Face authentication method and system based on Internet service

Abstract

The embodiment of the invention provides a face authentication method and system based on Internet services, and the method comprises the steps: carrying out the basic feature analysis of a face collection frame sequence through a basic neural network, generating basic semantic vectors, and further fusing the vectors with an original frame sequence, and generating advanced semantic vectors containing richer information. And then, deep analysis is performed on the advanced semantic vector by using an advanced neural network, so that the accuracy of false face recognition is effectively improved, and the safety and reliability of face authentication in Internet services are remarkably enhanced.

Description

Technical Field

[0001] The present application relates to the field of Internet technology, and in particular to a face recognition method and system based on Internet services. Background Art

[0002] With the rapid development of Internet technology, Internet services have penetrated into every aspect of people's lives. In order to ensure the security of Internet services and the privacy of user information, identity authentication technology is particularly important. Among them, face authentication, as a convenient and intuitive identity authentication method, is widely used in various Internet services.

[0003] However, current face recognition technology still faces many challenges. On the one hand, due to the large number of fake face images in the Internet environment, how to accurately and quickly identify these fake faces has become a problem that needs to be solved in face recognition technology. On the other hand, traditional face recognition methods often rely only on a single feature extraction and analysis method, which is difficult to cope with complex and changing Internet business scenarios. Summary of the invention

[0004] In view of this, the purpose of this application is to provide a face recognition method and system based on Internet services.

[0005] In conjunction with the first aspect of the present application, a face recognition method based on Internet services is provided, which is applied to a face recognition system based on Internet services, and the method comprises:

[0006] Obtain a face acquisition frame sequence of a target Internet service;

[0007] Loading the face acquisition frame sequence into a basic neural network for basic feature analysis to generate basic semantic vectors corresponding to a plurality of face acquisition segments, wherein the plurality of face acquisition segments are generated by semantic segmentation of the face acquisition frame sequence;

[0008] Fusing the basic semantic vectors corresponding to the face acquisition frame sequence and the multiple face acquisition segments to generate an advanced semantic vector;

[0009] The advanced semantic vector is loaded into an advanced neural network for advanced feature analysis to generate a judgment result of whether the face acquisition frame sequence is a false face.

[0010] In a possible implementation manner of the first aspect, the basic feature analysis includes:

[0011] Performing semantic segmentation on the face acquisition frame sequence to generate the plurality of face acquisition segments;

[0012] Performing semantic coding on each of the face collection segments to generate a multi-dimensional semantic vector corresponding to each of the face collection segments;

[0013] The multi-dimensional semantic vectors corresponding to each of the face collection segments are fused to generate the basic semantic vectors corresponding to each of the face collection segments.

[0014] In a possible implementation manner of the first aspect, semantically segmenting the face acquisition frame sequence to generate the plurality of face acquisition segments includes:

[0015] The face acquisition frame sequence is segmented in time sequence according to a preset time window size and a preset time length to generate the multiple face acquisition segments, and the multiple face acquisition segments have the same time length.

[0016] In a possible implementation of the first aspect, the advanced neural network includes a plurality of sequentially connected basic network units, an advanced network unit, and a third network unit, and loading the advanced semantic vector into the advanced neural network for advanced feature analysis to generate a judgment result of whether the face acquisition frame sequence is a false face includes:

[0017] Using multiple basic network units to represent the advanced semantic vector, generating a third semantic vector obtained by each of the basic network units, the loading data of the ath basic network unit among the multiple basic network units is the advanced semantic vector, the loading data of the ath basic network unit among the multiple basic network units is the generated data of the a-1th basic network unit, and a is an integer greater than 1;

[0018] Loading the plurality of third semantic vectors into the advanced network unit for feature fusion to generate a fourth semantic vector;

[0019] The fourth semantic vector is loaded into the third network unit for estimation to generate a judgment result of whether the face acquisition frame sequence is a false face.

[0020] In a possible implementation of the first aspect, the method further includes:

[0021] Obtain template face acquisition frame sequence;

[0022] Loading the template face acquisition frame sequence into a basic neural network for feature analysis, generating basic template semantic vectors and prior annotation knowledge corresponding to a plurality of template face acquisition segments, wherein the plurality of template face acquisition segments are generated by semantic segmentation of the template face acquisition frame sequence, and the prior annotation knowledge represents the false feature parts and non-false feature parts contained in the template face acquisition segments;

[0023] The network parameter information of the basic neural network is optimized according to the basic template semantic vector and prior annotation knowledge corresponding to the template face collection segment.

