Iris payment method, device, medium and product based on metaverse

Abstract

The embodiment of the application provides a kind of iris payment method, device, medium and product based on meta universe, it is related to artificial intelligence technical field.The method comprises: in the case where consumption object is in meta universe environment and carries out mobile payment, according to the payment gesture and / or body posture of the consumption object, the payment mode of the consumption object is determined;In the case where the payment mode is iris payment, the iris image collected by the virtual device corresponding to the consumption object is acquired;The identification result corresponding to the iris image is determined, and corresponding payment processing is carried out according to the identification result.The scheme of the application solves the problem that there is no natural interaction mode suitable for virtual scene for meta universe mobile payment, and it is difficult to realize reliable iris payment verification and processing due to the distortion of iris collected by virtual devices such as VR glasses and inaccurate identification.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, specifically to an iris payment method, device, medium, and product based on the metaverse. Background Technology

[0002] The metaverse is a collection of virtual spaces, created and linked using technological means, a virtual world that maps and interacts with the real world, a digital living space with a new social system. The metaverse is a parallel universe that runs parallel to human society and is carried in a digital form, bringing an immersive experience through augmented reality, virtual reality and the Internet.

[0003] Metaverse technologies such as Virtual Reality (VR) and Augmented Reality (AR) are developing rapidly, bringing huge market opportunities to iris recognition payment. In the future, iris recognition payment is expected to become one of the mainstream payment methods in metaverse scenarios. However, VR glasses are mainly used to create an immersive virtual environment for users, unlike common iris recognition devices that place the camera directly in front of the user's eyes. Moreover, users do not consciously look at the camera's position while using VR glasses. Therefore, the iris images captured by VR glasses may have iris distortion issues, leading to inaccurate iris recognition results and false recognition or rejection. Summary of the Invention

[0004] At least one embodiment of this application provides an iris payment method, apparatus, medium, and product based on the metaverse, which addresses the problem that metaverse mobile payment lacks a natural interaction method adapted to virtual scenarios, and that reliable iris payment verification and processing are difficult to achieve due to the easy distortion and inaccurate recognition of iris data collected by virtual devices such as VR glasses.

[0005] To solve the above-mentioned technical problems, this application is implemented as follows:

[0006] In a first aspect, embodiments of this application provide an iris payment method based on the metaverse, including:

[0007] When a consumer makes a mobile payment in a metaverse environment, the payment method of the consumer is determined based on the consumer's payment gesture and / or body posture.

[0008] When the payment method is iris payment, the iris image collected by the virtual device corresponding to the consumer is obtained;

[0009] The recognition result corresponding to the iris image is determined, and the corresponding payment processing is performed based on the recognition result.

[0010] Optionally, when the payment method is iris payment, before obtaining the iris image captured by the virtual device corresponding to the consumer, the method further includes:

[0011] When the consumer's account is logged in in the metaverse, an iris authentication request is sent to the server, and the authentication response data corresponding to the iris authentication request is received.

[0012] The identity information of the consumer is verified. If the verification is successful, the identity authentication activation information corresponding to the consumer is obtained and sent to the server. The identity authentication activation information includes the iris information and metaverse role information of the consumer. The metaverse role information is used to represent the virtual role of the consumer preset in the metaverse.

[0013] The server receives the result of verifying the identity authentication activation information, and determines whether the identity authentication corresponding to the iris authentication request is successful based on the result; if the iris authentication request is successful, the iris information of the consumer is stored in the server.

[0014] Optionally, determining the recognition result corresponding to the iris image includes:

[0015] The iris image is input into a preset iris quality grading model, and the quality grading result of the iris image is output.

[0016] Based on the quality grading results, determine the iris feature vector;

[0017] The iris feature vector is input into a preset iris recognition model to determine the recognition result corresponding to the iris image.

[0018] Optionally, based on the quality grading results, the iris feature vector is determined, including:

[0019] If the quality grading result indicates a first quality level, discard the iris image and proceed to the step of acquiring the iris image captured by the virtual device corresponding to the consumer object;

[0020] If the quality grading result indicates a second quality level, the iris image is automatically distorted to obtain a corrected image;

[0021] Based on the corrected image, or the iris image corresponding to the third quality level indicated by the quality grading result, an iris feature vector is determined; the intensity of the third quality level is greater than the intensity of the second quality level, and the intensity of the second quality level is greater than the intensity of the first quality level.

[0022] Optionally, if the quality grading result indicates a second quality level, automatic distortion correction is performed on the iris image to obtain a corrected image, including:

[0023] The iris image is subjected to noise suppression preprocessing and pixel adjustment preprocessing to obtain the processed first image;

[0024] Based on the first image, obtain the edge image;

[0025] The edge image is subjected to edge detection and edge enhancement processing to obtain the processed second image;

[0026] Ellipse fitting is performed on the second image to verify whether the fitted ellipse is the pupil region. If the fitting fails or is determined to be a non-unique pupil region, the image is discarded.

[0027] Verify that the ellipse fitted by the ellipse is the pupil region, and calculate the mapping formula from the ellipse to the circle based on the pupil parameters obtained from the ellipse fitting.

[0028] The iris image is corrected based on the mapping formula, and the black borders of the corrected image are filled and adjusted to a standard size to obtain the corrected image.

[0029] Optionally, before inputting the iris feature vector into a preset iris recognition model to determine the recognition result corresponding to the iris image, the method further includes:

[0030] Obtain iris image samples of the virtual device in the usage scenario, and construct a training sample set;

[0031] Extract the quality factor and left and right eye iris recognition scores for each iris image sample in the training sample set;

[0032] The quality factor is combined with the left and right eye iris recognition scores to form a multi-dimensional feature vector, which is used as a single sample.

[0033] Each sample is labeled with its corresponding recognition result label to form a training sample set;

[0034] A deep learning network architecture is constructed; the deep learning network architecture includes multiple fully connected modules and a classification layer; each fully connected module includes a fully connected layer, a batch normalization layer, an activation layer, and a random dropout layer; the classification layer is a fully connected network used to provide the judgment result;

[0035] The iris recognition model is constructed based on the training sample set and the deep learning network architecture.

[0036] Optionally, before inputting the iris image into a preset iris quality grading model and outputting the quality grading result of the iris image, the method further includes:

[0037] A convolutional neural network is constructed, comprising multiple cascaded convolutional pooling units and a fully connected layer at the end. Each convolutional pooling unit includes a convolutional layer, a normalization layer, an activation layer, and a pooling layer connected in sequence. Functionally, each convolutional pooling unit is divided into a low-level feature extraction unit, a mid-level feature abstraction unit, and a high-level semantic generation unit, used to extract edge and texture features of the sample image, construct iris structure features, and generate a semantic feature vector representing the iris quality level, respectively. The fully connected layer outputs the quality grading result of the sample image based on the semantic feature vector.

[0038] The convolutional neural network is trained using sample training data. If the decrease in the validation loss of a consecutive preset number of convolutional pooling units is less than a first preset threshold during the model training process, an initial model is obtained.

[0039] The initial model is embedded into the full-link iris recognition verification process, and the initial model is re-entered into the test set to verify it, thereby obtaining erroneous graded samples whose quality labels do not match the actual recognition results.

[0040] The initial model is retrained based on the error-graded samples. If the iris recognition false recognition rate of the initial model is lower than the second preset threshold in the full-link iris recognition verification process for a preset number of consecutive times, the model training is terminated, and the iris quality grading model is determined.

[0041] Secondly, embodiments of this application provide an iris payment device based on the metaverse, comprising:

[0042] The first determining module is used to determine the payment method of the consumer based on the consumer's payment gesture and / or body posture when the consumer is making a mobile payment in the metaverse environment.

