A method and system for detecting a live attack

Abstract

The present specification provides a live body attack detection method and system, and relates to the technical field of biometric identification. The live body attack detection method and system provided by the present specification can fully utilize the background information in the target image of the target object while performing live body attack detection using the biological characteristics of the target object and performing live body detection on the target object based on the background correlation degree. Double detection is used when performing face recognition on the target object, thereby further improving the security of live body attack detection.

Description

Technical Field

[0001] This specification relates to the field of biometric technology, and in particular to a method and system for detecting live attacks. Background Technology

[0002] Liveness detection attacks are one of the most common security risks faced by facial recognition systems. These attacks directly use photos or videos to impersonate individuals and attempt to steal user information and assets. To accurately detect liveness attacks, the biometric features of the target are typically used as the standard for determining whether the target is a liveness attacker. However, with the evolution of liveness attack methods, relying solely on biometric features for detection still presents numerous problems and challenges.

[0003] Therefore, there is a need to provide a more accurate and secure method and system for detecting live attacks.

[0004] The information in the background section is merely information known only to the inventor and does not imply that such information had entered the public domain before the date of this application, nor does it imply that it can be considered prior art in this disclosure. Summary of the Invention

[0005] The main purpose of this specification is to provide a method and system for detecting live attacks.

[0006] In a first aspect, this specification provides a liveness detection method, the liveness detection method comprising the following steps: obtaining a target image, the target image including a target image of a target object and a target background image excluding the target image; determining, at least based on the biometrics of the target object and the background correlation, whether the target image is a liveness detection result of a liveness attack image, the background correlation being the correlation between the target background image and a set of reference background images; and outputting the liveness attack detection result.

[0007] In some embodiments, the liveness attack detection method further includes: obtaining K candidate objects, where K is an integer greater than or equal to 1; determining N associated objects from the K candidate objects, where the historical correlation degree between each associated object and the target object is greater than a threshold, where N is an integer greater than or equal to 1; and determining the reference background image set corresponding to each of the N associated objects.

[0008] In some embodiments, the reference background image set includes background images of all locations of the N associated objects within the predetermined historical period.

[0009] In some embodiments, the determining the N associated objects comprises, for each candidate object, obtaining M historical positions of the target object in a predetermined historical period and a historical time ti corresponding to any one of the M historical positions Pi, wherein M is an integer greater than or equal to 1, and i is any integer in [1, M]; obtaining M target ranges corresponding to a preset position neighborhood of the M historical positions; obtaining M target time periods corresponding to a preset time neighborhood of the M historical times; determining N candidate objects in the M target time periods and appearing in the M target ranges more than a threshold number of times from the plurality of candidate objects as the N associated objects.

[0010] In some embodiments, the determining the liveness attack detection result of the target image based on at least the biological feature of the target object and the background correlation degree comprises: determining that a liveness attack probability of the target image is greater than a preset value; and determining the liveness attack detection result based on the background correlation degree.

[0011] In some embodiments, the determining the liveness attack detection result based on the background correlation degree comprises: decoupling the target image to obtain the target image and the target background image not including the target image; determining a target feature of the target background image and a plurality of reference features corresponding to a plurality of reference background images in the reference background image set; determining the background correlation degree of the target feature and the plurality of reference features; and determining the liveness attack detection result based on the background correlation degree.

[0012] In some embodiments, the target image is decoupled by a semantic segmentation network to obtain the target image and the target background image; and the target feature and the plurality of reference features are determined by a residual neural network.

[0013] In some embodiments, the determining the background correlation degree comprises: determining a background similarity set comprising a plurality of similarities of the target feature and the plurality of reference features; taking a maximum similarity in the plurality of similarities as a first similarity; taking an average value of the plurality of similarities as a second similarity; and determining a first correlation threshold and a second correlation threshold.

[0014] In some embodiments, the determining the similarity set comprises: obtaining a similarity between the target feature vector and each reference feature vector in the plurality of reference feature vectors based on a cosine similarity algorithm.

[0015] In some embodiments, the determining the living body detection result of the target object based on the background correlation degree comprises: when the first similarity is less than the first correlation threshold and the second similarity is less than the second correlation threshold, determining that the target object is a living body attack.

[0016] In some embodiments, the determining the living body detection result of the target object based on the background correlation degree comprises: when the first similarity is greater than the first correlation threshold, determining that the target object is a living body.

[0017] In some embodiments, the determining the living body detection result of the target object based on the background correlation degree comprises: when the second similarity is greater than the second correlation threshold, determining that the target object is a living body.

[0018] In some embodiments, the biological feature is at least one of a face feature, a fingerprint feature, a retina feature, a palmprint feature, a skeletal projection feature, and a dental feature.

[0019] In a second aspect, the present specification provides a living body attack detection system, comprising: at least one storage medium comprising at least one instruction set for implementing the living body detection method; and at least one processor in communication connection with the at least one storage medium, wherein when the system is running, the at least one processor reads the at least one instruction set and executes the method according to the instructions of the at least one instruction set.

[0020] In some embodiments, the living body attack detection system further comprises: an image acquisition unit in communication connection with the at least one processor, the image acquisition unit receiving an image acquisition instruction issued by the at least one processor to perform an image acquisition task according to the image acquisition instruction, the image acquisition task comprising acquiring a target image of a target object.

[0021] In some embodiments, the living body attack detection system further comprises: a positioning system in communication connection with the at least one processor, the positioning system being configured to upload position information of the image acquisition unit to the at least one processor after obtaining authorization of the target object.

[0022] In some embodiments, the living body attack detection system is in communication connection with a database to obtain historical positions of the target object in a predetermined historical period from the database during the living body detection process of the target object.

[0023] In some embodiments, the historical position information of the target object comprises a collection position of the image acquisition unit that has collected the target image in the predetermined historical period.

[0024] From the above technical solutions, the present specification provides a method for live body attack detection and a system for performing the method. The method for live body attack detection can fully utilize the background information in the target image of the target object and the information of the associated object of the target object by simultaneously utilizing the biological features of the target object and the background correlation degree for live body attack detection, thereby improving the accuracy of detecting whether the target object is a live body attack result.