[0024] In a possible implementation of the first aspect, optimizing the network parameter information of the basic neural network according to the basic template semantic vector corresponding to the template face collection segment and the prior annotation knowledge includes:

[0025] According to the basic template semantic vector corresponding to the template face collection segment, estimating the false state feature corresponding to the template face collection segment;

[0026] Determine the error of the false state feature corresponding to the template face collection segment relative to the priori annotation knowledge, and generate a basic error parameter;

[0027] Generate an advanced error parameter according to the feature distance between the false feature part and the non-false feature part in the semantic vector of the basic template;

[0028] According to the basic error parameter and the advanced error parameter, the network parameter information of the basic neural network is optimized.

[0029] In a possible implementation manner of the first aspect, the feature analysis includes:

[0030] Performing semantic segmentation on the template face acquisition frame sequence to generate the plurality of template face acquisition segments;

[0031] Generate prior annotation knowledge corresponding to each of the template face collection segments;

[0032] Performing semantic coding on each of the template face collection segments to generate a multi-dimensional semantic vector corresponding to each of the template face collection segments;

[0033] The multi-dimensional semantic vectors corresponding to each of the template face collection segments are fused to generate the basic template semantic vectors corresponding to each of the template face collection segments.

[0034] In a possible implementation of the first aspect, semantically segmenting the template face acquisition frame sequence to generate the plurality of template face acquisition segments includes:

[0035] According to a preset time window size and a preset time step, the template face acquisition frame sequence is segmented in time sequence semantically to generate the multiple template face acquisition segments, and the multiple template face acquisition segments have the same duration.

[0036] In a possible implementation of the first aspect, the method further includes:

[0037] Acquire a template face acquisition frame sequence and binary classification annotation data corresponding to the template face acquisition frame sequence, wherein the binary classification annotation data indicates whether the template face acquisition frame sequence is a false face;

[0038] Loading the template face acquisition frame sequence into an initialized basic neural network to generate basic template semantic vectors corresponding to a plurality of template face acquisition segments, wherein the plurality of template face acquisition segments are generated by semantically segmenting the template face acquisition frame sequence;

[0039] The template face acquisition frame sequence and the basic template semantic vectors corresponding to each of the plurality of template face acquisition segments are merged to generate an advanced template semantic vector;

[0040] Loading the advanced template semantic vector into an advanced neural network to generate a judgment result of whether the template face acquisition frame sequence is a false face;

[0041] Optimizing network parameter information of the advanced neural network based on the judgment result of the template face acquisition frame sequence and the binary classification labeling data;

[0042] The advanced neural network includes a plurality of sequentially connected basic network units, an advanced network unit and a third network unit, and the step of loading the advanced template semantic vector into the advanced neural network and generating a judgment result of whether the template face acquisition frame sequence is a false face includes:

[0043] Using multiple basic network units to represent the advanced template semantic vector, generating a third template semantic vector obtained by each of the basic network units, the loading data of the ath basic network unit among the multiple basic network units is the advanced template semantic vector, the loading data of the ath basic network unit among the multiple basic network units is the generated data of the a-1th basic network unit, and a is an integer greater than 1;

[0044] Loading the plurality of third template semantic vectors into the advanced network unit for feature fusion to generate a fourth template semantic vector;

[0045] Loading the fourth template semantic vector into the third network unit for estimation, and generating a judgment result of whether the template face acquisition frame sequence is a false face;

[0046] The step of fusing the basic template semantic vectors corresponding to the template face acquisition frame sequence and the multiple template face acquisition segments to generate an advanced template semantic vector includes:

[0047] Compressing the template face acquisition frame sequence to generate a compressed template face acquisition frame sequence, wherein the duration of the compressed template face acquisition frame sequence is the same as the duration of the template face acquisition segment;

[0048] The basic template semantic vectors corresponding to the compressed template face acquisition frame sequence and the multiple template face acquisition segments are fused to generate the advanced template semantic vector.

[0049] In combination with the second aspect of the present application, a face recognition system based on Internet business is provided, wherein the face recognition system based on Internet business includes a machine-readable storage medium and a processor, wherein the machine-readable storage medium stores machine-executable instructions, and when the processor executes the machine-executable instructions, the face recognition system based on Internet business implements the aforementioned face recognition method based on Internet business.

[0050] In conjunction with the third aspect of the present application, a computer-readable storage medium is provided, in which computer-executable instructions are stored. When the computer-executable instructions are executed, the aforementioned face recognition method based on Internet business is implemented.