[0043] The first acquisition module is used to acquire the iris image collected by the virtual device corresponding to the consumer when the payment method is iris payment;

[0044] The first processing module is used to determine the recognition result corresponding to the iris image and to make the corresponding payment based on the recognition result.

[0045] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method as described in any one of the first aspects.

[0046] Fourthly, embodiments of this application provide a computer program product including computer instructions that, when executed by a processor, implement the steps of the method as described in any one of the first aspects.

[0047] Compared with existing technologies, the iris payment method, device, medium, and product based on the metaverse provided in this application, on the one hand, determines the payment method based on payment gestures and / or body posture when the consumer makes a mobile payment in the metaverse, adapting to the virtual interaction scenario of the metaverse and providing a natural interaction method that conforms to the virtual environment; on the other hand, when the payment method is determined to be iris payment, the iris image collected by the virtual device corresponding to the consumer is directly acquired, and the recognition result of the iris image is determined and payment is processed accordingly, forming a complete iris payment process adapted to the virtual device. This realizes payment verification and processing based on the secure biometric feature of the iris in the metaverse scenario, solving the problem that there is a lack of natural interaction methods adapted to the virtual scenario for mobile payment in the metaverse, and that reliable iris payment verification and processing is difficult to achieve due to the easy distortion and inaccurate recognition of iris collected by virtual devices such as VR glasses. Attached Figure Description

[0048] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0049] Figure 1 A flowchart illustrating the iris payment method based on the metaverse provided in this application embodiment;

[0050] Figure 2 A flowchart illustrating the given rules for image quality labels provided in embodiments of this application;

[0051] Figure 3 A schematic diagram of the convolutional neural network structure of the iris quality grading model provided in the embodiments of this application;

[0052] Figure 4 A schematic diagram representing a distorted image;

[0053] Figure 5 A schematic diagram representing an undistorted image;

[0054] Figure 6 This is a schematic diagram showing the image changes of the first image after preprocessing.

[0055] Figure 7 This diagram illustrates the image changes performed during edge extraction and enhancement, and pupil region verification and screening.

[0056] Figure 8 This diagram illustrates the image changes after distortion correction and standardization.

[0057] Figure 9 A flowchart illustrating the overall process of the iris payment method based on the metaverse provided in this application embodiment;

[0058] Figure 10 This is a schematic diagram of the structure of the iris payment device based on the metaverse provided in this application embodiment. Detailed Implementation

[0059] The terms "first," "second," etc., used in this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, without limiting the number of objects; for example, the first object can be one or more. Furthermore, "or" in this application indicates at least one of the connected objects. For example, "A or B" covers three scenarios: Scenario 1: including A but not B; Scenario 2: including B but not A; Scenario 3: including both A and B. The character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0060] The term "instruction" in this application can be either a direct instruction (or explicit instruction) or an indirect instruction (or implicit instruction). A direct instruction can be understood as one in which the sender explicitly informs the receiver of specific information, the operation to be performed, or the requested result, etc.; an indirect instruction can be understood as one in which the receiver determines the corresponding information based on the instruction sent by the sender, or makes a judgment and determines the operation to be performed or the requested result, etc., based on the judgment result.

[0061] The pace of change in the payment era is beyond people's imagination. Payment 1.0 was cash transactions, Payment 2.0 was card payments, Payment 3.0 was QR code payments, and Payment 4.0 is biometric payments. Mobile payment methods based on biometric technology are becoming increasingly diversified. Biometric technology is characterized by rapid updates, short product technology lifecycles, and diversified customer needs.

[0062] Metaverse payment capabilities are a crucial component of the metaverse economic system, with mainstream scenarios including product purchases, game transactions, social interactions, and tipping. Iris recognition payment technology can seamlessly integrate into metaverse scenarios. The rapid development of metaverse technologies such as Virtual Reality (VR) and Augmented Reality (AR) presents a significant market opportunity for iris recognition payment. In the future, iris recognition payment is expected to become one of the mainstream payment methods in metaverse scenarios.

[0063] To enable those skilled in the art to better understand the embodiments of this application, the following description is provided first:

[0064] Misidentification refers to the act of incorrectly classifying a sample that does not belong to a certain category as belonging to that category. For example, identifying A as B.

[0065] False rejection refers to the act of incorrectly classifying a sample that should belong to a certain category as not belonging to that category. For example, sample A has been registered but was rejected during the identification process.

[0066] Within a class is used to indicate the differences or similarities between samples within the same category. For example, A and A' belong to the same class.

[0067] Between-class terms are used to represent the differences or discriminative power between samples from different categories. For example, A and B belong to the between-class term.

[0068] VR glasses primarily create an immersive virtual environment for users, producing a lifelike stereoscopic visual effect. Unlike common iris recognition devices, the camera is not directly in front of the user's eyes, and users don't consciously look at the camera's position while using VR glasses. Therefore, the iris images captured by VR glasses suffer from iris distortion. This application proposes an improved iris algorithm to address the characteristics of iris images captured by VR glasses.

[0069] Additionally, due to camera placement, a high degree of user cooperation may be required to ensure successful binocular recognition. Considering that far fewer people use VR glasses than with common iris recognition devices, the recognition strategy needs to be adjusted accordingly to guarantee a better user experience.

[0070] As described in the background section, existing VR glasses using iris payment suffer from problems such as poor payment experience and low accuracy, which seriously affect the user experience. To solve at least one of the above problems, this application provides an iris payment method, device, medium, and product based on the metaverse, which can reduce or avoid the occurrence of the above situations, improve communication efficiency, and enhance the user experience.

[0071] This application provides an iris payment method, apparatus, medium, and product based on the metaverse. The method and apparatus are based on the same concept, and since the principles by which they solve problems are similar, their implementations can be mutually referenced; repeated details will not be repeated.

[0072] Please refer to Figure 1 This application provides an iris payment method based on the metaverse, comprising:

[0073] Step 11: When the consumer is making a mobile payment in the metaverse environment, determine the payment method of the consumer based on the consumer's payment gesture and / or body posture.

[0074] In this embodiment of the application, step 11 is the trigger and decision-making starting point of the entire payment process. In the virtual interaction scenario of the metaverse, the target payment method is selected through natural user behavior (gestures or body postures), which solves the problem of the compatibility between virtual interaction and payment method selection in the metaverse environment.

[0075] Consumers enter the metaverse virtual space, such as the metaverse marketplace or virtual service scenarios, through VR or AR devices and generate payment needs, such as purchasing virtual goods or paying for virtual services. At this point, a payment method selection process needs to be initiated. The core of the metaverse is immersive virtual interaction. Unlike clicking on a phone screen to select a payment method in reality, here the decision is triggered through natural interactive behaviors adapted to the virtual scene.

[0076] Example 1: Payment gestures, such as when a consumer makes a preset gesture in the metaverse of pointing two fingers together to the temple, it represents selecting iris payment; making a waving gesture represents selecting other payment methods;

[0077] Example 2: Body posture, such as when the consumer nods to confirm in the virtual scene, the system defaults to selecting the preset iris payment; shaking the head switches to other methods;

[0078] The two behaviors described above can be triggered by gestures or postures alone, or by a combination of gesture and posture confirmation, improving the flexibility and accuracy of the interaction. The Metaverse system identifies the characteristics of gestures or postures and matches them with preset interaction-payment method mapping rules, ultimately outputting a clear payment method.

[0079] Step 12: If the payment method is iris payment, obtain the iris image captured by the virtual device corresponding to the consumer.

