Biometric recognition method and system, training method and system of recognition model

Abstract

The biometric feature recognition method and system provided in the specification obtains a broadband light image of a target object and determines an initial recognition result corresponding to the broadband light image. Based on the initial recognition result, a target scheme is determined and executed, the target scheme being one of multiple schemes, the multiple schemes including scheme one: predicting a best monochromatic light corresponding to biometric feature recognition of the target object based on the broadband light image; obtaining a target monochromatic light image of the target object under the best monochromatic light, and determining a target recognition result of the target object based on the broadband light image and the target monochromatic light image. In this way, the best monochromatic light of the target object is determined using the broadband light image, so that accurate biometric feature recognition is completed by combining the broadband light image and the target monochromatic light image under the best monochromatic light, without the need for the user to perform multiple coordinated actions, thereby improving the user experience.

Description

TECHNICAL FIELD

[0001] The present specification relates to the field of data processing, and in particular, to a biometric feature recognition method and system, a biometric feature recognition model training method and system. BACKGROUND

[0002] Biometric feature recognition systems, such as face recognition systems, have been widely applied in recent years and have achieved success in scenarios such as face payment, face station entry, face attendance, etc. In order to ensure the security of face recognition, users are usually required to cooperate with a series of actions, such as facing the front, turning the head to the left, turning the head to the right, etc., to capture the front face, left face, and right face of the user, etc. As an important part of face recognition systems, live body recognition is usually an interactive live body recognition method. This method prompts the user to cooperate with a series of actions such as blinking, opening the mouth, shaking the head, etc. through video and voice, and then detects whether the user is a live body or an attack according to the degree of completion of the user's actions. It can be seen that biometric feature recognition (such as face recognition) requires the user to cooperate with a series of actions, and the time-consuming of completing the actions is relatively long, which will lead to a poor user experience. Therefore, a method is needed to successfully recognize biometric features without the user having to cooperate with multiple actions. SUMMARY

[0003] The biometric feature recognition method and system provided by the present specification determine the best monochromatic light of a target object based on a broad spectrum light image, so as to combine the broad spectrum light image and the target monochromatic light image under the best monochromatic light to complete accurate recognition of the face, without the user having to cooperate with multiple actions, thereby improving the user experience.

[0004] In a first aspect, the present specification provides a biometric feature recognition method, comprising: obtaining a broad spectrum light image of a target object, and determining an initial recognition result corresponding to the broad spectrum light image; and based on the initial recognition result, determining and executing a target scheme to complete recognition of the biometric feature of the target object and output a target recognition result, the target scheme being one of a plurality of schemes, the plurality of schemes including scheme one: predicting a best monochromatic light corresponding to biometric feature recognition of the target object based on the broad spectrum light image, the biometric feature recognition accuracy of the target object under the best monochromatic light being higher than the biometric feature recognition accuracy under a non-best monochromatic light, obtaining a target monochromatic light image of the target object under the best monochromatic light, and determining the target recognition result of the target object based on the broad spectrum light image and the target monochromatic light image.

[0005] In some embodiments, the determining the initial recognition result corresponding to the broad spectrum light image comprises: performing region segmentation on the broad spectrum light image to obtain a plurality of part images of the target object; performing feature extraction on each part image in the plurality of part images respectively to obtain a plurality of part features; fusing the plurality of part features to obtain a first fused feature; and determining the initial recognition result based on the first fused feature.

[0006] In some embodiments, the predicting the best monochromatic light corresponding to the biometric recognition of the target object based on the broad spectrum light image comprises: inputting the plurality of part features and the first fused feature into a monochromatic light prediction model for prediction, and outputting the best monochromatic light.

[0007] In some embodiments, the obtaining the target monochromatic light image of the target object under the best monochromatic light comprises: sending an instruction signal to an image acquisition device, the instruction signal instructing the image acquisition device to take a photo of the target object under the best monochromatic light; and obtaining the target monochromatic light image of the target object.

[0008] In some embodiments, the determining the target recognition result of the target object based on the broad spectrum light image and the target monochromatic light image comprises: performing feature encoding on the broad spectrum light image to obtain a broad spectrum light feature map of the biometric feature; performing feature encoding on the target monochromatic light image to obtain a monochromatic light feature map of the biometric feature; fusing the broad spectrum light feature map and the monochromatic light feature map to obtain a second fused feature; and determining the target recognition result based on the second fused feature.

[0009] In some embodiments, the determining and executing a target scheme based on the initial recognition result comprises: determining that the initial recognition result is a living body; and executing the scheme one.

[0010] In some embodiments, the plurality of schemes at least comprises a scheme two, and the scheme two comprises: the target recognition result is the initial recognition result.

[0011] In some embodiments, the determining and executing a target scheme based on the initial recognition result comprises: determining that the initial recognition result is an attack; and executing the scheme two.

[0012] According to the technical solution, the biological feature recognition method and system provided by the present specification obtains a broadband light image of a target object and determines an initial recognition result corresponding to the broadband light image. Based on the initial recognition result, a target scheme is determined and executed to complete the biological feature recognition of the target object and output a target recognition result. The target scheme is one of multiple schemes, and the multiple schemes include scheme one: predicting a best monochromatic light corresponding to the biological feature recognition of the target object based on the broadband light image, the biological feature recognition accuracy of the target object under the best monochromatic light being higher than the biological feature recognition accuracy under a non-best monochromatic light; obtaining a target monochromatic light image of the target object under the best monochromatic light, and determining the target recognition result of the target object based on the broadband light image and the target monochromatic light image. In this way, the best monochromatic light of the target object is determined by using the broadband light image, so that the biological feature is accurately recognized by combining the broadband light image and the target monochromatic light image under the best monochromatic light, without the need for the user to perform multiple cooperative actions, thereby improving the user experience. Moreover, compared with the biological feature recognition based only on the broadband light image, the present specification combines the monochromatic light image and the broadband light image, so that the biological feature recognition utilizes more information and further improves the recognition accuracy.

[0013] In a second aspect, the present specification also provides a biological feature recognition system, which includes an image acquisition device, at least one storage medium storing at least one set of instructions for performing biological feature recognition, and at least one processor in communication connection with the at least one storage medium and the image acquisition device. When the target biological feature recognition system is running, the at least one processor executes the at least one set of instructions and determines an initial recognition result corresponding to a broadband light image of a target object received from the image acquisition device according to the instructions of the at least one set of instructions. Based on the initial recognition result, a target scheme is determined and executed to complete the target biological feature recognition of the target object and output a target recognition result. The target scheme is one of multiple schemes, and the multiple schemes include scheme one: predicting a best monochromatic light corresponding to the biological feature recognition of the target object based on the broadband light image, the biological feature recognition accuracy of the target object under the best monochromatic light being higher than the biological feature recognition accuracy under a non-best monochromatic light, obtaining a target monochromatic light image of the target object under the target filter from the image acquisition device, and determining the target recognition result of the target object based on the broadband light image and the target monochromatic light image.

[0014] In some embodiments, the image acquisition device comprises an image capturing device and a filter device, the filter device comprising at least one monochromatic filter, and the at least one processor is configured to: send an instruction signal to the image acquisition device; and the image acquisition device is configured to: set a target filter among the at least one monochromatic filter on an imaging light path of the image capturing device, and capture an image of the target object to obtain a target monochromatic light image of the target object under the target filter, the target filter being a monochromatic filter corresponding to the optimal monochromatic light, and send the target monochromatic light image to the at least one processor.

[0015] In some embodiments, different monochromatic filters among the at least one monochromatic filter are capable of transmitting monochromatic light of different colors.

[0016] In some embodiments, the filter device further comprises a filter driving device configured to set one of the at least one monochromatic filter on the imaging light path of the image capturing device in operation, and the at least one processor is configured to: determine the target filter corresponding to the optimal monochromatic light from the at least one monochromatic filter, and send the instruction signal to the filter driving device; and the filter driving device is configured to receive the instruction signal and set the target filter on the imaging light path.

