WO2022222957A1 - 一种目标识别的方法和系统 - Google Patents

一种目标识别的方法和系统 Download PDF

Info

Publication number
WO2022222957A1
WO2022222957A1 PCT/CN2022/087915 CN2022087915W WO2022222957A1 WO 2022222957 A1 WO2022222957 A1 WO 2022222957A1 CN 2022087915 W CN2022087915 W CN 2022087915W WO 2022222957 A1 WO2022222957 A1 WO 2022222957A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
shooting
authenticity
client
target
Prior art date
Application number
PCT/CN2022/087915
Other languages
English (en)
French (fr)
Inventor
程博
张天明
Original Assignee
北京嘀嘀无限科技发展有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京嘀嘀无限科技发展有限公司 filed Critical 北京嘀嘀无限科技发展有限公司
Publication of WO2022222957A1 publication Critical patent/WO2022222957A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation
    • G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements

Definitions

  • This specification relates to the field of image processing, in particular to a method and system for target recognition.
  • Target recognition is a recognition technology based on images acquired by image acquisition elements. In order to improve the accuracy of target recognition, it is necessary to determine the authenticity of the image.
  • the target identification method includes: acquiring at least one shooting parameter related to a shooting frame; sending a shooting instruction to a client, the shooting instruction instructing the client to display the shooting frame based on the at least one shooting parameter;
  • the client receives at least one image; and based on the at least one shooting parameter, determining the authenticity of the at least one image.
  • the method for target identification includes: receiving a shooting instruction from a server, where the shooting instruction includes at least one shooting parameter related to a shooting frame; displaying the shooting frame based on the at least one shooting parameter; acquiring at least one shooting frame based on an image acquisition element. sending the at least one captured image to the server to judge the authenticity of the at least one captured image.
  • the target recognition system includes: a parameter acquisition module for acquiring at least one shooting parameter related to a shooting frame; an instruction sending module for sending a shooting instruction to a client, where the shooting instruction is used to instruct the client Displaying the shooting frame based on the at least one shooting parameter; an image receiving module for receiving at least one image from the client; and a authenticity determining module for determining the at least one shooting parameter based on the at least one shooting parameter authenticity of an image.
  • the target recognition system includes: an instruction receiving module for receiving a shooting instruction from a server, the shooting instruction including at least one shooting parameter related to a shooting frame; a shooting frame display module for based on the at least one shooting parameter , displaying the shooting frame; an image acquisition module, used for acquiring at least one shot image based on the image acquisition element; an image sending module, used for sending the at least one shot image to the server for the at least one shot image The authenticity of the captured image is judged.
  • FIG. 1 is a schematic diagram of an application scenario of a target recognition system according to some embodiments of the present specification
  • FIG. 2 is an exemplary flowchart of a target identification method applied to a server according to some embodiments of the present specification
  • FIG. 3 is a flowchart of sending a shooting instruction to a client according to some embodiments of this specification
  • FIG. 4 is a schematic diagram of displaying a shooting frame according to some embodiments of the present specification.
  • FIG. 5 is an exemplary flowchart of determining the authenticity of an image according to some embodiments of the present specification
  • FIG. 6 is a schematic diagram of an image comparison model according to some embodiments of the present specification.
  • FIG. 7 is an exemplary flowchart of a target identification method applied to a client according to some embodiments of the present specification.
  • system means for distinguishing different components, elements, parts, parts or assemblies at different levels.
  • device means for converting components, elements, parts, parts or assemblies to different levels.
  • Target recognition is a technology based on the target acquired by the image acquisition element.
  • the target may be a human face, a fingerprint, a palm print, a pupil, a non-living body, or the like.
  • object recognition may be applied to authorization verification.
  • authorization verification For example, access control authority authentication and account payment authority authentication.
  • target recognition may also be used for authentication. For example, employee attendance certification and self-registration identity security certification.
  • the target recognition may be based on matching the image of the target collected in real time by the image acquisition element and the pre-acquired biometrics, so as to verify the identity of the target.
  • the image capture element can be hacked or hijacked, and attackers can upload fake images to pass authentication.
  • attacker A can directly upload the image of user B after attacking or hijacking the image capture element.
  • the target recognition system performs recognition based on the image of user B and the biometric features of user B obtained in advance, thereby passing user B's identity verification. Therefore, in order to ensure the security of target recognition, it is necessary to determine the authenticity of the image, that is, to determine that the image is collected in real time by the image acquisition element during the target recognition process.
  • FIG. 1 is a schematic diagram of an application scenario of a target recognition system according to some embodiments of the present specification.
  • the object recognition system 100 may include a server 110 , a network 120 , a client 130 and a storage device 140 .
  • Server 110 may be used to process data and/or information from at least one component of object recognition system 100 or an external data source (eg, a cloud data center). For example, the server 110 may acquire at least one shooting parameter related to the shooting frame, and determine the authenticity of the at least one image sent by the client 130 based on the at least one shooting parameter. For another example, the server 110 may perform preprocessing (eg, object detection, quality analysis, etc.) on at least one image acquired from the client 130 to obtain the preprocessed at least one image.
  • preprocessing eg, object detection, quality analysis, etc.
  • the server 110 may obtain data (eg, instructions) from the storage device 140 or save data (eg, at least one image) to the storage device 140 , or may read data from other sources such as the client 130 through the network 120 (eg, photographing environment information) or output data (eg, photographing instructions) to the client 130 .
  • data e.g, instructions
  • save data e.g, at least one image
  • other sources such as the client 130 through the network 120 (eg, photographing environment information) or output data (eg, photographing instructions) to the client 130 .
  • server 110 may be a single server or a group of servers.
  • the server group may be centralized or distributed (eg, server 110 may be a distributed system).
  • server 110 may be regional or remote.
  • server 110 may be implemented on a cloud platform, or provided in a virtual fashion.
  • cloud platforms may include private clouds, public clouds, hybrid clouds, community clouds, distributed clouds, internal clouds, multi-tier clouds, etc., or any combination thereof.
  • Network 120 may connect components of object recognition system 100 and/or connect object recognition system 100 with external components.
  • the network 120 enables communication between the various components of the object recognition system 100, and/or between the object recognition system 100 and external components, facilitating the exchange of data and/or information.
  • the network 120 may be any one or more of a wired network or a wireless network.
  • the network 120 may include a cable network, a fiber optic network, a telecommunications network, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a public switched telephone network (PSTN) , Bluetooth network, ZigBee network (ZigBee), near field communication (NFC), intra-device bus, intra-device line, cable connection, etc. or any combination thereof.
  • the network connection between the various parts of the system may be in one of the above-mentioned manners, or may be in multiple manners.
  • the network 120 may be of various topologies such as point-to-point, shared, centralized, or a combination of topologies.
  • network 120 may include one or more network access points.
  • network 120 may include wired or wireless network access points, such as base stations and/or network switching points 120-1, 120-2, . . . , through which one or more components of object recognition system 100 may Connect to network 120 to exchange data and/or information.
  • the client 130 may implement the interaction between the user and the object recognition system 100 .
  • client 130 may include an image capture element (eg, camera, camera) for capturing image data (images and/or video).
  • the client 130 eg, the screen of the client 130
  • the client 130 may display information instructing the user to photograph while the image capture element is photographing.
  • the client 130 may receive or determine one or more shooting parameters related to the shooting frame, and display the shooting frame on its screen based on the one or more shooting parameters, so as to guide the user to place the target in the shooting frame to shoot.
  • the client 130 may communicate with the server 110 through the network 120 and send the captured at least one image to the server 110 .
  • the client 130 may be a mobile device 130-1, a tablet computer 130-2, a laptop computer 130-3, other devices with input and/or output capabilities, the like, or any combination thereof.
  • the above examples are only used to illustrate the breadth of the scope of the client 130 device and not to limit its scope.
  • the storage device 140 may be used to store data (eg, a standard image of an object, at least one reference image of a qualified object, etc.) and/or instructions.
  • the storage device 140 may include one or more storage components, and each storage component may be an independent device or a part of other devices.
  • storage device 140 may include random access memory (RAM), read only memory (ROM), mass storage, removable memory, volatile read-write memory, the like, or any combination thereof.
  • mass storage may include magnetic disks, optical disks, solid state disks, and the like.
  • storage device 140 may be implemented on a cloud platform.
  • cloud platforms may include private clouds, public clouds, hybrid clouds, community clouds, distributed clouds, internal clouds, multi-tier clouds, etc., or any combination thereof.
  • storage device 140 may be integrated or included in one or more other components of object recognition system 100 (eg, server 110, client 130, or other possible components).
  • the server 110 may include a parameter acquisition module, an instruction transmission module, an image reception module, a authenticity determination module, and a model acquisition module.
  • the parameter acquisition module may be configured to acquire at least one shooting parameter related to the shooting frame. In some embodiments, the parameter acquisition module may randomly generate at least one shooting parameter. In some embodiments, the parameter acquisition module may determine a shooting difficulty coefficient based on the reference information; and determine the at least one shooting parameter based on the shooting difficulty coefficient.
  • the instruction sending module may be configured to send a shooting instruction to the client, where the shooting instruction is used to instruct the client to display a shooting frame based on at least one shooting parameter.
  • the instruction sending module may acquire a template image of the target; and based on the at least one shooting parameter, adjust the template image to generate a comparison template image, wherein the shooting instruction further instructs the client to The comparison template image is displayed in the shooting frame.
  • the image receiving module may be configured to receive at least one image from the client.
  • the authenticity determination module may be configured to determine the authenticity of the at least one image based on the at least one shooting parameter.
  • the authenticity determination module may preprocess the at least one image to generate the preprocessed at least one image; and based on the at least one shooting parameter and the preprocessed at least one image, The authenticity of the at least one image is determined.
  • the pre-processing of the at least one image by the authenticity determination module includes performing at least one of the following operations on each of the at least one image: subjecting the image to object detection, determining Whether the image contains a target object; perform quality analysis on the image to determine whether the image meets the quality requirements; or perform image segmentation on the image to generate a segmented image corresponding to the shooting frame.
  • the authenticity determination module may determine a first authenticity of the at least one image based on the at least one shooting parameter, the first authenticity reflecting whether the at least one image is of the customer The end is based on the image captured by the shooting instruction; the authenticity determination module may determine the second authenticity of the at least one image based on the at least one shooting parameter and at least one reference image of the at least one qualified target, and the second The authenticity reflects whether the at least one image is an image of one of the at least one eligible target.
  • the model acquisition module is used to acquire one or more machine learning models, such as an image comparison model, a difficulty coefficient determination model, and a shooting parameter determination model.
  • the model acquisition module may acquire one or more machine learning models from other sources in object recognition system 100 (eg, storage device 140 ) or external sources.
  • the client 130 may include a photographing instruction receiving module, a photographing frame displaying module, an image acquiring module, and an image sending module.
  • the instruction receiving module may be configured to receive a shooting instruction from the server, where the shooting instruction includes at least one shooting parameter related to the shooting frame.
  • the shooting frame display module may be configured to display the shooting frame based on at least one shooting parameter.
  • the shooting frame display module may display the contrast template image of the target in the shooting frame.
  • the image acquisition module may be configured to acquire at least one captured image based on the image acquisition element.
  • the image sending module may be configured to send the at least one captured image to the server to judge the authenticity of the at least one captured image.
  • the above description of the candidate item display and determination system and its modules is only for the convenience of description, and does not limit the description to the scope of the illustrated embodiments. It can be understood that for those skilled in the art, after understanding the principle of the system, various modules may be combined arbitrarily, or a subsystem may be formed to connect with other modules without departing from the principle.
  • the multiple modules disclosed in FIG. 1 may be different modules in a system, or one module may implement the functions of the above-mentioned two or more modules. For example, each module may share one storage module, and each module may also have its own storage module.
  • FIG. 2 is an exemplary flowchart of a method for object recognition according to some embodiments of the present specification. As shown in FIG. 2, the process 200 includes the following steps.
  • Step 210 Acquire at least one shooting parameter related to the shooting frame.
  • step 210 may be performed by a parameter acquisition module of a server (eg, server 110).
  • the shooting frame refers to a specific area displayed on the screen of the client (eg, the client 130 ), and the user of the client can be guided to place the target in the specific area when shooting.
  • the photographing frame may have any shape, for example, a rectangle, a circle, an ellipse, and the like.
  • the client in order to facilitate the user to identify the shooting frame, can mark the shooting frame on the screen. For example, the edges of the shot frame can be marked with a specific color. For another example, the shooting frame may be filled with a color different from the display area of the screen.
  • the user of the client refers to the user who uses the client for target recognition.
  • the target refers to the object that needs to be recognized.
  • the target can be the user's face, fingerprint, palm print, or pupil.
  • the target may be a non-living body (eg, a car).
  • the target refers to the face of a user who needs to be authenticated and/or authenticated.
  • the platform needs to verify whether the driver who takes the order is a registered driver user reviewed by the platform, and the target object is the driver's face.
  • the payment system needs to verify the payment authority of the payer, and the target object is the payer's face.
  • the shooting parameters may include any parameters related to the shape, size, position, display manner, etc. of the shooting frame.
  • Exemplary shooting parameters may include shooting angle, shooting distance, shooting center point, display parameters, and the like.
  • the shooting angle is the angle of the shooting frame relative to the reference direction (such as the length direction of the client screen). Changes in the shooting angle can lead to changes in the relative orientation of the shooting frame and the client screen. For example, suppose the shooting frame is a rectangle. When the shooting angle is 0°, the length direction of the shooting frame is parallel to the length direction of the screen; when the shooting angle is 30°, the included angle between the length direction of the shooting frame and the length direction of the screen is 30°.
  • the shooting distance refers to the estimated distance between the target and the image capture element of the client when the user places the target in the shooting frame for shooting. Changes in the shooting distance can cause the size ratio of the shooting frame to the screen to change. For example, when the shooting distance is 0.5m, the ratio of the shooting frame to the screen is 0.8:1; when the shooting distance is 1m, the ratio of the shooting frame to the screen is 0.6:1.
  • the shooting center point is the positioning point of the shooting frame.
  • the shooting center point may be a position point located at the center of the shooting frame, a position point located on the border of the shooting frame, or the like. Changing the position of the shooting center point on the screen can cause the position of the shooting frame to change on the screen.
  • the display parameter is a mode parameter related to how the frame is displayed.
  • the display parameters may include the shape of the shooting frame, the fill color, the border color, whether to flash the display, and the like.
  • the parameter acquisition module may randomly generate at least one shooting parameter. For example, for a certain shooting parameter, the parameter acquisition module may randomly determine the value of the shooting parameter within the value range of the shooting parameter preset by the target recognition system 100 .
  • the shooting parameters obtained in this embodiment are relatively random, which can improve the difficulty of cheating by the user, thereby improving the accuracy of target recognition.
  • the parameter acquisition module may determine the shooting parameters according to the default settings of the target recognition system 100 .
  • the parameter acquisition module may acquire pre-stored shooting parameters corresponding to the target from the storage device 140 according to the type of the target.
  • the parameter acquisition module may acquire the shooting parameters set by the user according to experience from the terminal device.
  • the parameter acquisition module may determine the shooting parameters through data analysis.
  • the parameter acquisition module may determine the shooting parameters according to the device information received from the client.
  • the parameter acquisition module may determine the shooting difficulty coefficient based on the reference information.
  • the parameter acquisition module may further determine shooting parameters based on the shooting difficulty coefficient.
  • the reference information may reflect the likelihood and/or difficulty of cheating in object recognition by the user of the client.
  • the reference information may include shooting environment information, historical behavior information of historical users corresponding to the client, personal information of historical users corresponding to the client, etc., or any combination thereof.
  • the photographing environment information is information related to the photographing environment of the image pickup element of the client.
  • the photographing environment information may include ambient light information, such as light intensity information, light type information, and the like.
  • the shooting environment information may include environment background information, for example, background static and dynamic information, background texture information, and the like.
  • the parameter acquisition module may receive shooting environment information from the client.
  • the client may determine photographing environment information based on image data photographed by the image capture element.
  • the client may include a sensor (eg, a photosensitive sensor) for detecting the shooting environment, for detecting shooting environment information. In general, the better the shooting environment (eg, the better the ambient lighting), the less difficult it is for the user to cheat.
  • the historical users corresponding to the client may include users who have a binding relationship with the client, historical users who have used the client, and the like.
  • the historical user corresponding to the client may be a driver who uses the client to register on the transportation service platform.
  • the historical user corresponding to the client may be the same as or different from the user currently using the client for target recognition.
  • the historical behavior information of the historical user may be related to the historical behavior of the historical user, such as the historical recognition behavior.
  • the historical behavior information of the historical user may include the number of historical recognition failures of the historical user, the reasons for the historical recognition failure, and the like.
  • the reasons for the failure of the history identification may include user cheating, user misoperation, and the like.
  • the parameter acquisition module may acquire the usage record of the client from the client or the storage device to determine the historical behavior information of the historical user.
  • the greater the number of historical recognition failures and/or the number of historical cheating of the historical user the higher the possibility that the user of the client side cheats in this target recognition.
  • Personal information of historical users is information related to historical users, such as historical user identification and historical user attributes.
  • the historical user ID is a symbol for distinguishing historical users.