[0080] Here, step 12 is the data input stage for iris payment. Following the decision result of step 11, it collects key biometric features, namely iris images, through a dedicated device in the metaverse scenario, providing core data support for subsequent identity verification and payment processing. If step 11 determines that another payment method is used, this step is skipped, and the process for the corresponding payment method begins. The virtual device here refers to an interactive terminal with image acquisition capabilities used by the consumer in the metaverse, such as VR glasses. VR glasses integrate a miniature infrared camera, adapted to the lighting requirements of iris recognition, and serve as a dedicated hardware carrier for collecting user iris images in the metaverse environment. After the payment method is determined to be iris payment, the system sends an iris acquisition command to the virtual device; the infrared camera of the virtual device automatically starts and adjusts its angle; the camera of the virtual device acquires the user's real-time iris image and transmits the image data to the recognition module of the payment system through the secure transmission channel of the metaverse; the payment system receives and stores the image data, completing the acquisition operation and preparing for recognition in step 13.

[0081] Step 13: Determine the recognition result corresponding to the iris image, and perform the corresponding payment processing based on the recognition result.

[0082] In this embodiment, the acquired iris image is subjected to feature extraction and comparison to determine whether the recognition result corresponding to the iris image is successful; corresponding operations are performed based on the recognition result. If the recognition is successful, the payment process is completed, such as deduction and voucher generation; if the recognition result corresponding to the iris image fails, the payment is terminated and the result is fed back, such as prompting a retry.

[0083] This application establishes a complete process for metaverse iris payment by defining iris payment, acquiring iris images, recognizing and processing payments. It adapts virtual interactive behavior to the metaverse scenario, ensures payment security through iris biometrics, and ultimately achieves reliable virtual payment through a closed loop of recognition and processing, thus solving the core issues of interactive adaptability and payment security in the metaverse environment.

[0084] Optionally, before step 12, the method further includes:

[0085] When the consumer's account is logged in in the metaverse, an iris authentication request is sent to the server, and the authentication response data corresponding to the iris authentication request is received.

[0086] The identity information of the consumer is verified. If the verification is successful, the identity authentication activation information corresponding to the consumer is obtained and sent to the server. The identity authentication activation information includes the iris information and metaverse role information of the consumer. The metaverse role information is used to represent the virtual role of the consumer preset in the metaverse.

[0087] The server receives the result of verifying the identity authentication activation information, and determines whether the identity authentication corresponding to the iris authentication request is successful based on the result; if the iris authentication request is successful, the iris information of the consumer is stored in the server.

[0088] In this embodiment, this step is the iris information entry and authentication activation process when a user first enables iris payment in the Metaverse. It involves the initial collection of iris features, binding them to a virtual identity, and storing them to provide benchmark comparison data for subsequent iris recognition during payments. Specifically, based on the login status, an iris authentication request is initiated. The consumer has logged into the Metaverse using an account and password, ensuring that the currently operating virtual character is associated with the real user account. After the user triggers the activation of the iris payment function, the system automatically sends an iris authentication request to the server and receives response data from the server, such as request acceptance confirmation and subsequent operation instructions. The iris authentication request is used to apply for iris information entry to activate this payment method.

[0089] By locking down the user's identity associated with the current Metaverse account through login status, the system ensures that subsequent iris information is bound to this account, preventing information confusion. The system first performs basic identity verification on the user, such as verifying login password, mobile verification code, and email verification, to confirm that the user is the account holder and prevent unauthorized use of the account to activate iris payment. After successful verification, the system collects the user's iris information. For the initial entry, a clear iris image is captured using a VR device's camera, and features are extracted. Simultaneously, the system obtains the user's virtual character information in Metaverse, such as the character ID and the account information bound to the character, to identify the real user corresponding to the virtual character in Metaverse. Both are packaged into identity authentication activation information and sent to the server. This achieves the binding of real iris features with the Metaverse virtual character. Subsequent iris payments initiated by this virtual character in Metaverse must match the iris information entered this time, ensuring that virtual behavior corresponds to a real identity.

[0090] After receiving the identity authentication activation information, the server performs dual verification: the validity of the iris information, such as whether the features are clear and whether they conform to the storage format; and the consistency between the virtual character information and the login account, ensuring that the iris information is bound to the correct metaverse account.

[0091] The system receives the verification result returned by the server. If it passes, the iris authentication request is confirmed to be successful; if it fails, such as due to blurry iris information or mismatched role information, the user is prompted to try again.

[0092] After the verification is approved, the server encrypts and stores the user's iris information (feature template) as the benchmark data for iris recognition comparison in the subsequent step 13. This is the core purpose of the initial entry, to provide a comparison template for each subsequent iris payment and ensure that the recognition is executable.

[0093] In this application, during the initialization process of using iris payment for the first time, a closed loop is established, which includes login status confirmation, basic identity verification, submission of iris and virtual role information, and server verification and storage. This completes the initial entry of iris information and binding of virtual identity, solving the problem of the lack of benchmark feature data for iris payment in the metaverse. This provides the necessary prerequisite for the subsequent iris image acquisition, recognition and comparison in steps 12 and 13, namely, there are pre-stored features for comparison, ensuring the security and feasibility of the first and subsequent iris payments.

[0094] Optionally, in step 13, determining the recognition result corresponding to the iris image includes:

[0095] The iris image is input into a preset iris quality grading model, and the quality grading result of the iris image is output.

[0096] Based on the quality grading results, determine the iris feature vector;

[0097] The iris feature vector is input into a preset iris recognition model to determine the recognition result corresponding to the iris image.

[0098] In this embodiment, the accuracy of iris recognition is improved through a progressive logic of quality screening, feature extraction, and accurate identification, especially adapting to distorted images that may be captured by VR devices. The iris image obtained in step 12, which may have distortion or blurring issues due to VR device capture, is input into a preset iris quality grading model. This preset iris quality grading model can be a deep learning-based classification model or a traditional image quality assessment algorithm. The preset iris quality grading model analyzes key image quality indicators, such as iris region integrity, sharpness, illumination uniformity, and distortion degree, and outputs a quality grading result. This quality grading result can be represented as high-quality data, medium-quality data, or low-quality data. The iris quality grading model first filters out low-quality images, such as severely distorted inferior images, to avoid direct use in identification and subsequent errors. Simultaneously, it provides a quality adaptation basis for subsequent feature extraction; different feature extraction strategies are required for images of different qualities.

[0099] Different feature extraction methods are used for different quality grading results:

[0100] For high-quality images, high-precision algorithms are used to directly extract detailed iris features and determine the iris feature vector. For medium-quality images with slight distortion or blurring, image enhancement processing, such as deblurring and distortion correction, is performed first before feature extraction to ensure the effectiveness of the feature vector. For low-quality images, a second acquisition may be triggered, prompting the user to adjust their posture and reacquire the image, or a simplified feature extraction logic may be used to focus on the core stable region of the iris and avoid interference from invalid features. This application preferably reacquires the image. The feature extraction strategy is dynamically adjusted according to the image quality to ensure that the generated iris feature vector accurately reflects the user's iris features and reduces feature distortion caused by image quality issues.

[0101] The iris feature vector obtained in the previous step is input into a preset iris recognition model. If the iris recognition model is a feature-matching algorithm, it compares the feature vector with the user's pre-stored iris feature template (i.e., the feature vector stored during initial registration) on the server, calculating the similarity between the two. Based on a preset similarity threshold, the recognition result is output. If the match is successful, it confirms the user's identity and the recognition passes; if the match fails, it indicates the user is not the owner or the features do not match, and the recognition fails. Through standardized feature comparison logic, image features are transformed into a clear identity verification result, providing a decision-making basis for subsequent payment processing.