[0051] In combination with any of the above aspects, the embodiment of the present application performs basic feature analysis on the face acquisition frame sequence through a basic neural network to generate basic semantic vectors, and further fuses these vectors with the original frame sequence to generate advanced semantic vectors containing richer information. Subsequently, the advanced semantic vectors are deeply analyzed using an advanced neural network, which effectively improves the accuracy of false face recognition and significantly enhances the security and reliability of face authentication in Internet services. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained by combining these drawings without paying creative work.

[0053] Figure 1 A flowchart of a face recognition method based on Internet services provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. According to the embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] The terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, device, product or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units inherent to these processes, methods, products or ends.

[0056] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0057] Figure 1 The flowchart of the face recognition method based on Internet services provided by the embodiment of the present application is shown. It should be understood that in other embodiments, the order of some steps of the face recognition method based on Internet services of this embodiment can be shared with each other according to actual needs, or some steps can be omitted or maintained. The face recognition method based on Internet services includes:

[0058] Step S110, obtaining a face acquisition frame sequence of a target Internet service.

[0059] The server in this embodiment is a server that provides security for an online financial service platform. This platform provides users with services such as online bank transfers and online loan applications. When users perform these high-risk business operations, in order to prevent identity fraud, especially when operating with fake faces, the server needs to obtain a facial capture frame sequence.

[0060] In this scenario, when a user initiates a transfer operation on a client device (such as a mobile phone or computer), the camera of the client device will be activated. The camera will start to collect the user's facial image at a certain frame rate (for example, 30 frames per second) to form a facial capture frame sequence. These frame sequences contain static images of the user's face at different times, and each image records the facial expression, posture, facial features under lighting, and other information at that time.

[0061] The server receives the face acquisition frame sequence from the client through the network. The network transmission may use a secure encryption protocol to ensure the integrity and confidentiality of the data. For example, using the SSL / TLS protocol, the face acquisition frame sequence is sent from the client to the server in the form of encrypted data packets. After receiving these data packets, the server will decrypt and parse the data format to convert them into a format that can be processed in subsequent steps.

[0062] In this process, some challenges may be encountered. For example, poor network conditions may cause packet loss or transmission delays. The server needs to have a certain fault tolerance mechanism, such as requesting the client to resend some lost frames, or tolerating delays to a certain extent, while ensuring that the final received face acquisition frame sequence can still meet the requirements of subsequent analysis.

[0063] In addition, the camera performance of different client devices may vary, which will result in different quality of the collected face acquisition frame sequences. Some devices may capture images with low resolution or have more noise in low light conditions. The server needs to be able to process such uneven quality face acquisition frame sequences to ensure the accuracy of analysis.

[0064] Step S120, loading the face acquisition frame sequence into a basic neural network for basic feature analysis, generating basic semantic vectors corresponding to a plurality of face acquisition segments, wherein the plurality of face acquisition segments are generated by semantic segmentation of the face acquisition frame sequence.

[0065] After receiving the facial capture frame sequence, the server loads it into the basic neural network, which is pre-trained and specifically used for analyzing facial features.

[0066] First, semantic segmentation is performed on the face acquisition frame sequence. Taking the online financial service scenario mentioned above as an example, suppose we perform temporal segmentation on the face acquisition frame sequence according to the preset time window size of 5 seconds (i.e., every 5 seconds as a segment) and the preset time length (here is 5 seconds). If the total length of the face acquisition frame sequence is 30 seconds, it can be segmented into 6 face acquisition segments.

[0067] For each generated face collection segment, the basic neural network begins to perform semantic encoding representation. For example, in the first face collection segment, the user's face may have a slight head turn within 5 seconds, and the expression changes from calm to slightly frowning. The neural network will extract facial features in each frame of this segment, such as the shape of the eyes, the outline of the nose, the degree of opening of the mouth, etc. Then, these features are converted into multi-dimensional semantic vectors. This semantic vector may be a high-dimensional vector, in which each dimension represents a specific facial feature or feature combination.

[0068] Taking the eye feature as an example, one dimension of the vector may represent the degree of eye opening and closing, and another dimension may represent the horizontal width ratio of the eye. There are also corresponding dimensions for nose features, mouth features, etc. Then, the multi-dimensional semantic vectors corresponding to all frames in the face collection segment are fused. Weighted averaging or other fusion algorithms can be used to finally generate the basic semantic vector corresponding to the face collection segment.