  • the historical user attributes may include the historical user's age, education, gender, credit history, and the like. Exemplarily, the better the credit record of the historical user, the lower the possibility of the user of the client cheating in this target recognition.
  • the parameter acquisition module may acquire historical user personal information from a client, a storage device, or an external source.
  • the client can collect personal information during historical user registration, and store the personal information in a storage device.
  • the shooting difficulty coefficient represents the difficulty for the user of the client to place the target in the shooting frame for shooting. In some embodiments, the less difficult it is for the user to place the target in the shooting frame for shooting, the greater the shooting difficulty factor.
  • the parameter acquisition module may determine the shooting difficulty coefficient based on the reference information. Exemplarily, the greater the light intensity, the easier it is for the user to place the target in the shooting frame. At this time, the parameter acquisition module can determine a larger shooting difficulty coefficient to prevent the user from cheating. For another example, in the user's historical behavior, the more historical target recognition failures with the target recognition failure cause of "user fraud", the higher the probability of fraud by the user in this target recognition process. At this time, the parameter acquisition module can determine a larger shooting difficulty coefficient to prevent the user from cheating. For another example, the worse the user's credit record is, the higher the probability of cheating in the target identification process of the user is. At this time, the parameter acquisition module can determine a larger shooting difficulty coefficient to prevent the user from cheating.
  • the parameter acquisition module may determine the shooting difficulty coefficient according to the first rule.
  • the first rule is related to the relationship between one or more kinds of reference information and the shooting difficulty factor.
  • the first rule may include that when the light intensity is less than 30 lux, the shooting difficulty coefficient is 0.1; when the light intensity is greater than 30 lux and less than 100 lux, the shooting difficulty coefficient is 0.3; when the light intensity is greater than 100 lux, the shooting difficulty coefficient is 0.6.
  • the first rule may include that when the number of historical recognition failures is greater than 10, the shooting difficulty coefficient is 0.6; when the number of historical recognition failures is less than 10 and greater than 3, the shooting difficulty coefficient is 0.3; when the number of historical recognition failures is small, the shooting difficulty coefficient is 0.3. When it is less than 3 times, the difficulty factor of shooting is 0.1.
  • the parameter acquisition module may determine a shooting difficulty coefficient based on each of various kinds of reference information.
  • the parameter acquisition module may further determine the final shooting difficulty coefficient based on the multiple shooting difficulty coefficients.
  • the final shooting difficulty coefficient may be determined by summing, weighted summing, averaging, etc. of multiple shooting difficulty coefficients.
  • the parameter acquisition module may also determine the shooting difficulty coefficient through the difficulty coefficient determining model, specifically, the input of the difficulty coefficient determining model is reference information, and the output of the difficulty coefficient determining model is the shooting difficulty coefficient.
  • the difficulty factor determination model may include, but is not limited to, a deep neural network model, a recurrent neural network model, and the like.
  • the parameter acquisition module may determine at least one shooting parameter based on the shooting difficulty coefficient. For example, the greater the shooting difficulty factor, the greater the shooting angle, the farther the shooting distance, and the farther the shooting center is from the center of the client's screen. As mentioned above, when the reference information shows that the client user is more likely to cheat and/or the difficulty of cheating is low, the shooting difficulty factor will be higher. By setting the value of at least one shooting parameter, the difficulty for the user to place the target in the shooting frame can be improved, thereby achieving a higher shooting difficulty factor.
  • the parameter acquisition module may determine at least one shooting parameter based on the second rule.
  • the second rule is related to the relationship between the shooting difficulty factor and at least one shooting parameter.
  • the shooting parameters may include: a shooting angle of 0°, a shooting distance of 0.1 m, and the shooting center point coincides with the center point of the screen; when the shooting difficulty coefficient is 0.6, the shooting parameters may include: The shooting angle is 40°, the shooting distance is 0.3m, and the shooting center point is 0.05m above the center point of the screen.
  • the parameter acquisition module may further determine at least one shooting parameter according to the shooting parameter determination model.
  • the input of the shooting parameter determination model is the shooting difficulty coefficient
  • the output is at least one shooting parameter.
  • the shooting parameter determination model may include, but is not limited to, a deep neural network model, a recurrent neural network model, and the like.
  • the difficulty factor determination model and the shooting parameter determination model may be different layers of the same model.
  • the at least one shooting parameter includes multiple shooting parameters
  • the parameter obtaining module may obtain the multiple shooting parameters in the same or different manners.
  • the parameter acquisition module may generate each of the plurality of shooting parameters at any time.
  • part of the multiple parameters may be randomly generated, and another part may be determined based on reference information.
  • Step 220 sending a shooting instruction to the client.
  • step 220 may be performed by an instruction sending module of the server.
  • the shooting instruction is an instruction instructing the client to display the shooting frame according to the shooting parameters.
  • the shooting instruction may include at least one shooting parameter, and is sent to the client via the network by the instruction sending module.
  • the client may display a shooting frame based on at least one shooting parameter. See step 720 for a description about how the client terminal displays a shooting frame based on at least one shooting parameter, and details are not repeated here.
  • the shooting instruction may further include a comparison template image, which is used to further instruct the client to display the comparison template image in the shooting frame.
  • the comparison template image is a reference image that can guide the user to adjust the position of the target and place the target in the shooting frame.
  • the comparison template image can be a real or virtual object image.
  • the shooting instruction may instruct the client to display the image of the target in the shooting frame, so as to guide the user to place the target in the shooting frame to shoot.
  • Step 230 receiving at least one image from the client.
  • Step 230 may be performed by the image receiving module of the server.
  • the image receiving module may accept at least one image from the client over the network.
  • the client may send the at least one image to a storage device for storage, and the image receiving module may acquire the at least one image from the storage device.
  • the at least one image may not contain or contain an object.
  • the at least one image may be captured by an image capturing element of the client, or determined based on data (for example, video or image) uploaded by the user.
  • the at least one image may be a real image captured by an image capture element of the client.
  • the client may display a photographing frame and/or a comparison template image based on the photographing instruction, and guide the user to photograph the face. The user adjusts the position of the face under the guidance of the client-side shooting frame and/or the comparison template image, so that the face is located in the shooting frame, and presses the shooting button to shoot the target image.
  • the hijacker can upload images or videos through the client device.
  • the uploaded image or video may or may not contain the object.
  • the uploaded images or videos may be historical images or videos taken by the client or other clients, or synthesized images or videos.
  • the client or other computing device eg, server 110
  • the client or other computing device may determine the at least one image based on the uploaded image or video.
  • a hijacked client can extract at least one image from the uploaded image or video.
  • the at least one image is a fake image uploaded by the hijacker, not the real image taken by the user when the client displays the shooting frame and/or the comparison template.
  • the target in the image received by the image receiving module is not completely within the shooting frame.
  • the image receiving module may preprocess the at least one image.
  • the preprocessing may include one or more operations of object detection, quality analysis, image segmentation, image noise reduction, image transformation, and the like.
  • the preprocessing may include at least one of object detection, quality analysis, and image segmentation.
  • Object detection is used to determine whether an image contains an object. For example, if the target is a user's face, the target detection can identify the image. If the user's face is identified in the image, the image contains the target; if the user's face does not exist in the image, the image does not contain the target.
  • the image receiving module may exclude images that do not contain the target from the at least one image based on the result of the target detection. For example, users may take images without objects due to misoperation. Removing these images can reduce the calculation amount and time of subsequent authenticity analysis, and improve analysis efficiency.
  • object detection may be performed based on an object detection algorithm.
  • the object detection may be implemented based on an object detection model.
  • Object detection models may include, but are not limited to, a Visual Geometry Group Network model, an Inception NET model, a Fully Convolutional Network model, a segmentation network model, and a Mask-Convolutional Neural Network (Mask- Region Convolutional Neural Network) model, etc.
  • the image receiving module may use a plurality of labeled images as training data when training the object detection model based on a machine learning algorithm (eg, gradient descent algorithm). Alternatively, the object detection model can be trained in another device or module.
  • a machine learning algorithm eg, gradient descent algorithm
  • Quality analysis is used to determine whether an image meets quality requirements.
  • the quality of an image can be measured by one or more image parameters of noise ratio, brightness, resolution, contrast, sharpness, and the like.
  • the quality analysis may analyze one or more image parameters of the image, so as to determine whether the quality of the image meets the requirements. For example, if the resolution of the image is greater than 1024 x 768, the image meets the quality requirements, and if the resolution of the image is less than 1024 x 768, the image does not meet the quality requirements.
  • the image receiving module may, based on the result of the quality analysis, remove images that do not meet the quality requirements from at least one image, so as to reduce the calculation amount of subsequent authenticity analysis and improve the analysis efficiency.
  • the quality analysis may be implemented based on a quality analysis model.
  • the quality analysis model may receive an input image and output a value characterizing the image quality or a determination result of whether the image quality meets quality requirements.
  • the quality analysis model may be, but is not limited to, a combination of one or more of a convolutional neural network model, a recurrent neural network model, and a long short-term memory network model.
  • the image segmentation can be used to segment an area within the photographing frame from the image (called a segmented image corresponding to the photographing frame).
  • the image receiving module may segment the segmented images from the images based on at least one capture parameter.
  • Image segmentation can reduce the interference of shooting out-of-frame images to determine the authenticity of the analysis, and improve the accuracy of target recognition.
  • segmenting the segmented image corresponding to the shooting frame can reduce the amount of computation for subsequent authenticity analysis and improve computation efficiency.
  • any one or more of the aforementioned object detection, quality analysis, and image segmentation may be performed in any order or simultaneously.
  • the image receiving module may first perform target detection on the image, and then perform quality analysis on the image containing the target after screening out the image containing the target; or may first perform quality analysis on the image, and screen out the images that meet the quality requirements. After the image, the target detection is performed on the image that meets the quality requirements.
  • Step 240 Determine authenticity of at least one image based on at least one shooting parameter. Step 240 may be performed by an authenticity determination module.
  • the authenticity of the at least one image includes the first authenticity and/or the second authenticity of each image.
  • the first authenticity may reflect whether the image is an image captured by the client based on the capturing instruction. For example, when the terminal has not been hijacked or attacked, the client will display the shooting frame. Based on the displayed shooting frame, the user moves the position of the target object so that the target object is located in the shooting frame, and performs image shooting. At this point, the image has a first authenticity. For another example, when the terminal is hijacked or attacked, the image is obtained based on the image or video uploaded by the attacker. At this point, the image does not have the first authenticity. In some embodiments, the first authenticity of the image can be used to determine whether the client's camera has been hijacked by an attacker.
  • the image acquisition element of the client is hijacked.
  • the image capturing element of the client is hijacked.
  • the second authenticity may reflect whether the image is an image of one of the at least one eligible target. For example, if the image is an image of a qualified target, the image has the second authenticity, otherwise it does not have the second authenticity.
  • the image authenticity determination module may determine the authenticity of the at least one image based on the preprocessed at least one image. For example, the image authenticity determination module may determine the first and/or second authenticity of the preprocessed at least one image as the first and/or second authenticity of the at least one image. In some embodiments, image segmentation processing may be performed on each of the at least one image to generate a segmented image corresponding to the shooting frame. The image authenticity determination module may determine the authenticity of the at least one image based on the at least one segmented image.
  • Some embodiments of this specification guide the user to locate the image of the target in the shooting frame through the shooting frame displayed by the client, and obtain the image from the client. Further, the authenticity of the image can be judged based on the shooting parameters of the shooting frame in the image, which can effectively determine whether the client is hijacked, thereby ensuring the authenticity of the image.
  • the difficulty coefficient may be determined based on the reference information, and then the shooting parameters of the shooting frame may be determined based on the difficulty coefficient. For example, for a user who has cheated more times, a shooting parameter corresponding to a higher difficulty coefficient can be set, thereby increasing the difficulty of cheating by the user.
  • FIG. 3 is a flowchart of sending a shooting instruction to a client according to some embodiments of the present specification.
  • the instruction sending module may send a shooting instruction including the comparison template image to the client, so as to instruct the client to display the comparison template image in the shooting frame.
  • the shooting instruction may be generated and sent using the process 300 shown in FIG. 3 .
  • Process 300 may include the following steps.
  • Step 310 acquiring a template image of the target.
  • step 310 may be performed by an instruction sending module of the server.
  • the template image is a target image generated based on standard shooting parameters.
  • the template image may be obtained by accessing a storage device, or the template image may be obtained by external input, invoking a related interface, or other methods.
  • the template image can be generated by the instruction sending module.
  • the instruction sending module may determine the position information of at least one key point of the target based on the standard image set of the target.
  • the standard image set of the object is a set containing a plurality of standard images of the object. Among them, the standard image of the target is an image that meets the standard conditions.
  • the standard conditions may include that the target is facing the image capturing element, the image size of the target is 50 mm ⁇ 50 mm, and the distance between the target and the image capturing element is 0.4 m.
  • the standard image set may include a plurality of object images that meet the standard conditions.
  • the key points of the target may include representative parts of the target.
  • keypoints can be eyes, nose, mouth, etc. in standard images.
  • the keypoints may include one or more of left eye center, right eye center, nose center, left mouth corner, right mouth corner, mouth center, etc. in a standard image.
  • the keypoint can also be any location in the standard image.
  • a keypoint can be the center position of a standard image.
  • the location information of key points can characterize their locations in multiple standard images.
  • the location information of the key points may be the average location coordinates of the key points in the standard image. Taking the left eye center as an example, the instruction sending module can determine the coordinates of the left eye center in each standard image, and determine the average coordinates of the left eye center in multiple standard images as the position information of the left eye center.
  • the standard shooting parameters are parameters for generating a template image of the target object.
  • the standard shooting parameters may include one or more of a standard shooting angle, a standard shooting distance, a standard shooting center point, and the like.
  • the standard shooting angle refers to the standard value of the shooting angle.
  • the standard shooting angle may be a shooting angle of 0°.
  • the standard shooting distance refers to the standard value of the shooting distance.
  • the standard shooting distance may be a shooting distance of 0.1 m.
  • the standard shooting center point is the standard position point of the shooting center point.
  • the standard photographing center point may be the position point of the center of the standard image, or the like.
  • the instruction sending module may generate a template image of the target based on at least one standard shooting parameter and position information of at least one key point. For example, the instruction sending module may generate a simulated target image conforming to standard shooting parameters based on the position information of the key points of the target, as the target template image. For another example, the instruction sending module may adjust a certain standard target image according to the position information of at least one key point and standard shooting parameters to generate a target template image.
  • key points such as the center of the left eye and the center of the right eye in the standard image can be adjusted according to their corresponding position information; the orientation, size and position of the target in the standard image can be based on the standard shooting angle and standard shooting distance respectively. and the standard shooting center point to be adjusted.
  • the instruction sending module may directly acquire a standard image set of the target object that conforms to standard shooting parameters, and determine a template image based on the standard image set. For example, the instruction sending module can arbitrarily select a standard image from the standard image set as a template image. For another example, the instruction sending module may determine the position information of at least one key point based on the standard image set, and determine the template image based on the position information of the key point and the standard image set (or a part thereof).
  • Step 320 Adjust the template image based on at least one shooting parameter to generate a comparison template image.
  • step 320 may be performed by an instruction sending module of the server.
  • the standard shooting parameters include a standard shooting angle of 0°, a standard shooting distance of 0.1 m, and a standard shooting center point (ie, the center point on the client screen). If the shooting parameters include the shooting angle of 30°, the shooting distance of 0.3m, and the position of the shooting center point at the upper left corner of the screen (x1, y1), the command sending module can first reduce the template based on the ratio between the standard shooting distance of 0.1m and the shooting distance of 0.3m.
  • Image 410 obtain the first comparison template image 420; then rotate the first comparison template image 420 by the shooting angle 30° to obtain the second comparison template image 430; then move the center position of the second comparison template image 430 to the standard shooting center point, A comparison template image 440 is acquired. It should be understood that the above rotation, scaling and movement of the template image can be performed in any order or simultaneously, which is not limited herein.
  • the instruction sending module may generate the comparison template image based on any one or more of the shooting angle, the shooting distance, the shooting center point, and/or other shooting parameters. For example, the instruction sending module first adjusts the template image based on the shooting center point, and then adjusts the adjusted template image again based on the shooting angle.
  • Step 330 Send a shooting instruction to the client, where the shooting instruction instructs the client to display the comparison template image in the shooting frame.
  • step 330 may be performed by an instruction sending module of the server.
  • the instruction sending module may send the template image and shooting parameters to the client.
  • the client can adjust the template image according to the shooting parameters to generate a comparison template image.
  • the client can further display the shooting frame and the comparison template image in the shooting frame.
  • FIG. 5 is an exemplary flowchart for determining the authenticity of an image according to some embodiments of the present specification.
  • the process 500 shown in FIG. 5 may be performed by a server authenticity determination module.
  • Step 510 Determine the first authenticity of the at least one image based on the at least one shooting parameter.
  • the first authenticity of the image may represent whether the image is an image captured by the client based on the capturing instruction.