[0102] This application specifically addresses potential quality issues in iris image acquisition by VR devices, such as distortion and blurring. It filters valid images through quality grading, retains key information through adaptive feature extraction, and finally achieves accurate identity verification through a recognition model, laying a reliable foundation for "payment processing based on recognition results" in step 13.

[0103] Optionally, based on the quality grading results, the iris feature vector is determined, including:

[0104] If the quality grading result indicates a first quality level, discard the iris image and proceed to the step of acquiring the iris image captured by the virtual device corresponding to the consumer object;

[0105] If the quality grading result indicates a second quality level, the iris image is automatically distorted to obtain a corrected image;

[0106] Based on the corrected image, or the iris image corresponding to the third quality level indicated by the quality grading result, an iris feature vector is determined; the intensity of the third quality level is greater than the intensity of the second quality level, and the intensity of the second quality level is greater than the intensity of the first quality level.

[0107] In this application, the iris feature vector is determined based on the quality grading results. If it is the first quality level (lowest level): the iris image is discarded and a new iris image acquired by the virtual device is obtained; if it is the second quality level (medium level): the iris image is first automatically distorted to obtain a corrected image, and then the iris feature vector is determined based on the corrected image; if it is the third quality level (highest level): the iris feature vector is directly determined based on the iris image.

[0108] It's important to note that due to the spatial layout of VR devices, unlike common iris recognition devices where the camera is directly in front of the viewer's eyes, the captured iris images may be distorted or lose important iris information, affecting recognition performance. For VR devices, iris quality is categorized into three levels: high-quality data, medium-quality data, and low-quality data. Images with severe distortion and missing important iris information are marked as low-quality data and discarded, triggering a re-capture. Images with slight distortion retain most of the iris information and are marked as medium-quality data. A distortion correction algorithm is then activated to reduce the impact of distortion on recognition performance and ensure efficiency. The remaining images are marked as high-quality data and directly enter the fast matching process. The first quality level represents low-quality data; the second quality level represents medium-quality data; and the third quality level represents high-quality data.

[0109] The iris image quality grading standard constructed in this application adopts a dynamic labeling method driven by recognition performance. Specifically, the grading system is scientifically constructed through the following technical path: images with high recognition rates are labeled as high-quality data, while data with very high false recognition or false rejection rates are labeled as low-quality samples. Simultaneously, a correction algorithm is used to distinguish between medium and low-quality data. Ultimately, a three-level image quality grading system (high, medium, and low) is established, aiming to construct a scientific and standardized iris image quality assessment system and provide clean and reliable data for iris image quality models.

[0110] In one alternative implementation, refer to Figure 2 As shown, iris quality labels are determined through multiple iterations of iris recognition results. The iris recognition algorithm is a pre-defined existing algorithm. The iris comparison score ranges from 0 to 100, with higher scores indicating higher iris similarity. The given rules for iris quality labeling are as follows: Iris comparison is performed on all acquired VR iris images, and images with no misidentification and an in-class score exceeding 60 are selected. Selected images are manually screened to ensure that images exceeding 60 are not strabismus images, and the screened images are labeled as high-quality data. High-quality data is used as registration data, and the remaining data, after distortion correction, is used as recognition data. The two sets of data are compared; recognition data with no misidentification and an in-class score exceeding 60 is labeled as medium-quality data. Recognition data with an in-class score below 60 and misidentifications are labeled as low-quality data.

[0111] After obtaining the iris quality labels through the above steps, the number of samples in each category may vary significantly. Excessive differences in data between categories can lead to poor model training performance, rendering the model unusable in practice. To balance the data across the three categories (e.g., by randomly sampling the category with the larger number of samples to make the sample sizes of the three categories similar), the data is then fed into the deep learning model for training.

[0112] Optionally, before inputting the iris image into a preset iris quality grading model and outputting the quality grading result of the iris image, the method further includes:

[0113] A convolutional neural network is constructed, comprising multiple cascaded convolutional pooling units and a fully connected layer at the end. Each convolutional pooling unit includes a convolutional layer, a normalization layer, an activation layer, and a pooling layer connected in sequence. Functionally, each convolutional pooling unit is divided into a low-level feature extraction unit, a mid-level feature abstraction unit, and a high-level semantic generation unit, used to extract edge and texture features of the sample image, construct iris structure features, and generate a semantic feature vector representing the iris quality level, respectively. The fully connected layer outputs the quality grading result of the sample image based on the semantic feature vector.

[0114] The convolutional neural network is trained using sample training data. If the decrease in the validation loss of a consecutive preset number of convolutional pooling units is less than a first preset threshold during the model training process, an initial model is obtained.

[0115] The initial model is embedded into the full-link iris recognition verification process, and the initial model is re-entered into the test set to verify it, thereby obtaining erroneous graded samples whose quality labels do not match the actual recognition results.

[0116] The initial model is retrained based on the error-graded samples. If the iris recognition false recognition rate of the initial model is lower than the second preset threshold in the full-link iris recognition verification process for a preset number of consecutive times, the model training is terminated, and the iris quality grading model is determined.

[0117] The iris quality grading model in this application is constructed using a deep learning model, fully leveraging its technological advantages in the field of image processing. With its end-to-end automated feature learning capabilities, the deep learning model can extract multi-dimensional features layer by layer from iris images, from low-level texture details to high-level semantic features, effectively addressing the quality assessment needs of VR devices.

[0118] The iris quality assessment model designed in this application adopts a hierarchical convolutional neural network architecture. Its core structure consists of six cascaded convolutional pooling units and fully connected layers at the end. The overall design follows the principles of hierarchical feature extraction, dynamic dimensionality compression, and semantic abstraction decision-making. Specifically, refer to... Figure 3As shown, the technical details are as follows: The model consists of multiple convolutional pooling units and fully connected layers. The convolutional pooling units include: connected convolutional layers, BatchNorm layers, activation layers (such as ReLU and Sigmoid), and max-pooling layers. The convolutional pooling units are used to extract semantic-level features, and the fully connected layers are used to provide quality judgment results.

[0119] Convolutional layers extract low-level features such as edges and gradients of the iris texture through local sliding windows. The number of channels increases layer by layer, achieving progressive abstraction of features from the pixel level to the semantic level. The BatchNorm layer normalizes the convolutional output, improving the network's training convergence speed. Activation layers alleviate gradient vanishing through sparse activation properties, improving the stability of deep network training. Max pooling layers reduce the feature map size, achieving spatial dimensionality compression and translation invariance.

[0120] The choice of the number of units in the model is related to engineering considerations. Based on the depth-accuracy balance theory of image classification tasks and considering the feature complexity of iris images, the number of units is anchored in the range of 5-10. This application uses 6 convolutional pooling units. The first 3 units focus on low-level features (edge ​​detection to texture combination), the middle 2 units complete mid-level abstraction (part recognition to structural modeling), and the last unit generates high-level semantics (quality level feature vector), forming a complete hierarchical representation. Comparative experiments show that the 6-unit scheme has lower inference time on VR devices while maintaining high accuracy, meeting the needs of real-time interaction. This application's 6-unit convolutional pooling architecture, through modular design, depth optimization, and engineering adaptation, achieves a balance between feature extraction efficiency, model generalization ability, and edge computing performance in VR iris image quality evaluation tasks.

[0121] The closed-loop iterative training mechanism constructed in this application achieves high robustness of the iris quality assessment model through a dynamic optimization strategy driven by quantitative indicators. The specific optimization scheme is as follows:

[0122] (1) Loss Detection and Process-Level Validation Trigger. When the validation loss decreases by less than 1% for 5 consecutive epochs during training, the full-link recognition validation process is automatically triggered. The current model is embedded into the complete iris recognition process (collection, quality assessment, feature extraction, template matching) and validated on a cross-scene test set containing more than 2000 samples. Two types of key recognition data are recorded during the validation process:

[0123] a. Incorrectly graded samples: The quality label does not match the actual recognition result. The quality label of the image needs to be adjusted before it is sent to the model for training.