[0025] Other functions of the method and system for live body attack detection provided by the present specification will be partially listed in the following description. According to the description, the following numbers and examples will be apparent to those of ordinary skill in the art. The creative aspects of the method and system for live body attack detection provided by the present specification can be fully explained by practicing or using the methods, devices and combinations described in the following detailed examples. BRIEF DESCRIPTION OF DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present specification, the following will briefly introduce the drawings used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present specification, and those of ordinary skill in the art can also obtain other drawings according to these drawings without creative labor.

[0027] Figure 1A A schematic diagram of an application scenario of a method for live body attack detection according to some embodiments of the present specification is shown;

[0028] Figure 1B A schematic diagram of a scenario of live body attack detection of a target object according to some embodiments of the present specification is shown;

[0029] Figure 2 A schematic diagram of the structure of a device provided according to some embodiments of the present specification is shown;

[0030] Figure 3 A method flowchart of a method for live body attack detection provided according to some embodiments of the present specification is shown;

[0031] Figure 4 A historical trajectory diagram of a target object and an associated object provided according to some embodiments of the present specification is shown; and

[0032] Figure 5 A method flowchart of an associated object of a target object provided according to some embodiments of the present specification is shown. DETAILED DESCRIPTION

[0033] The following description provides specific examples of the application and the requirements of the specification, in order to enable a person skilled in the art to manufacture and use the content of the specification. Various local modifications of the disclosed embodiments are obvious to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of the specification. Therefore, the specification is not limited to the embodiments shown, but is consistent with the widest scope of the claims.

[0034] The terms used herein are only for the purpose of describing specific example embodiments, and are not limiting. For example, unless the context clearly indicates otherwise, as used herein, the singular forms "a", "an", and "the" can also include the plural forms. When used in the specification, the terms "comprise", "contain" and / or "include" mean that the associated integers, steps, operations, elements and / or components exist, but do not exclude the presence of one or more other features, integers, steps, operations, elements, components and / or groups that can be added to the system / method.

[0035] In view of the following description, these features of the specification and other features, as well as the operation and function of related elements of the structure, and the economy of the combination and manufacture of components can be significantly improved. Referring to the drawings, all of which form part of the specification. However, it should be clearly understood that the drawings are for illustrative and descriptive purposes only and are not intended to limit the scope of the specification. It should also be understood that the drawings are not drawn to scale.

[0036] The flowchart used in the specification shows the operation of the system implementation according to some embodiments of the specification. It should be clearly understood that the operations of the flowchart can not be implemented in sequence. On the contrary, the operations can be implemented in reverse order or simultaneously. In addition, one or more other operations can be added to the flowchart. One or more operations can be removed from the flowchart.

[0037] For the convenience of description, the terms appearing in the specification are first explained as follows:

[0038] Live detection: also known as live anti-attack detection, refers to the technology of detecting and intercepting live attacks by using artificial intelligence models or other technical means, etc. Live detection is mainly carried out by identifying physiological feature information on the live body, and taking physiological feature information as a biological feature to distinguish biological features that are forged by non-biological substances such as photos, silica gel and plastic. When the biological feature information is obtained from a legal user, it can be identified that the biological feature is obtained from a legal user with a biological live body. The biological feature can include but is not limited to one or more of face features, fingerprint features, iris features, palm print features, dental features, and skeletal features.

[0039] Live attack: refers to an attack means that presents the biometric features (such as a human face) of a human body by using a screen, paper, a photo, a mask, etc., to try to bypass a human biometric feature recognition system (such as a human face recognition system), for example, a live attack by displaying a human face image through a mobile phone screen, a live attack by printing a human face image on paper.

[0040] Before the specific embodiments of the present specification are described, the application scenarios of the present specification are introduced as follows:

[0041] The live attack detection method provided by the present specification can be applied to a live attack detection scene in any biometric feature recognition process, for example, in a human face payment, human face recognition, or palm print payment scene, the original image of the biometric feature of the user to be paid or to be recognized collected can be detected by the live attack detection method of the present specification; in an identity verification scene, the original image of the biometric feature of the user collected can be detected by the live detection method of the present specification; the live detection method can also be applied to other any live attack detection scene, which will not be described one by one here. For the convenience of description, the live attack detection method will be described by taking the example of applying the live attack detection method to the human face in the human face recognition scene.

[0042] Those skilled in the art should understand that the live attack detection method and system described in the present specification also apply to other use scenarios within the protection scope of the present specification.

[0043] Figure 1A An application scenario schematic diagram of a live attack detection system 100 according to some embodiments of the present specification is shown. The live attack detection system 100 (hereinafter referred to as system 100) can be applied to live / live attack detection in any scene, such as live detection / live attack detection in a human face payment scene, live detection / live attack detection in an identity verification scene, live detection / live attack detection in other human face recognition scenes, etc. The system 100 can include a server 110, a client 120, a network 130, a database 140, and a target object 150.

[0044] The network 130 facilitates the exchange of information and / or data. The network 130 can be any type of wired or wireless network, or a combination thereof. For example, the network 130 may include a cable network, wired network, fiber optic network, telecommunications network, intranet, internet, local area network (LAN), wide area network (WAN), wireless local area network (WLAN), metropolitan area network (MAN), public switched telephone network (PSTN), Bluetooth network, ZigBee network, near field communication (NFC) network, or similar networks. The network 130 may include one or more network access points. For example, the network 130 may include wired or wireless network access points, such as base stations and / or internet switching points 130-1, 130-2, ... . Through this access point, one or more components of the client 120, the server 110, and the database 140 can connect to the network 130 to exchange data and / or information, wherein the target object 150 can connect to the network 130 through the client 120. Figure 1A As shown, the server 110, the client 120, and the database 140 can be connected to the network 130 and transmit information and / or data to each other through the network 130. For example, the client 120 can obtain services from the server 110 through the network 130.

[0045] The server 110 may be a server providing various preset services corresponding to the client 120. For example, it may be a backend server supporting liveness detection of the target image of the target object 150 acquired on the client 120. The liveness / liveness attack detection method described in this application can be executed on the server 110. In this case, the server 110 may store data or instructions for executing the liveness detection / liveness attack detection method described in this specification, and may execute or be used to execute the data or instructions. The server 110 may include hardware devices with data information processing functions and the necessary programs required to drive the hardware devices to work.

