Face background similarity recognition method, device, medium and system

Abstract

The application provides a face background similarity recognition method, device, medium and system. The method is different from the method for feature representation based on a classical deep neural network in the past. The method uses an encoder and a decoder architecture to realize segmentation of a face and a background. The application uses a background image vector for comparison. A comparison object is a fraud mapping relationship stored in a vector database. Therefore, the accuracy of determining whether a real-time face image represents a fraud scene is improved. Similarity comparison is performed between a real-time vector data set and a fraud mapping relationship stored in a vector database. The purpose of real-time monitoring is achieved under the premise of ensuring accuracy. Therefore, the problem that a clustering algorithm cannot be used for real-time monitoring in the face image background detection of an existing scheme is solved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and more specifically, to a method, apparatus, medium, and system for recognizing the similarity of a face background. Background Technology

[0002] Facial recognition, as an easy-to-use biometric identification technology, is currently widely used in industries such as finance, government affairs, and security. From smart city construction to the use of smart devices in daily life, facial recognition technology can be seen everywhere.

[0003] In the financial industry, facial recognition has become a crucial component of identity verification in various financial business scenarios, including identity recognition / verification, fraud prevention, offline payments, and smart branches in internet finance. The application of this technology has effectively improved user experience and risk management capabilities. However, facial recognition technology is facing increasingly serious new attack threats, such as adversarial sample attacks and facial expression manipulation attacks.

[0004] Currently, commonly used forgery detection methods mainly focus on backend algorithm detection. These algorithms learn the characteristics of common attacks / forged data and are trained in a targeted manner to perform adversarial example detection, deep fake detection, ghost image detection, and logo detection. This approach has the following limitations:

[0005] (1) Iterative learning is carried out on relevant data of cases that have occurred. The improvement of defense capabilities lags behind the development of attack technology, and proactive defense cannot be achieved.

[0006] (2) The clustering algorithm is used for the detection of the background of the face image, which cannot achieve real-time monitoring and can only be used for post-event inspection, and does not meet the anti-fraud requirements of in-process interception.

[0007] (3) The clustering algorithm is used for the detection of the background of the face image. It only integrates the data at the data level and does not combine the business scenario, customer transaction habits and other information to make a comprehensive judgment. It fails to use global data for mining.

[0008] In other words, the existing solutions for detecting the background of face images using clustering algorithms cannot achieve real-time monitoring. Summary of the Invention

[0009] The main objective of this application is to provide a method, apparatus, medium, and system for recognizing the similarity of facial backgrounds, so as to at least solve the problem that the detection of facial image backgrounds using clustering algorithms in existing solutions cannot achieve real-time monitoring.

[0010] To achieve the above objectives, according to one aspect of this application, a method for recognizing the similarity of a face background is provided, the method comprising:

[0011] Acquire real-time data, including real-time face images, real-time timestamp data, and real-time address data;

[0012] An encoder and decoder architecture is used to perform face image segmentation processing on the real-time face image in the real-time data to obtain a real-time background image. An image vector model is used to vectorize the real-time background image to obtain a real-time background image vector. A text vector model is used to vectorize the real-time timestamp data and the real-time address data to obtain a real-time timestamp vector and a real-time address vector, respectively.

[0013] The similarity is compared between the real-time vector dataset and the fraud mapping relationship stored in the vector database to obtain the final similarity. Based on the magnitude of the final similarity, it is determined whether the real-time face image represents a fraud scene. The real-time vector dataset includes the real-time background image vector, the real-time timestamp vector, and the real-time address vector. The fraud mapping relationship is a mapping relationship between the fraud scene image vector, the timestamp vector, and the address vector.

[0014] Optionally, a similarity comparison is performed between the real-time vector dataset and the fraud mapping relationship stored in the vector database to obtain the final similarity score, including:

[0015] The real-time background image vector is compared with the similarity of all the fraud scene image vectors in the fraud mapping relationship to obtain multiple first similarities. The real-time background image vector is compared with the similarity of all the fraud scene image vectors in the fraud mapping relationship to obtain multiple second similarities. The real-time background image vector is compared with the similarity of all the fraud scene image vectors in the fraud mapping relationship to obtain multiple third similarities.

[0016] The first target similarity is determined to be the maximum value among all the first similarities, the second target similarity is determined to be the maximum value among all the second similarities, and the third target similarity is determined to be the maximum value among all the third similarities;

[0017] The first target similarity, the second target similarity, and the third target similarity are weighted and summed to obtain the final similarity.

[0018] Optionally, determining whether the real-time face image represents a fraud scenario based on the final similarity score includes:

[0019] If the final similarity is greater than or equal to the first similarity threshold, the real-time face image is determined to represent a fraud scenario.

[0020] Optionally, the vector database also stores security mapping relationships, which are mapping relationships between security scene image vectors, timestamp vectors, and address vectors. After obtaining the final similarity, the method further includes:

[0021] When the final similarity is greater than or equal to the second similarity threshold and the final similarity is less than the first similarity threshold, the real-time background image is compared with all the security scene image vectors in the security mapping relationship stored in the vector database to obtain multiple fourth similarities. Based on the magnitude of all the fourth similarities, the fraud mapping relationship and the security mapping relationship in the vector database are corrected to obtain the corrected fraud mapping relationship and the corrected security mapping relationship.

[0022] If the final similarity is less than the second similarity threshold, the real-time face image is determined to represent a safe scenario.