[0017] In some embodiments, an incident surface of the target filter is parallel to an incident surface of a broad-spectrum light lens of the image capturing device.

[0018] The biometric identification system provided in the specification comprises an image acquisition device, at least one storage medium, and at least one processor. The at least one processor receives a broad-spectrum light image of a target object from the image acquisition device, determines an optimal monochromatic light of the target object by using the broad-spectrum light image, and thus completes accurate identification of a biometric feature by combining the broad-spectrum light image and a target monochromatic light image under the optimal monochromatic light. Moreover, the image acquisition device comprises an image capturing device and a filter device, the filter device comprising at least one monochromatic filter, and a target filter is set on an imaging light path of the image capturing device to obtain a target monochromatic light image. In this way, by introducing various monochromatic filters, i.e., by arranging one or more different monochromatic filters that can be switched in front of the image capturing device, low-cost modification is realized at the hardware level, and the performance of biometric identification is improved by low-cost hardware modification.

[0019] In a third aspect, the present specification also provides a method for training a biometric recognition model, comprising: obtaining a broad-spectrum light training image and a training sample pair, wherein the broad-spectrum light training image comprises a training subject, the training sample pair comprises a broad-spectrum light training image corresponding to the same training subject Figure One and a monochromatic light training image Figure Two ; training a broad-spectrum light recognition model based on the broad-spectrum light training image, outputting an image feature and a first recognition result; obtaining the image feature output by the trained broad-spectrum light recognition model as an image feature training image, and obtaining a monochromatic light label corresponding to the image feature training image; training a monochromatic light prediction model based on the image feature training image and the monochromatic light label, and outputting a monochromatic light prediction result predicted for the training subject; and training a target recognition model based on the training sample pair, and outputting a second recognition result.

[0020] In some embodiments, the broad-spectrum light recognition model comprises: a segmentation network configured to perform region segmentation on the broad-spectrum light training image, outputting a plurality of part images, each part image corresponding to a part of the training subject; a plurality of part feature extraction networks configured to respectively perform feature extraction and classification on the plurality of part images, outputting a plurality of part features and a plurality of part classification results corresponding thereto; and a first fusion network configured to perform feature fusion on the plurality of part features, outputting a first fusion feature and the first recognition result, wherein the image feature comprises the plurality of part features and the first fusion feature.

[0021] In some embodiments, the training of the broad-spectrum light recognition model comprises: training the broad-spectrum light recognition model with a constraint target that a first loss is less than a first preset loss value, wherein the first loss comprises: a part classification loss determined based on the plurality of part classification results, and a first classification loss determined based on the first recognition result.

[0022] In some embodiments, the training of the monochromatic light prediction model comprises: training the monochromatic light prediction model with a constraint target that a second loss is less than a second preset loss value, wherein the second loss is determined based on the monochromatic light label and the monochromatic light prediction result.

[0023] In some embodiments, the target recognition model comprises: a broad-spectrum light encoding network configured to perform feature extraction and classification on the broad-spectrum light training Figure One image, outputting a broad-spectrum light feature and a broad-spectrum light classification result; a monochromatic light encoding network configured to perform feature extraction and classification on the monochromatic light training Figure TwoThe feature extraction and classification are performed, and monochromatic light features and monochromatic light classification results are output; and the second fusion network is configured to perform feature fusion on the broadband light features and the monochromatic light features, and output second fusion features and the second recognition result.

[0024] In some embodiments, the training of the target recognition model comprises: training the target recognition model with a third loss being less than a third preset loss value as a constraint, the third loss comprising: a broadband light classification loss determined based on the broadband light classification result, a monochromatic light classification loss determined based on the monochromatic light classification result, and a second classification loss determined based on the second recognition result.

[0025] In a fourth aspect, the present specification also provides a biological feature recognition model training system, comprising: at least one storage medium storing at least one set of instructions for implementing training of a biological feature recognition model; and at least one processor in communication connection with the at least one storage medium, wherein when the biological feature recognition model training system is running, the at least one processor reads the at least one set of instructions and implements the biological feature recognition model training method of the third aspect.

[0026] The biological feature recognition model training method and system provided by the present specification include a broadband light recognition model, a monochromatic light prediction model, and a target recognition model. A broadband light training image and a training sample pair are obtained, the broadband light recognition model is trained based on the broadband light training image, and image features and a first recognition result are output. The image features output by the trained broadband light recognition model are obtained, which are used as an image feature training image, and a monochromatic light label corresponding to the image feature training image is obtained, the monochromatic light prediction model is trained based on the image feature training image and the monochromatic light label, and a monochromatic light prediction result predicted for the training object is output. The target recognition model is trained based on the training sample pair, and a second recognition result is output. The monochromatic light prediction model trained by this training method can accurately predict the best monochromatic light for the target object when applied, and the target recognition model trained based on the training sample pair can accurately recognize the biological features of the target object when applied.

[0027] Other functions of the biological feature recognition method, the biological feature recognition model training method and system provided by the present specification will be partially listed in the following description. According to the description, the following numbers and examples will be apparent to those of ordinary skill in the art. The creative aspects of the biological feature recognition method, the biological feature recognition model training method and system provided by the present specification can be fully explained by practicing or using the methods, devices and combinations described in the following detailed examples. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present specification, the drawings needed to be used in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present specification, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0029] Figure 1 A schematic diagram of an application scenario of a biometric identification system is shown according to some embodiments of the present specification;

[0030] Figure 2 A hardware structure diagram of a computing device is shown according to some embodiments of the present specification;

[0031] Figure 3 A flowchart of a biometric identification method is shown according to some embodiments of the present specification;

[0032] Figure 4 A flowchart of a training method of a biometric identification model is shown according to some embodiments of the present specification; and

[0033] Figure 5 A schematic diagram of a living body identification method based on a monochrome filter system is shown according to some embodiments of the present specification. DETAILED DESCRIPTION

[0034] The following description provides specific application scenarios and requirements of the present specification, so as to enable those skilled in the art to manufacture and use the contents in the present specification. Various local modifications of the disclosed embodiments are obvious to those skilled in the art, and the general principles defined herein can be applied to other embodiments and applications without departing from the spirit and scope of the present specification. Therefore, the present specification is not limited to the shown embodiments, but is consistent with the widest scope of the claims.

[0035] The terms used herein are only for the purpose of describing specific example embodiments, and are not limiting. For example, unless the context clearly indicates otherwise, as used herein, the singular forms "a", "an", and "the" can also include the plural forms. When used in the present specification, the terms "include", "contain" and / or "have" mean that the associated integers, steps, operations, elements and / or components exist, but do not exclude the existence of one or more other features, integers, steps, operations, elements, components and / or groups or can add other features, integers, steps, operations, elements, components and / or groups in the system / method.

[0036] These and other features, and characteristics of the present specification, as well as the methods of operation and functions of the related elements of structure and the combination of parts and economies of manufacture, will become more apparent upon consideration of the following description and the appended claims with reference to the accompanying drawings. The current description will be better understood with reference to the drawings in which:

[0037] The flowcharts used in the present specification show the operations of system implementations according to some embodiments in the present specification. It should be clearly understood that the operations of the flowcharts can not be implemented in sequence. Instead, the operations can be implemented in reverse order or simultaneously. In addition, one or more other operations can be added to the flowcharts. One or more operations can be removed from the flowcharts.