[0046] The database 140 can store data and / or instructions. The database 140 can store data and / or instructions for performing the live detection methods described in this specification. The server 110 and the client 120 can have access to the database 140. The server 110 and the client 120 can access data or instructions stored in the database 140 through the network 130. The database 140 can be directly connected to the server 110 and the client 120. Among them, the database 140 can be part of the server 200. The database 140 can include mass storage, removable storage, volatile read-write memory, read-only memory (ROM), or the like, or any combination thereof. An example of mass storage can include a non-transitory storage medium such as a disk, an optical disk, a solid state drive, etc. An example of removable storage can include a flash drive, a floppy disk, an optical disk, a memory card, a zip disk, a magnetic tape, etc. A typical volatile read-write memory can include a random access memory (RAM). An example of RAM can include dynamic RAM (DRAM), double date rate synchronous dynamic RAM (DDR SDRAM), static RAM (SRAM), thyristor RAM (T-RAM), and zero-capacitor RAM (Z-RAM), etc. An example of ROM can include mask ROM (MROM), programmable ROM (PROM), virtually programmable ROM (PEROM), electronically programmable ROM (EEPROM), compact disc (CD-ROM), and digital versatile disk ROM, etc. The database 140 can be implemented on a cloud platform. Just for example, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an inter-cloud, or the like, or any combination thereof. The storage medium can include a remote independent storage medium, such as the above-mentioned disk, hard disk, and solid state disk, etc. The server 110 / client 120 can access or update the data stored in the remote independent storage medium through a mobile device with access rights. The storage medium can include a local storage medium. For example, the data storage hard disk of the local server 140 or the IOT device itself, etc.

[0047] Figure 1B A schematic diagram of a scenario of live attack detection on a target object according to some embodiments of the present specification is shown. The target object 150 can be an object that needs to perform biometric recognition, or an object that is performing biometric recognition. The target object 150 can be an object detected by the system 100. The target object 150 can initiate a biometric recognition operation, thereby triggering the system 100 to perform live attack detection on the target object 150.

[0048] The client 120 can be any type of device capable of responding to and performing liveness detection / liveness attack detection operations on the target object 150. For example, the client 120 can be an automated payment vending machine in the unmanned vending field, such as... Figure 1B The IoT device shown is an example. For instance, client 120 may include mobile devices, tablets, laptops, built-in devices in motor vehicles, or similar content, or any combination thereof. In some embodiments, the mobile device may include smart home devices, smart mobile devices, virtual reality devices, augmented reality devices, or similar devices, or any combination thereof. In some embodiments, the smart home device may include smart TVs, desktop computers, etc., or any combination thereof. In some embodiments, the smart mobile device may include smartphones, personal digital assistants, gaming devices, navigation devices, etc., or any combination thereof. Client 120 may include an image acquisition unit. The image acquisition unit can acquire target images of the target object 150. The image acquisition unit may be a two-dimensional image acquisition device (e.g., an RGB camera) or a depth image acquisition device (e.g., a 3D structured light camera, a laser detector, etc.). Client 120 may also include a positioning system, which can be used to locate the position of client 120 after obtaining authorization from the target object 150. For example, when client 120 includes a smartphone, the target object 150 can initiate or respond to a liveness detection request from any location. After the target object 150 authorizes the client 120 to perform location positioning, the client 150 can obtain the location of the target object 150 during liveness detection and upload the location to the database 140. When the client 120 includes a fixed smart device, such as an IoT device, an automated payment vending machine in the unmanned vending field, etc., the location of the client 120 can be pre-stored in the database 140 when the staff sets up such a device. Therefore, the location information of the client 120 can be uploaded to the database 140 without authorization from the target object 150.

[0049] The client 120 can be installed with a target application (APP). The target APP can provide the target object 150 with the ability to interact with the outside world through the network 130 and an interface. The target APP includes but is not limited to: a web browser type APP program, a search type APP program, a chat type APP program, a shopping type APP program, a video type APP program, a financial type APP program, an instant messaging tool, a mailbox terminal device, a social platform software, and the like. The target APP can instruct the client 120 to use the image acquisition unit to acquire a target image having a biological feature of the target object 150, such as a face image. The target image can be used for live body detection / live body attack detection. The target object 150 can also trigger a live body detection / live body attack detection request through the target APP. The target APP can respond to the live body detection request and perform the live body detection / live body attack detection method described in the specification. The live body detection / live body attack detection method will be described in detail in the following.

[0050] It should be understood that Figure 1A The number of the client 120 and the server 110 in the above description is only illustrative. According to the needs of implementation, there can be any number of the client 120 and the server 110.

[0051] It should be noted that the live body detection / live body attack detection method can be completely executed on the client 120, completely executed on the server 110, partially executed on the client 120, and partially executed on the server 110.

[0052] For the convenience of description, the specification will focus on the description of the technical solutions by taking face recognition / detection as an example in the following introduction. Of course, those skilled in the art can fully understand that the methods and systems introduced in the specification can also be applied to the recognition of other biological features. For example, live body detection / live body attack detection by palm print, live body detection / live body attack detection by iris, live body detection / live body attack detection by fingerprint, and the like.

[0053] Figure 2 A structural schematic diagram of a device 200 according to an embodiment of the specification is shown. The device 200 can perform the live body attack detection method described in the specification. The device 200 can be the server 110 or the client 120.

[0054] As Figure 2As shown, the device 200 includes at least one storage medium 230 and at least one processor 220. In some embodiments, the device 200 can also include a communication port 250 and an internal communication bus 210. Meanwhile, the device 200 can also include an I / O component 260.

[0055] The internal communication bus 210 can connect different system components, including the storage medium 230 and the processor 220. The I / O component 260 supports input / output between the device 200 and other components.

[0056] The storage medium 230 can include a data storage device. The data storage device can be a non-transitory storage medium or a transitory storage medium. For example, the data storage device can include one or more of a disk 232, a read-only memory (ROM) 234, or a random access memory (RAM) 236. The storage medium 230 also includes at least one instruction set stored in the data storage device. The instructions are computer program codes, which can include programs, routines, objects, components, data structures, processes, modules, etc. that perform the method of living body detection of the target object 150 provided in the present specification.

[0057] The communication port 250 is used for data communication between the device 200 and the outside world. For example, the device 200 can connect to the network 130 through the communication port 250. The at least one processor 220 and the at least one storage medium 230 are communicatively connected through the internal communication bus 210. The at least one processor 220 is used to execute the at least one instruction set described above. When the system 100 is running, the at least one processor 220 reads the at least one instruction set and performs the living body detection method 300 provided in the present specification according to the instructions of the at least one instruction set.