[0069] In this process, the basic neural network may encounter some complex situations. For example, the user may wear glasses in a certain acquisition clip, which will interfere with the extraction of eye features. The neural network needs to have a certain robustness to be able to identify and correctly handle this situation. For another example, there may be short-term occlusion during the face acquisition process, such as hair blocking part of the face. The basic neural network needs to generate the basic semantic vector as accurately as possible through other visible facial features.

[0070] Step S130: fusing the basic semantic vectors corresponding to the face acquisition frame sequence and the multiple face acquisition segments to generate an advanced semantic vector.

[0071] Continuing with the scenario of online financial services as an example, after the server obtains the face acquisition frame sequence and the basic semantic vector corresponding to each acquisition segment, it starts the fusion operation.

[0072] First, the server processes the face acquisition frame sequence and may perform some compression operations on it to facilitate fusion with the basic semantic vector. For example, the resolution of each frame image in the face acquisition frame sequence is adjusted to reduce the data volume while still retaining key facial feature information.

[0073] Assume that the compressed face acquisition frame sequence is represented as a data structure in a specific format, which contains the coordinates of key facial feature points of each frame, overall facial contour information, etc. Then, the compressed face acquisition frame sequence is fused with the basic semantic vectors corresponding to the multiple face acquisition segments generated previously.

[0074] The fusion method may be to associate and combine the feature data in the face acquisition frame sequence with the corresponding features in the basic semantic vector. For example, the eye feature of a frame in the face acquisition frame sequence is fused with the eye feature part in the basic semantic vector, which may be done by weighted addition or other complex combination algorithms. The same operation is performed for other facial features such as the nose and mouth.

[0075] To take a specific example, if the eye feature part in the basic semantic vector represents the degree of eye opening as 0.8 (here is the normalized value), and the eye feature of a frame in the face acquisition frame sequence is analyzed to have a degree of opening as 0.7, during fusion, the fused eye opening characteristic value may be calculated as 0.74 based on certain weights (such as the weight of the basic semantic vector is 0.6, and the weight of the face acquisition frame sequence is 0.4).

[0076] By performing such a fusion operation on all facial features, an advanced semantic vector is finally generated. This advanced semantic vector combines the original features of the face acquisition frame sequence and the semantic features in the basic semantic vector, and can more comprehensively describe the feature information of the face.

[0077] During this fusion process, you may encounter problems with data matching and weight selection. Different facial features may require different weights for fusion to achieve the best results. The server needs to determine these weights based on a large amount of training data and experience, and be able to handle data mismatches or anomalies that may occur during the fusion process.

[0078] Step S140, loading the advanced semantic vector into an advanced neural network for advanced feature analysis, and generating a judgment result of whether the face acquisition frame sequence is a false face.

[0079] The advanced neural network includes a plurality of sequentially connected basic network units, an advanced network unit and a third network unit.

[0080] The server loads the generated advanced semantic vector into the advanced neural network. First, multiple basic network units are used to represent the features of the advanced semantic vector.

[0081] For example, after receiving the advanced semantic vector, the first basic network unit begins to analyze it. This basic network unit may be a convolutional neural network layer, which performs convolution operations on the features in the advanced semantic vector to extract more advanced features. Assuming that the advanced semantic vector contains comprehensive facial feature information such as eyes, nose, and mouth, the first basic network unit slides the convolution kernel on this vector to identify some local feature patterns, such as the muscle texture pattern around the eyes, the relative position relationship pattern between the nose and mouth, etc., and generates a new third semantic vector.

[0082] Then, this new third semantic vector will be passed to the next basic network unit (here we assume it is the second basic network unit). The second basic network unit will continue to extract features based on the output of the first basic network unit. It may use different convolution kernels or other operations to further explore deeper features and generate a new third semantic vector again. This process will be carried out in multiple basic network units in sequence, and each basic network unit will perform deeper feature extraction based on the previous unit.

[0083] Next, the third semantic vectors obtained by multiple basic network units are loaded into the advanced network unit for feature fusion. For example, suppose there are 3 basic network units, which generate 3 different third semantic vectors. The advanced network unit will fuse these 3 vectors, perhaps by splicing them together and then performing a fully connected layer operation to generate a fourth semantic vector. This fourth semantic vector combines the different levels of features extracted by the previous basic network units.

[0084] Finally, the fourth semantic vector is loaded into the third network unit for estimation. The third network unit may be a classifier, such as a logistic regression classifier. It will judge whether the face acquisition frame sequence is a false face based on the feature information in the fourth semantic vector. If some feature values ​​in the fourth semantic vector conform to the feature patterns of false faces (these feature patterns are learned through a large amount of training data), then the third network unit will output the judgment result that the face acquisition frame sequence is a false face; otherwise, it is judged to be a real face.