[0038] Before the specific embodiments of the present specification are described, the application scenarios of the present specification are introduced as follows:

[0039] The biometric feature recognition method and the training method of the biometric feature recognition model provided in the present specification can be applied in various scenarios requiring biometric feature recognition. The biometric feature recognition may, for example, be face recognition, fingerprint recognition, pupil recognition, and the like. In the face recognition scenario, it includes face-based identity recognition, face-based liveness recognition, face-based expression recognition, face-based payment willingness recognition, and the like. In identity recognition, the identity information of the face being recognized can be determined based on a face image, for example, in face recognition for access control, the identity of the person who swipes the face needs to be determined. In expression recognition, the expression on the face being recognized can be determined based on a face image, for example, the degree of love for an APP of a user group can be determined according to the facial expressions of the user group. In liveness recognition, it can be detected based on a face image whether the face being recognized is a living body or an attack, for example, in face recognition payment, it is detected whether the object being recognized is a living body. In willingness recognition, it can be determined based on a face image whether the face being recognized currently has the intention to be recognized, for example, in face recognition payment in a store queue, in order to avoid the use of other people's faces for payment by mistake, the payment willingness of the collected face can be recognized. It should be understood that the biometric feature recognition method and the training method of the biometric feature recognition model provided in the present specification can also be applied to other scenarios, and are not limited to the above scenarios. For ease of description, the face-based liveness recognition scenario will be mainly described hereinafter.

[0040] For ease of description, the terms that will appear in the following description of the present specification are explained as follows:

[0041] Attack / liveness attack: non-living body, refers to an attack means presented to the recognition system, including a photo displayed on a mobile phone screen, a printed paper photo, a high-precision mask, a mold, a prosthesis, and the like.

[0042] Liveness detection: In general, in a face recognition system, an algorithm technology for detecting and intercepting an attack by judging whether a user is a living body or an attack.

[0043] Figure 1 An application scenario diagram of a biometric system 001 according to some embodiments of the present specification is shown. The biometric system 001 can implement a biometric recognition method, and can also implement a training method of a biometric recognition model. As shown in Figure 1 The system 001 can include a target object 100, a client 200, a server 300, and a network 400.

[0044] The target object 100 can be any user who uses the client 200 for biometric recognition.

[0045] The client 200 can include an image acquisition device for acquiring images of the target object 100, such as face images, body images, pupil images, fingerprint images, etc. Of course, the image acquisition device can also acquire the sound, behavior, gait, etc. of the target object.

[0046] The image acquisition device can include an image capturing device and a filter device, and the filter device includes at least one monochromatic filter. The image capturing device can capture a broad-spectrum light image of the target object. The image capturing device includes a broad-spectrum light lens and an image sensor. The broad-spectrum light lens is a transparent lens without any color. The broad-spectrum light lens can be an RGB camera, a 3D structured light camera, etc. When the image acquisition device captures the target object in a daylight environment, the reflected light generated by the sunlight shining on the target object passes through the broad-spectrum light lens and forms a broad-spectrum light image of the target object on the image sensor. When the image acquisition device captures the target object in a light environment, the reflected light generated by the light shining on the target object passes through the broad-spectrum light lens and forms a broad-spectrum light image of the target object on the image sensor. The color of the light can be yellow light, warm white light, white light, cold light, etc. The broad-spectrum light can be referred to as visible light, which refers to the part of the electromagnetic wave that can be perceived by the human eye. The wavelength of visible light is generally 380-700 nm. The broad-spectrum light image refers to an image captured by the image capturing device in the ambient light (such as sunlight, light) without the filter device (i.e. without any filter), i.e. the ambient light directly enters the broad-spectrum light lens without any color filter.

[0047] Each monochromatic filter can transmit one monochromatic light and filter out other spectrum lights in the ambient light. Different monochromatic filters can transmit different colors of monochromatic light. For example, a red filter transmits red light, a green filter transmits green light, a blue filter transmits blue light, a yellow filter transmits green light, and so on. The monochromatic light in the present specification can be light of a single frequency (or wavelength), such as light of a wavelength of 0.6328 microns, or narrow spectrum light, i.e. light of a very narrow wavelength range, such as red light in the range of 0.77-0.622 microns. The at least one monochromatic filter can be a red filter, an orange filter, a yellow filter, a green filter, a cyan filter, a blue filter, a purple filter, or other monochromatic filters. The number of each monochromatic filter can be one or more. If the monochromatic light transmitted by one monochromatic filter still contains light of other colors, the number of each monochromatic filter can be increased, such as two, three, or the like, in order to obtain purer monochromatic light. In one possible embodiment, if the combination of different monochromatic filters can transmit a new monochromatic light, the combination of monochromatic filters is also within the protection scope of the present application.

[0048] The light filtering device further comprises a filter driving device. The filter driving device can be connected to each monochromatic filter, such as physically connected. The filter driving device can control the opening and closing of the monochromatic filter. The opening means that the filter driving device drives the monochromatic filter to be placed in the imaging light path of the image capturing device, and the incident surface of the monochromatic filter is parallel to the incident surface of the broadband light lens. The imaging light path refers to the light path of the incident light of the broadband light lens. In this way, the light first passes through the monochromatic filter and then enters the broadband light lens, thereby forming a monochromatic light image. That is, the image acquisition device can acquire a monochromatic light image when the monochromatic filter is opened. The closing means that the filter driving device drives the monochromatic filter not to be placed in the imaging light path of the image capturing device, such as moving the monochromatic filter away from the broadband light lens. At this time, the light does not pass through the monochromatic filter, and the image acquisition device can capture a broadband light image.

[0049] The monochromatic filter can be installed at a position beside the imaging light path of the broadband light lens without blocking the imaging light path of the broadband light lens, thereby facilitating the opening / closing of the monochromatic filter. When the monochromatic filter needs to be opened, the filter driving device drives the monochromatic filter to move / move to the front of the broadband light lens; when the monochromatic filter needs to be closed, the filter driving device drives the monochromatic filter to move away from the front of the broadband light lens.

[0050] The client 200 can include at least one processor. The processor can send an indication signal to the image acquisition device to instruct the image acquisition device to capture the broad spectrum light image and / or the monochromatic light image. For example, the processor can send an indication signal to the filter driving device to open the red filter, so that the filter driving device drives the red filter to open, i.e., the red filter is arranged in the imaging light path of the image capturing device. The method for biometric recognition implemented by the processor will be described in detail later.

[0051] In some embodiments, the biometric recognition method can be executed on the client 200. At this time, the client 200 can store data or instructions for executing the biometric recognition method described in the specification, and can execute or be used to execute the data or instructions. In some embodiments, the client 200 can include a hardware device with data information processing function and necessary programs required to drive the hardware device to work. As shown in Figure 1 The client 200 can be in communication connection with the server 300. In some embodiments, the biometric recognition method can be partially executed on the server 300 and partially executed on the client 200. At this time, the server 300 and the client 200 can store data or instructions for executing the biometric recognition method described in the specification, and can execute or be used to execute the data or instructions. In some embodiments, the server 300 and the client 200 can include a hardware device with data information processing function and necessary programs required to drive the hardware device to work. As shown in Figure 1 The client 200 can be in communication connection with the server 300. In some embodiments, the server 300 can be in communication connection with multiple clients 200. In some embodiments, the client 200 can interact with the server 300 through the network 400 to receive or send messages, etc., such as receiving or sending face images or label information. Of course, the biometric recognition method can also be executed on the server 300.

[0052] In some embodiments, the client 200 can include a mobile device, a tablet, a notebook, a built-in device of a motor vehicle, or the like, a payment Alipay's dragonfly device, a vending machine, a vending cabinet, or any combination thereof. In some embodiments, the mobile device can include a smart home device, a smart mobile device, a virtual reality device, an augmented reality device, or the like, or any combination thereof. In some embodiments, the smart home device can include a smart television, a desktop computer, or the like, or any combination thereof. In some embodiments, the smart mobile device can include a smart phone, a personal digital assistant, a game device, a navigation device, or the like, or any combination thereof. In some embodiments, the virtual reality device or the augmented reality device can include a virtual reality headset, a virtual reality glasses, a virtual reality patch, an augmented reality headset, an augmented reality glasses, an augmented reality patch, or the like, or any combination thereof. For example, the virtual reality device or the augmented reality device can include Google glasses, a head-mounted display, VR, or the like. In some embodiments, the built-in device in the motor vehicle can include an on-board computer, an on-board television, or the like. In some embodiments, the client 200 can be a device with positioning technology for positioning the location of the client 200.

