Iris recognition method, device, medium and equipment based on uncertainty modeling

Abstract

The present disclosure relates to the field of iris recognition technology and provides an iris recognition method, apparatus, medium, and device based on uncertainty modeling, comprising: obtaining iris image data and inputting it into a pre-built iris recognition model; performing pixel-level modeling on the image data using an uncertainty modeling module to extract the mean and variance features of each pixel, and generating an uncertainty feature representation and an iris feature template based on the extraction results and a re-parameter sampling strategy; generating a variance scaling map based on a dynamic mask generation module and image data, and generating a dynamic feature mask using the variance scaling map and a dynamic threshold judgment strategy; and determining target identity information based on a preset iris template database, the iris feature template, and the dynamic feature mask to complete identity recognition. By combining the iris feature template and the dynamic feature mask of the iris image, the present disclosure improves the accuracy and robustness of iris recognition, maintaining high recognition accuracy under varying conditions such as different lighting conditions.

Description

Technical Field

[0001] The present disclosure relates to the technical field of iris recognition, and in particular to an iris recognition method, apparatus, medium, and device based on uncertainty modeling. Background Art

[0002] Iris recognition technology, due to its high stability, uniqueness, and contactless nature, has become a core technology in the biometric identification field, widely used in fields such as identity authentication, security monitoring, and financial payments. As a unique and relatively stable biometric feature of the human body, the iris provides relatively accurate identification. Its contactless nature improves the security of the identification process while maintaining a good user experience. Traditional iris recognition methods primarily rely on extracting deterministic features from iris images through techniques such as filtering, phase encoding, or deep learning, and then matching them against a binary template.

[0003] However, in practical applications, iris images are easily affected by factors such as lighting changes, equipment differences, eye movement, and occlusion, which leads to a large dispersion in the distribution of iris samples of the same identity in the feature space, affecting the accuracy and stability of recognition.

[0004] Related technologies mostly rely on fixed thresholds for feature matching and mask generation, and fail to effectively deal with uncertainty factors under different acquisition conditions (such as changes in lighting, differences in imaging angles, etc.). This makes the system less robust, and the recognition accuracy often drops significantly in cross-device or cross-database tests. Secondly, existing methods usually separate the feature extraction and mask generation processes of iris images, and the quality of the iris mask directly affects the matching accuracy. Improper mask processing may lead to the loss of important information, thereby affecting the final recognition effect. In addition, although deep learning technology has made certain progress in iris recognition in recent years, existing methods often ignore the uncertainty factors introduced in the acquisition process, resulting in deep learning models still having difficulty improving their generalization ability and robustness when facing different data sets. Summary of the Invention

[0005] The embodiments of the present disclosure at least provide an iris recognition method, apparatus, medium, and device based on uncertainty modeling. By combining the feature template and dynamic feature mask of the iris image, the accuracy and robustness of iris recognition can be improved, while maintaining a high recognition accuracy under changing conditions such as different lighting and angles.

[0006] The present disclosure provides an iris recognition method based on uncertainty modeling, including:

[0007] Obtain iris image data;

[0008] Inputting the iris image data into a pre-built iris recognition model; wherein the iris recognition model includes an uncertainty modeling module and a dynamic mask generation module;

[0009] Performing pixel-level modeling on the iris image data based on the uncertainty modeling module, extracting mean features and variance features of each pixel in the iris image data, generating an uncertainty feature representation corresponding to the iris image data based on the extraction results and a re-parameter sampling strategy, and determining an iris feature template based on the uncertainty feature representation;

[0010] generating a variance scaling map corresponding to the iris image data based on the dynamic mask generation module and the iris image data, and generating a dynamic feature mask corresponding to the iris image data based on the variance scaling map and a dynamic threshold determination strategy;

[0011] Acquire a preset iris template database; and determine target identity information corresponding to the iris image data based on the preset iris template database, the iris feature template, and the dynamic feature mask to complete iris identity recognition.

[0012] In some possible embodiments, acquiring iris image data includes:

[0013] Acquire an initial iris image, and perform image segmentation on the initial iris image to obtain an initial iris binary mask for separating the iris area from the non-iris area;

[0014] Performing edge fitting processing on the iris outer boundary and pupil boundary in the initial iris binary mask based on an edge detection algorithm to obtain a circular iris region;

[0015] The circular iris area is mapped into the iris image data of uniform size.

[0016] In some possible embodiments, the iris recognition model further includes a high-dimensional feature extraction module; after inputting the iris image data into the pre-built iris recognition model, the process further includes:

[0017] Performing high-dimensional feature extraction on the iris image data based on the high-dimensional feature extraction module to obtain iris high-dimensional features;

[0018] Accordingly, performing pixel-level modeling on the iris image data based on the uncertainty modeling module includes:

[0019] Performing pixel-level modeling on the iris high-dimensional features based on the uncertainty modeling module;

[0020] Accordingly, the generating of a variance scaling map corresponding to the iris image data based on the dynamic mask generating module and the iris image data includes:

[0021] A variance scaling map corresponding to the iris image data is generated based on the dynamic mask generation module and the iris high-dimensional features.

[0022] In some possible embodiments, the uncertainty modeling module includes a mean extraction submodule, a variance extraction submodule, and a reparameterization sampling submodule;

[0023] The mean extraction submodule is used to extract the mean feature map of each pixel from the iris high-dimensional features;

[0024] The variance extraction submodule is used to extract the variance feature map of each pixel from the iris high-dimensional features;

[0025] The reparameterized sampling submodule is used to execute the reparameterized sampling strategy according to the mean feature map and the variance feature map of each pixel to introduce disturbance, and generate the pixel-level uncertainty feature representation with distribution characteristics based on the execution result; and determine the iris feature template based on the uncertainty feature representation.