[0085] In this process, advanced neural networks may encounter some challenges. For example, as the technology for making fake faces continues to develop, some new means of forgery may appear, resulting in the existing feature patterns being insufficient for accurate judgment. Servers need to continuously update and optimize advanced neural networks to adapt to new security threats. At the same time, when training advanced neural networks, a large amount of sample data of real and fake faces is required, and the diversity and accuracy of this data must be ensured, otherwise it may cause overfitting or underfitting of the neural network, affecting the accuracy of the judgment results.

[0086] Based on the above steps, the embodiment of the present application uses a basic neural network to perform basic feature analysis on the face acquisition frame sequence, generates basic semantic vectors, and further fuses these vectors with the original frame sequence to generate advanced semantic vectors containing richer information. Subsequently, the advanced semantic vectors are deeply analyzed using an advanced neural network, which effectively improves the accuracy of false face recognition and significantly enhances the security and reliability of face authentication in Internet services.

[0087] In a possible implementation, the basic feature analysis includes:

[0088] The face acquisition frame sequence is semantically segmented to generate the plurality of face acquisition segments.

[0089] Each of the face collection segments is semantically encoded to generate a multi-dimensional semantic vector corresponding to each of the face collection segments.

[0090] The multi-dimensional semantic vectors corresponding to each of the face collection segments are fused to generate the basic semantic vectors corresponding to each of the face collection segments.

[0091] In a possible implementation manner, the semantic segmentation of the face acquisition frame sequence to generate the plurality of face acquisition segments includes:

[0092] The face acquisition frame sequence is segmented in time sequence according to a preset time window size and a preset time length to generate the multiple face acquisition segments, and the multiple face acquisition segments have the same time length.

[0093] In a possible implementation, the advanced neural network includes a plurality of sequentially connected basic network units, an advanced network unit, and a third network unit, and the loading of the advanced semantic vector into the advanced neural network for advanced feature analysis to generate a judgment result of whether the face acquisition frame sequence is a false face includes:

[0094] The advanced semantic vector is characterized by using multiple basic network units to generate a third semantic vector obtained by each basic network unit, the loading data of the ath basic network unit among the multiple basic network units is the advanced semantic vector, the loading data of the ath basic network unit among the multiple basic network units is the generated data of the a-1th basic network unit, and a is an integer greater than 1.

[0095] Loading the plurality of third semantic vectors into the advanced network unit for feature fusion to generate a fourth semantic vector.

[0096] The fourth semantic vector is loaded into the third network unit for estimation to generate a judgment result of whether the face acquisition frame sequence is a false face.

[0097] In a possible implementation, the method further includes:

[0098] Get the template face acquisition frame sequence.

[0099] The template face acquisition frame sequence is loaded into the basic neural network for feature analysis to generate basic template semantic vectors and prior annotation knowledge corresponding to multiple template face acquisition segments, wherein the multiple template face acquisition segments are generated by semantic segmentation of the template face acquisition frame sequence, and the prior annotation knowledge represents the false feature parts and non-false feature parts contained in the template face acquisition segments.

[0100] The network parameter information of the basic neural network is optimized according to the basic template semantic vector and prior annotation knowledge corresponding to the template face collection segment.

[0101] In a possible implementation, optimizing the network parameter information of the basic neural network according to the basic template semantic vector corresponding to the template face collection segment and the prior annotation knowledge includes:

[0102] According to the basic template semantic vector corresponding to the template face collection segment, the false state feature corresponding to the template face collection segment is estimated.

[0103] The error of the false state feature corresponding to the template face collection segment relative to the priori annotation knowledge is determined to generate a basic error parameter.

[0104] An advanced error parameter is generated according to a feature distance between a false feature part and a non-false feature part in the basic template semantic vector.

[0105] According to the basic error parameter and the advanced error parameter, the network parameter information of the basic neural network is optimized.

[0106] In a possible implementation, the feature analysis includes:

[0107] The template face collection frame sequence is semantically segmented to generate the plurality of template face collection segments.

[0108] Generate the prior annotation knowledge corresponding to each of the template face collection segments.

[0109] Each of the template face collection segments is semantically encoded to generate a multi-dimensional semantic vector corresponding to each of the template face collection segments.