  • a target object when a target object is included in the shooting frame of the image, it may be considered that the image is an image captured by the client after the user adjusts the target to be within the shooting frame during the process of displaying the shooting frame, that is, the at least An image has a first authenticity.
  • the authenticity determination module may determine the segmented image corresponding to the shooting frame from the image according to at least one shooting parameter.
  • the authenticity determination module can further detect whether the segmented image contains at least a part of the target object (for example, a representative part or outline), so as to determine whether the target object is contained in the shooting frame of the image. For example, if the target is a user's face, if it is detected that the segmented image contains the user's facial features or facial contours, it means that the frame of the at least one image contains the user's face, so that it can be determined that the image has the first authenticity.
  • Step 520 Determine the second authenticity of the at least one image based on the at least one shooting parameter and the at least one reference image of the at least one qualified target.
  • the target contained in the image is a qualified target, it means that the image has the second authenticity.
  • Qualified targets are pre-validated targets.
  • the qualified target can be the driver's face that has been reviewed by the car-hailing platform during the driver registration process.
  • the qualified target may be the pupil of the payment personnel whose payment authority has been verified by the payment platform.
  • the reference image is the image that contains the qualified target.
  • the reference image may be pre-stored in a storage device, and the authenticity determination module may be based on retrieving the reference image from the storage device over a network.
  • the authenticity determination module may determine the second authenticity of the image based on each reference image in the storage device. For example, in a car-hailing application scenario, the authenticity determination module may determine the second authenticity of the image received from the client based on the reference images of all drivers in the storage device.
  • the authenticity determination module may retrieve a reference image of the user corresponding to the client based on the identification information of the client, and determine the second authenticity of the image based on the reference image of the user.
  • the authenticity determination module may retrieve a reference image of a driver bound to the client from the storage device based on the identification of the client, for confirming the second authenticity of the image received from the client.
  • the authenticity determination module may generate a first image corresponding to the image and a first image corresponding to each of the at least one reference image based on the at least one shooting parameter Second image.
  • the shooting parameters corresponding to the first image and the second image are the same or similar.
  • the client user may rotate or move the head in order to place the target in the shooting frame.
  • the image captured at this time corresponds to the shooting parameters of the shooting frame.
  • the reference image is usually shot under the preset shooting parameters.
  • a reference image of a driver on a car-hailing platform may be an image captured under preset parameters.
  • at least one image and at least one image need to be consistent or standardized so that they correspond to the same or similar shooting parameters.
  • the authenticity determination module may use the image or a segmented image corresponding to a segmented frame in the image as a first image.
  • the authenticity determination module may obtain at least one second image by adjusting each of the at least one reference image based on at least one shooting parameter (or a portion thereof). For example, if the shooting parameters include a shooting angle of 15°, the authenticity determination module may adjust the reference image based on the shooting angle so that the included angle between the reference image and the length direction of the screen is 15°. For another example, if the shooting parameters include the shooting center point with the position coordinates (25, 25), the authenticity determination module may move the position point of the center of the reference image to the coordinates (25, 25).
  • the authenticity determination module may be based on reducing the reference image by a factor of 5.
  • the authenticity determination module may take the adjusted reference image as its corresponding second image.
  • the authenticity determination module may use each of the at least one reference image as the second image. For each of the at least one image, the authenticity determination module may adjust the image or a segmented image corresponding to the image based on at least one shooting parameter (or a portion thereof) to generate the first image. For example, if the shooting parameters include a shooting angle of 15° and a shooting angle of the reference image at 0°, the authenticity determination module can rotate the image by -15° so that the shooting angle of the adjusted image is the same as the shooting angle of the reference image. The authenticity determination module may take the adjusted image as its corresponding first image.
  • the authenticity determination module first determines the segmented image corresponding to the shooting frame, and then adjusts the segmented image to generate the first image. In some embodiments, the authenticity determination module may adjust the at least one image and the at least one reference image respectively, so that both the first image and the second image correspond to standard shooting parameters or other same shooting parameters.
  • the authenticity determination module may determine the second authenticity of the image based on the similarity between the corresponding first image and the at least one second image. For example, the authenticity determination module may determine the similarity between the first image feature of the first image and the second image feature of each second image to determine the authenticity of the image.
  • Image features of an image may include color features, texture features, shape features, depth features, etc., or any combination thereof.
  • the similarity between the first image feature of the first image and the second image feature of the second image can be calculated by vector similarity, for example, determined by Euclidean distance, Manhattan distance, cosine similarity and the like.
  • the similarity between the first image feature of the first image and the second image feature of a certain second image exceeds a certain threshold, it can be considered that the first image and the second image are images of the same object. That is, the image corresponding to the first image is an image of a qualified target, that is, the image corresponding to the first image has the second authenticity. If the similarity between the first image feature of the first image and the second image features of all the second images does not exceed a certain threshold, it may be considered that the image corresponding to the first image does not have the second authenticity.
  • the authenticity determination module may process each of the first image and the at least one second image based on the image comparison model to determine the second authenticity of the image. For example, the authenticity determination module may compare one of the input images of the first image and the second image to the model, and the image comparison model may output the similarity between the first image and the second image and/or the first image and the second image The judgment result of whether the images are similar.
  • FIG. 6 shows an exemplary structural diagram of an image comparison model. As shown in the figure, the image comparison model 600 may include a feature extraction layer 605 , a similarity calculation layer 608 and a discrimination layer 610 .
  • the image 601 may include at least an image 601-1 and an image 601-m
  • the first image 603 generated according to the image 601 may include at least a first image 603-1 and a first image 603-m
  • the image 601 corresponds to
  • the reference image 602 may include at least a reference image 602-1 and a reference image 602-n
  • the second image 604 generated based on the reference image 602 may include at least a second image 604-1 and a second image 604-n.
  • one first image and at least one second image may form an image pair.
  • the image comparison model 600 may analyze the image pair to determine whether the first image and the second image in the image pair are similar. For example, as shown in FIG. 6 , an image pair consisting of a first image 603 - m and a second image 604 - n may be input into the image comparison model 600 .
  • the feature extraction layer 605 may be used to process the first image and the second image to obtain the first image features 606 of the first image 603-m and the second image features 607 of the second image 604-n.
  • the type of feature extraction layer 605 may include a convolutional neural network model such as ResNet, ResNeXt, SE-Net, DenseNet, MobileNet, ShuffleNet, RegNet, EfficientNet, or Inception, or a recurrent neural network model.
  • the first image 603-m and the second image 604-n may be stitched into the feature extraction layer 605.
  • the output of the feature extraction layer 605 may be a feature vector obtained by concatenating the first image features 606 of the first image 603-m and the second image features 607 of the second image 604-n.
  • the similarity calculation layer 608 may be used to determine the similarity 609 of the first image feature 606 and the second image feature 607 .
  • the discrimination layer 610 may be configured to output a determination result of whether the first image 603-m and the second image 604-n are similar based on the similarity 609 . For example, the discrimination layer 610 may compare the similarity 609 to a similarity threshold. If the similarity between the first image feature 606 of the first image 603-m and the second image feature 607 of the second image 604-n exceeds the similarity threshold, the first image 603-m and the second image 604-n resemblance.
  • the first image 603-m may be determined based on the judgment result 611 of whether the first image 603-m corresponding to the image 601-m and each second image 604 corresponding to each reference image 602 are similar Corresponds to the second reality of image 601-m. For example, if the first image 603-m is not similar to each of the second images 604, the first image does not have the second authenticity. For another example, if the first image 603-m and the second image 604-1 are similar, the first image has the second authenticity.
  • the authenticity determination module may input multiple image pairs consisting of the first image and the second image together into the image comparison model 600 .
  • the image comparison model 600 can simultaneously output the similarity determination result of each of the multiple pairs of image pairs.
  • the image comparison model 600 is a machine learning model with preset parameters. Preset parameters refer to the model parameters learned during the training of the machine learning model. Taking a neural network as an example, the model parameters include weight and bias.
  • the preset parameters of the image comparison model 600 are generated through a training process. For example, the model acquisition module can train an initial image comparison model based on multiple training samples with labels to obtain an image comparison model.
  • Training samples include one or more sample image pairs with labels.
  • Each sample image pair includes a first sample image and a second sample image.
  • the first sample image and the second sample image may be images of the same or different objects.
  • the labels of the training samples can indicate whether the first sample image and the second sample image are similar (or whether they are pictures of the same object).
  • the image comparison model may be pre-trained by the processing device or a third party and stored in the storage device, and the processing device may directly call the image comparison model from the storage device.
  • the authenticity determination module may determine the authenticity of the image based only on the first authenticity. For example, if at least one image has the first authenticity, the at least one image is considered to be authentic, passing object recognition. In some embodiments, the authenticity determination module may determine the authenticity of the image based on the first authenticity and the second authenticity. For example, it may be determined that at least one image has the first authenticity, and then it is determined whether the target in the at least one image is a qualified target. If the image has the second authenticity, then at least one image is considered to be authentic, passing target recognition. The first authenticity analysis is performed based on simple features of the image, which is simpler and requires less computing resources than the second authenticity analysis.
  • the efficiency of target recognition can be improved, the steps of target recognition can be simplified, and the waste of computing resources (for example, avoiding the use of computing
  • the resource conducts a second authenticity analysis of the fake images uploaded by the hijackers).
  • the authenticity determination module may determine the authenticity of the image directly based on the second authenticity. For example, if at least one image has the first authenticity, the at least one image is considered to be authentic, passing object recognition.
  • the authenticity determination module may select different methods for target recognition based on the reference information. For example, if the user's history of cheating times is greater than a certain threshold, the authenticity determination module may select a target recognition method in which the first authenticity is judged first, and then the second authenticity is judged. For another example, if the user's history of cheating is 0 times, the authenticity determination module may select a target recognition method for directly judging the second authenticity.
  • FIG. 7 is an exemplary flowchart of a target identification method applied to a client according to some embodiments of the present specification. As shown in FIG. 7, the process 700 includes the following steps.
  • Step 710 receiving a shooting instruction from the server.
  • step 710 may be performed by an instruction receiving module of the client.
  • the shooting instruction refers to an instruction for the client to display the shooting frame.
  • the photographing instruction may include photographing parameters related to the photographing frame, for example, the shape, size, position, and display parameters of the photographing frame.
  • the photographing instruction may further include a comparison template image of the target, which is used to instruct the client to display the comparison template image in the photographing frame.
  • the capture instructions include a template image of the target. The client can generate a comparison template image based on the template image and shooting parameters.
  • the shooting instruction reference may be made to the descriptions in other parts of this application, for example, step 220 .
  • Step 720 displaying a shooting frame based on at least one shooting parameter.
  • step 720 may be performed by a capture frame display module of the client.
  • the shooting frame refers to a specific area displayed on the screen of the client (eg, the client 130 ), and the user of the client can be guided to place the target in the specific area when shooting.
  • the shooting parameters may include any parameters related to the shape, size, position, display manner, etc. of the shooting frame.
  • the shooting frame display module may generate a shooting frame based on at least one shooting parameter, and instruct the client to display the shooting frame in the screen display area.
  • the shooting frame display parameters may generate the shooting frame based on the shape, size, and position parameters, and instruct the client to display the shooting frame in a specific manner (eg, in a specific color, blinking frequency) based on the display parameters.
  • the client can obtain a preset shooting frame, and the preset shooting frame has a specific shape, size, position, and the like.
  • the shooting frame display module may rotate, zoom, and translate the preset shooting frame based on the shooting parameters to generate the shooting frame.
  • the process of adjusting the shooting frame based on the shooting parameters is similar to the process of adjusting the template image based on the shooting parameters described in FIGS. 3 and 4 , and details are not repeated here.
  • the shooting frame display module may instruct the client to display the comparison template image in the shooting frame, similar to that shown in FIG. 4 .
  • the client side displays the comparison template image
  • the user can align the outline of the target object with the outline of the comparison template image to improve the accuracy of template recognition.
  • the target is a user's face
  • the user can align the facial contour with the target image template (ie, compare the template image).
  • the shooting frame display module may instruct the client to display at least one key point of the target in the shooting frame or the comparison template image. Taking the user's face as an example, the left eye, right eye, nose tip, left mouth corner, and right mouth corner of the face can be further displayed in the target image template (ie, the comparison template image). The corners of the mouth and the right corner of the mouth are respectively aligned with the corresponding key points in the target image template.
  • the shooting frame display module may directly acquire the comparison template image and display it on the screen of the client.
  • the edge of the contrast template image can be regarded as a shot frame.
  • Step 730 Acquire at least one captured image based on the image capturing element.
  • step 730 may be performed by an image acquisition module.
  • the captured image is the image acquired by the image acquisition element of the client. Objects may or may not be included in the captured image.
  • the image acquisition module may acquire at least one captured image based on the video captured by the image capturing element. Specifically, the image acquisition module may extract at least one frame of image from the video as at least one shot image. For example, the image acquisition module randomly selects n frames from the video shot by the image acquisition element as shot images. For another example, the image acquisition module may first identify the video frames containing the target object in the shooting frame, and extract n frames of the video containing the target object as the shot image.
  • the image acquisition module may instruct the image acquisition element to acquire the captured image based on the confirmation instruction.
  • the confirmation instruction is a shooting instruction triggered by the user through a confirmation operation.
  • the confirmation operation may be a manual input operation, a voice input operation, or the like.
  • the image acquisition module can capture images.
  • the image acquisition module may automatically instruct the image acquisition element to capture images. For example, when it is detected that there is a target in the shooting frame, the image acquisition module can automatically instruct the image acquisition element to capture an image.
  • Step 740 Send the at least one captured image to the server to judge the authenticity of the at least one captured image.
  • step 740 may be performed by an image sending module.
  • the image sending module may send the at least one captured image acquired by the image capturing element as an image to the server through the network, so as to determine the authenticity of the at least one captured image (eg, the second authenticity determination) .
  • the hijacker may upload images or videos through the client device. In this case, step 730 may be omitted.
  • the client can send the image or video uploaded by the hijacker to the server, and the server can judge the first authenticity and/or the second authenticity of the image or video.
  • the image sending module will preprocess the captured image, and send the preprocessed captured image to the server for further analysis.
  • the preprocessing of the captured image is similar to the preprocessing of the image by the image receiving module of the server, and will not be repeated here. Refer to FIG. 5 and related descriptions for a detailed description of the server's determination of the authenticity of the at least one captured image, which will not be repeated here.
  • the client can also receive the authenticity determination result of the captured image from the server through the network.
  • the client may display the guidance information based on the authenticity determination result.
  • Guidance information is information that prompts the user to perform the next step based on the authenticity judgment result.
  • the guidance information may include voice information, text information, image information, and the like.
  • the guidance information can be the voice message "Approved, please start driving".
  • the guidance information may be the text information displayed on the screen display area of the client terminal "Failed to pass the review, please identify again”.
  • the client may further determine the guide information based on the reference information, where the reference information may include shooting environment information, historical behavior of the user, personal information of the user, and the like.
  • the client can determine the guidance information as "Failed to pass the review, please go to a brighter environment based on the shooting environment information "light intensity ⁇ 10lux" Identify again".
  • the client can determine the guidance information as "The audit has not been approved” based on the user's historical behavior "the number of historical target identification failures caused by user fraud > 10 times".
  • Some embodiments of this specification determine the guidance information based on the reference information, and can provide guidance, prompts or warnings according to different target recognition intentions and different operation behaviors of the user, so as to improve the pertinence and effectiveness of the guidance information, thereby improving the accuracy of target recognition. sex.
  • the possible beneficial effects of the embodiments of this specification include, but are not limited to: (1) Guide the user to adjust the target object to the shooting frame to acquire an image through the shooting frame displayed by the client, and determine the authenticity of the image based on the shooting parameters of the shooting frame Judgment can effectively determine whether the client is hijacked and/or whether the image is an image of a qualified target; (2) determine the difficulty coefficient based on the reference information, and then determine the shooting parameters of the shooting frame based on the difficulty coefficient, so that by targeting different Different shooting parameters can be set for different scenes, which can improve the accuracy of authenticity judgment, and improve the applicability and flexibility of target recognition; (3)
  • the first authenticity analysis is based on the simple features of the image, which is similar to the second authenticity analysis. It is simpler and saves computing resources.
  • the efficiency of target recognition can be improved, the steps of target recognition can be simplified, and the waste of computing resources (for example, avoiding the use of computing (4)
  • selecting different methods for target recognition can improve the adaptability and efficiency of target recognition.