[0058] The processor 220 can perform all steps of the method 300 of performing liveness detection on the target object 150. The processor 220 can be in the form of one or more processors. The processor 220 can issue execution instructions. For example, the processor 220 can be in communication connection with the image acquisition unit, and issue image acquisition instructions to the image acquisition unit to instruct the image acquisition unit to perform image acquisition tasks according to the instructions, and acquire target images of the target object 150. The processor 220 can also be in communication connection with the positioning system, and receive position information from the positioning system. The position information can be the position information of the image acquisition unit. For example, the mobile device is provided with an image acquisition unit and a positioning system. The processor 220 can receive the position information of the image acquisition unit of the processor 220 from the positioning system. The processor 220 can include one or more hardware processors, such as a microcontroller, a microprocessor, a reduced instruction set computer (RISC), an application-specific integrated circuit (ASIC), an application-specific instruction set processor (ASIP), a central processing unit (CPU), a graphics processing unit (GPU), a physics processing unit (PPU), a microcontroller unit, a digital signal processor (DSP), a field programmable gate array (FPGA), an advanced RISC machine (ARM), a programmable logic device (PLD), any circuit or processor capable of performing one or more functions, or the like, or any combination thereof.

[0059] For the sake of illustration only, only one processor 220 is described in the device 200 in the present specification. However, it should be noted that the device 200 in the present specification can also include multiple processors, and therefore, the operations and / or method steps disclosed in the present specification can be performed by one processor as described in the present specification, or jointly performed by multiple processors. For example, if the processor 220 of the device 200 in the present specification performs step A and step B, it should be understood that step A and step B can also be performed jointly or separately by two different processors 220 (for example, a first processor performs step A, and a second processor performs step B, or the first and second processors jointly perform steps A and B).

[0060] Figure 3 A flowchart of a liveness attack detection method 300 according to an embodiment of the present specification is shown. The liveness attack detection method 300 (hereinafter referred to as method 300) can be executed locally on the client 120, or can be executed by the server 110 alone, or can be executed by the server 110 and the client 120 in cooperation. Among them, the server 110 / client 120 can have the functions as described above. Figure 2The structure of device 200 is shown. Specifically, the storage medium in server 110 / client 120 may store at least one set of instructions for liveness detection. Processor 220 may be communicatively connected to storage medium 230. Processor 220 may read the instruction set stored in its local storage medium and / or database 140, and then execute method 300 according to the instructions.

[0061] For ease of description only, this specification describes method 300 using the example of server 110 and client 120 working together. That is, client 120 receives instructions from server 110 to execute some steps of method 300, and then server 110 remotely completes the other steps of method 300. Of course, those skilled in the art will fully understand that the steps in method 300 can be completed independently by client 120.

[0062] Specifically, the method 300 may include execution via processor 220:

[0063] S310: Obtain the target image.

[0064] The target image may include a target image of the target object 150 and a target background image excluding the target image. The target object 150 may be an object or user to be subjected to liveness detection or currently undergoing liveness detection. The target image may be acquired by the client device 120. The client device 120 may be the aforementioned terminal device, such as a smartphone, IoT device, etc. The target object 150 may complete login or further operations on the client device 120 by performing liveness verification or facial verification on the client device 120.

[0065] The client 120 may include an image acquisition unit. When the client 120 receives a face verification request, it activates the image acquisition unit to acquire the target image. The verification request may be a normal request from a real user, or it may be a liveness attack launched by criminals using fake facial information. For example, when the target 150 pays for goods, they choose face recognition payment. In this case, the IoT device can activate the image acquisition unit to acquire an image of the target 150 to detect liveness attacks and complete the payment request based on the detection result. As another example, when the target user 150 withdraws money from a bank ATM, identity verification is required. After the target 150 authorizes identity verification, the camera (image acquisition unit) on the ATM acquires an image of the target 150 to detect liveness attacks and completes the withdrawal request based on the detection result.

[0066] In a real user scenario, the target object 150 triggers a live body verification or face verification request. The request can be uploaded to the server 110. The processor 220 in the server 110 can receive and respond to the request, that is, send an instruction to the image acquisition unit to perform image acquisition. After the client 120 receives the instruction, it can control the image acquisition unit to perform image acquisition on the target object 150. The target image obtained by the client 120 contains not only the real target image of the target object 150 (such as the image of the face), but also the target background image of the environment in which the target image is located. The target image can be the part of the target object 150 itself mapped on the target image by the image acquisition unit. The target background image can include the part of the target image remaining after the target image is removed. The target background image can reflect the environmental information or geographical location information of the target object 150 when performing live body detection. For example, a consumer performs face payment in front of an IOT device in a convenience store, and the IOT device becomes the client 120, and the consumer becomes the target object 150. The target image obtained by the IOT device includes the target image composed of the face area and the target background image composed of the interior space and the display area of the convenience store.

[0067] In a live body attack scenario using false face information, the live body verification or face verification request is not issued by the target object 150. Criminals try to show the face image of the target object 150 in front of the image acquisition unit by false means (such as using a mobile phone screen or a photo, etc.). At this time, the request is uploaded to the server 110. The processor 220 in the server 110 can receive and respond to the request, and send an instruction to the image acquisition unit to perform image acquisition. After the client 120 receives the instruction, it can control the image acquisition unit to perform image acquisition on the false means. The target image obtained by the client 120 contains the false image of the target object 150 and the false background image. The false background image cannot reflect the real environmental information or geographical location information of the client 120.

[0068] S330: Determine a live body attack detection result of whether the target image is a live body attack image based on at least the biological feature of the target object 150 and the background correlation degree.

[0069] The client 150 / the server 110 can detect a live body attack based on the biological feature of the target object 150, and determine whether the target image is a live body attack image based on the background correlation degree.

[0070] Specifically, S330 can include the following steps:

[0071] S331: decoupling the target image to obtain the target image and the target background image.