[0023] Optionally, based on the magnitude of all the fourth similarities, the fraud mapping relationship and the security mapping relationship in the vector database are modified to obtain the modified fraud mapping relationship and the modified security mapping relationship, including:

[0024] The target security scene image vector is determined to be the security scene image vector corresponding to the fourth similarity that is greater than or equal to the third similarity threshold;

[0025] When the current proportion is greater than or equal to the preset proportion, the target security scene image vector, the timestamp vector and the address vector corresponding to the target security scene image vector are stored in the fraud mapping relationship to obtain the corrected fraud mapping relationship. The current proportion is the proportion of the fourth similarity that is greater than or equal to the third similarity threshold in all the fourth similarities.

[0026] The corrected security mapping relationship is obtained by deleting the target security scene image vector, the timestamp vector, and the address vector corresponding to the target security scene image vector from the security mapping relationship.

[0027] Optionally, after determining that the real-time face image represents a fraud scenario, the method further includes:

[0028] An alarm message is generated to alert the user that the real-time facial image represents the fraud scenario.

[0029] Optionally, the encoder in the encoder and decoder architecture adopts an inverted bottleneck structure, and the decoder in the encoder and decoder architecture adopts a distributed offset convolution structure.

[0030] According to another aspect of this application, a face background similarity recognition device is provided, the device comprising:

[0031] The acquisition unit is used to acquire real-time data, which includes real-time face images, real-time timestamp data, and real-time address data.

[0032] The first processing unit is used to perform face image segmentation processing on the real-time face image in the real-time data using an encoder and decoder architecture to obtain a real-time background image, and to perform vectorization processing on the real-time background image using an image vector model to obtain a real-time background image vector, and to perform vectorization processing on the real-time timestamp data and the real-time address data using a text vector model to obtain a real-time timestamp vector and a real-time address vector, respectively.

[0033] The second processing unit is used to compare the similarity between the real-time vector dataset and the fraud mapping relationship stored in the vector database to obtain the final similarity, and to determine whether the real-time face image represents a fraud scene based on the magnitude of the final similarity. The real-time vector dataset includes the real-time background image vector, the real-time timestamp vector and the real-time address vector, and the fraud mapping relationship is a mapping relationship between the fraud scene image vector, the timestamp vector and the address vector.

[0034] According to another aspect of this application, a computer-readable storage medium is provided, the computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform any of the aforementioned face background similarity recognition methods.

[0035] According to another aspect of this application, a facial background similarity recognition system is provided, the system comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs comprising methods for performing any of the facial background similarity recognition methods described above.

[0036] The technical solution of this application differs from previous methods based on classic deep neural networks for feature representation. It achieves face and background segmentation by using an encoder and decoder architecture. Furthermore, this application uses background image vector comparison, with the comparison object being the fraud mapping relationship stored in the vector database. This improves the accuracy of determining whether the real-time face image represents a fraudulent scenario. Moreover, by comparing the similarity between the real-time vector dataset and the fraud mapping relationship stored in the vector database, the goal of real-time monitoring is achieved while ensuring accuracy. This solves the problem that existing solutions using clustering algorithms for face image background detection cannot achieve real-time monitoring. Attached Figure Description

[0037] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0038] Figure 1 A flowchart illustrating a method for facial background similarity recognition according to an embodiment of this application is shown.

[0039] Figure 2 A schematic diagram of the architecture of an anti-fraud system provided according to an embodiment of this application is shown;

[0040] Figure 3 A flowchart illustrating another face background similarity recognition method provided according to an embodiment of this application is shown.

[0041] Figure 4 A structural block diagram of a face background similarity recognition device provided according to an embodiment of this application is shown. Detailed Implementation

[0042] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0043] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0044] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0045] For ease of description, the following explains some of the nouns or terms used in the embodiments of this application:

[0046] Adversarial example attacks refer to input samples created by intentionally adding perturbations to a dataset, which cause the model to give incorrect outputs.

[0047] Deepfake: A machine learning model called "Generative Adversarial Network" (GAN) merges or overlays images or videos onto source images or videos. It uses neural network technology to learn from a large number of samples and splices together a person's voice, facial expressions, and body movements to synthesize fake content.

[0048] Clustering algorithms divide a dataset into different classes or clusters based on a specific criterion (such as distance), maximizing the similarity of data objects within the same cluster and maximizing the differences between data objects in different clusters. In other words, after clustering, data of the same class should be grouped together as much as possible, while data of different classes should be separated as much as possible.

[0049] Vector data: a mathematical representation that uses an ordered set of numerical values ​​(usually floating-point numbers) to represent an object or data point. It is commonly used to represent the location, features, or attributes of data points in multidimensional space.

[0050] Instance segmentation model: A model algorithm that predicts the category of each pixel in an input image and distinguishes each instance based on this.

[0051] Distributed Offset Convolution (DSConv): This is a deep convolutional neural network model used for sequence modeling. It extracts features from the input sequence through multiple layers of convolutional neural networks (CNN) and performs classification or regression through fully connected layers.

[0052] Bottleneck structure: This is a special deep learning network structure, mainly used to reduce the number of parameters and computational cost while maintaining the network's expressive power.

[0053] Background extraction technology: The person and the background are identified and extracted separately. The goal is to distinguish the foreground and background of the image at the pixel level, which can be achieved by segmentation technology.

[0054] Vue + ElementUI is a front-end development technology combination. Vue is a popular JavaScript framework for building user interfaces, while ElementUI is a UI component library based on Vue.js, providing rich components and styles to help developers quickly build modern web application interfaces. By combining Vue and ElementUI, developers can quickly build web applications with excellent user experiences. Vue + ElementUI is widely used in various web development projects.