Abstract

本说明书实施例提供一种目标识别的方法和系统。该方法包括：获取与拍摄框有关的至少一个拍摄参数；向客户端发送拍摄指令，拍摄指令指示客户端基于至少一个拍摄参数显示拍摄框；从客户端接收至少一幅图像；以及基于至少一个拍摄参数，确定至少一幅图像的真实性。

Description

一种目标识别的方法和系统
优先权声明
本申请要求2021年04月20日提交的中国专利申请号202110424645.8的优先权，其内容全部通过引用并入本文。
技术领域
本说明书涉及图像处理领域，特别涉及一种目标识别的方法和系统。
背景技术
目标识别是基于图像采集元件获取的图像进行识别的技术。目标物为了提高目标识别的准确性，需要确定图像的真实性。
因此，希望提供一种目标识别的方法和系统，可以确定图像的真实性。
发明内容
本说明书实施例之一提供一种目标识别的方法。所述目标识别的方法包括：获取与拍摄框有关的至少一个拍摄参数；向客户端发送拍摄指令，所述拍摄指令指示所述客户端基于所述至少一个拍摄参数显示所述拍摄框；从所述客户端接收至少一幅图像；以及基于所述至少一个拍摄参数，确定所述至少一幅图像的真实性。
本说明书实施例之一提供一种目标识别的方法。所述目标识别的方法包括：从服务器接收拍摄指令，所述拍摄指令包括与拍摄框有关的至少一个拍摄参数；基于所述至少一个拍摄参数，显示所述拍摄框；基于图像采集元件获取至少一幅拍摄图像；将所述至少一幅拍摄图像发送给所述服务器以对所述至少一幅拍摄图像的真实性进行判别。
本说明书实施例之一提供一种目标识别的系统。所述的目标识别的系统包括：参数获取模块，用于获取与拍摄框有关的至少一个拍摄参数；指令发送模 块，用于向客户端发送拍摄指令，所述拍摄指令用于指示所述客户端基于所述至少一个拍摄参数显示所述拍摄框；图像接收模块，用于从所述客户端接收至少一幅图像；以及真实性确定模块，用于基于所述至少一个拍摄参数，确定所述至少一幅图像的真实性。
本说明书实施例之一提供一种目标识别的系统。所述的目标识别的系统包括：指令接收模块，用于从服务器接收拍摄指令，所述拍摄指令包括与拍摄框有关的至少一个拍摄参数；拍摄框显示模块，用于基于所述至少一个拍摄参数，显示所述拍摄框；图像获取模块，用于基于图像采集元件获取至少一幅拍摄图像；图像发送模块，用于将所述至少一幅拍摄图像发送给所述服务器以对所述至少一幅拍摄图像的真实性进行判别。
附图说明
本说明书将以示例性实施例的方式进一步说明，这些示例性实施例将通过附图进行详细描述。这些实施例并非限制性的，在这些实施例中，相同的编号表示相同的结构，其中：
图1是根据本说明书一些实施例所示的目标识别系统的应用场景示意图；
图2是根据本说明书一些实施例所示的应用于服务器的目标识别方法的示例性流程图；
图3是根据本说明书一些实施例所示的向客户端发送拍摄指令的流程图；
图4是根据本说明书一些实施例所示的显示拍摄框的示意图；
图5是根据本说明书一些实施例所示的确定图像真实性的示例性流程图；
图6是根据本说明书一些实施例所示的图像对比模型的示意图；
图7是根据本说明书一些实施例所示的应用于客户端的目标识别方法的示例性流程图。
具体实施方式
为了更清楚地说明本说明书实施例的技术方案，下面将对实施例描述中 所需要使用的附图作简单的介绍。显而易见地，下面描述中的附图仅仅是本说明书的一些示例或实施例，对于本领域的普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图将本说明书应用于其它类似情景。除非从语言环境中显而易见或另做说明，图中相同标号代表相同结构或操作。
应当理解，本文使用的“系统”、“装置”、“单元”和/或“模块”是用于区分不同级别的不同组件、元件、部件、部分或装配的一种方法。然而，如果其他词语可实现相同的目的，则可通过其他表达来替换所述词语。
如本说明书和权利要求书中所示，除非上下文明确提示例外情形，“一”、“一个”、“一种”和/或“该”等词并非特指单数，也可包括复数。一般说来，术语“包括”与“包含”仅提示包括已明确标识的步骤和元素，而这些步骤和元素不构成一个排它性的罗列，方法或者设备也可能包含其它的步骤或元素。
本说明书中使用了流程图用来说明根据本说明书的实施例的系统所执行的操作。应当理解的是，前面或后面操作不一定按照顺序来精确地执行。相反，可以按照倒序或同时处理各个步骤。同时，也可以将其他操作添加到这些过程中，或从这些过程移除某一步或数步操作。
目标识别是基于图像采集元件获取的目标物进行识别的技术。在一些实施例中，目标物可以是人脸、指纹、掌纹、瞳孔、非生物体等。在一些实施例中，目标识别可以应用于权限验证。例如，门禁权限认证和账户支付权限认证等。在一些实施例中，目标识别还可以用于身份验证。例如，员工考勤认证和本人注册身份安全认证。仅作为示例，目标识别可以基于图像采集元件实时采集到的目标物的图像和预先获取的生物特征进行匹配，从而验证目标物的身份。
然而，图像采集元件可能被攻击或劫持，攻击者可以上传虚假的图像通过身份验证。例如，攻击者A可以在攻击或劫持图像采集元件后，直接上传用户B的图像。目标识别系统基于用户B的图像和预先获取的用户B的生物特征进行识别，从而通过用户B的身份验证。因此，为了保证目标识别的安全性，需要确定图像的真实性，即确定图像是图像采集元件在目标识别过程中实时采集 到的。
图1是根据本说明书一些实施例所示的目标识别系统的应用场景示意图。
如图1所示，目标识别系统100可以包括服务器110、网络120、客户端130和存储设备140。
服务器110可以用于处理来自目标识别系统100的至少一个组件或外部数据源(例如，云数据中心)的数据和/或信息。例如，服务器110可以获取与拍摄框有关的至少一个拍摄参数，并基于所述至少一个拍摄参数确定客户端130发送的至少一幅图像的真实性等。又例如，服务器110可以对从客户端130获取的至少一幅图像进行预处理(例如，目标检测、质量分析等)，得到预处理后的至少一幅图像。在处理过程中，服务器110可以从存储设备140获取数据(如指令)或将数据(例如，至少一幅图像)保存到存储设备140，也可以通过网络120从客户端130等其他来源读取数据(例如，拍摄环境信息)或将数据(例如，拍摄指令)输出至客户端130。
在一些实施例中，服务器110可以是单一服务器或服务器组。该服务器组可以是集中式或分布式的(例如，服务器110可以是分布式系统)。在一些实施例中，服务器110可以是区域的或者远程的。在一些实施例中，服务器110可以在云平台上实施，或者以虚拟方式提供。仅作为示例，云平台可以包括私有云、公共云、混合云、社区云、分布云、内部云、多层云等或其任意组合。
网络120可以连接目标识别系统100的各组成部分和/或连接目标识别系统100与外部部分。网络120使得目标识别系统100各组成部分之间，和/或目标识别系统100与外部部分之间可以进行通讯，促进数据和/或信息的交换。在一些实施例中，网络120可以是有线网络或无线网络中的任意一种或多种。例如，网络120可以包括电缆网络、光纤网络、电信网络、互联网、局域网络(LAN)、广域网络(WAN)、无线局域网络(WLAN)、城域网(MAN)、公共交换电话网络(PSTN)、蓝牙网络、紫蜂网络(ZigBee)、近场通信(NFC)、设备内总线、设备内线路、线缆连接等或其任意组合。在一些实施例中，系统各部分之 间的网络连接可以采用上述一种方式，也可以采取多种方式。在一些实施例中，网络120可以是点对点的、共享的、中心式的等各种拓扑结构或者多种拓扑结构的组合。在一些实施例中，网络120可以包括一个或以上网络接入点。例如，网络120可以包括有线或无线网络接入点，例如基站和/或网络交换点120-1、120-2、…，通过这些网络接入点，目标识别系统100的一个或多个组件可连接到网络120以交换数据和/或信息。
客户端130可以实现用户和目标识别系统100之间的交互。在一些实施例中，客户端130可以包括图像采集元件(例如，摄像头、照相机)，用于拍摄图像数据(图像和/或视频)。在一些实施例中，图像采集元件在拍摄时，客户端130(例如，客户端130的屏幕)可以显示指导用户进行拍摄的信息。例如，客户端130可以接收或确定与拍摄框有关的一个或多个拍摄参数，并基于所述一个或多个拍摄参数在其屏幕上显示拍摄框，以引导用户将目标物放置在拍摄框内进行拍摄。在一些实施例中，客户端130可以通过网络120与服务器110通信，并将拍摄的至少一幅图像发送到服务器110。在一些实施例中，客户端130可以是移动设备130-1、平板计算机130-2、膝上型计算机130-3、其他具有输入和/或输出功能的设备等或其任意组合。上述示例仅用于说明所述客户端130设备范围的广泛性而非对其范围的限制。
存储设备140可以用于存储数据(如目标物的标准图像、合格目标物的至少一幅参考图像等)和/或指令。存储设备140可以包括一个或多个存储组件，每个存储组件可以是一个独立的设备，也可以是其他设备的一部分。在一些实施例中，存储设备140可包括随机存取存储器(RAM)、只读存储器(ROM)、大容量存储器、可移动存储器、易失性读写存储器等或其任意组合。示例性地，大容量储存器可以包括磁盘、光盘、固态磁盘等。在一些实施例中，存储设备140可在云平台上实现。仅作为示例，云平台可以包括私有云、公共云、混合云、社区云、分布云、内部云、多层云等或其任意组合。在一些实施例中，存储设备140可以集成或包括在目标识别系统100的一个或多个其他组件(例如，服务器110、 客户端130或其他可能的组件)中。
在一些实施例中，服务器110可以包括参数获取模块、指令发送模块、图像接收模块、真实性确定模块和模型获取模块。
参数获取模块可以用于获取与拍摄框有关的至少一个拍摄参数。在一些实施例中，参数获取模块可以随机生成至少一个拍摄参数。在一些实施例中，参数获取模块可以基于参考信息，确定拍摄难度系数；基于所述拍摄难度系数，确定所述至少一个拍摄参数。
指令发送模块可以用于向客户端发送拍摄指令，拍摄指令用于指示客户端基于至少一个拍摄参数显示拍摄框。在一些实施例中，指令发送模块可以获取目标物的模板图像；以及基于所述至少一个拍摄参数，对模板图像进行调整，以生成对比模板图像，其中所述拍摄指令进一步指示所述客户端在所述拍摄框内显示所述对比模板图像。
图像接收模块可以用于从客户端接收至少一幅图像。
真实性确定模块可以用于基于至少一个拍摄参数，确定至少一幅图像的真实性。在一些实施例中，真实性确定模块可以对所述至少一幅图像进行预处理，生成预处理后的至少一幅图像；以及基于所述至少一个拍摄参数和预处理后的至少一幅图像，确定所述至少一幅图像的真实性。在一些实施例中，真实性确定模块对所述至少一幅图像进行预处理包括对所述至少一幅图像中的每一幅执行以下操作中的至少一个：对所述图像进行目标检测，确定所述图像是否包含目标物；对所述图像进行质量分析，确定所述图像是否满足质量要求；或者对所述图像进行图像分割，生成所述拍摄框对应的分割图像。
在一些实施例中，真实性确定模块可以基于所述至少一个拍摄参数，确定所述至少一幅图像的第一真实性，所述第一真实性反映所述至少一幅图像是否为所述客户端基于拍摄指令拍摄的图像；真实性确定模块可以基于所述至少一个拍摄参数和至少一个合格目标物的至少一张参考图像，确定所述至少一幅图像的第二真实性，所述第二真实性反映所述至少一幅图像是否为所述至少一个 合格目标物中的一个的图像。
模型获取模块用于获取一个或多个机器学习模型，例如图像对比模型、难度系数确定模型和拍摄参数确定模型等。在一些实施例中，模型获取模块可以从目标识别系统100中的其他原件(例如，存储设备140)或外部来源获取一个或多个机器学习模型。
在一些实施例中，客户端130可以包括拍摄指令接收模块、拍摄框显示模块、图像获取模块及图像发送模块。指令接收模块可以用于从服务器接收拍摄指令，拍摄指令包括与拍摄框有关的至少一个拍摄参数。拍摄框显示模块可以用于基于至少一个拍摄参数，显示拍摄框。在一些实施例中，拍摄框显示模块可以在所述拍摄框内显示所述目标物的对比模板图像。图像获取模块可以用于基于图像采集元件获取至少一幅拍摄图像。图像发送模块可以用于将至少一幅拍摄图像发送给服务器以对至少一幅拍摄图像的真实性进行判别。
需要注意的是，以上对于候选项显示、确定系统及其模块的描述，仅为描述方便，并不能把本说明书限制在所举实施例范围之内。可以理解，对于本领域的技术人员来说，在了解该系统的原理后，可能在不背离这一原理的情况下，对各个模块进行任意组合，或者构成子系统与其他模块连接。在一些实施例中，图1中披露的多个模块可以是一个系统中的不同模块，也可以是一个模块实现上述的两个或两个以上模块的功能。例如，各个模块可以共用一个存储模块，各个模块也可以分别具有各自的存储模块。
图2是根据本说明书一些实施例所示的目标识别方法的示例性流程图。如图2所示，流程200包括下述步骤。
步骤210，获取与拍摄框有关的至少一个拍摄参数。在一些实施例中，步骤210可以由服务器(例如，服务器110)的参数获取模块执行。
拍摄框是指在客户端(例如，客户端130)屏幕上显示的特定区域，可以引导客户端的用户在进行拍摄时将目标物放置于所述特定区域中。在一些实施例中，拍摄框可以为任意形状，例如，矩形、圆形、椭圆形等。在一些实施例中， 为了便于用户识别拍摄框，客户端可以在屏幕上标记出拍摄框。例如，拍摄框的边缘可以用特定的颜色标记。又例如，可以用不同于屏幕显示区域的颜色填充拍摄框。客户端的用户指的是使用客户端进行目标识别的用户。