[0124] b. Samples with blurred boundaries: These are critical cases where the quality score is near the threshold (60±5), causing fluctuations in the recognition results. To avoid affecting the training effect of the model, these samples are discarded.

[0125] Training iterations terminate when the false recognition rate falls below the preset requirement in three consecutive full-process tests. The training mechanism in this application transforms human experience into a standardized iterative process through three core stages: quantitative monitoring, process verification, and targeted correction. This avoids blindly tuning based solely on loss values ​​and enhances the model's adaptability to complex scenarios through in-depth utilization of false recognition samples. This solution ensures performance for engineering deployment while providing a sustainable optimization path for iris quality assessment models, making it particularly suitable for edge computing scenarios like VR devices that demand high real-time performance and robustness.

[0126] In VR device iris data acquisition, the spatial layout characteristics of the camera array lead to iris image distortion, which is a typical phenomenon. For slightly distorted iris images, although the core features are not completely destroyed, it can directly lead to a decrease in the accuracy of effective iris region segmentation, thereby affecting the accuracy of subsequent feature encoding. This application proposes an adaptive correction method for VR iris images. By preserving the core feature information of the original image and specifically repairing geometric distortion, it avoids the waste of samples caused by directly discarding slightly distorted data, and provides high-precision input for subsequent feature extraction and template generation, effectively solving the typical distortion problem in VR device iris acquisition.

[0127] In an optional embodiment, an adaptively corrected iris recognition model needs to be constructed before adaptive correction. Further, before inputting the iris feature vector into the preset iris recognition model to determine the recognition result corresponding to the iris image, the method further includes:

[0128] Obtain iris image samples of the virtual device in the usage scenario, and construct a training sample set;

[0129] Extract the quality factor and left and right eye iris recognition scores for each iris image sample in the training sample set;

[0130] The quality factor is combined with the left and right eye iris recognition scores to form a multi-dimensional feature vector, which is used as a single sample.

[0131] Each sample is labeled with its corresponding recognition result label to form a training sample set;

[0132] A deep learning network architecture is constructed; the deep learning network architecture includes multiple fully connected modules and a classification layer; each fully connected module includes a fully connected layer, a batch normalization layer, an activation layer, and a random dropout layer; the classification layer is a fully connected network used to provide the judgment result;

[0133] The iris recognition model is constructed based on the training sample set and the deep learning network architecture.

[0134] In this embodiment, a model capable of accurately recognizing irises captured by virtual devices is obtained by collecting scenario-based samples, extracting key features, designing a network architecture, and training the model. Specifically, scenario-based samples are acquired by constructing an initial training set that collects iris image samples captured by virtual devices (such as VR glasses) in actual usage scenarios. This ensures that the samples cover image features that may occur in the scenario (such as distortion, angular deviation, etc.), providing raw data close to real-world applications for model training. Key features are extracted: quality factors and left / right eye recognition scores are extracted from each sample image. Two types of core features are extracted: the quality factor reflects the quality of the iris image, such as sharpness, degree of distortion, and iris region integrity, used by the model to judge image reliability; the left / right eye iris recognition score is a quantitative score used to distinguish whether a sample belongs to the left or right eye, helping the model eliminate the interference of left / right eye confusion on the recognition results. Features are fused to form sample input: the extracted quality factors and left / right eye recognition scores are integrated into a multi-dimensional feature vector, which serves as the input feature for a single sample. Through multi-dimensional feature fusion, the model can utilize both image quality and eye-specific information simultaneously, improving the robustness of recognition. Labeling and constructing a supervised training set involves labeling each sample with a corresponding recognition result label, such as successful match or failed match. The feature vectors are then associated with the labels to form a complete supervised learning training set, enabling the model to learn the mapping relationship between features and recognition results through the labels.

[0135] Building a deep learning network architecture and designing a network structure adapted to the task includes:

[0136] The system consists of multiple fully connected modules and a classification layer. Each fully connected module comprises a fully connected layer (responsible for feature mapping), a batch normalization layer (accelerates training and stabilizes parameters), an activation layer (introduces non-linearity to enhance fitting ability), and a dropout layer (randomly discards some neurons to prevent overfitting), used to progressively extract and optimize features. The classification layer uses a fully connected network to output the final recognition result (such as whether it is the target user), completing the classification task.

[0137] The model is trained and an iris recognition model is constructed using a labeled training sample set and algorithms such as backpropagation to train the aforementioned deep learning network. The network parameters (such as the weights of fully connected layers) are adjusted so that the model can accurately output iris recognition results based on the input multidimensional feature vector, and finally form an iris recognition model adapted to virtual device scenarios.

[0138] The entire process takes into account the characteristics of iris images acquired by virtual devices (such as susceptibility to distortion and special scene conditions). Through scene-specific samples, multi-feature fusion, and anti-overfitting network design, it ensures that the model can achieve accurate iris recognition in the metaverse scene.

[0139] Specifically, this application designs an iris recognition scheme tailored to the unique usage scenarios of VR devices and the advantages of iris biometrics. The highly specialized nature of VR device user groups—VR devices (such as those for home or personal office use) are typically used frequently by 1-3 fixed users—is a significant departure from iris recognition devices in public settings like airport security scanners and time clocks. In virtual reality (VR) scenarios, the accuracy and stability of iris recognition face severe challenges due to the complex optical environment of head-mounted devices, the dynamic differences in user wearing, and the uncontrollability of eye movements. Traditional judgment methods based on fixed thresholds are ill-suited to the complex and ever-changing recognition conditions in VR scenarios, easily leading to misjudgments or missed judgments. Therefore, this VR iris recognition strategy model was developed to effectively circumvent the limitations of manually set thresholds through an intelligent decision-making mechanism.

[0140] The input layer design of this iris recognition model comprehensively incorporates key information from five core dimensions. In terms of image quality assessment, it integrates four quality factors: image brightness, sharpness, occlusion, and distortion. These indicators accurately characterize the usability of the iris image. Simultaneously, it introduces iris recognition scores for both the left and right eyes. These scores, generated by the iris algorithm, directly reflect the similarity between the iris to be recognized and the template iris. This score includes the recognition scores of images with good iris quality assessment and corrected iris images. Through the collaborative input of multi-dimensional data, the model constructs a complete basis for recognition decisions.

[0141] In terms of model operation mechanism, an advanced deep learning architecture is adopted. Through training and optimization with massive amounts of data, the model can automatically learn the nonlinear mapping relationship between each input factor and the recognition result. Compared with the simple decision-making mode of manually setting a single threshold in traditional methods, this iris recognition model can dynamically weigh the weights of different quality factors and adaptively adjust the judgment criteria of the comparison score. For example, when the image clarity is low but the brightness and occlusion are within an acceptable range, the model will appropriately lower the threshold requirement for the comparison score; while when the image distortion is severe, it will raise the comparison score standard to ensure recognition accuracy. This intelligent decision-making process not only effectively avoids the subjective arbitrariness of manually setting thresholds, but also significantly improves the robustness and adaptability of the recognition system in complex VR environments.

[0142] Ultimately, the model's output layer visually presents the success or failure of the comparison as a binary result (0 or 1), providing reliable decision support for VR device authentication and access management. Through this data-driven intelligent recognition strategy, the VR iris recognition system overcomes the limitations of traditional threshold judgment, achieving more accurate and stable identity authentication and providing a solid guarantee for information security and user experience in VR application scenarios.