[0072] After the server 110 / the client 120 obtains the target image, the server 110 / the client 120 can send the target image to a preset neural network to extract the target image and the target background image from the target image. The target image can contain the biological features of the target object 150, such as facial features. For example, the target image is an image that only includes the face of the target object. The target background image does not include the target image, for example, the target background image is the part of the target image excluding the target image. The neural network used to extract the target image and the target background image can be a semantic segmentation neural network, for example, U-Net16, U-Net128, etc.

[0073] After decoupling the target image, the server 110 / the client 120 will use the target image for the above-mentioned biological feature-based live body detection; and use the target background image for the background correlation-based live body detection.

[0074] S333: determining that the live body attack probability of the target image is greater than a preset value.

[0075] The client 120 / the server 110 can extract the biological features of the target object 150 from the target image. Biological features are important information for live body detection. Taking face recognition as an example, face recognition is a biometric technology that identifies the identity based on the facial feature information of a person. Live body detection is mainly performed by recognizing the physiological feature information on the live body, and the physiological feature information is used as a biological feature to distinguish the biological features that are forged by non-biological substances such as photos, silica gel, and plastic. When the biological feature information is obtained from a legal user, it can be identified that the biological feature is obtained from a legal user with a biological live body. The biological features can include, but are not limited to, one or more of facial features, fingerprint features, iris features, palm print features, dental features, and skeletal features. After the server 110 / the client 120 obtains the target image, the server 110 / the client 120 can send the target image to a preset neural network to extract the biological features of the target object 150 from the target image. The neural network used to extract the biological features can be a residual neural network or any other neural network capable of achieving the purpose of the present application.

[0076] After the live detection is performed on the biological features contained in the target image, a probability of live attack of the target image can be obtained. Since the biological features have stability and singularity, it also means that the biological features are important basis for distinguishing whether the target image is a live body or a live attack. Therefore, when the probability of live attack of the target image is less than a preset value, it means that the target image passes the verification with a high degree of association with the identity information in the live detection process, and it can be considered that the target image is not a live attack image, that is, the target object 150 providing the target image is a real target object and is not a live attack. If the probability of live attack of the target image is greater than the preset value, it means that the target image may be a live attack image. However, due to the environment of the live detection, which may have a negative impact on the result of the live detection, and other factors, it is necessary to further detect the target image that may be a live attack.

[0077] S335: analyzing the target background image, and determining the live attack detection result based on the background association degree.

[0078] In step S333, the client 120 / the server 110 judges whether the target image is a live attack based on the probability of live attack. If the probability of live attack is greater than a preset value, it is judged that the background image information needs to be further analyzed. In step S335, the client 120 / the server 110 uses the target background image to perform live detection based on the background association degree.

[0079] The background association degree can include an association degree of the target background image with a reference background image set. The reference background image set is from an associated object of the target object 150. The associated object can be obtained by the server 110 / the client 120 analyzing a historical trajectory of the target object 150.

[0080] The historical trajectory of the target object 150 is a moving trajectory of the target object collected by the system 100. For example, under the premise of obtaining authorization of the target object, when the target object takes money at an ATM, the ATM can upload the geographical position of the ATM to the system 100, so that the system 100 can obtain the position of the target object at the time of taking money and record it; for another example, when the target object scans a code to pay at a POS machine through a mobile phone, under the premise of obtaining authorization of the target object, the mobile phone or the POS machine can upload the geographical position of the payment to the system 100, so that the system 100 can obtain the position of the target object at the time of taking money, record it, and store it in the database 140. The above positions are accumulated to become the historical trajectory of the target object 150. Figure 4 A historical trajectory diagram of a target object and an associated object is shown according to some embodiments of the present specification.Figure 4 The solid line in FIG. 13 shows the historical trajectory of the target object. Similarly, the system 100 can obtain the historical trajectories of other users and store them in the database 140 under authorization.

[0081] As mentioned above, the client contacted by each user of the system 100 (including the target object) can obtain the location information of the user through the positioning system and upload it to the database 140. Therefore, the database 140 can include the historical trajectories of the target object 150 and all other users of the system 100. The server 110 / the client 120 can select a plurality of candidate objects from the entire set of users according to certain criteria to form a candidate object set. For example, the candidate object set can be the entire set of users of the system 100.

[0082] After determining the candidate object set, the server 110 / the client 120 can filter the associated objects of the target object 150 from the candidate object set based on the historical trajectories of each person, thereby obtaining the reference background image set.

[0083] The associated objects can be users of the system 100 who have visited the same location or appeared near the location multiple times within a predetermined time period. Figure 5 The dashed line in FIG. 13 shows the historical trajectory of an associated object. Through statistics and research, it is found that for a user who has visited a location, the probability of subsequently continuing to visit the same location is as high as 15%. That is, for the target object 150, there is a high probability that the location where the live detection is currently performed is a location that has been visited by one of its associated objects before. The more times the associated object has visited the location, the stronger the association between the associated object and the target object. Therefore, as long as the number of associated user groups mastered by the system 100 is sufficient, there are some people in the group who have a high probability of appearing at the same location or nearby locations recently.

[0084] Figure 5 FIG. 13 shows a method flowchart P400 for determining the associated objects of the target object 150 according to some embodiments of the present specification. P400 can be performed by the server 110 or the client 120. For ease of description, the steps in P400 are performed by the server 110. Further, the server 110 performing P400 can have the same configuration as the server 110 described above. Figure 2The structure of the device 200 is shown. Specifically, the storage medium in the server 110 can store at least one set of instructions for determining the associated object of the target object 150. The processor 220 can be communicatively connected with the storage medium 230. The processor 220 can read the instruction set stored in its local storage medium and / or the database 140, and then execute the P400 provided in the present specification according to the provisions of the instruction set. The P400 can include the following steps executed by at least one processor 220:

[0085] P410: Obtain K candidate objects.

[0086] The K candidate objects can constitute a candidate object set. Wherein the historical trajectory of each object in the candidate object set can be obtained by the server 110 from the database 140. Wherein the K candidate objects can be all users in the database 140 who have historical trajectories. Screening the associated users from the candidate object set composed of all users can ensure that there is no omission when screening the associated users, so that the reference background image set data obtained is more accurate. The K is an integer greater than or equal to 1. For example, K can be 15, 20, 25, 30, etc.

[0087] P430: Determine N associated objects from the candidate object set.

