以既包括计算机设备8的内部存储单元也包括其外部存储设备。本实施例中,存储器81通常用于存储安装于计算机设备8的操作系统和各类应用软件,例如基于深度学习的人脸图像评估方法的计算机可读指令等。此外,存储器81还可以用于暂时地存储已经输出或者将要输出的各类数据。
处理器82在一些实施例中可以是中央处理器(Central Processing Unit,CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器82通常用于控制计算机设备8的总体操作。本实施例中,处理器82用于运行存储器81中存储的计算机可读指令或者处理数据,例如运行上述基于深度学习的人脸图像评估方法的计算机可读指令,以实现基于深度学习的人脸图像评估方法的各种实施例。
网络接口83可包括无线网络接口或有线网络接口,该网络接口83通常用于在计算机设备8与其他电子设备之间建立通信连接。
本申请还提供了另一种实施方式,即提供一种计算机可读存储介质,计算机可读存储介质存储有计算机的计算机可读指令,计算机的计算机可读指令可被至少一个处理器执行,以使至少一个处理器执行如上述的一种基于深度学习的人脸图像评估方法的步骤。
通过以上的实施方式的描述,本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现,当然也可以通过硬件,但很多情况下前者是更佳的实施方式。基于这样的理解,本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中,包括若干指令用以使得一台终端设备(可以是手机,计算机,服务器,空调器,或者网络设备等)执行本申请各个实施例的方法。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain),本质上是一个去中心化的数据库,是一串使用密码学方法相关联产生的数据块,每一个数据块中包含了一批次网络交易的信息,用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
显然,以上所描述的实施例仅仅是本申请一部分实施例,而不是全部的实施例,附图中给出了本申请的较佳实施例,但并不限制本申请的专利范围。本申请可以以许多不同的形式来实现,相反地,提供这些实施例的目的是使对本申请的公开内容的理解更加透彻全面。尽管参照前述实施例对本申请进行了详细的说明,对于本领域的技术人员来而言,其依然可以对前述各具体实施方式所记载的技术方案进行修改,或者对其中部分技术特征进行等效替换。凡是利用本申请说明书及附图内容所做的等效结构,直接或间接运用在其他相关的技术领域,均同理在本申请专利保护范围之内。

Claims (20)