[0055] As described in the background section, currently available forgery detection methods primarily focus on backend algorithm detection. These algorithms learn the characteristics of common attacks / forged data and are trained specifically to perform adversarial example detection, deep fake detection, ghost image detection, and logo detection. This approach has the following shortcomings: iterative learning based on data from past cases means that the improvement in defense capabilities lags behind the development of attack techniques, making proactive defense impossible; the clustering algorithm used for detecting facial image backgrounds cannot achieve real-time monitoring and can only be used for post-event checks, failing to meet the anti-fraud requirements of real-time interception; and the clustering algorithm used for facial image background detection only integrates data at the data level, without combining business scenarios, customer transaction habits, and other information for comprehensive judgment, failing to utilize global data for mining. To address the problem that existing solutions using clustering algorithms for facial image background detection cannot achieve real-time monitoring, embodiments of this application provide a method, apparatus, medium, and system for facial background similarity recognition.

[0056] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0057] This embodiment provides a method for facial background similarity recognition that runs on a mobile terminal, computer terminal, or similar computing device. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0058] Figure 1 This is a flowchart illustrating a method for facial background similarity recognition according to an embodiment of this application. Figure 1 As shown, the method includes the following steps:

[0059] Step S101: Obtain real-time data, including real-time face images, real-time timestamp data, and real-time address data;

[0060] Step S102: Using an encoder and decoder architecture, the real-time face image in the real-time data is processed by face image segmentation to obtain a real-time background image. The real-time background image is then vectorized using an image vector model to obtain a real-time background image vector. The real-time timestamp data and the real-time address data are then vectorized using a text vector model to obtain a real-time timestamp vector and a real-time address vector, respectively.

[0061] In one embodiment of this application, the encoder in the encoder and decoder architecture described above adopts an inverted bottleneck structure, and the decoder in the encoder and decoder architecture described above adopts a distributed offset convolution structure.

[0062] Face background segmentation models in background extraction techniques employ an encoder-decoder architecture. The encoder uses an inverted bottleneck structure to effectively extract face image features, while the decoder uses distributed offset convolution (DSConv). This approach can easily replace standard neural network architectures and achieves lower memory usage and higher computational speed. DSConv decomposes the traditional convolutional kernel into two components: a variable quantization kernel (VQK) and a distributed offset. It achieves lower memory usage and higher speed by storing only integer values ​​in the VQK, while maintaining the same output as the original convolution by applying kernel-based and channel-based distributed offsets.

[0063] The face background segmentation model takes a face image captured by the front end as input, processes it, and outputs a mask vector of the segmented face image. The mask vector uses black and white to represent the segmented face image; the black portion represents the background, and the white portion represents the face. Then, the black portion of the mask vector in the original face image is taken as the face image background.

[0064] Step S103: The similarity is compared between the real-time vector dataset and the fraud mapping relationship stored in the vector database to obtain the final similarity. Based on the magnitude of the final similarity, it is determined whether the real-time face image represents a fraud scene. The real-time vector dataset includes the real-time background image vector, the real-time timestamp vector, and the real-time address vector. The fraud mapping relationship is the mapping relationship between the fraud scene image vector, the timestamp vector, and the address vector.

[0065] Unlike previous methods that used classic deep neural networks for feature representation, this application uses an encoder and decoder architecture to segment the face and background. Furthermore, it compares background image vectors with fraud mapping relationships stored in a vector database. This improves the accuracy of determining whether the real-time face image represents a fraudulent scenario. By comparing the similarity between the real-time vector dataset and the fraud mapping relationships stored in the vector database, real-time monitoring is achieved while maintaining accuracy. This solves the problem that existing solutions using clustering algorithms for face image background detection cannot achieve real-time monitoring.

[0066] By combining vector database technology and comprehensively considering the similarity of facial image backgrounds with transaction information, the system ensures the real-time and efficient monitoring of transaction risks. It deeply integrates a lightweight facial background segmentation model with efficient vector database retrieval technology to achieve rapid response to fraudulent transactions and effectively ensure the real-time nature of fraud detection. By combining facial backgrounds with vector databases, the system aims to efficiently handle highly similar facial backgrounds in risk detection, thereby significantly improving the efficiency and accuracy of risk detection.

[0067] The mapping relationship in the vector database is implemented as follows: An image vector model (MP) is used to process the security background dataset and the fraud background dataset into vector form (V_Back). A text vector model (MT) is used to process the timestamp dataset and the address information dataset into vector form (V_Time, V_Loc). The processed image vectors and text vectors are then mapped one-to-one and stored in the vector database. The data format is [serial number, account number, image vector, time text vector (i.e., timestamp vector), location text vector (i.e., address vector), fraud identifier].

[0068] The sequence number is an identifier starting from 1; the account is the account to which the face data belongs; the image vector (V_Back) is the vector obtained by processing the background image through the image vector model; the time text vector (V_Time) is the vector obtained by processing the timestamp data through the text vector model; the address text vector (V_Loc) is the vector obtained by processing the address information data through the text vector model; and the fraud label (Label) indicates whether this data is fraudulent.

[0069] Step S103 involves comparing the similarity between the real-time vector dataset and the fraud mapping relationship stored in the vector database to obtain the final similarity score, including:

[0070] The real-time background image vector is compared with the similarity of all the fraud scene image vectors in the fraud mapping relationship to obtain multiple first similarities. The real-time background image vector is compared with the similarity of all the fraud scene image vectors in the fraud mapping relationship to obtain multiple second similarities. The real-time background image vector is compared with the similarity of all the fraud scene image vectors in the fraud mapping relationship to obtain multiple third similarities.