[0053] In some embodiments, the client 200 can be installed with one or more applications (APPs). The APPs can provide the target object 110 with the ability to interact with the outside world through the network 400 and an interface. The APPs include, but are not limited to, web browser type APP programs, search type APP programs, chat type APP programs, shopping type APP programs, video type APP programs, financial type APP programs, instant messaging tools, email clients, social platform software, and the like. In some embodiments, the client 200 can be installed with a target APP. The target APP can be used to collect biological images for the client 200. In some embodiments, the target APP can also be used to identify facial images. The target object 100 can trigger a biological feature recognition request through the target APP. The target APP can respond to the biological feature recognition request and execute the biological feature recognition method.

[0054] The server 300 can be a server that provides various services, such as a background server that provides support for a page displayed on the client 200. In some embodiments, the method of training the biometric model can be executed on the server 300. At this time, the server 300 can store data or instructions for executing the method of training the biometric model described in the specification and can execute or be used to execute the data or instructions. In some embodiments, the server 300 can include a hardware device having a data information processing function and a necessary program for driving the hardware device to operate. In some embodiments, the method of training the biometric model can be partially executed on the server 300 and partially executed on the client 200. At this time, the server 300 and the client 200 can store data or instructions for executing the method of training the biometric model described in the specification and can execute or be used to execute the data or instructions. In some embodiments, the server 300 and the client 200 can include a hardware device having a data information processing function and a necessary program for driving the hardware device to operate. Of course, the method of training the biometric model can also be executed on the client 200.

[0055] The network 400 is a medium for providing a communication connection between the client 200 and the server 300. The network 400 can facilitate the exchange of information or data. As shown, the client 200 and the server 300 can be connected to the network 400 and transmit information or data to each other through the network 400. In some embodiments, the network 400 can be any type of wired or wireless network, or a combination thereof. For example, the network 400 can include a cable network, a wired network, a fiber optic network, a telecommunication network, an intranet, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a wide area network (WAN), a public switched telephone network (PSTN), a Bluetooth network, a ZigBee network, a near-field communication (NFC) network, or the like. In some embodiments, the network 400 can include one or more network access points. For example, the network 400 can include wired or wireless network access points, such as base stations or Internet exchange points, through which one or more components of the client 200 and the server 300 can connect to the network 400 to exchange data or information. Figure 1

[0056] It should be understood that the number of clients 200, servers 300, and networks 400 in FIG. 1 is merely illustrative. Any number of clients 200, servers 300, and networks 400 can be provided as needed for implementation. Figure 1

[0057] Figure 2 ​​A hardware structure diagram of a computing device 600 is shown, which is provided according to some embodiments of the present specification. The computing device 600 can execute the biometric recognition method and / or the training method of the biometric recognition model described in the present specification. The biometric recognition method and / or the training method of the biometric recognition model are introduced in other parts of the present specification. When the biometric recognition method and / or the training method of the biometric recognition model is executed on the client 200, the computing device 600 can be the client 200. When the biometric recognition method and / or the training method of the biometric recognition model is executed on the server 300, the computing device 600 can be the server 300. When the biometric recognition method and / or the training method of the biometric recognition model can be partially executed on the client 200 and partially executed on the server 300, the computing device 600 can be the client 200 and the server 300.

[0058] As shown in Figure 2 The computing device 600 can include at least one storage medium 630 and at least one processor 620. In some embodiments, the computing device 600 can further include a communication port 650 and an internal communication bus 610. Meanwhile, the computing device 600 can also include an I / O component 660.

[0059] The internal communication bus 610 can connect different system components, including the storage medium 630, the processor 620 and the communication port 650.

[0060] The I / O component 660 supports input / output between the computing device 600 and other components.

[0061] The communication port 650 is used for data communication between the computing device 600 and the outside world, for example, the communication port 650 can be used for data communication between the computing device 600 and the network 400. The communication port 650 can be a wired communication port or a wireless communication port.

[0062] The storage medium 630 can include a data storage device. The data storage device can be a non-transitory storage medium or a transitory storage medium. For example, the data storage device can include one or more of a disk 632, a read-only memory (ROM) 634, or a random access memory (RAM) 636. The storage medium 630 can store at least one set of instructions for implementing the biometric recognition method and / or the training method of the biometric recognition model. The instructions are computer program codes, which can include programs, routines, objects, components, data structures, processes, modules, and the like that perform the biometric recognition method and / or the training method of the biometric recognition model provided in the specification. The storage medium 630 can store the biometric recognition model for implementing the biometric recognition method. The model can be one or more sets of instructions stored in the storage medium 630 that execute corresponding instructions and are executed by the processor 620 in the computing device 600. In some embodiments, the model can also be a part of a circuit, a hardware device, or a module in the computing device 600. For example, the visible light recognition model can be a hardware device / module in the computing device 600 that implements recognition of visible light images, the monochromatic light prediction model can be a hardware device / module in the computing device 600 that implements prediction of monochromatic light, and the like.

[0063] The at least one processor 620 can be communicatively connected with the at least one storage medium 630 and the communication port 650 through the internal communication bus 610. The at least one processor 620 is configured to execute the at least one set of instructions described above. When the computing device 600 is running, the at least one processor 620 can read the at least one set of instructions and execute the biometric recognition method and / or the training method of the biometric recognition model provided in the present specification according to the instructions of the at least one set of instructions. The processor 620 can execute all steps included in the biometric recognition method and / or the training method of the biometric recognition model. The processor 620 can be in the form of one or more processors, and in some embodiments, the processor 620 can include one or more hardware processors, such as a microcontroller, a microprocessor, a reduced instruction set computer (RISC), an application-specific integrated circuit (ASIC), an application-specific instruction set processor (ASIP), a central processing unit (CPU), a graphics processing unit (GPU), a physics processing unit (PPU), a microcontroller unit, a digital signal processor (DSP), a field programmable gate array (FPGA), an advanced RISC machine (ARM), a programmable logic device (PLD), any circuit or processor capable of executing one or more functions, or the like, or any combination thereof. For the sake of illustration only, only one processor 620 is described in the computing device 600 in the present specification. However, it should be noted that the computing device 600 in the present specification can also include multiple processors, and thus, the operations and / or method steps disclosed in the present specification can be executed by one processor as described in the present specification, or jointly executed by multiple processors. For example, if the processor 620 of the computing device 600 in the present specification executes step A and step B, it should be understood that step A and step B can also be executed jointly or separately by two different processors 620 (e.g., a first processor executes step A, and a second processor executes step B, or the first and second processors jointly execute steps A and B).

[0064] Figure 3 A flowchart of a biometric recognition method P100 provided according to some embodiments of the present specification is shown. The following description is made by way of example with the processor of the client executing the biometric recognition method P100 described in the present specification.

[0065] As shown in Figure 3 , the method P100 can include:

[0066] S120: obtaining a broad-spectrum light image of a target object, and determining an initial recognition result corresponding to the broad-spectrum light image.

[0067] When the target object is in front of the client, the image acquisition device on the client can acquire a broad spectrum light image of the target object. The target object can be a real user, or an attack such as a paper photo, a mobile phone screen, a mask, etc. The image acquisition device can open a broad spectrum light lens to capture the target object, so that the processor of the client can obtain a broad spectrum light image of the target object from the image acquisition device. At this time, the monochrome filter is in a closed state. For example, the filter driving device controls the monochrome filter to move away from the front of the broad spectrum light lens, so as to avoid affecting the acquisition of the broad spectrum light image. In face-based recognition, the broad spectrum light image can include a face region of the target object, and can also include a background region other than the face region. In other biometric recognition based on fingerprints or pupils, other regions such as fingerprint regions, pupil regions, etc. are included in the broad spectrum light image.