[0110] The multi-dimensional semantic vectors corresponding to each of the template face collection segments are fused to generate the basic template semantic vectors corresponding to each of the template face collection segments.

[0111] In a possible implementation, the semantic segmentation of the template face acquisition frame sequence to generate the plurality of template face acquisition segments includes:

[0112] According to a preset time window size and a preset time step, the template face acquisition frame sequence is segmented in time sequence semantically to generate the multiple template face acquisition segments, and the multiple template face acquisition segments have the same duration.

[0113] In a possible implementation, the method further includes:

[0114] A template face acquisition frame sequence and binary classification annotation data corresponding to the template face acquisition frame sequence are obtained, wherein the binary classification annotation data indicates whether the template face acquisition frame sequence is a false face.

[0115] The template face acquisition frame sequence is loaded into an initialized basic neural network to generate basic template semantic vectors corresponding to multiple template face acquisition segments, respectively. The multiple template face acquisition segments are generated by semantically segmenting the template face acquisition frame sequence.

[0116] The basic template semantic vectors corresponding to the template face acquisition frame sequence and the multiple template face acquisition segments are fused to generate an advanced template semantic vector.

[0117] The advanced template semantic vector is loaded into the advanced neural network to generate a judgment result of whether the template face acquisition frame sequence is a false face.

[0118] The network parameter information of the advanced neural network is optimized according to the judgment result of the template face acquisition frame sequence and the binary classification labeling data.

[0119] The advanced neural network includes a plurality of sequentially connected basic network units, an advanced network unit and a third network unit, and the step of loading the advanced template semantic vector into the advanced neural network and generating a judgment result of whether the template face acquisition frame sequence is a false face includes:

[0120] The advanced template semantic vector is characterized by using multiple basic network units to generate a third template semantic vector obtained by each basic network unit, the loading data of the ath basic network unit among the multiple basic network units is the advanced template semantic vector, the loading data of the ath basic network unit among the multiple basic network units is the generated data of the a-1th basic network unit, and a is an integer greater than 1.

[0121] Loading the plurality of third template semantic vectors into the advanced network unit for feature fusion to generate a fourth template semantic vector.

[0122] The fourth template semantic vector is loaded into the third network unit for estimation to generate a judgment result of whether the template face acquisition frame sequence is a false face.

[0123] The step of fusing the basic template semantic vectors corresponding to the template face acquisition frame sequence and the multiple template face acquisition segments to generate an advanced template semantic vector includes:

[0124] The template face acquisition frame sequence is compressed to generate a compressed template face acquisition frame sequence, wherein the duration of the compressed template face acquisition frame sequence is the same as the duration of the template face acquisition segment.

[0125] The basic template semantic vectors corresponding to the compressed template face acquisition frame sequence and the multiple template face acquisition segments are fused to generate the advanced template semantic vector.

[0126] In the above embodiments, the Internet-based face authentication system for executing the above method embodiments has at least one processor, a control module (chip set) coupled to at least one of the (at least one) processors, a memory coupled to the control module, a non-volatile memory (NVM) / storage device coupled to the control module, at least one input / output device coupled to the control module, and a network interface coupled to the control module.

[0127] The processor may include at least one single-core or multi-core processor, and the processor may include any combination of general-purpose processors or dedicated processors (such as graphics processors, application processors, baseband processors, etc.). For some alternative implementations, the face recognition system based on Internet services can be used as a face recognition system device based on Internet services such as the gateway described in the embodiments of the present application.

[0128] For some alternative embodiments, a face recognition system based on Internet business may include at least one computer-readable medium (e.g., a memory or NVM / storage device) having instructions and at least one processor integrated with the at least one computer-readable medium and configured to execute instructions to implement a module to perform the actions described in the present disclosure.

[0129] For one embodiment, the control module may include any suitable interface controller to provide any suitable interface to at least one of the processor(s) and / or any suitable device or component in communication with the control module.

[0130] The control module may include a memory controller module to provide an interface to the memory. The memory controller module may be a hardware module, a software module, and / or a firmware module.

[0131] The memory may be used, for example, to load and store data and / or instructions for a face authentication system based on an Internet service. For one embodiment, the memory may include any suitable volatile memory, for example, a suitable DRAM.

[0132] For one embodiment, the control module may include at least one input / output controller to provide an interface to the NVM / storage device and (at least one) input / output device.

[0133] For example, the NVM / storage device may be used to store data and / or instructions. The NVM / storage device may include any suitable non-volatile memory (e.g., flash memory) and / or may include any suitable (at least one) non-volatile storage device (e.g., at least one hard disk drive (HDD), at least one compact disk (CD) drive, and / or at least one digital versatile disk (DVD) drive).