  1. 一种基于深度学习的人脸图像评估方法,包括:
    获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域;
    将每个所述图像识别区域进行灰度处理,得到每个所述图像识别区域对应的灰度图;
    计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据;
    根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量;
    对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型;
    获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果。
  2. 根据权利要求1所述的基于深度学习的人脸图像评估方法,其中,所述根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量,包括:
    根据所述人脸评估模型,提取每个所述图像识别区域的深度特征;
    通过均值池化的方式,对所述深度特征进行池化处理,得到所述深度特征对应的列向量,并将所述列向量作为所述基础向量。
  3. 根据权利要求1所述的基于深度学习的人脸图像评估方法,其中,所述对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型,包括:
    对所述基础向量进行降维处理,得到目标向量;
    对所述目标向量进行sigmoid函数计算,得到计算结果;
    基于所述标注数据,对所述计算结果进行损失函数计算,得到图像识别区域对应的损失值;
    根据所述图像识别区域对应的损失值,对所述人脸评估模型的参数进行更新,得到所述训练好的人脸评估模型。
  4. 根据权利要求3所述的基于深度学习的人脸图像评估方法,其中,所述根据所述图像识别区域对应的损失值,对所述人脸评估模型的参数进行更新,得到所述训练好的人脸评估模型,包括:
    将所有的所述图像识别区域对应的损失值进行相加,得到目标损失值;
    按照梯度回传的方式,将所述目标损失值进行梯度回传,对所述人脸评估模型的参数进行更新;
    当所述目标损失值达到预设值时,停止对所述人脸评估模型的参数的更新,得到所述训练好的人脸评估模型。
  5. 根据权利要求1所述的基于深度学习的人脸图像评估方法,其中,所述获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域,包括:
    获取用于训练的人脸图像;
    将所述人脸图像向中心缩放预设倍数,得到取样区域;
    按照预设数量,将所述取样区域划分为多个相同大小的区域,作为所述图像识别区域。
  6. 根据权利要求1所述的基于深度学习的人脸图像评估方法,其中,所述计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据,包括:
    按照预设的梯度计算方式,计算所述灰度图的梯度值,得到所述灰度图对应的梯度值;
    设定梯度阈值,将所述灰度值与所述梯度阈值进行比较,得到所述标注数据,其中,若所述梯度值大于所述梯度阈值,则标注数据为1,若所述梯度值小于或等于所述梯度阈值,则标注数据为0。
  7. 根据权利要求1所述的基于深度学习的人脸图像评估方法,其中,所述获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果,包括:
    获取待评估的人脸图像,并通过所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应图像识别区域的得分值;
    将所述得分值与预设模糊阈值进行对比,得到所述待评估的人脸图像对应的评估结果。
  8. 一种基于深度学习的人脸图像评估装置,包括:
    图像识别区域获取模块,用于获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域;
    图像识别区域处理模块,用于将每个所述图像识别区域进行灰度处理,得到每个所述图像识别区域对应的灰度图;
    标注数据获取模块,用于计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据;
    基础向量获取模块,用于根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量;
    人脸评估模型训练模块,用于对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型;
    人脸图像评估模块,用于获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果。
  9. 一种计算机设备,包括存储器和处理器,所述存储器中存储有计算机可读指令,其中,所述处理器执行所述计算机可读指令时实现如下步骤:
    获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域;
    将每个所述图像识别区域进行灰度处理,得到每个所述图像识别区域对应的灰度图;
    计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据;
    根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量;
    对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型;
    获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果。
  10. 根据权利要求9所述的计算机设备,其中,所述根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量,包括:
    根据所述人脸评估模型,提取每个所述图像识别区域的深度特征;
    通过均值池化的方式,对所述深度特征进行池化处理,得到所述深度特征对应的列向量,并将所述列向量作为所述基础向量。
  11. 根据权利要求9所述的计算机设备,其中,所述对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型,包括:
    对所述基础向量进行降维处理,得到目标向量;
    对所述目标向量进行sigmoid函数计算,得到计算结果;
    基于所述标注数据,对所述计算结果进行损失函数计算,得到图像识别区域对应的损失值;
    根据所述图像识别区域对应的损失值,对所述人脸评估模型的参数进行更新,得到所 述训练好的人脸评估模型。
  12. 根据权利要求11所述的计算机设备,其中,所述根据所述图像识别区域对应的损失值,对所述人脸评估模型的参数进行更新,得到所述训练好的人脸评估模型,包括:
    将所有的所述图像识别区域对应的损失值进行相加,得到目标损失值;
    按照梯度回传的方式,将所述目标损失值进行梯度回传,对所述人脸评估模型的参数进行更新;
    当所述目标损失值达到预设值时,停止对所述人脸评估模型的参数的更新,得到所述训练好的人脸评估模型。
  13. 根据权利要求9所述的计算机设备,其中,所述获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域,包括:
    获取用于训练的人脸图像;
    将所述人脸图像向中心缩放预设倍数,得到取样区域;
    按照预设数量,将所述取样区域划分为多个相同大小的区域,作为所述图像识别区域。
  14. 根据权利要求9所述的计算机设备,其中,所述计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据,包括:
    按照预设的梯度计算方式,计算所述灰度图的梯度值,得到所述灰度图对应的梯度值;
    设定梯度阈值,将所述灰度值与所述梯度阈值进行比较,得到所述标注数据,其中,若所述梯度值大于所述梯度阈值,则标注数据为1,若所述梯度值小于或等于所述梯度阈值,则标注数据为0。
  15. 根据权利要求9所述的计算机设备,其中,所述获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果,包括:
    获取待评估的人脸图像,并通过所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应图像识别区域的得分值;
    将所述得分值与预设模糊阈值进行对比,得到所述待评估的人脸图像对应的评估结果。
  16. 一种计算机可读存储介质,其中,所述计算机可读存储介质存储有计算机可读指令,所述计算机可读指令被一种处理器执行时使得所述一种处理器执行如下步骤:
    获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域;
    将每个所述图像识别区域进行灰度处理,得到每个所述图像识别区域对应的灰度图;
    计算所述灰度图的梯度值,得到灰度图对应的梯度值,并根据所述梯度值,得到标注数据;
    根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量;
    对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型;
    获取待评估的人脸图像,并将所述待评估的人脸图像输入到所述训练好的人脸评估模型中,输出所述待评估的人脸图像对应的评估结果。
  17. 根据权利要求16所述的计算机可读存储介质,其中,所述根据人脸评估模型,对所述图像识别区域进行向量提取,得到基础向量,包括:
    根据所述人脸评估模型,提取每个所述图像识别区域的深度特征;
    通过均值池化的方式,对所述深度特征进行池化处理,得到所述深度特征对应的列向量,并将所述列向量作为所述基础向量。
  18. 根据权利要求16所述的计算机可读存储介质,其中,所述对所述基础向量进行降维处理,得到目标向量,并根据所述目标向量和所述标注数据对所述人脸评估模型的参数进行更新,得到训练好的人脸评估模型,包括:
    对所述基础向量进行降维处理,得到目标向量;
    对所述目标向量进行sigmoid函数计算,得到计算结果;
    基于所述标注数据,对所述计算结果进行损失函数计算,得到图像识别区域对应的损失值;
    根据所述图像识别区域对应的损失值,对所述人脸评估模型的参数进行更新,得到所述训练好的人脸评估模型。
  19. 根据权利要求18所述的计算机可读存储介质,其中,所述根据所述图像识别区域对应的损失值,对所述人脸评估模型的参数进行更新,得到所述训练好的人脸评估模型,包括:
    将所有的所述图像识别区域对应的损失值进行相加,得到目标损失值;
    按照梯度回传的方式,将所述目标损失值进行梯度回传,对所述人脸评估模型的参数进行更新;
    当所述目标损失值达到预设值时,停止对所述人脸评估模型的参数的更新,得到所述训练好的人脸评估模型。
  20. 根据权利要求16所述的计算机可读存储介质,其中,所述获取用于训练的人脸图像,并按照预设数量,将所述人脸图像划分为多个相同大小的区域,作为图像识别区域,包括:
    获取用于训练的人脸图像;
    将所述人脸图像向中心缩放预设倍数,得到取样区域;
    按照预设数量,将所述取样区域划分为多个相同大小的区域,作为所述图像识别区域。
PCT/CN2021/083750 2020-12-18 2021-03-30 基于深度学习的人脸图像评估方法、装置、设备及介质 WO2022126917A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011509136.7A CN112529888B (zh) 2020-12-18 基于深度学习的人脸图像评估方法、装置、设备及介质
CN202011509136.7 2020-12-18