目标物指需要进行目标识别的物体。例如，目标物可以是用户的面部、指纹、掌纹或瞳孔等。又例如，目标物可以是非生物体(如汽车)。在一些实施例中，所述目标物指需要进行身份验证和/或权限认证的用户的面部。例如，在网约车应用场景中，平台需要验证接单司机是否为平台审核过的注册司机用户，则所述目标物是司机的面部。又例如，在支付应用场景中，支付系统需要验证支付人员的支付权限，则所述目标物是支付人员的面部。
拍摄参数可以包括与拍摄框的形状、大小、位置、显示方式等相关的任何参数。示例性的拍摄参数可以包括拍摄角度、拍摄距离、拍摄中心点、显示参数等。
拍摄角度是拍摄框相对于参考方向(如客户端屏幕的长度方向)的角度。拍摄角度的改变可以导致拍摄框和客户端屏幕相对方向的改变。例如，假设拍摄框为矩形。当拍摄角度为0°时，拍摄框的长度方向与屏幕的长度方向平行；当拍摄角度为30°时，拍摄框的长度方向与屏幕的长度方向之间的夹角为30°。
拍摄距离是指当用户将目标物放置于拍摄框内进行拍摄时，目标物与客户端的图像采集元件之间的预估距离。拍摄距离的改变可以导致拍摄框与屏幕的大小比例发生改变。例如，当拍摄距离为0.5m时，拍摄框与屏幕的比例为0.8：1；当拍摄距离为1m时，拍摄框与屏幕的比例为0.6：1。
拍摄中心点是拍摄框的定位点。例如，拍摄中心点可以是位于拍摄框的中心的位置点、位于拍摄框的边框上的位置点等。拍摄中心点在屏幕上的位置改变可以导致拍摄框在屏幕上的位置改变。
显示参数是与拍摄框的显示方式相关的模式参数。在一些实施例中，显示参数可以包括拍摄框的形状、填充颜色、边框颜色、是否闪烁显示等。
在一些实施例中，参数获取模块可以随机生成至少一个拍摄参数。例如， 对于某一拍摄参数，参数获取模块可以在目标识别系统100的预先设定的该拍摄参数的数值范围内，随机确定该拍摄参数的数值。本实施例获取的拍摄参数随机性较大，可以提高用户作弊难度，从而提高目标识别的准确性。
在一些实施例中，参数获取模块可以根据目标识别系统100的默认设定确定所述拍摄参数。例如，参数获取模块可以根据目标物的类型，从存储设备140获取预先存储的与该目标物对应的拍摄参数。在一些实施例中，参数获取模块可以从终端设备获取用户根据经验设定的拍摄参数。在一些实施例中，参数获取模块可以通过数据分析确定所述拍摄参数。例如，参数获取模块可以根据从客户端接收的设备信息确定拍摄参数。
在一些实施例中，参数获取模块可以基于参考信息，确定拍摄难度系数。参数获取模块可以进一步基于所述拍摄难度系数，确定拍摄参数。参考信息可以反映客户端的用户在目标识别中作弊的可能性和/或难度。例如，参考信息可以包括拍摄环境信息、客户端对应的历史用户的历史行为信息、客户端对应的历史用户的个人信息等，或其任意组合。
拍摄环境信息是与客户端的图像采集元件的拍摄环境相关的信息。例如，拍摄环境信息可以包括环境光照信息，例如，光照强度信息、光照类型信息等。又例如，拍摄环境信息可以包括环境背景信息，例如，背景静动态信息、背景纹理信息等。在一些实施例中，参数获取模块可以从客户端接收拍摄环境信息。例如，客户端可以基于图像采集元件拍摄的图像数据确定拍摄环境信息。又例如，客户端可以包含检测拍摄环境的传感器(如光敏传感器)，用于检测拍摄环境信息。通常，拍摄环境越好(例如，环境光照越好)，用户作弊的难度越小。
客户端对应的历史用户可以包括与客户端具有绑定关系的用户、曾经使用过客户端的历史用户等。例如，客户端对应的历史用户可以是利用该客户端在交通服务平台上进行注册的司机。客户端对应的历史用户可以与当前使用客户端进行目标识别的用户相同或者不同。
历史用户的历史行为信息可以与历史用户的历史行为，如历史识别行为 有关。例如，历史用户的历史行为信息可以包括历史用户的历史识别失败次数、历史识别失败的原因等。其中，历史识别失败的原因可以包括用户作弊、用户误操作等。在一些实施例中，参数获取模块可以从客户端或存储设备获取客户端的使用记录，以确定历史用户的历史行为信息。通常，历史用户的历史识别失败次数和/或历史作弊次数越多，客户端的用户在本次目标识别中作弊的可能性越高。
历史用户的个人信息是与历史用户相关的信息，如历史用户标识和历史用户属性。其中，历史用户标识是区分历史用户的符号。例如，历史用户的身份证ID、驾驶证ID等。历史用户属性可以包括历史用户的年龄、学历、性别、信用记录等。示例性地，历史用户的信用记录越好，客户端的用户在本次目标识别中作弊的可能性越低。在一些实施例中，参数获取模块可以从客户端、存储设备或外部来源获取历史用户的个人信息。例如，客户端可以在历史用户注册时采集个人信息，并将个人信息存储到存储设备。
拍摄难度系数是表征客户端的用户将目标物放置在拍摄框中进行拍摄的难度。在一些实施例中，用户将目标物放置在拍摄框中进行拍摄的难度越小，则拍摄难度系数越大。
在一些实施例中，参数获取模块可以基于参考信息确定拍摄难度系数。示例性地，光照强度越大，用户越容易将目标物放置在拍摄框中。此时，参数获取模块可以确定较大的拍摄难度系数，以防止用户作弊。又例如，用户历史行为中，目标识别失败原因为“用户欺诈”的历史目标识别失败次数越多，用户在本次目标识别过程中的欺诈可能性越高。此时，参数获取模块可以确定较大的拍摄难度系数，以防止用户作弊。又一示例性地，用户的信用记录越差，说明用户在本次目标识别过程中的作弊可能性越高。此时，参数获取模块可以确定较大的拍摄难度系数，以防用户作弊。
在一些实施例中，参数获取模块可以根据第一规则确定拍摄难度系数。第一规则与一种或多种参考信息与拍摄难度系数之间的关系有关。例如，第一规则可以包括当光照强度小于30lux时，拍摄难度系数为0.1；当光照强度大于30lux， 小于100lux时，拍摄难度系数为0.3；当光照强度大于100lux时，拍摄难度系数为0.6。又例如，第一规则可以包括当历史识别失败次数大于10次时，拍摄难度系数为0.6；当历史识别失败次数小于10次大于3次时，拍摄难度系数为0.3；当历史识别失败次数较少小于3次时，拍摄难度系数为0.1。
在一些实施例中，参数获取模块可以基于多种参考信息中的每一种确定一个拍摄难度系数。参数获取模块可以进一步基于多个拍摄难度系数确定最终的拍摄难度系数。例如，可以对多个拍摄难度系数进行求和、加权求和、求平均值等，来确定最终的拍摄难度系数。例如，参数获取模块分别基于光照强度40lux、历史识别失败次数次7次、用户信用等级好，确定难度系数为0.3、0.3和0.3，获取最终的拍摄难度系数为(0.3+0.3+0.3)/3＝0.3。
在一些实施例中，参数获取模块还可以通过难度系数确定模型确定拍摄难度系数，具体的，难度系数确定模型的输入为参考信息，难度系数确定模型的输出为拍摄难度系数。在一些实施例中，难度系数确定模型可以包括但不限于深度神经网络模型、循环神经网络模型等。
进一步地，参数获取模块可以基于拍摄难度系数，确定至少一个拍摄参数。例如，拍摄难度系数越大，则拍摄角度越大，拍摄距离越远，拍摄中心越远离客户端的屏幕中心。如上文所述，当参考信息显示客户端用户作弊的可能性较高和/或作弊的难度较小时，拍摄难度系数会较高。通过设置至少一个拍摄参数的值，可以使用户将目标物放置在拍摄框中的难度提高，由此实现较高的拍摄难度系数。
在一些实施例中，参数获取模块可以基于第二规则确定至少一个拍摄参数。第二规则与拍摄难度系数和至少一个拍摄参数之间的关系有关。示例性的，当拍摄难度系数为0.1时，拍摄参数可以包括：拍摄角度0°、拍摄距离0.1m、拍摄中心点与屏幕的中心点重合；当拍摄难度系数为0.6时，拍摄参数可以包括：拍摄角度40°、拍摄距离0.3m、拍摄中心点位于屏幕的中心点上方0.05m。
在一些实施例中，参数获取模块还可以根据拍摄参数确定模型确定至少 一个拍摄参数。具体的，拍摄参数确定模型的输入为拍摄难度系数，输出为至少一个拍摄参数。在一些实施例中，拍摄参数确定模型可以包括但不限于深度神经网络模型、循环神经网络模型等。在一些实施例中，所述难度系数确定模型和拍摄参数确定模型可以是同一模型的不同层。
在一些实施例中，所述至少一个拍摄参数包括多个拍摄参数，参数获取模块可以用相同或者不同的方式来获取所述多个拍摄参数。例如，参数获取模块可以随时生成多个拍摄参数中的每一个。又例如，多个参数中的部分可以随机生成，另一部分可以基于参考信息确定。
步骤220，向客户端发送拍摄指令。在一些实施例中，步骤220可以由服务器的指令发送模块执行。
拍摄指令是指示客户端根据拍摄参数显示拍摄框的指令。例如，拍摄指令可以包括至少一个拍摄参数，并由指令发送模块经由网络被发送至客户端。进一步地，客户端可以基于至少一个拍摄参数，显示拍摄框。关于客户端基于至少一个拍摄参数，显示拍摄框的相关描述参见步骤720，在此不再赘述。
在一些实施例中，拍摄指令还可以包括对比模板图像，用于进一步指示客户端在拍摄框内显示对比模板图像。对比模板图像是可以引导用户调整目标物的位置、将目标物放置在拍摄框中的参考图像。例如，在目标物识别中，对比模板图像可以是真实的或虚拟的目标物图像。拍摄指令可以指示客户端在拍摄框中显示目标物图像，以引导用户将目标物放置在拍摄框中进行拍摄。关于对比模板图像的详细描述可以参考本说明书其他部分，例如图4及其相关描述。
步骤230，从客户端接收至少一幅图像。步骤230可以由服务器的图像接收模块执行。
在一些实施例中，图像接收模块可以通过网络从客户端接受至少一幅图像。或者，客户端可以将所述至少一幅图像发送至存储设备中进行存储，所述图像接收模块可以从所述存储设备中获取所述至少一幅图像。所述至少一幅图像中可能不包含或包含目标物。所述至少一幅图像可以是客户端的图像采集元件 拍摄采集，也可以是基于用户上传的数据(例如，视频或图像)确定。
在一些实施例中，当客户端未被劫持时，所述至少一幅图像可以是客户端的图像采集元件拍摄的真实图像。例如，客户端会基于拍摄指令显示拍摄框和/或对比模板图像，引导用户拍摄面部。用户在客户端拍摄框和/或对比模板图像的引导下调整面部位置，使得面部位于拍摄框内，并按下拍摄键拍摄目标物图像。
当所述客户端被劫持时，劫持者可以通过客户端设备上传图像或视频。所述上传的图像或视频可以包含或者不包含目标物。所述上传的图像或视频可以是由所述客户端或者其他客户端拍摄的历史图像或视频，或者是合成的图像或视频。所述客户端或其他计算设备(例如服务器110)可以基于所述上传的图像或视频确定所述至少一幅图像。例如，被劫持的客户端可以从所述上传的图像或视频中抽取至少一幅图像。此时，所述至少一幅图像是被劫持者上传的虚假图像，而非所述用户基于客户端显示拍摄框和/或对比模板时拍摄的真实图像。可以理解，当所述客户端被劫持时，图像中的目标物通常会有至少一部分位于拍摄框外。例如，目标物为用户面部，当客户端被劫持或攻击时，图像接收模块接收的图像中目标物没有完全位于拍摄框内。
在一些实施例中，图像接收模块可以对至少一幅图像进行预处理。例如，所述预处理可以包括目标检测、质量分析、图像分割、图像降噪、图像转化等中的一个或多个操作。在一些实施例中，所述预处理可以包括目标检测、质量分析和图像分割中的至少一个。
目标检测用于确定图像是否包含目标物。例如，目标物为用户面部，目标检测可以对图像进行识别，若识别出图像内存在用户面部，则图像内包含有目标物；若图像内不存在用户面部，则图像内不包含有目标物。在一些实施例中，图像接收模块可以基于目标检测的结果，从至少一幅图像中剔除不包含目标物的图像。例如，用户可能会因为误操作拍摄不含目标物的图像，剔除这些图像可以减少后续真实性分析的计算量和计算时间、提高分析效率。在一些实施例中，目标检测可以基于目标检测算法执行。
在一些实施例中，所述目标检测可以基于目标检测模型实现。目标检测模型可以包括但不限于视觉几何群网络(Visual Geometry Group Network)模型、Inception NET模型、全卷积神经网络(Fully Convolutional Network)模型、分割网络模型和掩模-卷积神经网络(Mask-Region Convolutional Neural Network)模型等。在一些实施例中，图像接收模块可以使用多个带有标签的图像为训练数据，基于机器学习算法(例如，梯度下降算法)训练目标检测模型时。或者，目标检测模型可以在另外的设备或模块中被训练。
质量分析用于确定图像是否满足质量要求。例如，图像的质量可以用噪声比率、亮度、分辨率、对比度、清晰度等中的一个或多个图像参数来衡量。所述质量分析可以对图像的一个或多个图像参数进行分析，从而判断图像的质量是否满足要求。例如，若图像的分辨率大于1024 x 768，则图像满足质量要求，若图像的分辨率小于1024 x 768，则图像不满足质量要求。在一些实施例中，图像接收模块可以基于质量分析的结果，从至少一幅图像中剔除不满足质量要求的图像，以减少后续真实性分析的计算量、提高分析效率。
在一些实施例中，所述质量分析可以基于质量分析模型实现。例如，质量分析模型可以接收输入的图像，输出表征图像质量的值或者图像质量是否满足质量要求的确定结果。在一些实施例中，质量分析模型可以是但不限于卷积神经网络模型、循环神经网络模型和长短期记忆网络模型中的一种或多种的组合。
所述图像分割可以用于从图像中分割出拍摄框内的区域(称为拍摄框对应的分割图像)。在一些实施例中，图像接收模块可以基于至少一个拍摄参数，从图像中分割出分割图像。图像分割可以减少拍摄框外图像对确定真实性分析的干扰，提高目标识别的准确度。另一方面，分割出拍摄框对应的分割图像可以减少后续真实性分析的计算量，提高计算效率。
在一些实施例中，上述目标检测、质量分析和图像分割中的任意一种或多种可以以任意顺序或同时进行。示例性地，图像接收模块可以先对图像进行目标检测，筛选出包含目标物的图像后，再对包含目标物的图像进行质量分析；也可 以先对图像进行质量分析，筛选出满足质量要求的图像后，再对满足质量要求的图像进行目标检测。