[0068] After the client obtains the broad-spectrum light image, the client can determine an initial recognition result corresponding to the broad-spectrum light image, such as by using a broad-spectrum light recognition model to obtain the initial recognition result. In some embodiments, the client can perform region segmentation on the broad-spectrum light image to obtain a plurality of part images. For example, the client can input the broad-spectrum light image into a segmentation network to perform region segmentation, segment the face region into a plurality of part regions such as eyes, nose, mouth, ears, and cheeks, and output a plurality of part images corresponding to the plurality of part regions, each part image corresponding to a face part. The segmentation network can be Unet, HourGlass Net, etc. Of course, the fingerprint region, pupil region, human body region, etc. can also be segmented. The client can perform feature extraction on the plurality of part images respectively to obtain a plurality of part features. For example, the client can input the plurality of part images into a plurality of part feature extraction networks respectively for feature extraction, and output a part feature map corresponding to each part image. The plurality of part feature extraction networks can be N ResNet18, N DenseNet Transformer, etc. The client can fuse the plurality of part features to obtain a first fused feature. For example, the client can input the plurality of part features into a first fusion network for feature fusion, and output the first fused feature. The first fusion network can be an MLP. The client can determine the initial recognition result corresponding to the first fused feature. Specifically, the client can determine a recognition probability corresponding to the first fused feature, and determine the initial recognition result according to the recognition probability. For example, in live body recognition, the client can determine a probability P1 that the target object is an attack according to the first fused feature, compare P1 with a first preset threshold T1, when P1 is greater than T1, determine that the initial recognition result is that the target object is an attack, and when P1 is less than T1, determine that the initial recognition result is that the target object is a live body. When P1 is equal to T1, the initial recognition result can be that it is uncertain whether the target object is an attack or a live body, in which case the client can re-shoot the target object. When P1 is equal to T1, the initial recognition result can also be an attack or a live body. For example, in identity recognition, the client can determine a probability that the identity information of the target object is incorrect according to the first fused feature, compare the incorrect probability with a preset identity threshold, when the incorrect probability is greater than the identity threshold, determine that the initial recognition result is that the identity of the target object is incorrect, and when the incorrect probability is less than the identity threshold, determine that the initial recognition result is that the identity of the target object is correct. When the incorrect probability is equal to the identity threshold, it is determined that the initial recognition result is uncertain whether the identity of the target object is correct, in which case the client can re-shoot the target object. For biological feature recognition in other scenarios (such as expression recognition), similar to live body recognition and identity recognition, the present application will not be described again.

[0069] In some embodiments, the client can not perform region segmentation on the broad spectrum light image, but perform feature extraction on the broad spectrum light image, and determine the initial recognition result based on the extracted features. The embodiments of the present application do not limit the process of obtaining the initial recognition result.

[0070] The client can determine and execute a target scheme based on the initial recognition result, so as to complete the biometric recognition of the target object and output a target recognition result. The present specification includes multiple schemes, and the target scheme is one of the schemes executed by the client when performing biometric recognition. The multiple schemes include scheme one (i.e., S141-S145) and scheme two (S161). For example, in living body recognition, if the initial recognition result is living body, it indicates that the preliminary judgment of the target object is passed, and in order to improve the accuracy of living body judgment, the client can execute scheme one, i.e., execute S141-S145. If the initial recognition result is attack, it indicates that the target object does not pass the preliminary judgment, and the client can directly perform interception and execute scheme two, i.e., take the initial recognition result as the target recognition result. In some embodiments, if the initial recognition result is attack, the client can also execute scheme one to verify the correctness of the initial recognition result. In some embodiments, if the initial recognition result is living body, scheme one can also not be executed, and the initial recognition result is taken as the target recognition result.

[0071] In some embodiments, the multiple schemes can also include other schemes, such as scheme three: predicting the best monochromatic light corresponding to the biometric recognition of the target object based on the broad spectrum light image, obtaining a target monochromatic light image of the target object under the best monochromatic light, and determining the target recognition result of the target object based on the target monochromatic light image.

[0072] S141: predicting the best monochromatic light corresponding to the biometric recognition of the target object based on the broad spectrum light image.

[0073] The client can obtain multiple part features and first fusion features from the broad spectrum light recognition model, input the multiple part features and the first fusion features into a monochromatic light prediction model, and output the best monochromatic light. The monochromatic light prediction model can be an MLP of L layers, such as an MLP of 3 layers. In some embodiments, the client can input the multiple part features or the first fusion features into the monochromatic light prediction model to predict the best monochromatic light. In some embodiments, the client can perform feature extraction on the broad spectrum light image, and input the extracted global features into the monochromatic light prediction model for prediction.

[0074] There are different materials on the human body, such as moles, bumps, depressions, wrinkles and other different skin materials on the human skin. Different skin materials respond differently to different colors of monochromatic light. For example, moles respond more strongly to blue light, and it is easier to identify moles on the face under blue light, thereby distinguishing between living bodies and attacks. If a user has a mole on his face, the best monochromatic light predicted by the monochromatic light prediction model for him may be blue light. For example, shadows (or depth information) respond more strongly to yellow light, and it is easier to identify shadows on the face, such as the shadows on the sides of the nose bridge, under yellow light. If a user's nose bridge is relatively high, the shadows on the sides of the nose bridge may be more obvious under the ambient light, and the best monochromatic light predicted by the monochromatic light prediction model for him may be yellow light. For example, wrinkles respond more strongly to red light, and it is easier to identify wrinkles on the face, such as wrinkles at the corners of the eyes, under red light. If a user has deep wrinkles at the corners of his eyes, the best monochromatic light predicted by the monochromatic light prediction model for him may be red light.

[0075] Different users present different attack characteristics (skin materials that are easy to distinguish between attacks and living bodies), such as user A having a mole on his face and user B having deep wrinkles at the corners of his eyes, and different colors of monochromatic light are directed at different attack characteristics. The present specification can predict different best monochromatic light for users with different attack characteristics, i.e., customizing the prediction of the monochromatic light most suitable for achieving living body recognition for each user, so that a higher accuracy rate can be achieved for each user in living body recognition.

[0076] For the same user (target object), different parts of the face have different characteristics, such as user A having a mole on his cheeks and wrinkles at the corners of his eyes, and different parts may respond differently to different monochromatic light. Therefore, the present specification predicts the best monochromatic light based on multiple part features and first fused features, fully utilizes the local information and global information of the face, and improves the accuracy of the prediction compared to predicting the best monochromatic light based on global features.

[0077] It should be noted that the best monochromatic light predicted by the present specification for the target object is the monochromatic light most suitable for living body recognition of the target object, i.e., the biological feature recognition accuracy of the target object under the best monochromatic light is higher than the biological feature recognition accuracy under non-best monochromatic light, which refers to any monochromatic light other than the best monochromatic light.

[0078] S143: Obtain a target monochromatic light image of the target object under the best monochromatic light.

[0079] The processor of the client can send an instruction signal to the image acquisition device, instructing the image acquisition device to capture the target object under the optimal monochromatic light, thereby obtaining a target monochromatic light image of the target object. Specifically, after determining the optimal monochromatic light corresponding to the target object, the processor can determine a target filter corresponding to the optimal monochromatic light from the at least one monochromatic filter, the color of the target filter being the same as the color of the optimal monochromatic light. The processor can send an instruction signal to the filter driving device, instructing the filter driving device to open the target filter, i.e., to place the target filter in the imaging light path of the image capturing device. After the filter driving device opens the target filter, it can send a response signal to the processor indicating that the target filter is successfully opened. After receiving the response signal, the processor can send a capture signal to the image capturing device to instruct the image capturing device to capture the target object, thereby obtaining a target monochromatic light image of the target object under the target filter. The image capturing device can send the target monochromatic light image to the processor, and the processor thereby obtains the target monochromatic light image. In some embodiments, the image capturing device can automatically start capturing upon detecting that the filter driving device has opened the target filter, without receiving the capture signal from the processor.