[0134] The NVM / storage device may include storage resources that are physically part of the device on which the face authentication system for Internet-based services is installed, or that may be accessible to the device without being part of the device. For example, the NVM / storage device may be accessed via (at least one) input / output device over a network.

[0135] (At least one) input / output device may provide an interface for the face authentication system based on Internet services to communicate with any other appropriate device, and the input / output device may include a communication component, a phonetic component, a sensor component, etc. The network interface may provide an interface for the face authentication system based on Internet services to predict network communication based on at least one user preference, and the face authentication system based on Internet services may wirelessly communicate with at least one component of a wireless network based on any of at least one wireless network priors and / or protocols, such as accessing a wireless network based on a communication prior.

[0136] For one embodiment, at least one of the (at least one) processor may be loaded together with the logic of at least one controller of the control module (e.g., a memory controller module). For one embodiment, at least one of the (at least one) processor may be loaded together with the logic of at least one controller of the control module to form a system level load. For one embodiment, at least one of the (at least one) processor may be fused on the same die with the logic of at least one controller of the control module. For one embodiment, at least one of the (at least one) processor may be fused on the same die with the logic of at least one controller of the control module to form a system on chip (SoC).

[0137] The embodiments of the present application are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

[0138] An embodiment of the present invention discloses a computer-readable storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute the steps of the face recognition method based on Internet services described in the aforementioned embodiment.

[0139] An embodiment of the present invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute the steps in the face recognition method based on Internet services described in the aforementioned embodiment.

[0140] The device embodiments described above are only illustrative, wherein the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple user preference prediction network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0141] Through the specific description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, a magnetic disk storage, a magnetic tape storage, or any other medium that can be used to be computer-readable with or store data.

[0142] Finally, it should be noted that what is disclosed above is only a preferred embodiment of the present invention, which is only used to illustrate the technical solution of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, it should be understood by those skilled in the art that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features therein can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A face recognition method based on Internet services, characterized in that: The method comprises: Obtain a face acquisition frame sequence of a target Internet service; Loading the face acquisition frame sequence into a basic neural network for basic feature analysis to generate basic semantic vectors corresponding to a plurality of face acquisition segments, wherein the plurality of face acquisition segments are generated by semantic segmentation of the face acquisition frame sequence; Fusing the basic semantic vectors corresponding to the face acquisition frame sequence and the multiple face acquisition segments to generate an advanced semantic vector; The advanced semantic vector is loaded into an advanced neural network for advanced feature analysis to generate a judgment result of whether the face acquisition frame sequence is a false face.

2. The face recognition method based on Internet business according to claim 1, characterized in that: The basic feature analysis includes: Performing semantic segmentation on the face acquisition frame sequence to generate the plurality of face acquisition segments; Performing semantic coding on each of the face collection segments to generate a multi-dimensional semantic vector corresponding to each of the face collection segments; The multi-dimensional semantic vectors corresponding to each of the face collection segments are fused to generate the basic semantic vectors corresponding to each of the face collection segments.

3. The face recognition method based on Internet business according to claim 2, characterized in that: The semantic segmentation of the face acquisition frame sequence to generate the plurality of face acquisition segments includes: The face acquisition frame sequence is segmented in time sequence according to a preset time window size and a preset time length to generate the multiple face acquisition segments, and the multiple face acquisition segments have the same time length.

4. The face recognition method based on Internet business according to claim 1, characterized in that: The advanced neural network includes a plurality of sequentially connected basic network units, an advanced network unit, and a third network unit. The loading of the advanced semantic vector into the advanced neural network for advanced feature analysis to generate a judgment result of whether the face acquisition frame sequence is a false face includes: Using multiple basic network units to represent the advanced semantic vector, generating a third semantic vector obtained by each of the basic network units, the loading data of the ath basic network unit among the multiple basic network units is the advanced semantic vector, the loading data of the ath basic network unit among the multiple basic network units is the generated data of the a-1th basic network unit, and a is an integer greater than 1; Loading the plurality of third semantic vectors into the advanced network unit for feature fusion to generate a fourth semantic vector; The fourth semantic vector is loaded into the third network unit for estimation to generate a judgment result of whether the face acquisition frame sequence is a false face.