步骤240，基于至少一个拍摄参数，确定至少一幅图像的真实性。步骤240可以由真实性确定模块执行。
在一些实施例中，至少一幅图像的真实性包括每幅图像的第一真实性和/或第二真实性。
所述第一真实性可以反映图像是否是为客户端基于拍摄指令拍摄的图像。例如，当终端未被劫持或攻击时，客户端会显示拍摄框。用户基于显示的拍摄框，移动目标物的位置，使目标物位于拍摄框内，并进行图像拍摄。此时，所述图像具有第一真实性。又例如，当终端被劫持或攻击时，所述图像是基于攻击者上传的图像或视频获取。此时，所述图像不具有第一真实性。在一些实施例中，图像的第一真实性可以用于确定客户端的摄像头是否被攻击者劫持。例如，多幅图像中若存在至少一幅不具有第一真实性，则说明客户端的图像采集元件被劫持。又例如，多幅图像中若超过预设数量的图像不具有第一真实性，则说明客户端的图像采集元件被劫持。
所述第二真实性可以反映图像是否为所述至少一个合格目标物中的一个的图像。例如，如果图像是合格目标物的图像，则所述图像具有第二真实性，否则不具有第二真实性。
在一些实施例中，图像真实性确定模块可以基于预处理后的至少一幅图像，确定所述至少一幅图像的真实性。例如，图像真实性确定模块可以确定预处理后的至少一幅图像的第一和/或第二真实性，以作为至少一幅图像的第一和/或第二真实性。在一些实施例中，可以对至少一幅图像中的每一幅图像进行图像分割处理，以生成拍摄框对应的分割图像。图像真实性确定模块可以基于至少一幅分割图像，确定所述至少一幅图像的真实性。
关于确定图像真实性的详细描述可以参见图5，在此不再赘述。
本说明书的一些实施例通过客户端显示的拍摄框引导用户将目标物的图 像位于拍摄框内，并从客户端获取图像。进一步地，可以基于图像中拍摄框的拍摄参数对图像的真实性进行判断，能够有效判断客户端是否被劫持，从而确保图像的真实性。此外，在一些实施例中，可以基于参考信息确定难度系数，再基于难度系数确定拍摄框的拍摄参数。例如，对于作弊次数较多的用户可以设置较高的难度系数对应的拍摄参数，由此可以提高该用户作弊的难度。通过针对不同的用户设置不同的拍摄参数，可以提高真实性判断的准确性，并提高目标识别的适用性和灵活性。
图3是根据本说明书一些实施例所示的向客户端发送拍摄指令的流程图。
如步骤220所述，指令发送模块可以给客户端发送包括对比模板图像的拍摄指令，用于指示客户端在拍摄框内显示对比模板图像。在一些实施例中，所述拍摄指令可以用图3所示的流程300生成与发送。流程300可以包括下述步骤。
步骤310，获取目标物的模板图像。在一些实施例中，步骤310可以由服务器的指令发送模块执行。
模板图像是基于标准拍摄参数生成的目标物图像。在一些实施例中，可以通过访问存储设备获取模板图像，也可以通过外部输入、调用相关接口或其他方式获取模板图像。或者，模板图像可以由指令发送模块生成。例如指令发送模块可以基于目标物的标准图像集合，确定目标物的至少一个关键点的位置信息。目标物的标准图像集合是包含目标物的多个标准图像的集合。其中，目标物的标准图像是符合标准条件的图像。示例性地，标准条件可以包括目标物正对图像采集元件、目标物图像尺寸为50mm×50mm、目标物和图像采集元件距离为0.4m。例如，标准图像集合可以包括符合标准条件的多张目标物图像。
目标物的关键点可以包括目标物的有代表性的部位。例如，关键点可以是标准图像中的眼睛、鼻子和嘴等。在一些实施例中，关键点可以包括标准图像中的左眼中心、右眼中心、鼻子中心、左嘴角、右嘴角、嘴中心等中的一个或多个。在一些实施例中，关键点也可以是标准图像中的任意位置。例如，关键点可以是 标准图像的中心位置。关键点的位置信息可以表征其在多张标准图像中的位置。例如，关键点的位置信息可以是关键点在标准图像中的平均位置坐标。以左眼中心为例，指令发送模块可以确定左眼中心在每幅标准图像中的坐标，并确定左眼中心在多幅标准图像中的平均坐标，以作为左眼中心的位置信息。
标准拍摄参数是用于生成目标物的模板图像的参数。例如，标准拍摄参数可以包括标准拍摄角度、标准拍摄距离、标准拍摄中心点等中的一个或多个。标准拍摄角度是指拍摄角度的标准值。例如，标准拍摄角度可以是拍摄角度为0°。标准拍摄距离是指拍摄距离的标准值。例如，标准拍摄距离可以是拍摄距离为0.1m。标准拍摄中心点是拍摄中心点的标准位置点。例如，标准拍摄中心点可以标准图像的中心的位置点等。
进一步地，指令发送模块可以基于至少一个标准拍摄参数和至少一个关键点的位置信息，生成目标物的模板图像。例如，指令发送模块可以基于目标物关键点的位置信息，生成符合标准拍摄参数的模拟目标物图像，以作为目标物模板图像。又例如，指令发送模块可以将某一标准目标物图像根据至少一个关键点的位置信息和标准拍摄参数进行调整，以生成目标物模板图像。仅作为示例，标准图像中的左眼中心、右眼中心等关键点可以按照其对应的位置信息进行位置调整；标准图像中的目标物朝向、大小和位置可以分别基于标准拍摄角度、标准拍摄距离和标准拍摄中心点进行调整。在一些实施例中，指令发送模块可以直接获取符合标准拍摄参数的目标物的标准图像集合，并基于所述标准图像集合确定模板图像。例如，指令发送模块可以从标准图像集合中任意选择一张标准图像作为模板图像。又例如，指令发送模块可以基于标准图像集合确定至少一个关键点的位置信息，并基于关键点的位置信息和标准图像集合(或其一部分)确定模板图像。
步骤320，基于至少一个拍摄参数，对模板图像进行调整，以生成对比模板图像。在一些实施例中，步骤320可以由服务器的指令发送模块执行。
仅作为实例，如图4所示，假设模板图像符合标准拍摄参数的目标物图 像。所述标准拍摄参数包括标准拍摄角度0°、标准拍摄距离为0.1m及标准拍摄中心点(即位于客户端屏幕的中心点)。如果拍摄参数包括拍摄角度为30°、拍摄距离0.3m、拍摄中心点位置为屏幕左上角(x1,y1)，指令发送模块可以先基于标准拍摄距离0.1m和拍摄距离0.3m的比值，缩小模板图像410，获取第一对比模板图像420；然后将第一对比模板图像420旋转拍摄角度30°获取第二对比模板图像430；再将第二对比模板图像430的中心位置移动到标准拍摄中心点，获取对比模板图像440。应该可以理解，上述对模板图像的旋转、缩放和移动可以以任意顺序或同时进行，在此不作限制。
在一些实施例中，指令发送模块可以基于拍摄角度、拍摄距离、拍摄中心点和/或其他拍摄参数中的任意一个或多个生成对比模板图像。例如，指令发送模块先基于拍摄中心点对模板图像进行调整，再对调整后的模板图像基于拍摄角度再次进行调整。
步骤330，向客户端发送拍摄指令，所述拍摄指令指示客户端在拍摄框内显示对比模板图像。在一些实施例中，步骤330可以由服务器的指令发送模块执行。
在一些实施例中，指令发送模块可以向客户端发送模板图像和拍摄参数。客户端可以根据拍摄参数对模板图像进行调整，以生成对比模板图像。客户端可以进一步显示拍摄框以及拍摄框内的对比模板图像。
图5是根据本说明书一些实施例所示的确定图像真实性的示例性流程图。在一些实施例中，图5所示的流程500可以由服务器的真实性确定模块执行。
步骤510，基于所述至少一个拍摄参数，确定所述至少一幅图像的第一真实性。
如前所述，图像的第一真实性可以表征该图像是否是为客户端基于拍摄指令拍摄的图像。在一些实施例中，当图像的拍摄框中包含目标物时，可以认为该图像是客户端在显示拍摄框的过程中、用户将目标物调整至拍摄框内后拍摄的图像，即所述至少一幅图像具有第一真实性。
在一些实施例中，真实性确定模块可以根据至少一个拍摄参数从图像中确定拍摄框对应的分割图像。真实性确定模块可以进一步检测分割图像中是否包含目标物的至少一部分(如有代表性的部位或者轮廓)，从而确定图像的拍摄框中是否包含目标物。例如，目标物为用户面部，若检测到分割图像中包含用户的五官或面部轮廓，则说明所述至少一幅图像的拍摄框中包含用户面部，从而可以确定所述图像具有第一真实性。
步骤520，基于所述至少一个拍摄参数和至少一个合格目标物的至少一张参考图像，确定所述至少一幅图像的第二真实性。
当图像中包含的目标物为合格目标物，则说明该图像具有第二真实性。合格目标物是预先通过验证的目标物。例如，在网约车应用场景中，合格目标物可以是司机注册过程中，通过网约车平台审核的司机面部。又例如，在支付应用场景中，合格目标物可以是支付平台验证过支付权限的支付人员瞳孔。
参考图像是包含合格目标物的图像。在一些实施例中，参考图像可以预先存储在存储设备中，真实性确定模块可以基于通过网络从存储设备调取参考图像。在一些实施例中，真实性确定模块可以基于存储设备中的每一幅参考图像，确定图像的第二真实性。例如，在网约车应用场景中，真实性确定模块可以基于存储设备中所有司机的参考图像，确定从客户端接收的图像的第二真实性。或者，真实性确定模块可以基于客户端的标识信息调取与该客户端对应的用户的参考图像，并基于该用户的参考图像确定图像的第二真实性。示例性地，真实性确定模块可以基于客户端的标识从存储设备调取与该客户端绑定的司机的参考图像，用于确认从客户端接收的图像的第二真实性。
在一些实施例中，对至少一幅图像中的每一幅，真实性确定模块可以基于至少一个拍摄参数，生成图像对应的第一图像和与至少一幅参考图像中的每一幅对应的第二图像。第一图像和第二图像对应的拍摄参数相同或相似。客户端用户在拍摄图像时，为了将目标物放置在拍摄框中，可能会对头部进行旋转、位置移动等。此时拍摄的图像对应拍摄框的拍摄参数。而参考图像通常是在预设的拍 摄参数下拍摄的。例如，网约车平台上的司机的参考图像可能是预设参数下拍摄的图像。为了避免拍摄参数不同影响真实性判断结果，需要对至少一幅图像和至少一幅图像进行一致化或标准化处理，使其对应相同或相似的拍摄参数。
在一些实施例中，对于至少一幅图像中的每一幅，真实性确定模块可以将该图像或者该图像中分割框对应的分割图像作为一幅第一图像。真实性确定模块可以将至少一幅参考图像中的每一幅基于至少一个拍摄参数(或其一部分)进行调整，获取至少一幅第二图像。例如，若拍摄参数包括拍摄角度为15°，真实性确定模块可以基于该拍摄角度对参考图像进行调整，使参考图像与屏幕的长度方向之间的夹角为15°。又例如，若拍摄参数包括位置坐标为(25，25)的拍摄中心点，真实性确定模块可以基于将参考图像的中心的位置点移动到坐标(25，25)。还例如，若参考图像的拍摄距离为0.1m而拍摄参数包括拍摄距离为0.5m，真实性确定模块可以基于将参考图像缩小5倍。真实性确定模块可以将调整后的参考图像作为其对应的第二图像。
在一些实施例中，真实性确定模块可以将至少一幅参考图像中的每一幅作为第二图像。对至少一幅图像中的每一幅，真实性确定模块可以基于至少一个拍摄参数(或其一部分)对图像或该图像对应的分割图像进行调整，以生成第一图像。例如，若拍摄参数包括拍摄角度为15°而参考图像的拍摄角度为0°，真实性确定模块可以将图像旋转-15°，使调整后的图像的拍摄角度与参考图像的拍摄角度相同。真实性确定模块可以将调整后的图像作为其对应的第一图像。在一些实施例中，对于至少一幅图像中的每一幅，真实性确定模块会先确定拍摄框对应的分割图像，再对分割图像进行调整，以生成第一图像。在一些实施例中，真实性确定模块可以对至少一幅图像和至少一幅参考图像分别进行调整，使第一图像和第二图像均对应标准拍摄参数或其他相同的拍摄参数。
进一步地，对至少一幅图像中的每一幅，真实性确定模块可以基于其对应的第一图像和至少一幅第二图像的相似度，确定该图像的第二真实性。例如，真实性确定模块可以确定第一图像的第一图像特征和每幅第二图像的第二图像特 征之间的相似度，以确定图像的真实性。图像的图像特征可以包括颜色特征、纹理特征、形状特征、深度特征等，或其任意组合。第一图像的第一图像特征和第二图像的第二图像特征之间的相似度可以用向量相似度计算得到，例如，通过欧式距离、曼哈顿距离、余弦相似度等确定。若第一图像的第一图像特征和某一第二图像的第二图像特征之间的相似度超过特定阈值，可以认为该第一图像和第二图像是同一物体的图像。也就是说，第一图像对应的图像是某一合格目标物的图像，即第一图像对应的图像具有第二真实性。若第一图像的第一图像特征和所有第二图像的第二图像特征之间的相似度不超过特定阈值，可以认为第一图像对应的图像不具有第二真实性。
在一些实施例中，真实性确定模块可以基于图像对比模型处理第一图像和至少一幅第二图像中的每一幅，确定图像的第二真实性。例如，真实性确定模块可以将第一图像和第二图像中的一幅输入图像对比模型，图像对比模型可以输出第一图像与第二图像之间的相似度和/或第一图像与第二图像是否相似的判断结果。出于说明目的，图6示出了图像对比模型的示例性结构图。图所示，图像对比模型600可以包括特征提取层605、相似度计算层608和判别层610。
如图6所示，图像601可以至少包括图像601-1、图像601-m，根据图像601生成的第一图像603可以至少包括第一图像603-1、第一图像603-m，图像601对应的参考图像602可以至少包括参考图像602-1、参考图像602-n，基于参考图像602生成第二图像604可以至少包括第二图像604-1、第二图像604-n。在一些实施例中，一幅第一图像和至少一幅第二图像可以组成一个图像对。图像对比模型600可以该图像对进行分析，以确定该图像对中第一图像和第二图像是否相似。例如，如图6所示，第一图像603-m和第二图像604-n构成的图像对可以被输入图像对比模型600。