[0143] Optionally, the step of determining the iris recognition model may be:

[0144] Obtain training samples: Combine four quality factors (brightness, sharpness, occlusion, and distortion) and the scores of the left and right eyes into a sample, and label whether the recognition was successful.

[0145] Network Construction: The network consists of multiple fully connected modules and a classification layer. Each fully connected module includes a fully connected layer, a BatchNorm layer, an activation layer, and a Dropout layer. The classification layer is a fully connected network used to provide the judgment result.

[0146] Network training: Stop training when the validation set error rate is below one in a million.

[0147] It should be noted that, based on the degree of VR iris distortion, automatic distortion correction processing of the iris image is triggered by a trained iris recognition model. Furthermore, if the quality grading result indicates a second quality level, automatic distortion correction is performed on the iris image to obtain a corrected image, including:

[0148] The iris image is subjected to noise suppression preprocessing and pixel adjustment preprocessing to obtain the processed first image;

[0149] Based on the first image, obtain the edge image;

[0150] The edge image is subjected to edge detection and edge enhancement processing to obtain the processed second image;

[0151] Ellipse fitting is performed on the second image to verify whether the fitted ellipse is the pupil region. If the fitting fails or is determined to be a non-unique pupil region, the image is discarded.

[0152] Verify that the ellipse fitted by the ellipse is the pupil region, and calculate the mapping formula from the ellipse to the circle based on the pupil parameters obtained from the ellipse fitting.

[0153] The iris image is corrected based on the mapping formula, and the black borders of the corrected image are filled and adjusted to a standard size to obtain the corrected image.

[0154] It should be noted that, referring to Figure 4 and Figure 5For the corresponding distorted and undistorted image samples, the data with slight distortion, which retains most of the iris features, is corrected. The correction method uses the positional information located at the edge of the pupil, i.e., the black area in the center of the iris. There are two reasons for using the pupil's edge instead of the iris's edge: First, the probability of an image containing the complete iris region is very small, increasing the difficulty of subsequent ellipse fitting. If the complete iris region needs to be displayed, the user would have to widen their eyes, resulting in a poor user experience. Second, the pupil region appears completely in most images, and the pupil and iris are almost concentric circles; the degree of distortion in the pupil can be used to infer the degree of distortion in the entire iris.

[0155] The principle of correction is to first eliminate interference from noise, eyelashes, etc., then use an edge detection algorithm to locate the edge of the pupil, and then use an ellipse fitting algorithm to fit and locate the position of the pupil in the image. Based on the length and angle of the major and minor axes of the pupil ellipse, the formula for mapping the ellipse to a circle is calculated, thereby correcting the entire image. When the quality grading result is the second quality level (the iris image has some distortion but can be corrected), the core of automatic distortion correction is to eliminate distortion and standardize the image through preprocessing, edge extraction, pupil fitting, and mapping correction. The specific steps and objectives are as follows:

[0156] (1), reference Figure 6 As shown, after noise suppression and pixel adjustment preprocessing, the preprocessed first image can be obtained. Specifically, Gaussian filtering is applied to the iris image to reduce image noise and prevent noise interference with subsequent edge detection; pixels with a value higher than 50 are set to 255 (since the pixel value of the iris region in VR scenes is usually lower than 50, this operation weakens interference from bright areas such as eyelashes); simultaneously, pixels within 80 pixels around the perimeter of the image are set to 255 (referring to the minimum iris radius in relevant standards; if the pupil is close to the edge, the iris is incomplete, and such invalid areas are excluded in advance). After the above processing, a first image with a prominent iris region and reduced interference is obtained.

[0157] Reference Figure 7 As shown, edge extraction and enhancement, pupil region verification and screening are performed.

[0158] (2) Edge extraction and enhancement include: using an edge detector (such as the Canny operator) to perform edge detection on the first image, extracting the boundary contours of the pupil and iris to obtain an edge image; performing morphological dilation on the edge image to enhance edge continuity, making the pupil edge clearer, and finally obtaining a second image with prominent edge features.

[0159] (3) Pupil region verification and screening, comprising: since the pupil in a VR image is likely to be elliptical due to viewing angle deviation, performing ellipse fitting on the second image to obtain parameters such as the major axis and minor axis of the ellipse; if the fitting fails (no valid ellipse is obtained), discarding the image. Further, verifying whether the ellipse is a pupil. The judgment is made that the average pixel value inside the ellipse is less than 40 (the pupil region is darker in color and has lower pixel values), and only the only ellipse meeting the condition is retained (to eliminate interference from false edges); if a unique pupil region cannot be determined, the image is also discarded.

[0160] (4) With reference to Figure 8 , distortion correction and standardization are performed. According to the fitted pupil ellipse parameters, a mapping formula from the ellipse to a standard circle is derived, and the distorted elliptical pupil is corrected into a regular circle through geometric transformation to eliminate distortion caused by the viewing angle; the corrected image may have black edges due to rotation, the black edges are filled by adjacent pixel values, and the image is adjusted to a standard size (e.g., 640x480), so as to obtain a corrected image meeting the identification requirements, which lays a foundation for subsequent iris feature extraction.

[0161] The entire process addresses the problems that irises in VR images are prone to distortion and have multiple interferences. Through layered preprocessing, accurate edge extraction and geometric correction, low-quality but retrievable iris images are converted into standardized images, which improves the accuracy of subsequent identification.

[0162] Further, the steps from ellipse fitting to circle are as follows: obtaining ellipse parameters: center coordinates ( ), lengths of the major axis and the minor axis (a, b), and rotation angle θ. Calculating a scaling ratio according to the major axis and the minor axis: setting the length of the major axis of the ellipse as a and the length of the minor axis as b. If a > b, then = , = 1. If a < b, then = 1, = . Rotating the ellipse to the horizontal or vertical direction, the point ( ) after affine transformation is obtained from (x, y) through affine transformation:

[0163] .

[0164] In an alternative embodiment, with reference to Figure 9The overall process diagram shown illustrates that in VR scenarios, due to the complex optical structure of head-mounted devices, differences in wearing angles, and the uncontrollability of eye movements, iris distortion becomes a key factor restricting the accuracy and stability of iris recognition technology. To address this challenge, this application proposes an automatically corrected iris feature extraction method, aiming to overcome the limitations of traditional technologies and achieve high-precision iris recognition. This method constructs a two-dimensional evaluation system based on conventional iris image quality assessment: on the one hand, it uses traditional indicators such as sharpness, illumination uniformity, and occlusion ratio to ensure that the input image meets basic recognition requirements; on the other hand, it introduces a VR iris distortion-specific judgment, using a deep learning model to quantitatively evaluate the degree of distortion. Based on the distortion evaluation results, the system automatically triggers an adaptive correction mechanism. The corrected image then enters the feature extraction module to extract stable and robust iris feature vectors, ultimately achieving high-precision identity authentication. This scheme effectively reduces the impact of iris distortion on recognition results in VR environments, providing reliable technical support for biometric recognition in VR scenarios. This application provides a complete strategy for iris recognition in VR glasses, including an optimized method for iris feature extraction in VR glasses and a payment strategy for VR devices. This can improve user experience while ensuring recognition accuracy.

[0165] Compared with the prior art, this application has at least the following advantages:

[0166] 1) On the one hand, existing technologies use different score fitting methods. Assuming the score range is 0-100, using score fitting is equivalent to dividing the image into 101 categories. Dividing the image into different categories is more effective than simply providing scores. This application divides the image into 3 categories, which is more suitable from the perspective of model training performance. On the other hand, this application uses similarity scores as image quality labels. Quality labels are directly related to the recognition results, and the label provision method used in this application is more efficient.