[0071] The first target similarity is determined to be the maximum value among all the aforementioned first similarities, the second target similarity is determined to be the maximum value among all the aforementioned second similarities, and the third target similarity is determined to be the maximum value among all the aforementioned third similarities;

[0072] The first target similarity, the second target similarity, and the third target similarity are weighted and summed to obtain the final similarity.

[0073] Specifically, the first target similarity, the second target similarity, and the third target similarity can be 0.7, 0.15, and 0.15, respectively. Determining the first target similarity as the maximum value among all the first similarities can minimize fraud scenarios. The same applies to the second target similarity and the third target similarity, so they will not be elaborated on here.

[0074] Step S103, determining whether the real-time face image represents a fraud scenario based on the final similarity score, includes:

[0075] If the final similarity is greater than or equal to the first similarity threshold, the above real-time face image is determined to represent a fraud scenario.

[0076] Specifically, if the final similarity is greater than or equal to the first similarity threshold, the real-time face image is identified as representing a fraudulent scenario.

[0077] In one embodiment of this application, the vector database further stores security mapping relationships, which are mapping relationships between security scene image vectors, timestamp vectors, and address vectors. After obtaining the final similarity, the method further includes:

[0078] When the final similarity is greater than or equal to the second similarity threshold and less than the first similarity threshold, the real-time background image is compared with all the security scene image vectors in the security mapping relationship stored in the vector database to obtain multiple fourth similarities. Based on the magnitude of all the fourth similarities, the fraud mapping relationship and the security mapping relationship in the vector database are corrected to obtain the corrected fraud mapping relationship and the corrected security mapping relationship.

[0079] If the final similarity is greater than or equal to the second similarity threshold and less than the first similarity threshold, the real-time background image is compared with all the security scene image vectors in the security mapping relationship stored in the vector database to obtain multiple fourth similarities, which are used to further avoid fraud scenarios.

[0080] Specifically, based on the magnitude of all the aforementioned fourth similarities, the fraud mapping relationships and security mapping relationships in the aforementioned vector database are modified to obtain the modified fraud mapping relationships and modified security mapping relationships, including:

[0081] The target security scene image vector is determined to be the security scene image vector corresponding to the fourth similarity that is greater than or equal to the third similarity threshold.

[0082] When the current proportion is greater than or equal to the preset proportion, the above target security scene image vector, the above timestamp vector and the above address vector corresponding to the above target security scene image vector are stored in the above fraud mapping relationship to obtain the above corrected fraud mapping relationship. The above current proportion is the proportion of the above fourth similarity that is greater than or equal to the third similarity threshold in all the above fourth similarities.

[0083] Additionally, real-time background images, real-time timestamp vectors, and real-time address vectors can be stored in the fraud mapping relationship.

[0084] Delete the target security scene image vector, the timestamp vector, and the address vector corresponding to the target security scene image vector from the above security mapping relationship to obtain the above-corrected security mapping relationship.

[0085] If the final similarity is less than the second similarity threshold, the real-time face image representation security scenario is determined.

[0086] Specifically, the first similarity threshold can be 85%, the second similarity threshold can be 75%, and the preset proportion can be 1 / 2. When the current proportion is greater than or equal to the preset proportion, the target security scene image vector, the timestamp vector and the address vector corresponding to the target security scene image vector are stored in the fraud mapping relationship to obtain the corrected fraud mapping relationship. The target security scene image vector, the timestamp vector and the address vector corresponding to the target security scene image vector are deleted from the security mapping relationship to obtain the corrected security mapping relationship. This can further avoid fraud scenarios and update the stored mapping relationship of the vector database to achieve the purpose of real-time updating of the vector database, thereby improving the accuracy of judgment.

[0087] In one embodiment of this application, after determining the aforementioned real-time face image representation fraud scenario, the method further includes:

[0088] An alarm message is generated to indicate that the real-time facial image represents the aforementioned fraud scenario.

[0089] Specifically, by generating alarm information to alert users to the aforementioned real-time facial image representing the fraud scenario, fraud can be avoided.

[0090] like Figure 2 As shown, the anti-fraud system adopts a microservice architecture to ensure its robustness, flexibility, and scalability. In this architecture, functional components are deployed and maintained as microservices, achieving a highly modular and loosely coupled system structure. The system can flexibly respond to changes in business scale, effectively balancing resource utilization and performance requirements through a dynamic scaling mechanism. The system architecture mainly consists of a presentation layer, a control layer, a service layer, and a storage layer. The following sections will provide a detailed introduction to each layer of the system architecture.

[0091] The presentation layer uses Vue + ElementUI technology for front-end page display, and uses JSON format data for front-end and back-end interaction. The back-end uses API interfaces encapsulated by the Java development platform to provide JSON data to the front-end. Vue adopts a pure front-end and back-end separation front-end technology framework, and automatically updates data to interface elements through data binding. Front-end developers obtain data by requesting back-end interfaces and bind it to property variables in Vue components, eliminating the need to directly manipulate DOM elements (DOM elements are elements in an HTML document, i.e., nodes in the Document Object Model. Each element represents a part of the document, such as paragraphs, headings, links, etc. Through the DOM, these elements can be manipulated and modified using JavaScript, such as adding new elements, modifying element styles, and getting element content. Common DOM elements include div, p, span, and a), simplifying interface operations.