特征提取层605可以用于处理第一图像和第二图像，获取第一图像603-m的第一图像特征606和第二图像604-n的第二图像特征607。在一些实施例中，特征提取层605的类型可以包括ResNet、ResNeXt、SE-Net、DenseNet、 MobileNet、ShuffleNet、RegNet、EfficientNet或Inception等卷积神经网络模型，或循环神经网络模型。在一些实施例中，可以将第一图像603-m和第二图像604-n拼接后输入特征提取层605。特征提取层605的输出可以是第一图像603-m的第一图像特征606和第二图像604-n的第二图像特征607拼接后的特征向量。
相似度计算层608可以用于确定第一图像特征606和第二图像特征607的相似度609。判别层610可以用于基于相似度609，输出第一图像603-m和第二图像604-n是否相似的判断结果。例如，判别层610可以比较相似度609和相似度阈值。如果第一图像603-m的第一图像特征606和第二图像604-n的第二图像特征607之间的相似度超过相似度阈值，则第一图像603-m和第二图像604-n相似。在一些实施例中，可以基于图像601-m对应的第一图像603-m和每一张参考图像602对应的每一张第二图像604是否相似的判断结果611，确定第一图像603-m对应图像601-m的第二真实性。例如，第一图像603-m和每一张第二图像604都不相似，则第一图像不具有第二真实性。又例如，第一图像603-m和第二图像604-1相似，则第一图像具有第二真实性。
在一些实施例中，真实性确定模块可以将第一图像和第二图像组成的多对图像对一起输入图像对比模型600。图像对比模型600可以同时输出所述多对图像对中每一对的相似性判别结果。在一些实施例中，图像对比模型600为预置参数的机器学习模型。预置参数是指机器学习模型训练过程中，学习到的模型参数。以神经网络为例，模型参数包括权重(Weight)和偏置(bias)等。图像对比模型600的所述预置参数通过训练过程生成。例如，模型获取模块可以基于带有标签的多个训练样本训练初始图像对比模型，以得到图像对比模型。
训练样本包括带有标签的一个或多个样本图像对。每个样本图像对包括第一样本图像和第二样本图像。其中，第一样本图像和第二样本图像可以是相同或者不同对象的图像。训练样本的标签可以说明第一样本图像和第二样本图像是否相似(或是否为同一对象的图片)。
在一些实施例中，图像对比模型可以由处理设备或第三方预先训练后保 存在存储设备中，处理设备可以从存储设备中直接调用图像对比模型。
在一些实施例中，真实性确定模块可以只基于第一真实性确定图像的真实性。例如，如果至少一幅图像具有第一真实性，则认为至少一幅图像具有真实性，通过目标识别。在一些实施例中，真实性确定模块可以基于第一真实性和第二真实性确定图像的真实性。例如，可以确定先确定至少一幅图像具有第一真实性，再确定至少一幅图像中目标物是否为合格目标物。如果该图像具有第二真实性，则认为至少一幅图像具有真实性，通过目标识别。第一真实性分析基于图像的简单特征进行，与第二真实性分析相比更加简单、所需计算资源更小。通过只进行第一真实性分析，或者判断图像具有第一真实性后再进行第二真实性分析，可以提高目标识别的效率，简化目标识别的步骤，减少计算资源的浪费(例如，避免使用计算资源对劫持者上传的虚假图片进行第二真实性分析)。
在一些实施例中，真实性确定模块可以直接基于第二真实性确定图像的真实性。例如，如果至少一幅图像具有第一真实性，则认为至少一幅图像具有真实性，通过目标识别。
在一些实施例中，真实性确定模块可以基于参考信息，选择不同的方式进行目标识别。例如，若用户的历史作弊次数大于特定阈值，则真实性确定模块可以选择先判断第一真实性，再判断第二真实性的目标识别方式。又例如，用户的历史作弊次数为0次，则真实性确定模块可以选择直接判断第二真实性的目标识别方式。
图7是根据本说明书一些实施例所示的应用于客户端的目标识别方法的示例性流程图。如图7所示，流程700包括下述步骤。
步骤710，从服务器接收拍摄指令。在一些实施例中，步骤710可以由客户端的指令接收模块执行。
如前所述，拍摄指令是指让客户端显示拍摄框的指令。在一些实施例中，拍摄指令可以包括拍摄框有关的拍摄参数，例如，拍摄框的形状、大小、位置和显示参数等。在一些实施例中，拍摄指令还可以包括目标物的对比模板图像，用 于指示客户端在拍摄框内显示对比模板图像。在一些实施例中，拍摄指令包括目标物的模板图像。客户端可以基于从模板图像和拍摄参数，生成对比模板图像。关于拍摄指令的具体描述可以参见本申请其他部分的描述，例如，步骤220。
步骤720，基于至少一个拍摄参数，显示拍摄框。在一些实施例中，步骤720可以由客户端的拍摄框显示模块执行。
如前所述，拍摄框是指在客户端(例如，客户端130)屏幕上显示的特定区域，可以引导客户端的用户在进行拍摄时将目标物放置于所述特定区域中。
在一些实施例中，拍摄参数可以包括与拍摄框的形状、大小、位置、显示方式等相关的任何参数。拍摄框显示模块可以基于至少一个拍摄参数生成拍摄框，并指示客户端在屏幕显示区域内显示拍摄框。例如，拍摄框显示参数可以基于形状、大小、位置参数生成拍摄框，并基于显示参数指示客户端以特定的方式(例如，以特定的颜色、闪烁频率)显示拍摄框。在一些实施例中，客户端可以获取预设拍摄框，该预设拍摄框具有特定的形状、大小、位置等。拍摄框显示模块可以基于拍摄参数对预设拍摄框进行旋转、缩放、平移等以生成拍摄框。基于拍摄参数对拍摄框进行调整的过程与图3和4所述的基于拍摄参数对模板图像进行调整的过程类似，在此不再赘述。
在一些实施例中，拍摄框显示模块可以指示客户端在拍摄框内显示对比模板图像，类似图4所示。当客户端显示对比模板图像时，用户可以将目标物轮廓对准对比模板图像轮廓，提高模板识别的准确率。例如，目标物为用户面部，用户可以将面部轮廓对准目标物图像模板(即，对比模板图像)。附加的或可选的，拍摄框显示模块可以指示客户端在拍摄框或对比模板图像中显示至少目标物的一个关键点。以用户面部为例，可以进一步在目标物图像模板(即，对比模板图像)中显示面部的左眼、右眼、鼻尖、左嘴角和右嘴角，用户可以将左眼、右眼、鼻尖、左嘴角和右嘴角分别对准目标物图像模板中的对应关键点。
在一些实施例中，拍摄框显示模块可以直接获取对比模板图像，将其显示在客户端的屏幕上。此时，对比模板图像的边缘可以视为拍摄框。
步骤730，基于图像采集元件获取至少一幅拍摄图像。在一些实施例中，步骤730可以由图像获取模块执行。
拍摄图像是客户端的图像采集元件获取的图像。拍摄图像中可以包含或不包含目标物。在一些实施例中，图像获取模块可以基于图像采集元件拍摄的视频，获取至少一幅拍摄图像。具体的，图像获取模块可以从视频中抽取至少一帧图像作为至少一幅拍摄图像。例如，图像获取模块从图像采集元件拍摄的视频中随机抽取n帧作为拍摄图像。又例如，图像获取模块可以先识别出拍摄框内包含目标物的视频帧，将抽取包含目标物的视频n帧作为拍摄图像。
在一些实施例中，图像获取模块可以基于确认指令指示图像采集元件获取拍摄图像。确认指令是用户通过确定操作触发的拍摄指令。例如，确认操作可以为手动输入操作、语音输入操作等。具体的，当用户将目标物调整至拍摄框内后，可以通过确定操作触发确认指令。图像获取模块接收确认指令后，图像采集元件可以拍摄图像。在一些实施例中，图像获取模块可以自动指示图像采集元件拍摄图像。例如，当检测到拍摄框中存在目标物时，图像获取模块可以自动指示图像采集元件拍摄图像。
步骤740，将所述至少一幅拍摄图像发送给所述服务器以对所述至少一幅拍摄图像的真实性进行判别。在一些实施例中，步骤740可以由图像发送模块执行。
在一些实施例中，图像发送模块可以通过网络将图像采集元件获取的至少一幅拍摄图像作为图像发送给服务器，以对至少一幅拍摄图像的真实性进行判别(例如，第二真实性判别)。在一些实施例中，若客户端被劫持，在接收拍摄指令后，劫持者可能通过客户端设备上传图像或视频。此时，步骤730可以省略。客户端可以将劫持者上传的图像或视频发送给服务器，服务器可以对所述图像或视频进行第一真实性和/或第二真实性判断。在一些实施例中，图像发送模块会对拍摄图像进行预处理，并将预处理后的拍摄图像发送给服务器进行进一步分析。对拍摄图像的预处理与服务器的图像接收模块对图像的预处理类似，在 此不再赘述。关于服务器对所述至少一幅拍摄图像进行真实性判别的详细描述参见图5及其相关描述，在此不再赘述。
在一些实施例中，客户端还通过网络可以从服务器接收拍摄图像的真实性判别结果。可选地，客户端可以基于真实性判别结果，显示引导信息。引导信息是基于真实性判别结果，提示用户下一步操作的信息。引导信息可以包括语音信息、文字信息和图像信息等。例如，在网约车应用场景中，真实性判别结果反映目标通过识别时，即平台验证接单司机甲为平台注册的司机用户时，引导信息可以为语音信息“审核通过，请开始驾驶”。又例如，在支付应用场景中，真实性判别结果反映目标未通过识别时，引导信息可以为显示在客户端的屏幕显示区域的文字信息“审核未通过，请再次识别”。
在一些实施例中，客户端可以进一步基于参考信息确定引导信息，其中，参考信息可以包括拍摄环境信息、用户的历史行为和用户的个人信息等。例如，在网约车应用场景中，若真实性判别结果反映目标未通过识别时，客户端可以基于拍摄环境信息“光照强度＜10lux”确定引导信息为“审核未通过，请前往更加明亮的环境再次识别”。又例如，在支付的应用场景中，当真实性判别结果反映目标未通过识别时，客户端可以基于用户的历史行为“用户欺诈导致历史目标识别失败次数＞10次”确定引导信息为“审核未通过，请本人进行识别”。本说明书的一些实施例基于参考信息确定引导信息，可以针对用户不同的目标识别意图和不同的操作行为，进行引导、提示或警告，提高引导信息的针对性和有效性，从而提高目标识别的准确性。
本说明书实施例可能带来的有益效果包括但不限于：(1)通过客户端显示的拍摄框引导用户将目标物调整至拍摄框内获取图像，并基于拍摄框的拍摄参数对图像的真实性进行判断，能够有效判断客户端是否被劫持和/或所述图像是否为合格目标物的图像；(2)基于参考信息确定难度系数，再基于难度系数确定拍摄框的拍摄参数，从而通过针对不同的场景设置不同的拍摄参数，可以提高真实性判断的准确性，并提高目标识别的适用性和灵活性；(3)第一真实性 分析基于图像的简单特征进行，与第二真实性分析相比更加简单、节省计算资源。通过只进行第一真实性分析，或者判断图像具有第一真实性后再进行第二真实性分析，可以提高目标识别的效率，简化目标识别的步骤，减少计算资源的浪费(例如，避免使用计算资源对劫持者上传的虚假图片进行第二真实性分析)；(4)基于参考信息，选择不同的方式进行目标识别，可以提高目标识别的适应性和效率。
上文已对基本概念做了描述，显然，对于本领域技术人员来说，上述详细披露仅仅作为示例，而并不构成对本说明书的限定。虽然此处并没有明确说明，本领域技术人员可能会对本说明书进行各种修改、改进和修正。该类修改、改进和修正在本说明书中被建议，所以该类修改、改进、修正仍属于本说明书示范实施例的精神和范围。
同时，本说明书使用了特定词语来描述本说明书的实施例。如“一个实施例”、“一实施例”、和/或“一些实施例”意指与本说明书至少一个实施例相关的某一特征、结构或特点。因此，应强调并注意的是，本说明书中在不同位置两次或多次提及的“一实施例”或“一个实施例”或“一个替代性实施例”并不一定是指同一实施例。此外，本说明书的一个或多个实施例中的某些特征、结构或特点可以进行适当的组合。
此外，除非权利要求中明确说明，本说明书所述处理元素和序列的顺序、数字字母的使用、或其他名称的使用，并非用于限定本说明书流程和方法的顺序。尽管上述披露中通过各种示例讨论了一些目前认为有用的发明实施例，但应当理解的是，该类细节仅起到说明的目的，附加的权利要求并不仅限于披露的实施例，相反，权利要求旨在覆盖所有符合本说明书实施例实质和范围的修正和等价组合。例如，虽然以上所描述的系统组件可以通过硬件设备实现，但是也可以只通过软件的解决方案得以实现，如在现有的服务器或移动设备上安装所描述的系统。
同理，应当注意的是，为了简化本说明书披露的表述，从而帮助对一个或 多个发明实施例的理解，前文对本说明书实施例的描述中，有时会将多种特征归并至一个实施例、附图或对其的描述中。但是，这种披露方法并不意味着本说明书对象所需要的特征比权利要求中提及的特征多。实际上，实施例的特征要少于上述披露的单个实施例的全部特征。
一些实施例中使用了描述成分、属性数量的数字，应当理解的是，此类用于实施例描述的数字，在一些示例中使用了修饰词“大约”、“近似”或“大体上”来修饰。除非另外说明，“大约”、“近似”或“大体上”表明所述数字允许有±20％的变化。相应地，在一些实施例中，说明书和权利要求中使用的数值参数均为近似值，该近似值根据个别实施例所需特点可以发生改变。在一些实施例中，数值参数应考虑规定的有效数位并采用一般位数保留的方法。尽管本说明书一些实施例中用于确认其范围广度的数值域和参数为近似值，在具体实施例中，此类数值的设定在可行范围内尽可能精确。
针对本说明书引用的每个专利、专利申请、专利申请公开物和其他材料，如文章、书籍、说明书、出版物、文档等，特此将其全部内容并入本说明书作为参考。与本说明书内容不一致或产生冲突的申请历史文件除外，对本说明书权利要求最广范围有限制的文件(当前或之后附加于本说明书中的)也除外。需要说明的是，如果本说明书附属材料中的描述、定义、和/或术语的使用与本说明书所述内容有不一致或冲突的地方，以本说明书的描述、定义和/或术语的使用为准。
最后，应当理解的是，本说明书中所述实施例仅用以说明本说明书实施例的原则。其他的变形也可能属于本说明书的范围。因此，作为示例而非限制，本说明书实施例的替代配置可视为与本说明书的教导一致。相应地，本说明书的实施例不仅限于本说明书明确介绍和描述的实施例。