[0080] S145: determining the target recognition result of the target object based on the broad-spectrum light image and the target monochromatic light image.

[0081] The client can input the broad-spectrum light image and the target monochromatic light image into the target recognition model, and output a target recognition result of the target object. Specifically, the client can perform feature encoding on the broad-spectrum light image, such as inputting the broad-spectrum light image into a broad-spectrum light encoding network to obtain a broad-spectrum light feature map of the biometric feature. The client can perform feature encoding on the target monochromatic light image, such as inputting the target monochromatic light image into a monochromatic light encoding network to obtain a monochromatic light feature map of the biometric feature. The client can fuse the broad-spectrum light feature map and the monochromatic light feature map, such as inputting the broad-spectrum light feature map and the monochromatic light feature map into a second fusion network to obtain a second fusion feature. Further, the client can determine the target recognition result based on the second fusion feature. For example, the client can calculate the probability P2 that the second fusion feature belongs to an attack through a classifier, compare P2 with a second preset threshold T2, when P2 is greater than T2, the target recognition result is an attack. When P2 is less than T2, the target recognition result is a live body. When P2 is equal to T2, the target recognition result can be that the target object is an attack or a live body, at this time, the client can re-capture the target monochromatic light image of the target object, or re-capture the broad-spectrum light image and the target monochromatic light image of the target object. When P2 is equal to T2, the target initial recognition result can also be an attack or a live body.

[0082] In some embodiments, after the client obtains the target monochromatic light image, the target monochromatic light image can be only feature coded to obtain a monochromatic light feature map, and the target recognition result can be determined based on the monochromatic light feature map, which is not limited in the embodiments of the present specification.

[0083] S161: The target recognition result is the initial recognition result.

[0084] In summary, the biometric feature recognition method and system provided by the present specification use a wide-spectrum light image to determine the optimal monochromatic light of a target object, thereby combining the wide-spectrum light image and the target monochromatic light image under the optimal monochromatic light to accurately recognize the biometric feature, without requiring the user to perform multiple cooperative actions, thereby improving the user experience. Moreover, compared to biometric feature recognition based only on a wide-spectrum light image, the present specification combines a monochromatic light image with a wide-spectrum light image, so that the biometric feature recognition utilizes more information, thereby further improving the accuracy of the recognition. Furthermore, the present specification can customize the optimal monochromatic light for each user to achieve liveness recognition, so that liveness recognition for each user can achieve a higher accuracy. Furthermore, the present specification predicts the optimal monochromatic light based on multiple part features and a first fusion feature, thereby fully utilizing the local information and global information of the face, and compared to predicting the optimal monochromatic light based on global features, the accuracy of the prediction of the optimal monochromatic light is improved.

[0085] Figure 4 A flowchart of a biometric feature recognition model training method P200 is shown, which is provided according to some embodiments of the present specification. The following description is made by way of example with the processor of a server executing the biometric feature recognition model training method P200 described in the present specification. As shown in Figure 4 The method P200 can include the following steps:

[0086] S210: Obtain a wide-spectrum light training image and a training sample pair, the wide-spectrum light training image including a training object, and the training sample pair including a wide-spectrum light training image and a monochromatic light training image corresponding to the same training object. Figure One Figure Two .

[0087] The server can obtain a plurality of wide-spectrum light training images from a database, and can also obtain an identification label corresponding to each wide-spectrum light training image, such as liveness or attack, wherein the wide-spectrum light training image and the identification label are used to train a wide-spectrum light recognition model. The wide-spectrum light image can include a training object, such as a face, which includes a real user's face and a fake attack face.

[0088] The server can also obtain a plurality of training sample pairs. Each training sample pair includes a wide-spectrum light training image and a monochromatic light training image corresponding to the same training object. Figure One Figure Two The server can also obtain a plurality of training sample pairs. Each training sample pair includes a wide-spectrum light training image and a monochromatic light training image corresponding to the same training object.​For example, the training sample pairs include a combination of a broad-spectrum light training image and a red light training image, a combination of a broad-spectrum light training image and a blue light training image, a combination of a broad-spectrum light training image and a green light training image, a combination of a broad-spectrum light training image and a yellow light training image, and so on. Each training sample pair can have an identification label, such as "live" and "attack." The multiple training sample pairs are used to train the target recognition model.

[0089] S230: Training a broad-spectrum light recognition model based on the broad-spectrum light training image, and outputting image features and a first recognition result.

[0090] The server's processor can perform supervised training on the broad-spectrum light recognition model based on the broad-spectrum light training image and its identification labels, or it can perform unsupervised training on the broad-spectrum light recognition model based on the broad-spectrum light training image. The broad-spectrum light recognition model includes a segmentation network, multiple part feature extraction networks, and a first fusion network. Specifically, the processor can input the broad-spectrum light training image into the segmentation network for region segmentation training, causing it to output multiple part images, each corresponding to a part of the training object. For example, the segmentation network can be trained to segment a face image into five part images: eyes, nose, mouth, ears, and cheeks. The processor can input each of the multiple part images into multiple part feature extraction networks for feature extraction and classification training, extracting and classifying features for each part image, and outputting multiple part features and a part classification result corresponding to each part feature. The part classification result can be, for example, a probability that the training object is an attack, calculated based on the part features. Furthermore, the processor can input the multiple part features into the first fusion network for feature fusion training, causing it to output a first fused feature and a first recognition result. The first recognition result can be, for example, a probability that the training object is an attack, calculated based on the first fused feature.

[0091] The image features include multiple part features and a first fusion feature. It should be noted that the model involved in the training method has the same network structure as the model involved in the aforementioned recognition method, and will not be described in detail here.

[0092] The processor may train the broad-spectrum light recognition model with a constraint target that the first loss is less than a first preset loss value. The first loss includes a part classification loss and a first classification loss. Specifically, the processor may calculate the first loss by adding the part classification loss and the first classification loss using the following formula 1:

[0093] Formula 1: Loss total1 =Loss partial +Loss fision1

[0094] Among them, Loss total1 For the first loss, Loss partial is the part classification loss, Lossfusion1 is a first classification loss.

[0095] It should be noted that the part classification loss can be determined based on multiple part classification results. For example, the processor can calculate a first sub-loss corresponding to each part classification result according to each part classification result and the identification label of the broad-spectrum light training image, obtain multiple first sub-losses, and then take the average of the multiple first sub-losses as the part classification loss, or take the sum of the multiple first sub-losses as the part classification loss. The first classification loss can be determined based on the first identification result. For example, the processor can calculate the first classification loss according to the first identification result and the identification label of the broad-spectrum light training image.

[0096] The processor can train the broad-spectrum light identification model according to the above model structure and loss function until the model converges, and then the model training is completed. The broad-spectrum light identification model completed training can accurately identify the biological features of the target object in actual application.

[0097] S 250: obtaining the image features output by the broad-spectrum light identification model completed training as an image feature training image, and obtaining a monochromatic light label corresponding to the image feature training image.

[0098] The processor can obtain the image features output by the broad-spectrum light identification model completed training, such as the first fusion feature and the multiple part features, as an image feature training image, and obtain a monochromatic light label corresponding to the image feature training image, which can be obtained by manual annotation.

[0099] S270: training a monochromatic light prediction model based on the image feature training image and the monochromatic light label, and outputting a monochromatic light prediction result predicted by the training object.

[0100] The processor can input the image feature training image into the monochromatic light prediction model for prediction training, so that the monochromatic light prediction result predicted by the training object is output. The model structure of the monochromatic light prediction model can be an MLP of L layers.

[0101] The processor can train the monochromatic light prediction model with the constraint target of the second loss being less than the second preset loss value. The second loss can be determined based on the monochromatic light label and the monochromatic light prediction result, such as substituting the monochromatic light label and the monochromatic light prediction result into the classification loss function to calculate the second loss.