[0092] The control layer implements functions such as service routing, service rate limiting, service circuit breaking, application authentication, service degradation, routing policies, and capacity load balancing. It mainly uses Ribbon to achieve load balancing between services and Sentinel to achieve service rate limiting.

[0093] Ribbon is a load balancer based on HTTP and TCP that helps us achieve load balancing between services. By inserting Ribbon between the client and server, we can achieve load balancing for multiple service instances, improving system performance and availability.

[0094] Sentinel is an open-source traffic control and service protection component that helps us implement functions such as rate limiting, circuit breaking, and degradation for services. By adding Sentinel to the service call chain, we can effectively prevent traffic overload and protect the system from malicious attacks or abnormal traffic.

[0095] The service layer is the core of the anti-fraud system, encompassing various business logic and data processing functions. It utilizes a microservice platform developed based on Java, implementing specific business logic processing through modularization. A service registry and configuration center (Nacos is an open-source distributed configuration center and service discovery framework that helps developers implement dynamic configuration management, service discovery, and service registration. It provides a RESTful API-based operation interface (RESTful APIs are interfaces that perform CRUD operations on resources via HTTP requests), supports multiple languages ​​and environments, and can be integrated with various microservice frameworks, making it a very practical microservice infrastructure component) are used to implement registration and management between modules. The presentation layer and service layer are connected via Spring Cloud Gateway (Spring Cloud Gateway is an API gateway based on Spring Framework 5, Spring Boot 2, and Project Reactor, providing unified routing, filtering, rate limiting, and monitoring capabilities. Spring Cloud Gateway makes it easy to build and manage API gateways in microservice architectures, achieving a unified entry point and security control for external requests). Service calls within the system use the POST method for communication (POST is a way to send data over the network. In this method, when a client sends a request to the server, the requested data is included in the request body and transmitted encrypted. Compared to GET, POST is more secure and suitable for transmitting sensitive information. POST is typically used for submitting form data to the server, uploading files, etc.). An eventual consistency scheme is adopted for data transactions, using a compensation pattern to guarantee data consistency. Logs are implemented using ELK technology, receiving external system messages through a message broker for data synchronization notifications. In terms of software architecture, database operations are implemented through object-relational mapping, using a combination of JDBCTemplate and MyBatis Plus. (JDBCTemplate and MyBatis Plus are two different persistence frameworks, each providing different ways to operate on the database. JDBCTemplate is a simple JDBC operation template provided by the Spring framework, allowing direct JDBC database operations. MyBatis Plus is an enhancement tool for MyBatis, providing a more convenient way to operate on the database, encapsulating some commonly used operations, and simplifying development.)

[0096] The storage layer introduces a vector database to support face-background similarity detection, thereby determining whether a transaction constitutes fraud. The vector database is implemented using a pre-built Milvus library, which converts background images into feature vectors and uses these feature vectors to construct a vector dataset. Milvus is a vector database that can be built from user-input datasets, primarily used for vector similarity search to accelerate unstructured data retrieval.

[0097] The system uses PostgreSQL and Redis. PostgreSQL is used for application information, while Redis is mainly used to store streaming information, such as photos of faces to be identified.

[0098] PostgreSQL is an open-source relational database management system that supports advanced features and complex queries. It offers a rich set of features, including transaction support, full ACID compatibility, multi-version concurrency control, replication, and disaster recovery. PostgreSQL is commonly used to store structured data and is widely used in a variety of applications.

[0099] Redis is an open-source, in-memory data structure store, also known as a caching database. It supports various data structures such as strings, lists, sets, sorted sets, and hash tables, and provides fast read and write operations. Redis is typically used to cache frequently accessed data to improve application performance and responsiveness.

[0100] Combining PostgreSQL and Redis enables more powerful data management and storage capabilities. For example, structured data can be stored in PostgreSQL while using Redis as a caching layer to improve data read speed and performance. Furthermore, Redis can be used as a message queue or publish / subscribe system, and its integration with PostgreSQL allows for more complex data processing and application scenarios. Therefore, using PostgreSQL and Redis together provides applications with better performance, scalability, and flexibility.

[0101] To enable those skilled in the art to better understand the technical solution of this application, the implementation process of the face background similarity recognition method of this application will be described in detail below with reference to specific embodiments.

[0102] This embodiment relates to a specific method for facial background similarity recognition, such as... Figure 3 As shown, it includes:

[0103] Real-time data is acquired, including real-time face images, real-time timestamp data, and real-time address data. An encoder and decoder architecture is used to perform face image segmentation on the real-time face images in the real-time data to obtain real-time background images. An image vector model is then used to vectorize the real-time background images to obtain real-time background image vectors. A text vector model is then used to vectorize the real-time timestamp data and real-time address data to obtain real-time timestamp vectors and real-time address vectors, respectively.

[0104] The real-time background image vector is compared with the similarity of all fraud scene image vectors in the fraud mapping relationship to obtain multiple first similarity scores. The real-time background image vector is compared with the similarity of all fraud scene image vectors in the fraud mapping relationship to obtain multiple second similarity scores. The real-time background image vector is compared with the similarity of all fraud scene image vectors in the fraud mapping relationship to obtain multiple third similarity scores.

[0105] The first target similarity is determined as the maximum value among all first similarities, the second target similarity is determined as the maximum value among all second similarities, and the third target similarity is determined as the maximum value among all third similarities;

[0106] The weighted summation of the similarity scores of the first target, the second target, and the third target is used to obtain the final similarity score.

[0107] If the final similarity is greater than or equal to the first similarity threshold, the real-time face image representation fraud scenario is determined.