[0167] 2) Algorithm improvement scheme for VR glasses iris images: Improve the iris feature extraction algorithm to address the iris distortion problem in VR glasses and improve the accuracy and recognition rate of iris recognition in VR glasses;

[0168] 3) VR glasses iris payment strategy: Based on the unique user characteristics of VR devices, a VR glasses iris recognition strategy model is proposed, which effectively avoids the subjective arbitrariness of manually setting thresholds and significantly improves the robustness and adaptability of the recognition system in complex VR environments.

[0169] 4) Automatic labeling method for iris image quality labels acquired by VR glasses. In deep learning training, the labels of data are crucial for model training. This proposal puts forward a labeling method for iris image quality labels. This method can obtain a large number of reliable data labels with only a small amount of manual work, which is the basis for training iris quality models.

[0170] The various methods of the embodiments of this application have been described above. Apparatus for implementing the above methods will now be provided.

[0171] Please refer to Figure 10 This application also provides an iris payment device based on the metaverse, comprising:

[0172] The first determining module 101 is used to determine the payment method of the consumer based on the consumer's payment gesture and / or body posture when the consumer is making a mobile payment in the metaverse environment.

[0173] The first acquisition module 102 is used to acquire the iris image collected by the virtual device corresponding to the consumer when the payment method is iris payment;

[0174] The first processing module 103 is used to determine the recognition result corresponding to the iris image and to make corresponding payments based on the recognition result.

[0175] Optionally, the device further includes:

[0176] The second processing module is used to send an iris authentication request to the server and receive the authentication response data corresponding to the iris authentication request when the consumer's account is logged in in the metaverse.

[0177] The third processing module is used to verify the identity information of the consumer. If the verification is successful, the module obtains the identity authentication activation information corresponding to the consumer and sends the identity authentication activation information to the server. The identity authentication activation information includes the iris information and metaverse role information of the consumer. The metaverse role information is used to represent the virtual role that the consumer is preset in the metaverse.

[0178] The fourth processing module is used to receive the result of the verification of the identity authentication activation information sent by the server, and determine whether the identity authentication corresponding to the iris authentication request is successful based on the result; if the iris authentication request is successful, the iris information of the consumer is stored in the server.

[0179] Optionally, the first processing module 103 described above includes:

[0180] The first processing unit is used to input the iris image into a preset iris quality grading model and output the quality grading result of the iris image;

[0181] The first determining unit is used to determine the iris feature vector based on the quality grading result;

[0182] The second determining unit is used to input the iris feature vector into a preset iris recognition model to determine the recognition result corresponding to the iris image.

[0183] Optionally, the first determining unit described above is specifically used for:

[0184] If the quality grading result indicates a first quality level, discard the iris image and proceed to the step of acquiring the iris image captured by the virtual device corresponding to the consumer object;

[0185] If the quality grading result indicates a second quality level, the iris image is automatically distorted to obtain a corrected image;

[0186] Based on the corrected image, or the iris image corresponding to the third quality level indicated by the quality grading result, an iris feature vector is determined; the intensity of the third quality level is greater than the intensity of the second quality level, and the intensity of the second quality level is greater than the intensity of the first quality level.

[0187] Optionally, if the quality grading result indicates a second quality level, automatic distortion correction is performed on the iris image to obtain a corrected image, including:

[0188] The iris image is subjected to noise suppression preprocessing and pixel adjustment preprocessing to obtain the processed first image;

[0189] Based on the first image, obtain the edge image;

[0190] The edge image is subjected to edge detection and edge enhancement processing to obtain the processed second image;

[0191] Ellipse fitting is performed on the second image to verify whether the fitted ellipse is the pupil region. If the fitting fails or is determined to be a non-unique pupil region, the image is discarded.

[0192] Verify that the ellipse fitted by the ellipse is the pupil region, and calculate the mapping formula from the ellipse to the circle based on the pupil parameters obtained from the ellipse fitting.

[0193] The iris image is corrected based on the mapping formula, and the black borders of the corrected image are filled and adjusted to a standard size to obtain the corrected image.

[0194] Optionally, the above-mentioned device further includes:

[0195] The first construction module is used to acquire iris image samples of the virtual device in the usage scenario and construct a training sample set;

[0196] The extraction module is used to extract the quality factor and left and right eye iris recognition scores for each iris image sample in the training sample set;

[0197] The fifth processing module is used to combine the quality factor with the left and right eye iris recognition scores into a multi-dimensional feature vector as a single sample;

[0198] The sixth processing module is used to label each sample with the corresponding recognition result label to form a training sample set;

[0199] The second building module is used to build a deep learning network architecture; the deep learning network architecture includes multiple fully connected modules and a classification layer; each fully connected module includes a fully connected layer, a batch normalization layer, an activation layer, and a random deactivation dropout layer; the classification layer is a fully connected network used to provide the judgment result;

[0200] The third construction module is used to construct the iris recognition model based on the training sample set and the deep learning network architecture.

[0201] Optionally, the above-mentioned device further includes:

[0202] The fourth construction module is used to construct a convolutional neural network. The convolutional neural network includes multiple cascaded convolutional pooling units and fully connected layers at the ends. Each convolutional pooling unit contains a convolutional layer, a normalization layer, an activation layer, and a pooling layer connected in sequence. Functionally, each convolutional pooling unit is divided into a low-level feature extraction unit, a mid-level feature abstraction unit, and a high-level semantic generation unit, used to extract edge and texture features of the sample image, construct iris structural features, and generate semantic feature vectors representing iris quality levels, respectively. The fully connected layer outputs the quality grading results of the sample image based on the semantic feature vectors.

[0203] The second acquisition module is used to train the convolutional neural network using sample training data, and to acquire an initial model when the verification loss of a consecutive preset number of convolutional pooling units decreases less than a first preset threshold during model training.

[0204] The third acquisition module is used to embed the initial model into the full-link iris recognition verification process, re-input the test set to verify the initial model, and obtain erroneous graded samples whose quality labels do not match the actual recognition results.

[0205] The second determining module is used to retrain the initial model based on the error-graded samples. When the iris recognition false recognition rate corresponding to the initial model is lower than the second preset threshold in the full-link iris recognition verification process for a preset number of consecutive times, the model training is terminated and the iris quality grading model is determined.

[0206] It should be noted that the device in this embodiment corresponds to the method described above, and the implementation methods in each of the above embodiments are applicable to the embodiment of this device, achieving the same technical effect. The device provided in this application embodiment can implement all the method steps implemented in the above method embodiments and achieve the same technical effect. Therefore, the parts and beneficial effects that are the same as those in the method embodiments will not be described in detail here.

[0207] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the above-described iris payment method embodiment based on the metaverse, and achieves the same technical effect. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0208] This application also provides a computer program product, including computer instructions. When the computer instructions are executed by a processor, they implement the various processes of the above-described iris payment method embodiment based on the metaverse, and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0209] It should be noted that the collection, gathering, updating, analysis, processing, use, transmission, and storage of user personal information involved in this disclosed technical solution all comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken to prevent unauthorized access to user personal information data and to safeguard user personal information security and network security.