[0088] The server 110 can select the candidate object with a historical association degree greater than a threshold value with the target object 150 from the candidate object set as an associated user. The N is an integer greater than or equal to 1. Because the associated objects are determined from the candidate object set, the number of associated objects is less than or equal to the number of candidate objects. For example, when K is 15, N can be 15, 14, 13, 10, etc.

[0089] Specifically, P430 can include:

[0090] P431: Obtain M historical positions of the target object 150 within a predetermined historical period and a historical time ti corresponding to any one of the M historical positions Pi.

[0091] The predetermined historical period can be the time period prior to the current moment when the target object 150 is being detected for liveness. The period can be any time interval, such as 1 day, 15 days, 30 days, etc. The M historical locations can be the locations where the target object 150 has authorized location access or the locations that the server 110 can obtain within the predetermined historical period. Here, M is an integer greater than or equal to 1, and i is any integer between [1, M]. For example, M can be 3, 4, 5, 6, etc. When M is 4, i can be 1, 2, 3, 4. When M is 6, i can be 1, 2, 3, 4, 5, 6.

[0092] exist Figure 4 In the historical trajectory diagram, the candidate object set includes candidate object A. For example, taking day W as the predetermined historical period. During day W, the target object 150 sequentially stopped at 9 historical positions along the arrow direction. Therefore, M is 9, and along the arrow direction, the 9 historical positions Pi are P1 to P9, and the 9 historical moments are t1 to t9. Table 1 lists the historical moments corresponding to each of the 9 historical positions of the target object 150 during day W.

[0093] Table 1

[0094]

[0095] P433: Obtain M target ranges, wherein the M target ranges correspond to the preset location neighborhoods of the M historical locations.

[0096] The preset location area can be a range of locations centered at each of the M historical locations, with a preset distance from the center of each circle. For example, the preset location can be obtained by drawing a circle with each location as the center and a preset distance as the radius. Figure 4 As shown, each circular area represents the preset location area. The server 11 can obtain... Figure 4 Nine preset location areas are defined for nine historical locations. Each preset location area can serve as a target range. For example, a circular area with a radius of 1 km centered at location P1 is the target range for P1.

[0097] P435: Obtain M target time periods, wherein the M target time periods correspond to the preset time neighborhood of the M historical moments.

[0098] The preset time field can be a preset time period including each of the M historical time points. For example, the target object is located at position P1 at 8:30 on day W, and the preset time period is 15 minutes, then the target time period corresponding to P1 is 8:15-8:45; the target object is located at position P2 at 9:30 on day W, then the target time period corresponding to P2 is 9:15-9:45; the target object is located at position P3 at 11:30 on day W, then the target time period corresponding to P3 is 11:15-11:45. The remaining target time periods can be obtained by the server 110 in this way.

[0099] P437: determining N candidate objects in the plurality of candidate objects that appear in the M target ranges more than a threshold number of times within the M target time periods as the N associated objects.

[0100] The server 110 can obtain a historical trajectory of the candidate object. The historical trajectory can include position information and time information of arriving at the corresponding position of the candidate object on day W (i.e., the preset historical period). As shown in Table 2, the candidate object A has 8 historical positions on day W in the arrow direction. Figure 4

[0101] Table 2

[0102]

[0103] In Figure 4 ​The target object 150 is marked with 9 target ranges. When the threshold is 4 times, the server 110 can calculate the historical correlation degree between the candidate object A and the target object 150 according to the above data. The server 110 can first select the historical positions of the candidate object A in the 9 target ranges from the 8 historical positions of the candidate object A, including PA3, PA4, PA5, PA6, PA7, and PA8. Among them, PA3 is in the target range of P2, PA4 is in the target range of P5, PA5 is in the target range of P6, PA6 is in the target range of P3, PA7 is in the target range of P8, and PA8 is in the target range of P9. When the candidate object A meets the requirement of being located in the target range, it also meets the requirement of being located in the corresponding target time period. For example, the target time period corresponding to P2 is 9:15-9:45, and the time tA5 when the candidate object A is located at the PA3 position is 9:42. Therefore, the historical correlation times between the candidate object A and the target object 150 can be recorded as 1 time. For another example, the target time period corresponding to P3 is 11:15-11:45, and the time tA6 when the candidate object A is located at the PA6 position is 17:00, which is not in the target time period. The candidate object A meets the distance requirement but not the time requirement, so the candidate object A located at PA6 does not have a historical correlation with the target object 150. According to this method, the historical correlation times between the candidate object A and the target object 150 can be obtained as 5 times, which exceeds the threshold times, so the candidate object A can be used as the correlation object of the target object 150. Similarly, the server 110 can also select the time that meets the target time period from the historical time of the candidate object, and then select the position that meets the target range from the position corresponding to the time, and increase the historical correlation degree by 1.

[0104] It is necessary to know that Figure 4 Only to show how to determine the correlation object from the candidate object, in fact, the number of candidate objects in the candidate object set should be as large as possible. The above operation is performed for each candidate object to obtain M correlation objects. According to the above method, the historical correlation degree between each correlation object in the M correlation objects and the target object 150 is greater than the threshold.

[0105] After determining the N correlation objects of the target object 150, the server 110 can determine the reference background image set corresponding to each of the N correlation objects.

[0106] The set of reference background images is a collection of background images of each recorded position in the historical trajectory of the associated object of the target object 150. As mentioned above, each associated object of the target object 150 has a corresponding historical trajectory stored in the database 140. Each node of the historical trajectory is a historical position recorded by the associated object at a historical time. The position can correspond to one or more pictures containing reference background images. For example, the position can be the position information left by the associated object when taking money at an ATM, and the corresponding picture contains the background image reflecting the position, which is referred to as a reference background image. In this way, each historical trajectory of each associated object of the target object 150 corresponds to one or more position information, and at least part of the position information corresponds to one or more reference background images. All the reference background images are collectively referred to as the set of reference background images.

[0107] In some embodiments described above, the set of reference background images includes background images of all positions of the N associated objects in the predetermined historical period. For example, the candidate object A is an associated user of the target object 150. The set of reference background images can include background images of each position of PA1-PA8 of the candidate object A on the Wth day. The server 110 can also separate the image of the candidate object from the background image based on the semantic segmentation neural network, and extract the background image from the original image of the candidate object.