Publications (1)

Publication Number Publication Date
WO2022126917A1 true WO2022126917A1 (zh) 2022-06-23

Family

ID=75001907

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/083750 WO2022126917A1 (zh) 2020-12-18 2021-03-30 基于深度学习的人脸图像评估方法、装置、设备及介质

Country Status (1)

Country Link
WO (1) WO2022126917A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103839033A (zh) * 2012-11-20 2014-06-04 广东工业大学 一种基于模糊规则的人脸识别方法
CN105160336A (zh) * 2015-10-21 2015-12-16 云南大学 基于Sigmoid函数的人脸识别方法
CN108171254A (zh) * 2017-11-22 2018-06-15 北京达佳互联信息技术有限公司 图像标签确定方法、装置及终端
US20180204094A1 (en) * 2015-11-26 2018-07-19 Tencent Technology (Shenzhen) Company Limited Image recognition method and apparatus
CN112529888A (zh) * 2020-12-18 2021-03-19 平安科技(深圳)有限公司 基于深度学习的人脸图像评估方法、装置、设备及介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103839033A (zh) * 2012-11-20 2014-06-04 广东工业大学 一种基于模糊规则的人脸识别方法
CN105160336A (zh) * 2015-10-21 2015-12-16 云南大学 基于Sigmoid函数的人脸识别方法
US20180204094A1 (en) * 2015-11-26 2018-07-19 Tencent Technology (Shenzhen) Company Limited Image recognition method and apparatus
CN108171254A (zh) * 2017-11-22 2018-06-15 北京达佳互联信息技术有限公司 图像标签确定方法、装置及终端
CN112529888A (zh) * 2020-12-18 2021-03-19 平安科技(深圳)有限公司 基于深度学习的人脸图像评估方法、装置、设备及介质

Also Published As

Publication number Publication date
CN112529888A (zh) 2021-03-19

Similar Documents

Publication Publication Date Title
CN108416370B (zh) 基于半监督深度学习的图像分类方法、装置和存储介质
CN108764195B (zh) 手写模型训练方法、手写字识别方法、装置、设备及介质
CN109345553B (zh) 一种手掌及其关键点检测方法、装置和终端设备
CN109034069B (zh) 用于生成信息的方法和装置
US20190087683A1 (en) Method and apparatus for outputting information
CN109086654B (zh) 手写模型训练方法、文本识别方法、装置、设备及介质
US20200125836A1 (en) Training Method for Descreening System, Descreening Method, Device, Apparatus and Medium
CN113094478B (zh) 表情回复方法、装置、设备及存储介质
CN112966685B (zh) 用于场景文本识别的攻击网络训练方法、装置及相关设备
US11893773B2 (en) Finger vein comparison method, computer equipment, and storage medium
CN111639527A (zh) 英文手写文本识别方法、装置、电子设备及存储介质
CN115050064A (zh) 人脸活体检测方法、装置、设备及介质
CN113793256A (zh) 基于用户标签的动漫人物生成方法、装置、设备及介质
CN114282258A (zh) 截屏数据脱敏方法、装置、计算机设备及存储介质
CN112801186A (zh) 一种验证图像生成方法、装置及设备
CN114463798A (zh) 人脸识别模型的训练方法、装置、设备及存储介质
CN112632956A (zh) 文本匹配方法、装置、终端和存储介质
WO2022126917A1 (zh) 基于深度学习的人脸图像评估方法、装置、设备及介质
TWI803243B (zh) 圖像擴增方法、電腦設備及儲存介質
CN114241411B (zh) 基于目标检测的计数模型处理方法、装置及计算机设备
CN112529888B (zh) 基于深度学习的人脸图像评估方法、装置、设备及介质
CN111767710B (zh) 印尼语的情感分类方法、装置、设备及介质
CN113723359A (zh) 用户情绪识别方法、装置、计算机设备及可读存储介质
CN113111833B (zh) 人工智能系统的安全性检测方法、装置及终端设备
CN110956190A (zh) 图像识别方法及装置、计算机装置和计算机可读存储介质

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: 21904857

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: 21904857

Country of ref document: EP

Kind code of ref document: A1