5. The face recognition method based on Internet services according to any one of claims 1 to 4, characterized in that: The method further comprises: Obtain template face acquisition frame sequence; Loading the template face acquisition frame sequence into a basic neural network for feature analysis, generating basic template semantic vectors and prior annotation knowledge corresponding to a plurality of template face acquisition segments, wherein the plurality of template face acquisition segments are generated by semantic segmentation of the template face acquisition frame sequence, and the prior annotation knowledge represents the false feature parts and non-false feature parts contained in the template face acquisition segments; The network parameter information of the basic neural network is optimized according to the basic template semantic vector and prior annotation knowledge corresponding to the template face collection segment.

6. The face recognition method based on Internet service according to claim 5, characterized in that: The optimizing the network parameter information of the basic neural network according to the basic template semantic vector corresponding to the template face collection segment and the prior annotation knowledge includes: According to the basic template semantic vector corresponding to the template face collection segment, estimating the false state feature corresponding to the template face collection segment; Determine the error of the false state feature corresponding to the template face collection segment relative to the priori annotation knowledge, and generate a basic error parameter; Generate an advanced error parameter according to the feature distance between the false feature part and the non-false feature part in the semantic vector of the basic template; According to the basic error parameter and the advanced error parameter, the network parameter information of the basic neural network is optimized.

7. The face recognition method based on Internet services according to claim 5 or 6, characterized in that: The feature analysis includes: Performing semantic segmentation on the template face acquisition frame sequence to generate the plurality of template face acquisition segments; Generate prior annotation knowledge corresponding to each of the template face collection segments; Performing semantic coding on each of the template face collection segments to generate a multi-dimensional semantic vector corresponding to each of the template face collection segments; The multi-dimensional semantic vectors corresponding to each of the template face collection segments are fused to generate the basic template semantic vectors corresponding to each of the template face collection segments.

8. The face recognition method based on Internet service according to claim 7, characterized in that: The semantic segmentation of the template face acquisition frame sequence to generate the plurality of template face acquisition segments comprises: According to a preset time window size and a preset time step, the template face acquisition frame sequence is segmented in time sequence semantically to generate the multiple template face acquisition segments, and the multiple template face acquisition segments have the same duration.

9. The face recognition method based on Internet services according to any one of claims 5 to 8, characterized in that: The method further comprises: Acquire a template face acquisition frame sequence and binary classification annotation data corresponding to the template face acquisition frame sequence, wherein the binary classification annotation data indicates whether the template face acquisition frame sequence is a false face; Loading the template face acquisition frame sequence into an initialized basic neural network to generate basic template semantic vectors corresponding to a plurality of template face acquisition segments, wherein the plurality of template face acquisition segments are generated by semantically segmenting the template face acquisition frame sequence; The template face acquisition frame sequence and the basic template semantic vectors corresponding to each of the plurality of template face acquisition segments are merged to generate an advanced template semantic vector; Loading the advanced template semantic vector into an advanced neural network to generate a judgment result of whether the template face acquisition frame sequence is a false face; Optimizing network parameter information of the advanced neural network based on the judgment result of the template face acquisition frame sequence and the binary classification labeling data; The advanced neural network includes a plurality of sequentially connected basic network units, an advanced network unit and a third network unit, and the step of loading the advanced template semantic vector into the advanced neural network and generating a judgment result of whether the template face acquisition frame sequence is a false face includes: Using multiple basic network units to represent the advanced template semantic vector, generating a third template semantic vector obtained by each of the basic network units, the loading data of the ath basic network unit among the multiple basic network units is the advanced template semantic vector, the loading data of the ath basic network unit among the multiple basic network units is the generated data of the a-1th basic network unit, and a is an integer greater than 1; Loading the plurality of third template semantic vectors into the advanced network unit for feature fusion to generate a fourth template semantic vector; Loading the fourth template semantic vector into the third network unit for estimation, and generating a judgment result of whether the template face acquisition frame sequence is a false face; The step of fusing the basic template semantic vectors corresponding to the template face acquisition frame sequence and the multiple template face acquisition segments to generate an advanced template semantic vector includes: Compressing the template face acquisition frame sequence to generate a compressed template face acquisition frame sequence, wherein the duration of the compressed template face acquisition frame sequence is the same as the duration of the template face acquisition segment; The basic template semantic vectors corresponding to the compressed template face acquisition frame sequence and the multiple template face acquisition segments are fused to generate the advanced template semantic vector.

10. A face recognition system based on Internet services, characterized in that: It includes a processor and a computer-readable storage medium, wherein the computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are executed by a computer, the face recognition method based on Internet business described in any one of claims 1 to 9 is implemented.

Images