[0108] If the final similarity is less than the second similarity threshold, the real-time face image representation security scenario is determined.

[0109] If the final similarity is greater than or equal to the second similarity threshold and less than the first similarity threshold, the similarity is compared between the real-time background image and all security scene image vectors in the security mapping relationship stored in the vector database to obtain multiple fourth similarities.

[0110] The target security scene image vector is determined to be the security scene image vector corresponding to a fourth similarity greater than or equal to the third similarity threshold;

[0111] If the current proportion is greater than or equal to the preset proportion, the target security scene image vector and the timestamp vector and address vector corresponding to the target security scene image vector are stored in the fraud mapping relationship to obtain the corrected fraud mapping relationship. The current proportion is the proportion of the fourth similarity that is greater than or equal to the third similarity threshold among all fourth similarities. The target security scene image vector and the timestamp vector and address vector corresponding to the target security scene image vector are deleted from the security mapping relationship to obtain the corrected security mapping relationship. The real-time vector dataset includes real-time background image vector, real-time timestamp vector and real-time address vector, and the fraud mapping relationship is the mapping relationship between fraud scene image vector, timestamp vector and address vector.

[0112] After identifying a real-time facial image fraud scenario, an alarm message is generated to alert the user to the real-time facial image fraud scenario.

[0113] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0114] This application also provides a face background similarity recognition device. It should be noted that the face background similarity recognition device of this application can be used to execute the face background similarity recognition method provided in this application. This device is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0115] The following describes the face background similarity recognition device provided in the embodiments of this application.

[0116] Figure 4 This is a structural block diagram of a face background similarity recognition device provided according to an embodiment of this application.

[0117] like Figure 4 As shown, the device includes:

[0118] Acquisition unit 41 is used to acquire real-time data, including real-time face images, real-time timestamp data, and real-time address data;

[0119] The first processing unit 42 is used to perform face image segmentation processing on the real-time face image in the real-time data using an encoder and decoder architecture to obtain a real-time background image, and to perform vectorization processing on the real-time background image using an image vector model to obtain a real-time background image vector, and to perform vectorization processing on the real-time timestamp data and the real-time address data using a text vector model to obtain a real-time timestamp vector and a real-time address vector, respectively.

[0120] The second processing unit 43 is used to compare the similarity between the real-time vector dataset and the fraud mapping relationship stored in the vector database to obtain the final similarity, and to determine whether the real-time face image represents a fraud scene based on the magnitude of the final similarity. The real-time vector dataset includes the real-time background image vector, the real-time timestamp vector and the real-time address vector, and the fraud mapping relationship is the mapping relationship between the fraud scene image vector, the timestamp vector and the address vector.

[0121] Unlike previous methods that used classic deep neural networks for feature representation, the above-described device uses an encoder and decoder architecture to segment the face and background. Furthermore, this application compares background image vectors with fraud mapping relationships stored in a vector database. This improves the accuracy of determining whether the real-time face image represents a fraudulent scenario. By comparing the similarity between the real-time vector dataset and the fraud mapping relationships stored in the vector database, real-time monitoring is achieved while maintaining accuracy. This solves the problem that existing solutions using clustering algorithms for face image background detection cannot achieve real-time monitoring.

[0122] In one embodiment of this application, the second processing unit includes a first processing module, a first determining module, and a second processing module.

[0123] The first processing module is used to compare the real-time background image vector with all the fraud scene image vectors in the fraud mapping relationship to obtain multiple first similarities, and to compare the real-time background image vector with all the fraud scene image vectors in the fraud mapping relationship to obtain multiple second similarities, and to compare the real-time background image vector with all the fraud scene image vectors in the fraud mapping relationship to obtain multiple third similarities.

[0124] The first determining module is used to determine the first target similarity as the maximum value among all the aforementioned first similarities, and to determine the second target similarity as the maximum value among all the aforementioned second similarities, and to determine the third target similarity as the maximum value among all the aforementioned third similarities;

[0125] The second processing module is used to perform a weighted summation of the first target similarity, the second target similarity, and the third target similarity to obtain the final similarity.

[0126] In one embodiment of this application, the second processing unit includes a second determining module.

[0127] The second determining module is used to determine the above-mentioned real-time face image representation fraud scenario when the final similarity is greater than or equal to the first similarity threshold.

[0128] In one embodiment of this application, the above-mentioned device further includes a third processing unit and a fourth processing unit. The vector database also stores security mapping relationships, which are mapping relationships between security scene image vectors, timestamp vectors, and address vectors. After obtaining the final similarity,

[0129] The third processing unit is used to compare the similarity between the real-time background image and all the security scene image vectors in the security mapping relationship stored in the vector database when the final similarity is greater than or equal to the second similarity threshold and the final similarity is less than the first similarity threshold, to obtain multiple fourth similarities, and to correct the fraud mapping relationship and the security mapping relationship in the vector database according to the magnitude of all the fourth similarities, so as to obtain the corrected fraud mapping relationship and the corrected security mapping relationship.

[0130] The fourth processing unit is used to determine the real-time face image representation security scenario when the final similarity is less than the second similarity threshold.

[0131] In one embodiment of this application, the third processing unit includes a third determining module, a third processing module, and a fourth processing module.

[0132] The third determining module is used to determine the target security scene image vector as the security scene image vector corresponding to the fourth similarity that is greater than or equal to the third similarity threshold.

[0133] The third processing module is used to store the target security scene image vector, the timestamp vector and the address vector corresponding to the target security scene image vector into the fraud mapping relationship when the current proportion is greater than or equal to the preset proportion, so as to obtain the corrected fraud mapping relationship. The current proportion is the proportion of the fourth similarity that is greater than or equal to the third similarity threshold in all the fourth similarities.