[0210] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0211] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0212] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An iris payment method based on the metaverse, characterized in that, include: When a consumer makes a mobile payment in a metaverse environment, the payment method of the consumer is determined based on the consumer's payment gesture and / or body posture. When the payment method is iris payment, the iris image collected by the virtual device corresponding to the consumer is obtained; Determine the recognition result corresponding to the iris image, and perform the corresponding payment processing based on the recognition result; The determination of the recognition result corresponding to the iris image includes: The iris image is input into a preset iris quality grading model, and the quality grading result of the iris image is output. Determining the iris feature vector based on the quality grading result includes: if the quality grading result indicates a first quality level, discarding the iris image and obtaining the iris image captured by the virtual device corresponding to the consumer; if the quality grading result indicates a second quality level, automatically correcting the distortion of the iris image to obtain a corrected image; determining the iris feature vector based on the corrected image, or the iris image corresponding to the quality grading result indicating a third quality level; wherein the intensity of the third quality level is greater than the intensity of the second quality level, and the intensity of the second quality level is greater than the intensity of the first quality level. The iris feature vector is input into a preset iris recognition model to determine the recognition result corresponding to the iris image; Before inputting the iris feature vector into a preset iris recognition model to determine the recognition result corresponding to the iris image, the method further includes: acquiring iris image samples of the virtual device in the usage scenario to construct a training sample set; extracting the quality factor and left and right eye iris recognition scores of each iris image sample in the training sample set; combining the quality factor and the left and right eye iris recognition scores into a multi-dimensional feature vector as a single sample; labeling each sample with a corresponding recognition result label to form a training sample set; constructing a deep learning network architecture; the deep learning network architecture includes multiple fully connected modules and a classification layer; each fully connected module includes a fully connected layer, a batch normalization layer, an activation layer, and a random deactivation dropout layer; the classification layer is a fully connected network used to provide a judgment result; and constructing the iris recognition model based on the training sample set and the deep learning network architecture. Before inputting the iris image into a preset iris quality grading model and outputting the quality grading result of the iris image, the method further includes: A convolutional neural network is constructed, comprising multiple cascaded convolutional pooling units and a fully connected layer at the end. Each convolutional pooling unit includes a convolutional layer, a normalization layer, an activation layer, and a pooling layer connected in sequence. Functionally, each convolutional pooling unit is divided into a low-level feature extraction unit, a mid-level feature abstraction unit, and a high-level semantic generation unit, used to extract edge and texture features of the sample image, construct iris structure features, and generate a semantic feature vector representing the iris quality level, respectively. The fully connected layer outputs the quality grading result of the sample image based on the semantic feature vector. The convolutional neural network is trained using sample training data. If the decrease in the validation loss of a consecutive preset number of convolutional pooling units is less than a first preset threshold during the model training process, an initial model is obtained. The initial model is embedded into the full-link iris recognition verification process, and the initial model is re-entered into the test set to verify it, thereby obtaining erroneous graded samples whose quality labels do not match the actual recognition results. The initial model is retrained based on the error-graded samples. If the iris recognition false recognition rate of the initial model is lower than the second preset threshold in the full-link iris recognition verification process for a preset number of consecutive times, the model training is terminated, and the iris quality grading model is determined.

2. The method according to claim 1, characterized in that, When the payment method is iris payment, before obtaining the iris image captured by the virtual device corresponding to the consumer, the method further includes: When the consumer's account is logged in in the metaverse, an iris authentication request is sent to the server, and the authentication response data corresponding to the iris authentication request is received. The identity information of the consumer is verified. If the verification is successful, the identity authentication activation information corresponding to the consumer is obtained and sent to the server. The identity authentication activation information includes the iris information and metaverse role information of the consumer. The metaverse role information is used to represent the virtual role of the consumer preset in the metaverse. The server receives the result of verifying the identity authentication activation information, and determines whether the identity authentication corresponding to the iris authentication request is successful based on the result; if the iris authentication request is successful, the iris information of the consumer is stored in the server.

3. The method according to claim 1, characterized in that, When the quality grading result indicates a second quality level, automatic distortion correction is performed on the iris image to obtain a corrected image, including: The iris image is subjected to noise suppression preprocessing and pixel adjustment preprocessing to obtain the processed first image; Based on the first image, obtain the edge image; The edge image is subjected to edge detection and edge enhancement processing to obtain the processed second image; Ellipse fitting is performed on the second image to verify whether the fitted ellipse is the pupil region. If the fitting fails or is determined to be a non-unique pupil region, the image is discarded. Verify that the ellipse fitted by the ellipse is the pupil region, and calculate the mapping formula from the ellipse to the circle based on the pupil parameters obtained from the ellipse fitting. The iris image is corrected based on the mapping formula, and the black borders of the corrected image are filled and adjusted to a standard size to obtain the corrected image.

4. An iris payment device based on the metaverse, characterized in that, include: The first determining module is used to determine the payment method of the consumer based on the consumer's payment gesture and / or body posture when the consumer is making a mobile payment in the metaverse environment. The first acquisition module is used to acquire the iris image collected by the virtual device corresponding to the consumer when the payment method is iris payment; The first processing module is used to determine the recognition result corresponding to the iris image and to make the corresponding payment based on the recognition result; The first processing module includes: The first processing unit is used to input the iris image into a preset iris quality grading model and output the quality grading result of the iris image; The first determining unit is configured to determine an iris feature vector based on the quality grading result; specifically, the first determining unit is configured to: discard the iris image and perform the step of acquiring the iris image collected by the virtual device corresponding to the consumer when the quality grading result indicates a first quality level; automatically correct the distortion of the iris image to acquire a corrected image when the quality grading result indicates a second quality level; and determine the iris feature vector based on the corrected image, or the iris image corresponding to the third quality level when the quality grading result indicates a third quality level; wherein the intensity of the third quality level is greater than the intensity of the second quality level, and the intensity of the second quality level is greater than the intensity of the first quality level. The second determining unit is used to input the iris feature vector into a preset iris recognition model to determine the recognition result corresponding to the iris image; The device further includes: a first construction module for acquiring iris image samples of the virtual device in a usage scenario and constructing a training sample set; an extraction module for extracting the quality factor and left and right eye iris recognition scores for each iris image sample in the training sample set; a fifth processing module for combining the quality factor and the left and right eye iris recognition scores into a multi-dimensional feature vector as a single sample; a sixth processing module for labeling each sample with a corresponding recognition result label to form a training sample set; a second construction module for building a deep learning network architecture; the deep learning network architecture includes multiple fully connected modules and a classification layer; each fully connected module includes a fully connected layer, a batch normalization layer, an activation layer, and a random dropout layer; the classification layer is a fully connected network used to provide a judgment result; and a third construction module for constructing the iris recognition model based on the training sample set and the deep learning network architecture. The device further includes: The fourth construction module is used to construct a convolutional neural network. The convolutional neural network includes multiple cascaded convolutional pooling units and fully connected layers at the ends. Each convolutional pooling unit contains a convolutional layer, a normalization layer, an activation layer, and a pooling layer connected in sequence. Functionally, each convolutional pooling unit is divided into a low-level feature extraction unit, a mid-level feature abstraction unit, and a high-level semantic generation unit, used to extract edge and texture features of the sample image, construct iris structural features, and generate semantic feature vectors representing iris quality levels, respectively. The fully connected layer outputs the quality grading results of the sample image based on the semantic feature vectors. The second acquisition module is used to train the convolutional neural network using sample training data, and to acquire an initial model when the verification loss of a consecutive preset number of convolutional pooling units decreases less than a first preset threshold during model training. The third acquisition module is used to embed the initial model into the full-link iris recognition verification process, re-input the test set to verify the initial model, and obtain erroneous graded samples whose quality labels do not match the actual recognition results. The second determining module is used to retrain the initial model based on the error-graded samples. When the iris recognition false recognition rate corresponding to the initial model is lower than the second preset threshold in the full-link iris recognition verification process for a preset number of consecutive times, the model training is terminated and the iris quality grading model is determined.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 3.

6. A computer program product, characterized in that, It includes computer instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 3.

Images