[0102] The processor can train the monochromatic light prediction model according to the above model structure and loss function until the model converges, and then the model training is completed. The monochromatic light prediction model completed training can accurately predict the best monochromatic light of the target object in actual application.

[0103] S290: training the target recognition model based on the training sample pair, and outputting a second recognition result.

[0104] The processor can train the target recognition model supervisedly according to the training sample pair and the recognition label thereof, or unsupervisedly according to the training sample pair. The target recognition model includes a broadband light encoding network, a monochromatic light encoding network and a second fusion network. Specifically, the processor can input the broadband light training sample into the broadband light encoding network for feature extraction and classification training, so as to output broadband light features and a broadband light classification result. The broadband light classification result is, for example, a probability of the training object belonging to an attack calculated according to the broadband light features. The processor can input the monochromatic light training sample into the monochromatic light encoding network for feature extraction and classification training, so as to output monochromatic light features and a monochromatic light classification result, which is, for example, a probability of the training object belonging to an attack calculated according to the monochromatic light features. Further, the processor can input the broadband light features and the monochromatic light features into the second fusion network for feature fusion training, so as to output second fusion features and a second recognition result, which is, for example, a probability of the training object belonging to an attack calculated according to the second fusion features. Figure One Figure Two

[0105] The processor can train the target recognition model with a constraint that a third loss is less than a third preset loss value. The third loss includes a broadband light classification loss, a monochromatic light classification loss and a second classification loss. Specifically, the processor can calculate the third loss by adding the broadband light classification loss, the monochromatic light classification loss and the second classification loss according to the following Formula Two:

[0106] Formula Two: Loss total2 = Loss rgb + Loss single + Loss fusion2

[0107] wherein, Loss total2 is the third loss, Loss rgb is the broadband light classification loss, Loss single is the monochromatic light classification loss, and Loss fusion2 is the second classification loss.

[0108] ​​It should be noted that the broad-spectrum light classification loss can be determined based on the broad-spectrum light classification result. For example, the processor calculates the broad-spectrum light classification loss according to the broad-spectrum light classification result and the identification label of the training sample pair. The monochromatic light classification loss can be determined based on the monochromatic light classification result. For example, the processor calculates the monochromatic light classification loss according to the monochromatic light classification result and the identification label of the training sample pair. The second classification loss can be determined based on the second identification result. For example, the processor calculates the second classification loss according to the first identification result and the identification label of the broad-spectrum light training image.

[0109] The processor can train the target light identification model according to the above model structure and loss function until the model converges, and then the model training is completed. The target light identification model completed the training can accurately identify the biological characteristics of the target object according to the monochromatic light image and the broad-spectrum light image of the target object in actual application.

[0110] It should be noted that if the biological characteristics of the target object are identified based on the best monochromatic light in actual application, the processor can only use the monochromatic light training image to train the target identification model, and at this time, the target identification model can include a monochromatic light feature encoding module and a classification module.

[0111] In summary, the training method and system of the biological characteristic identification model provided in the specification, the monochromatic light prediction model trained can accurately predict the best monochromatic light for the target object in application, and the target identification model trained based on the training sample pair can accurately identify the biological characteristics of the target object in application.

[0112] Taking living body recognition as an example, Figure 5 A schematic diagram of a living body recognition method based on a monochromatic filter system is shown, which is provided according to some embodiments of the specification. The living body recognition method based on the monochromatic filter system is as follows:

[0113] 1. Visible light living body recognition model training (or broad-spectrum light living body recognition model): training a living body recognition model under visible light conditions (without any filter), which will be used for selection of monochromatic filters in subsequent steps.

[0114] 2. Best monochromatic light prediction: predicting the best filter according to the output features and results of the visible light living body recognition model.

[0115] 3. Living body recognition model training based on monochromatic light and visible light data: using the data and features of monochromatic light and visible light for final living body recognition model training.

[0116] 4. Model deployment and living body recognition: deploying the trained model and making corresponding living body recognition decisions.

[0117] In another aspect of the present specification, a non-transitory storage medium storing at least one set of executable instructions for performing the biometric recognition and / or training of the biometric recognition model is provided. When the executable instructions are executed by a processor, the executable instructions direct the processor to implement the steps of the biometric recognition method P100 and / or the training method P200 of the biometric recognition model described in the present specification. In some possible implementation manners, various aspects of the present specification can also be implemented in the form of a program product including program codes. When the program product is run on the biometric recognition system 001, the program codes are used to cause the biometric recognition system 001 to perform the steps of the biometric recognition method P100 and / or the training method P200 of the biometric recognition model described in the present specification. The program product for implementing the above method can include the program codes in a portable compact disc read-only memory (CD-ROM) and can be run on the biometric recognition system 001. However, the program product of the present specification is not limited to this, and in the present specification, the readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system. The program product can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium may, for example, be but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any suitable combination of the above. More specific examples of the readable storage medium include an electrical connection having one or more wires, a portable disc, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or a flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. The computer readable storage medium can include a data signal carried in a baseband or as part of a carrier wave, in which readable program codes are borne. Such a propagated data signal can take on multiple forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable storage medium can also be any readable medium that is not a storage medium and that can send, propagate or transmit programs for use by or in connection with an instruction execution system, apparatus or device. The program codes contained in the readable storage medium can be transmitted in any suitable medium, including but not limited to wireless, wired, optical fibers, RF, and the like, or any suitable combination of the above. The program codes for performing the operations of the present specification can be written in any combination of one or more programming languages, including an object-oriented programming language such as Java, C++, and the like, and a conventional procedural programming language such as the "C" language or a similar programming language.The program code can execute entirely on the biometric system 001, partly on the biometric system 001, as a stand-alone software package, partly on the biometric system 001 and partly on a remote computing device, or entirely on the remote computing device.

[0118] The above description of certain examples of the disclosure has been presented for the purposes of illustration and description. Other examples are within the scope and spirit of the claims. In some cases, the actions or steps recited in the claims can be performed in a different order and still accomplish the desired results. Also, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order to achieve the desired results. In certain implementations, multitasking and parallel processing can be advantageous.

[0119] In light of the above, it should be appreciated that the foregoing detailed description and specific examples of the present disclosure are presented for purposes of illustration and description only. The foregoing description and specific examples are not intended to limit the disclosure to the particular examples presented. Rather, the disclosure is intended to embrace all alternatives, modifications, and variations in addition to those shown in the examples. Further, the disclosure is intended to embrace all reasonable modifications and changes that fall within the spirit and scope of the disclosure.

[0120] In addition, certain terminology has been used to describe the embodiments of the disclosure. For example, the terms "one embodiment," "an embodiment” and / or "some embodiments” mean that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the disclosure. The appearances of the phrase "in one embodiment” or "in an embodiment” in various places in the specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0121] It should be understood that, in the foregoing description of embodiments of the disclosure, various features are sometimes grouped together in a single embodiment, figure, or description of a figure for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various aspects, embodiments, and / or features of the disclosure. This should not be interpreted as an intention that the features are to be necessarily grouped together in a single embodiment, or that the disclosure is intended to embrace a combination of features from different embodiments. Rather, the present disclosure is intended to embrace all combinations of features from different embodiments.

[0122] Each patent, patent application, publication of a patent application, and other material, for example articles, books, specifications, publications, documents, things, or the like which can be cited in the present document are hereby incorporated by reference in their entirety for all purposes to the same extent as if each were specifically and individually indicated to be incorporated by reference herein in its entirety for all purposes. In the event of any inconsistency between the terminology, description, definition, and / or use of a term associated with any of the incorporated material and the terminology, description, definition, and / or use of the term in the present document, the terminology, description, definition, and / or use of the term in the present document shall control.

[0123] Finally, it should be understood that the embodiments of the application disclosed herein are illustrative of the principles of the present specification. Other modifications that fall within the scope of the present specification can also be made. Accordingly, the present specification discloses embodiments only as examples. Substantially any arrangement, which is neither specifically described nor explicitly illustrated, can be substituted for the specific embodiments disclosed, without departing from the scope of the present specification. Accordingly, the present specification discloses embodiments only as examples.