[0108] As mentioned above, the target background image contains environmental information or geographical position information of the target object 150 when performing the live detection. Therefore, without obtaining the position information authorized by the target object 150, the possibility that the target object 150 is currently visiting the position corresponding to the target background image can still be determined by determining the correlation degree between the reference background image in the set of reference background images and the target background image.

[0109] After step P400 is performed, return to Figure 3 , the server 110 / the client 120 continues to perform Figure 3 the following steps:

[0110] S335-1: Determine the target feature of the target background image and the plurality of reference features corresponding to the plurality of reference background images in the set of reference background images.

[0111] The server 110 / the client 120 can extract a target feature from the target background of the target object 150. The target feature can represent the target background. The target feature can be a feature vector or a feature matrix. The server 110 can also extract a reference feature of a reference background image from the reference background image set. The reference feature of each reference background image can be extracted, and the reference features of all reference background images can form a reference feature set. The reference feature can be a feature vector or a feature matrix. The neural network used to extract the target feature and the reference feature can be a residual neural network, such as Res-Net18 or Res-Net34, etc.

[0112] S335-3: Determine the background correlation degree of the target feature and the plurality of reference features.

[0113] The reference background image set can include a plurality of reference background images. The server 110 can determine a plurality of similarities between the target feature and the plurality of reference features. The plurality of similarities between the target feature and the reference features can be that a similarity between the target feature and each reference feature in the reference feature set is determined. The number of the plurality of similarities is the same as the number of features in the reference feature set. The more similar the target feature and the reference features are, the more likely it is that the location where the target object 150 is performing the liveness detection is a location that the associated object has visited. Based on the above statistical rule, the probability that the target image of the target object 150 is a liveness attack image is smaller.

[0114] Therefore, the similarity between the target feature and the reference feature can be used as the background correlation degree. When the target feature and the plurality of reference features are feature vectors, the server 110 / the client 120 can use a cosine similarity algorithm to determine the similarity between the target feature vector and each reference feature vector in the plurality of reference feature vectors. The server 110 / the client 120 can also use a Euclidean geometry algorithm to determine the similarity between the target feature vector and each reference feature vector in the plurality of reference feature vectors. The plurality of similarities can form a similarity set.

[0115] The server 110 / the client 120 can select the maximum similarity from the plurality of similarities in the similarity set as a first similarity, and can select the average of the plurality of similarities in the similarity set as a second similarity. The server 110 can also determine a first association threshold and a second association threshold, and use the two as criteria for dividing whether the target image is a live attack image. The first association threshold and the second association threshold can be derived from the experience of staff. For example, the target user 150 performs a live attack detection in front of a smart cabinet of unmanned retail. Since the location of the smart cabinet is fixed, the area photographed by the camera on the smart cabinet is also fixed, so the background photographed when the target object 150 performs live detection by the smart cabinet is relatively determined, that is, the background image is basically consistent with the reference background image photographed by the smart cabinet on other associated objects. If the associated user of the target user 150 has visited this smart cabinet for face detection, there will be a reference background in the reference background set that has a high similarity with the background image of the target object 150, and the first association threshold and the second association threshold can be relatively high. If the associated object of the target object 150 has not visited this smart cabinet, but has performed face detection through a mobile device near the smart cabinet, the reference background image obtained through the mobile device is the closest reference background to the smart cabinet. Since the shooting angle of the mobile device on the associated object can be arbitrary, the reference background image obtained may not be the same as the target background image obtained through the smart cabinet. However, it can still be used as the background of the target object 150 visiting the smart cabinet, and the first association threshold and the second association threshold can be set relatively low at this time.

[0116] S335-5: Determine the live attack detection result based on the background association degree.

[0117] The server 110 can determine whether the target object 150 is likely to visit the location for live detection based on the background association degree, thereby determining the live attack detection result. For example:

[0118] When the first similarity is less than the first association threshold and the second similarity is less than the second association threshold, it means that the similarity of the background in the reference background image set with the background of the target object 150 performing live detection is not high, which means that the possibility of the target object 150 visiting the location where the target image is displayed is very small, and therefore the target object 150 can be determined to be a live attack.

[0119] When the first similarity is greater than the first correlation threshold or the second similarity is greater than the second correlation threshold, it indicates that the background in the reference background image set and the target object 150 have a high background similarity for live body detection, meaning that the target object 150 has a relatively high possibility of visiting the location where the target image is displayed, and thus it can be determined that the target object 150 is a live body.

[0120] S350: Output the live body attack detection result.

[0121] In the present specification, after the server 110 / client 120 determines the live body detection result of the target object 150, the result can be output to the client 120. For example, in a face recognition payment scenario, the target object 150 selects face recognition for payment, and uses an IOT device to perform live body attack detection. The IOT device obtains the detection result of the target object 150 based on the above execution steps. When the detection result is that the target image is a non-live body attack image, the IOT device completes the payment request of the target object 150, and displays the word "payment success" on the IOT device interaction interface; when the detection result is that the target image is a live body attack image, the IOT device refuses to complete the payment request of the target object 150, and displays the word "payment failure" on the IOT device interaction interface. The completion of the payment request can also be sent to the merchant to supervise the user to make payment. The target object 150 can know whether the current live body detection is successful through the client 120. If the live body detection result is safe, it means that the target object 150 passes the current live body detection, otherwise, it means that the target object 150 fails the current live body detection. If the target object 150 fails the current live body detection, the target object 150 can perform live body verification again through the client 120, and the server 110 can repeat the previous live body attack detection method to perform live body attack detection again.

[0122] In summary, the live body attack detection method 300 and the live body attack detection system 100 provided in the present specification use biological features to perform live body attack detection, and use background correlation to perform live body detection. In the process of using background correlation to perform live body detection, the server 110 / client 120 obtains the associated object of the target object 150 and the reference background image, and determines the live body attack detection result based on the correlation between the target background image of the target object 150 and the images in the reference background image set. If the correlation between the target background and the images in the reference background image set exceeds a threshold, it means that the target object 150 has a relatively high possibility of visiting the location where live body detection is performed, and it can be determined whether the target image of the target object 150 is a live body image attack, thereby improving the accuracy of live body attack detection.

[0123] Upon reading this detailed disclosure, those skilled in the art can appreciate that the foregoing detailed disclosure can be presented in a manner that is merely exemplary and that can not be limiting. Although the description herein can not specifically enumerate the description of certain features, those skilled in the art can understand that the description herein is intended to encompass various reasonable alterations, improvements and modifications of the embodiments. These alterations, improvements and modifications are intended to be presented by the description herein and are within the spirit and scope of the exemplary embodiments of the description.