[0134] The fourth processing module is used to delete the target security scene image vector, the timestamp vector, and the address vector corresponding to the target security scene image vector from the above security mapping relationship, so as to obtain the above-mentioned corrected security mapping relationship.

[0135] In one embodiment of this application, the above-mentioned device further includes a generation unit. After determining that the real-time face image represents a fraud scenario, the generation unit generates alarm information to alert that the real-time face image represents the fraud scenario.

[0136] In one embodiment of this application, the encoder in the encoder and decoder architecture described above adopts an inverted bottleneck structure, and the decoder in the encoder and decoder architecture described above adopts a distributed offset convolution structure.

[0137] The aforementioned face background similarity recognition device includes a processor and a memory. The acquisition unit, the first processing unit, and the second processing unit are all stored as program units in the memory, and the processor executes the program units stored in the memory to achieve the corresponding functions. All of the above modules are located in the same processor; alternatively, the modules may be located in different processors in any combination.

[0138] The processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and adjusting kernel parameters can address the issue that clustering algorithms used in existing face image background detection methods cannot achieve real-time monitoring.

[0139] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0140] This invention provides a computer-readable storage medium including a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the face background similarity recognition method.

[0141] This invention provides a processor for running a program, wherein the program executes the face background similarity recognition method.

[0142] This invention provides a device including a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs at least the following steps: acquiring real-time data, including real-time face images, real-time timestamp data, and real-time address data; using an encoder and decoder architecture, performing face image segmentation processing on the real-time face images in the real-time data to obtain a real-time background image, and using an image vector model to vectorize the real-time background image to obtain a real-time background image vector; and using a text vector model to vectorize the real-time timestamp data and the real-time address data to obtain a real-time timestamp vector and a real-time address vector, respectively; comparing the similarity between the real-time vector dataset and a fraud mapping relationship stored in a vector database to obtain a final similarity, and determining whether the real-time face image represents a fraud scenario based on the magnitude of the final similarity. The real-time vector dataset includes the real-time background image vector, the real-time timestamp vector, and the real-time address vector, and the fraud mapping relationship is a mapping relationship between fraud scenario image vectors, timestamp vectors, and address vectors. The device described herein can be a server, PC, PAD, mobile phone, etc.

[0143] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having at least the following method steps: acquiring real-time data, the real-time data including real-time face images, real-time timestamp data, and real-time address data; employing an encoder and decoder architecture to perform face image segmentation processing on the real-time face images in the real-time data to obtain a real-time background image, and employing an image vector model to vectorize the real-time background image to obtain a real-time background image vector, and employing a text vector model to vectorize the real-time timestamp data and the real-time address data respectively to obtain a real-time timestamp vector and a real-time address vector; comparing the similarity between the real-time vector dataset and the fraud mapping relationship stored in the vector database to obtain a final similarity, and determining whether the real-time face image represents a fraud scenario based on the magnitude of the final similarity, wherein the real-time vector dataset includes the real-time background image vector, the real-time timestamp vector, and the real-time address vector, and the fraud mapping relationship is a mapping relationship between fraud scenario image vectors, timestamp vectors, and address vectors.

[0144] This application also provides a face background similarity recognition system, comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include methods for performing any of the aforementioned face background similarity recognition methods. Unlike previous methods based on classic deep neural networks for feature representation, this application uses an encoder and decoder architecture to segment the face and background. Furthermore, it uses background image vector comparison, comparing against a fraud mapping relationship stored in a vector database. This improves the accuracy of determining whether the real-time face image represents a fraudulent scenario. By comparing the similarity between a real-time vector dataset and the fraud mapping relationship stored in the vector database, real-time monitoring is achieved while maintaining accuracy, thus solving the problem that existing solutions using clustering algorithms for face image background detection cannot achieve real-time monitoring.

[0145] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0146] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0147] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0148] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0149] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0150] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0151] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0152] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0153] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0154] As can be seen from the above description, the embodiments of this application achieve the following technical effects:

[0155] 1) The face background similarity recognition method of this application differs from previous methods based on classic deep neural networks for feature representation. It achieves face and background segmentation by using an encoder and decoder architecture. Furthermore, this application uses background image vector comparison, and the comparison object is the fraud mapping relationship stored in the vector database. This improves the accuracy of determining whether the real-time face image represents a fraud scene. Moreover, by using a real-time vector dataset and the fraud mapping relationship stored in the vector database for similarity comparison, the goal of real-time monitoring is achieved while ensuring accuracy. This solves the problem that existing solutions using clustering algorithms for face image background detection cannot achieve real-time monitoring.

[0156] 2) The face background similarity recognition device of this application differs from previous methods based on classic deep neural networks for feature representation. It achieves face and background segmentation by using an encoder and decoder architecture. Furthermore, this application uses background image vector comparison, and the comparison object is the fraud mapping relationship stored in the vector database. This improves the accuracy of determining whether the real-time face image represents a fraudulent scenario. Moreover, by using a real-time vector dataset and the fraud mapping relationship stored in the vector database for similarity comparison, the purpose of real-time monitoring is achieved while ensuring accuracy. This solves the problem that existing solutions using clustering algorithms for face image background detection cannot achieve real-time monitoring.