Claims

1. A method for biometric recognition, comprising: obtaining a broad spectrum light image of a target object; and performing a target scheme to complete biometric recognition of the target object and output a target recognition result, including: inputting a plurality of part features and / or a first fusion feature obtained based on the broad spectrum light image into a corresponding prediction model to predict an optimal monochromatic light corresponding to biometric recognition of the target object, the first fusion feature being obtained by fusing the plurality of part features, the biometric recognition accuracy of the target object under the optimal monochromatic light being higher than the biometric recognition accuracy under a non-optimal monochromatic light, and the optimal monochromatic light corresponding to different objects not always being the same, obtaining a target monochromatic light image of the target object under the optimal monochromatic light, and determining the target recognition result of the target object based on the target monochromatic light image. 2.The method of claim 1, wherein the plurality of part features are obtained by: segmenting the broad spectrum light image to obtain a plurality of part images of the target object, and extracting features from each of the plurality of part images to obtain the plurality of part features; and the determining of an initial recognition result corresponding to the broad spectrum light image includes: determining the initial recognition result based on the first fusion feature. 3.The method of claim 2, wherein the inputting of the plurality of part features and / or the first fusion feature obtained based on the broad spectrum light image into a corresponding prediction model to predict an optimal monochromatic light corresponding to biometric recognition of the target object includes: inputting the plurality of part features and the first fusion feature into a monochromatic light prediction model to predict and output the optimal monochromatic light. 4.The method of claim 1, wherein the obtaining of a target monochromatic light image of the target object under the optimal monochromatic light includes: sending an instruction signal to an image acquisition device, the instruction signal instructing the image acquisition device to capture the target object under the optimal monochromatic light; and obtaining the target monochromatic light image of the target object. 5.The method of claim 1, wherein the determining of a target recognition result of the target object based on the broad spectrum light image and the target monochromatic light image includes: performing feature coding on the broad spectrum light image to obtain a broad spectrum light feature map of the biometric feature; performing feature coding on the target monochromatic light image to obtain a monochromatic light feature map of the biometric feature; fusing the broad spectrum light feature map and the monochromatic light feature map to obtain a second fusion feature; and determining the target recognition result based on the second fusion feature. 6.The method of claim 1, further comprising: determining an initial recognition result corresponding to the broad spectrum light image; and the performing of the target scheme includes: determining that the initial recognition result is a living body and performing the target scheme. 7.A biometric recognition system, comprising: an image acquisition device; at least one storage medium storing at least one set of instructions for biometric recognition; and ​ ​ ​ ​ at least one processor, in communication connection with the at least one storage medium and the image acquisition device, wherein when the biometric feature recognition system is running, the at least one processor executes the at least one set of instructions and is instructed by the at least one set of instructions to: receive a broad-spectrum light image of a target object from the image acquisition device; and execute a target scheme to complete target biometric feature recognition of the target object and output a target recognition result: input a plurality of part features and / or a first fusion feature obtained based on the broad-spectrum light image into a corresponding prediction model, the first fusion feature being obtained by fusing the plurality of part features, the target biometric feature recognition accuracy of the target object under the best monochromatic light being higher than the target biometric feature recognition accuracy of the target object under a non-best monochromatic light, the best monochromatic light corresponding to different objects not always being the same, obtain a target monochromatic light image of the target object under the best monochromatic light received from the image acquisition device, and determine the target recognition result of the target object based on the target monochromatic light image.

8. The system of claim 7, wherein the image acquisition device comprises an image shooting device and a filter device, the filter device comprising at least one monochromatic filter, the at least one processor being instructed by the at least one set of instructions to: send an instruction signal to the image acquisition device; and the image acquisition device being configured to: set a target filter in the at least one monochromatic filter on an imaging light path of the image shooting device, and shoot the target object to obtain a target monochromatic light image of the target object under the target filter, the target filter being a monochromatic filter corresponding to the best monochromatic light, and send the target monochromatic light image to the at least one processor.

9. The system of claim 8, wherein different monochromatic filters in the at least one monochromatic filter can transmit monochromatic light of different colors.

10. The system of claim 8, wherein the filter device further comprises a filter driving device configured to set one of the at least one monochromatic filter on the imaging light path of the image shooting device when running, the at least one processor being instructed by the at least one set of instructions to: determine the target filter corresponding to the best monochromatic light from the at least one monochromatic filter, and send the instruction signal to the filter driving device; and the filter driving device being configured to receive the instruction signal and set the target filter on the imaging light path.

11. The system of claim 8, wherein an incident surface of the target filter is parallel to an incident surface of a broad-spectrum light lens of the image shooting device.

12. A training method of a biometric feature recognition model, comprising: obtaining a broad-spectrum light training image and a training sample pair, the broad-spectrum light training image comprising a training object, and the training sample pair comprising a broad-spectrum light training image and a monochromatic light training image corresponding to the same training object; training a broad-spectrum light recognition model based on the broad-spectrum light training graph, and outputting image features and a first recognition result; obtaining the image features output by the trained broad-spectrum light recognition model, taking the image features as an image feature training graph, and obtaining a monochromatic light label corresponding to the image feature training graph; training a monochromatic light prediction model based on the image feature training graph and the monochromatic light label, and outputting a monochromatic light prediction result predicted for the training object; and training a target recognition model based on the training sample pair, and outputting a second recognition result, wherein the trained biometric feature recognition model is used to implement the biometric feature recognition method of any one of claims 1-6.

13. The training method of claim 12, wherein the broad-spectrum light recognition model comprises: a segmentation network configured to perform region segmentation on the broad-spectrum light training graph, and output a plurality of part images, each part image corresponding to one part of the training object; a plurality of part feature extraction networks configured to respectively perform feature extraction and classification on the plurality of part images, and output a plurality of part features and a plurality of part classification results corresponding to the plurality of part features; and a first fusion network configured to perform feature fusion on the plurality of part features, and output first fusion features and the first recognition result, wherein the image features comprise the plurality of part features and the first fusion features. The broad-spectrum light recognition model is trained with a first loss being less than a first preset loss value as a constraint target, wherein the first loss comprises:

14. The training method of claim 12, wherein the training the broad spectrum light recognition model comprises: a part classification loss determined based on the plurality of part classification results, and a first classification loss determined based on the first recognition result.

15. The training method of claim 12, wherein the training of the monochromatic light prediction model comprises: training the monochromatic light prediction model with a second loss being less than a second preset loss value as a constraint target, wherein the second loss is determined based on the monochromatic light label and the monochromatic light prediction result.

16. The training method of claim 12, wherein the target recognition model comprises: a broad-spectrum light encoding network configured to perform feature extraction and classification on the broad-spectrum light training graph in the training sample pair, and output broad-spectrum light features and a broad-spectrum light classification result; a monochromatic light encoding network configured to perform feature extraction and classification on the monochromatic light training graph in the training sample pair, and output monochromatic light features and a monochromatic light classification result; and a second fusion network configured to perform feature fusion on the broad-spectrum light features and the monochromatic light features, and output second fusion features and the second recognition result. The target recognition model is trained with a third loss being less than a third preset loss value as a constraint target, wherein the third loss comprises: a broad-spectrum light classification loss determined based on the broad-spectrum light classification result, 17. The training method of claim 16, wherein the training the object recognition model comprises: a monochromatic light classification loss determined based on the monochromatic light classification result, and a second classification loss determined based on the second recognition result.

18. A biometric feature recognition model training system, comprising: ​ ​ at least one storage medium storing at least one set of instructions for implementing training of the biometric model; and at least one processor in communication with the at least one storage medium, wherein when the training system of the biometric model is running, the at least one processor reads the at least one set of instructions and implements the method of training of the biometric model of any one of claims 12-17.

Images