[0124] The foregoing detailed description of certain embodiments of the description has been presented. Other embodiments are within the scope of the following claims. In some instances, the acts or steps recited in the claims can be performed in a different order and still accomplish the desired results. Also, the ordering of processes depicted in the accompanying figures can not be required to accomplish the desired results. In certain implementations, multitasking and parallel processing can be advantageous or possible.

[0125] In addition, certain terminology has been used for the purpose of reference only. For example, "one embodiment", "an embodiment" and / or "some embodiments" means that a particular feature, structure or characteristic described in connection with the embodiment is included in at least one embodiment of the description. Thus, it is appreciated that two or more references to "embodiments" or "one embodiment" or "an alternative embodiment" in various portions of the description do not necessarily all refer to the same embodiment. Additionally, particular features, structures or characteristics can be combined in any suitable manner in one or more embodiments.

[0126] It is to be understood that in the foregoing detailed description of the embodiments of the description, various features are sometimes grouped together in a single embodiment, figure, or description of a figure for the purpose of streamlining the description. Other features can be spread across multiple embodiments, figures or descriptions of a figure. However, this is not intended to be limiting. Rather, the present description encompasses multiple alternate combinations for implementing the embodiments. That is, the present description is to be understood to encompass all possible combinations that can be derived from the multiple features described throughout the specification.

Claims

1. A method for detecting liveness attacks, comprising: Obtain a target image, wherein the target image includes a target image of the target object and a target background image excluding the target image; Based on the historical behavior records of the target object and candidate objects, the associated objects of the target object and the reference background image set corresponding to the associated objects are determined from the candidate objects. The candidate objects include users with historical behavior records in the database. Based on the feature similarity between the reference background image set and the target background image, a background correlation degree is determined, which is used to characterize the behavioral similarity between the target object and the associated object; At least based on the biometrics of the target object and the background correlation, a liveness detection result is obtained to determine whether the target image is a liveness attack image; and Output the results of the liveness attack detection.

2. The method of claim 1, wherein determining the associated object of the target object and the reference background image set corresponding to the associated object includes: Obtain K candidate objects, where K is an integer greater than or equal to 1; N associated objects are determined from the K candidate objects, and the historical association degree between each associated object and the target object is greater than a threshold, where N is an integer greater than or equal to 1; as well as Determine the reference background image set corresponding to each of the N associated objects.

3. The method as described in claim 2, wherein, The reference background image set includes background images of all locations of the N associated objects within a predetermined historical period.

4. The method of claim 2, wherein, Determining the N associated objects includes, for each candidate object: Obtain M historical positions of the target object within a predetermined historical period and the historical time ti corresponding to any one of the M historical positions Pi, where M is an integer greater than or equal to 1, and i is any integer variable between [1, M] used to distinguish different historical positions; M target ranges are obtained, and the M target ranges correspond to the preset location neighborhoods of the M historical locations; M target time periods are obtained, and the M target time periods correspond to a preset time neighborhood of the M historical moments; as well as The N candidate objects that appear more than a threshold number of times within the M target time periods are identified as the N associated objects.

5. The method of claim 1, wherein, The live attack detection result, which determines whether the target image is a live attack image based at least on the biometrics of the target object and the background correlation, includes: The probability of a liveness attack on the target image is determined to be greater than a preset value; and The liveness attack detection result is determined based on the background correlation.

6. The method of claim 5, wherein, Determining the liveness attack detection result based on the background correlation includes: The target image is decoupled to obtain the target image and the target background image excluding the target image; Determine the target features of the target background image and the multiple reference features corresponding to multiple reference background images in the reference background image set; Determine the background correlation degree between the target feature and the plurality of reference features; and Based on the background correlation, the liveness attack detection result is determined.

7. The method of claim 6, wherein, The target image and the target background image are obtained by decoupling the target image through a semantic segmentation network; and The target features and the plurality of reference features are determined by a residual neural network.

8. The method of claim 6, wherein, Determining the background relevance includes: Determine the background similarity set, including multiple similarities between the target feature and the multiple reference features; The largest similarity among the plurality of similarities is taken as the first similarity; The average of the multiple similarities is taken as the second similarity; and Determine the first association threshold and the second association threshold.

9. The method of claim 8, wherein, Determining the similarity set includes: Based on the cosine similarity algorithm, the similarity between the target feature vector and each of the multiple reference feature vectors is obtained.

10. The method of claim 8, wherein, The determination of the liveness attack detection result based on the background correlation includes: If the first similarity is less than the first association threshold and the second similarity is less than the second association threshold, the target object is determined to be a live attack.

11. The method of claim 8, wherein, The determination of the liveness attack detection result based on the background correlation includes: If the first similarity is greater than the first association threshold, the target object is determined to be a living being.

12. The method of claim 8, wherein, The determination of the liveness attack detection result based on the background correlation includes: If the second similarity is greater than the second association threshold, the target object is determined to be a living organism.

13. The method of claim 1, wherein, The biometric features include at least one of the following: facial features, fingerprint features, pupil features, palm print features, skeletal projection features, and dental print features.

14. A liveness attack detection system, comprising: At least one storage medium, including at least one instruction set, for implementation analysis of the liveness detection method; as well as At least one processor is communicatively connected to the at least one storage medium. When the system is running, the at least one processor reads the at least one instruction set and executes the method of any one of claims 1-13 according to the instructions of the at least one instruction set.

15. The system of claim 14, further comprising: An image acquisition unit is communicatively connected to the at least one processor and receives image acquisition instructions from the at least one processor to execute an image acquisition task according to the image acquisition instructions. The image acquisition task includes acquiring a target image of a target object.

16. The system of claim 15, further comprising: A positioning system, communicatively connected to the at least one processor, is configured to upload the location information of the image acquisition unit to the at least one processor after obtaining authorization from the target object.

17. The system of claim 16, wherein, The liveness detection system is connected to a database to obtain the historical location of the target object within a predetermined historical period from the database during the liveness detection process.

18. The system of claim 17, wherein, The historical location information of the target object includes the acquisition locations of the image acquisition units that have acquired images of the target object during the predetermined historical period.

Images