Claims (10)

  1. 一种目标识别的方法，其特征在于，包括：
    获取与拍摄框有关的至少一个拍摄参数；
    向客户端发送拍摄指令，所述拍摄指令指示所述客户端基于所述至少一个拍摄参数显示所述拍摄框；
    从所述客户端接收至少一幅图像；以及
    基于所述至少一个拍摄参数，确定所述至少一幅图像的真实性。
  2. 如权利要求1所述方法，其特征在于，进一步包括：
    获取目标物的模板图像；以及
    基于所述至少一个拍摄参数，对所述模板图像进行调整，以生成对比模板图像，其中所述拍摄指令进一步指示所述客户端在所述拍摄框内显示所述对比模板图像。
  3. 如权利要求1所述方法，所述获取与拍摄框有关的至少一个拍摄参数包括：
    随机生成所述至少一个拍摄参数；或者
    基于参考信息确定所述至少一个拍摄参数，所述基于参考信息确定至少一个拍摄参数包括基于参考信息，确定拍摄难度系数；以及基于所述拍摄难度系数，确定所述至少一个拍摄参数。
  4. 如权利要求1所述的方法，基于所述至少一个拍摄参数，确定所述至少一幅图像的真实性包括：
    对所述至少一幅图像进行预处理，生成预处理后的至少一幅图像；以及
    基于所述至少一个拍摄参数和预处理后的至少一幅图像，确定所述至少一幅图像的真实性。
  5. 如权利要求4所述方法，对所述至少一幅图像进行预处理包括对所述至 少一幅图像中的每一幅执行以下操作中的至少一个：
    对所述图像进行目标检测，确定所述图像是否包含目标物；
    对所述图像进行质量分析，确定所述图像是否满足质量要求；或者
    对所述图像进行图像分割，生成所述拍摄框对应的分割图像。
  6. 如权利要求1所述方法，所述基于所述至少一个拍摄参数，确定所述至少一幅图像的真实性，包括以下操作中的至少一个：
    基于所述至少一个拍摄参数，确定所述至少一幅图像的第一真实性，所述第一真实性反映所述至少一幅图像是否为所述客户端基于拍摄指令拍摄的图像；
    基于所述至少一个拍摄参数和至少一个合格目标物的至少一张参考图像，确定所述至少一幅图像的第二真实性，所述第二真实性反映所述至少一幅图像是否为所述至少一个合格目标物中的一个的图像。
  7. 一种目标识别的方法，其特征在于，包括：
    从服务器接收拍摄指令，所述拍摄指令包括与拍摄框有关的至少一个拍摄参数；
    基于所述至少一个拍摄参数，显示所述拍摄框；
    基于图像采集元件获取至少一幅拍摄图像；以及
    将所述至少一幅拍摄图像发送给所述服务器以对所述至少一幅拍摄图像的真实性进行判别。
  8. 如权利要求7所述的方法，其特征在于，所述拍摄指令进一步包括目标物的模板图像，所述方法进一步包括：
    在所述拍摄框内显示所述目标物的对比模板图像；
    从所述服务器接收所述拍摄图像的真实性判别结果；以及
    基于所述真实性判别结果，显示引导信息。
  9. 一种目标识别的系统，其特征在于，包括：
    参数获取模块，用于获取与拍摄框有关的至少一个拍摄参数；
    指令发送模块，用于向客户端发送拍摄指令，所述拍摄指令用于指示所述客户端基于所述至少一个拍摄参数显示所述拍摄框；
    图像接收模块，用于从所述客户端接收至少一幅图像；以及
    真实性确定模块，用于基于所述至少一个拍摄参数，确定所述至少一幅图像的真实性。
  10. 一种目标识别的系统，由客户端执行，其特征在于，包括：
    指令接收模块，用于从服务器接收拍摄指令，所述拍摄指令包括与拍摄框有关的至少一个拍摄参数；
    拍摄框显示模块，用于基于所述至少一个拍摄参数，显示所述拍摄框；
    图像获取模块，用于基于图像采集元件获取至少一幅拍摄图像；以及
    图像发送模块，用于将所述至少一幅拍摄图像发送给所述服务器以对所述至少一幅拍摄图像的真实性进行判别。
PCT/CN2022/087915 2021-04-20 2022-04-20 一种目标识别的方法和系统 WO2022222957A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110424645.8 2021-04-20
CN202110424645.8A CN113158893A (zh) 2021-04-20 2021-04-20 一种目标识别的方法和系统

Publications (1)

Publication Number Publication Date
WO2022222957A1 true WO2022222957A1 (zh) 2022-10-27

Family

ID=76869050

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/087915 WO2022222957A1 (zh) 2021-04-20 2022-04-20 一种目标识别的方法和系统

Country Status (2)

Country Link
CN (1) CN113158893A (zh)
WO (1) WO2022222957A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113158893A (zh) * 2021-04-20 2021-07-23 北京嘀嘀无限科技发展有限公司 一种目标识别的方法和系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105590043A (zh) * 2014-10-22 2016-05-18 腾讯科技（深圳）有限公司 身份验证方法、装置及系统
DE102018208604A1 (de) * 2018-05-30 2019-12-05 Siemens Aktiengesellschaft Ermitteln eines Aufnahmeverhaltens einer Aufnahmeeinheit
CN111161259A (zh) * 2019-12-31 2020-05-15 支付宝(杭州)信息技术有限公司 检测图像是否被篡改的方法及装置和电子设备
CN111881844A (zh) * 2020-07-30 2020-11-03 北京嘀嘀无限科技发展有限公司 一种判断图像真实性的方法及系统
CN112183270A (zh) * 2020-09-18 2021-01-05 支付宝实验室(新加坡)有限公司 身份验证的拍摄参数的优化方法、装置及电子设备
CN113158893A (zh) * 2021-04-20 2021-07-23 北京嘀嘀无限科技发展有限公司 一种目标识别的方法和系统

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108605085B (zh) * 2016-11-08 2020-12-01 华为技术有限公司 一种获取拍摄参考数据的方法、移动终端
CN107566654B (zh) * 2017-09-27 2020-03-17 Oppo广东移动通信有限公司 解锁控制方法及相关产品
CN110944109B (zh) * 2018-09-21 2022-01-14 华为技术有限公司 一种拍照方法、装置与设备
CN111159749B (zh) * 2019-12-31 2022-08-09 Oppo广东移动通信有限公司 拍照方法、拍照装置、移动终端及计算机可读存储介质
CN111654624B (zh) * 2020-05-29 2021-12-24 维沃移动通信有限公司 拍摄提示方法、装置及电子设备
CN112312021B (zh) * 2020-10-30 2022-04-15 维沃移动通信有限公司 拍摄参数调整方法及装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105590043A (zh) * 2014-10-22 2016-05-18 腾讯科技（深圳）有限公司 身份验证方法、装置及系统
DE102018208604A1 (de) * 2018-05-30 2019-12-05 Siemens Aktiengesellschaft Ermitteln eines Aufnahmeverhaltens einer Aufnahmeeinheit
CN111161259A (zh) * 2019-12-31 2020-05-15 支付宝(杭州)信息技术有限公司 检测图像是否被篡改的方法及装置和电子设备
CN111881844A (zh) * 2020-07-30 2020-11-03 北京嘀嘀无限科技发展有限公司 一种判断图像真实性的方法及系统
CN112183270A (zh) * 2020-09-18 2021-01-05 支付宝实验室(新加坡)有限公司 身份验证的拍摄参数的优化方法、装置及电子设备
CN113158893A (zh) * 2021-04-20 2021-07-23 北京嘀嘀无限科技发展有限公司 一种目标识别的方法和系统

Also Published As

Publication number Publication date
CN113158893A (zh) 2021-07-23

Similar Documents

Publication Publication Date Title
US11288504B2 (en) Iris liveness detection for mobile devices
EP3493088B1 (en) Security gesture authentication
US9082235B2 (en) Using facial data for device authentication or subject identification
JP5076563B2 (ja) 顔照合装置
CN103383723A (zh) 用于生物特征验证的电子欺骗检测的方法和系统
JP2006114018A (ja) セキュリティシステム
CN105141842A (zh) 一种防篡改的证照相机系统及方法
KR102145132B1 (ko) 딥러닝을 이용한 대리 면접 예방 방법
US20230041573A1 (en) Image processing method and apparatus, computer device and storage medium
CN107609364A (zh) 用户身份确认方法和装置
CN114511915B (zh) 一种基于移动客户端的可信任证件照采集系统及方法
WO2022222957A1 (zh) 一种目标识别的方法和系统
WO2022222585A1 (zh) 一种目标识别的方法和系统
WO2023034251A1 (en) Spoof detection based on challenge response analysis
CN113111810B (zh) 一种目标识别方法和系统
CN110837901A (zh) 云试驾预约审核方法及装置、存储介质、云服务器
US20240028698A1 (en) System and method for perfecting and accelerating biometric identification via evolutionary biometrics via continual registration
KR101725219B1 (ko) 디지털 이미지 판단방법 및 시스템, 이를 위한 애플리케이션 시스템, 및 인증 시스템
JP2021131737A (ja) データ登録装置、生体認証装置、およびデータ登録プログラム
KR20230110681A (ko) 응시자 단말의 정면 카메라와 보조 카메라를 사용하여 부정행위를 방지하는 안면윤곽선 인식 인공지능을 사용한 온라인 시험 시스템 및 그 방법
CN110516426A (zh) 身份认证方法、认证终端、装置及可读存储介质
CN111767845B (zh) 证件识别方法及装置
CN108764033A (zh) 身份验证方法和装置、电子设备、计算机程序和存储介质
JP2022100522A (ja) 本人確認方法、プログラム、及び情報システム
KR20210136771A (ko) 안면윤곽선 인식 인공지능을 사용한 ubt 시스템 및 그 방법

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22791063

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 22791063

Country of ref document: EP

Kind code of ref document: A1