[0157] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for recognizing the similarity of a face and its background, characterized in that, include: Acquire real-time data, including real-time face images, real-time timestamp data, and real-time address data; An encoder and decoder architecture is used to perform face image segmentation processing on the real-time face image in the real-time data to obtain a real-time background image. An image vector model is used to vectorize the real-time background image to obtain a real-time background image vector. A text vector model is used to vectorize the real-time timestamp data and the real-time address data to obtain a real-time timestamp vector and a real-time address vector, respectively. The similarity is compared between the real-time vector dataset and the fraud mapping relationship stored in the vector database to obtain the final similarity. Based on the magnitude of the final similarity, it is determined whether the real-time face image represents a fraud scene. The real-time vector dataset includes the real-time background image vector, the real-time timestamp vector, and the real-time address vector. The fraud mapping relationship is the mapping relationship between the fraud scene image vector, the timestamp vector, and the address vector. The method involves comparing the real-time vector dataset with the fraud mapping relationship stored in the vector database to obtain a final similarity score. This includes: comparing the real-time background image vector with all the fraud scene image vectors in the fraud mapping relationship to obtain multiple first similarities; comparing the real-time timestamp data with all the timestamp vectors in the fraud mapping relationship to obtain multiple second similarities; and comparing the real-time address vector with all the address vectors in the fraud mapping relationship to obtain multiple third similarities. A first target similarity score is determined as the maximum value among all the first similarities, a second target similarity score is determined as the maximum value among all the second similarities, and a third target similarity score is determined as the maximum value among all the third similarities. Finally, a weighted sum of the first, second, and third target similarities is performed to obtain the final similarity score.

2. The method according to claim 1, characterized in that, Determining whether the real-time facial image represents a fraud scenario based on the final similarity score includes: If the final similarity is greater than or equal to the first similarity threshold, the real-time face image is determined to represent a fraud scenario.

3. The method according to claim 1, characterized in that, The vector database also stores security mapping relationships, which are mapping relationships between security scene image vectors, timestamp vectors, and address vectors. After obtaining the final similarity, the method further includes: When the final similarity is greater than or equal to the second similarity threshold and the final similarity is less than the first similarity threshold, the real-time background image is compared with all the security scene image vectors in the security mapping relationship stored in the vector database to obtain multiple fourth similarities. Based on the magnitude of all the fourth similarities, the fraud mapping relationship and the security mapping relationship in the vector database are corrected to obtain the corrected fraud mapping relationship and the corrected security mapping relationship. If the final similarity is less than the second similarity threshold, the real-time face image is determined to represent a safe scenario.

4. The method according to claim 3, characterized in that, Based on the magnitudes of all the fourth similarities, the fraud mapping relationships and security mapping relationships in the vector database are corrected to obtain corrected fraud mapping relationships and corrected security mapping relationships, including: The target security scene image vector is determined to be the security scene image vector corresponding to the fourth similarity that is greater than or equal to the third similarity threshold; When the current proportion is greater than or equal to the preset proportion, the target security scene image vector, the timestamp vector and the address vector corresponding to the target security scene image vector are stored in the fraud mapping relationship to obtain the corrected fraud mapping relationship. The current proportion is the proportion of the fourth similarity that is greater than or equal to the third similarity threshold in all the fourth similarities. The corrected security mapping relationship is obtained by deleting the target security scene image vector, the timestamp vector, and the address vector corresponding to the target security scene image vector from the security mapping relationship.

5. The method according to any one of claims 1 to 4, characterized in that, After determining the real-time face image representation of the fraud scenario, the method further includes: An alarm message is generated to alert the user that the real-time facial image represents the fraud scenario.

6. The method according to any one of claims 1 to 4, characterized in that, The encoder in the encoder-decoder architecture adopts an inverted bottleneck structure, and the decoder in the encoder-decoder architecture adopts a distributed offset convolution structure.

7. A facial background similarity recognition device, characterized in that, include: The acquisition unit is used to acquire real-time data, which includes real-time face images, real-time timestamp data, and real-time address data. The first processing unit is used to perform face image segmentation processing on the real-time face image in the real-time data using an encoder and decoder architecture to obtain a real-time background image, and to perform vectorization processing on the real-time background image using an image vector model to obtain a real-time background image vector, and to perform vectorization processing on the real-time timestamp data and the real-time address data using a text vector model to obtain a real-time timestamp vector and a real-time address vector, respectively. The second processing unit is used to compare the similarity between the real-time vector dataset and the fraud mapping relationship stored in the vector database to obtain the final similarity, and to determine whether the real-time face image represents a fraud scene based on the magnitude of the final similarity. The real-time vector dataset includes the real-time background image vector, the real-time timestamp vector and the real-time address vector, and the fraud mapping relationship is a mapping relationship between the fraud scene image vector, the timestamp vector and the address vector. The second processing unit includes a first processing module, a first determining module, and a second processing module. The first processing module is used to compare the real-time background image vector with all the fraud scene image vectors in the fraud mapping relationship to obtain multiple first similarities, and to compare the real-time timestamp data with all the timestamp vectors in the fraud mapping relationship to obtain multiple second similarities, and to compare the real-time address vector with all the address vectors in the fraud mapping relationship to obtain multiple third similarities. The first determining module is used to determine the first target similarity as the maximum value among all the first similarities, the second target similarity as the maximum value among all the second similarities, and the third target similarity as the maximum value among all the third similarities. The second processing module is used to perform a weighted summation of the first target similarity, the second target similarity, and the third target similarity to obtain the final similarity.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the face background similarity recognition method according to any one of claims 1 to 6.

9. A facial background similarity recognition system, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including methods for performing a face background similarity recognition method according to any one of claims 1 to 6.

Images