[0026] In some possible embodiments, before generating the variance scaling map corresponding to the iris image data based on the dynamic mask generation module and the iris high-dimensional features, the method includes:

[0027] Normalization processing is performed on the iris image data, and pixel-level grayscale statistical processing is performed on the normalization processing result to obtain a global grayscale mean and a global grayscale standard deviation of the iris image data.

[0028] In some possible embodiments, generating a variance scaling map corresponding to the iris image data based on the dynamic mask generation module and the iris image data, and generating a dynamic feature mask corresponding to the iris image data based on the variance scaling map and a dynamic threshold determination strategy includes:

[0029] performing convolution processing on the iris high-dimensional features based on the dynamic mask generation module, and determining a variance scaling map corresponding to the iris image data based on the convolution processing result;

[0030] Determining the dynamic feature mask for the iris image data based on a dynamic mask generation formula, the global grayscale mean, the global grayscale standard deviation, and the variance scaling map;

[0031] The dynamic mask generation formula is expressed as:

[0032]

[0033] Among them, M ij Indicates the value of the dynamic feature mask at the i-th and j-th pixel positions, which can be 0 or 1; I represents the grayscale value of the i-th and j-th pixel positions in the normalized iris image; μ Represents the global grayscale mean, I σ Represents the global grayscale standard deviation, A ij Represents the value of the i-th and j-th pixels in the variance scaling map.

[0034] In some possible embodiments, the preset iris template database includes multiple iris templates of known identities; and determining the target identity information corresponding to the iris image data based on the preset iris template database, the iris feature template, and the dynamic feature mask includes:

[0035] performing dynamic weighting processing on the iris feature template based on the dynamic feature mask, and using the iris feature template after the dynamic weighting processing as a query template;

[0036] Comparing the query template with the iris templates of each known identity in the preset iris template database in sequence, and calculating the similarity values ​​between the query template and the iris templates of each known identity in the preset iris template database respectively;

[0037] The identity information corresponding to the iris template with the largest similarity value exceeding a preset threshold is determined as the target identity information corresponding to the iris image data.

[0038] The present disclosure provides an iris recognition device based on uncertainty modeling, including:

[0039] A data acquisition module, used for acquiring iris image data;

[0040] A data input module, configured to input the iris image data into a pre-built iris recognition model; wherein the iris recognition model includes an uncertainty modeling module and a dynamic mask generation module;

[0041] a template determination module, configured to perform pixel-level modeling on the iris image data based on the uncertainty modeling module, extract mean features and variance features of each pixel in the iris image data, generate an uncertainty feature representation corresponding to the iris image data based on the extraction results and a reparameter sampling strategy, and determine an iris feature template based on the uncertainty feature representation;

[0042] a mask determination module, configured to generate a variance scaling map corresponding to the iris image data based on the dynamic mask generation module and the iris image data, and generate a dynamic feature mask corresponding to the iris image data based on the variance scaling map and a dynamic threshold determination strategy;

[0043] The identity recognition module is used to obtain a preset iris template database; and determine the target identity information corresponding to the iris image data based on the preset iris template database, the iris feature template and the dynamic feature mask to complete iris identity recognition.

[0044] In some possible embodiments, the data acquisition module is specifically configured to:

[0045] Acquire an initial iris image, and perform image segmentation on the initial iris image to obtain an initial iris binary mask for separating the iris area from the non-iris area;

[0046] Performing edge fitting processing on the iris outer boundary and pupil boundary in the initial iris binary mask based on an edge detection algorithm to obtain a circular iris region;

[0047] The circular iris area is mapped into the iris image data of uniform size.

[0048] In some possible embodiments, the iris recognition model further includes a high-dimensional feature extraction module; and the data input module is further configured to:

[0049] Performing high-dimensional feature extraction on the iris image data based on the high-dimensional feature extraction module to obtain iris high-dimensional features;

[0050] Accordingly, the template determination module is specifically configured to:

[0051] Performing pixel-level modeling on the iris high-dimensional features based on the uncertainty modeling module;

[0052] Accordingly, the mask determination module is specifically configured to:

[0053] A variance scaling map corresponding to the iris image data is generated based on the dynamic mask generation module and the iris high-dimensional features.

[0054] In some possible embodiments, the uncertainty modeling module includes a mean extraction submodule, a variance extraction submodule, and a reparameterization sampling submodule;

[0055] The mean extraction submodule is used to extract the mean feature map of each pixel from the iris high-dimensional features;

[0056] The variance extraction submodule is used to extract the variance feature map of each pixel from the iris high-dimensional features;

[0057] The reparameterized sampling submodule is used to execute the reparameterized sampling strategy according to the mean feature map and the variance feature map of each pixel to introduce disturbance, and generate the pixel-level uncertainty feature representation with distribution characteristics based on the execution result; and determine the iris feature template based on the uncertainty feature representation.

[0058] In some possible embodiments, the mask determination module is further configured to:

[0059] Normalization processing is performed on the iris image data, and pixel-level grayscale statistical processing is performed on the normalization processing result to obtain a global grayscale mean and a global grayscale standard deviation of the iris image data.

[0060] In some possible embodiments, the mask determination module is specifically configured to:

[0061] performing convolution processing on the iris high-dimensional features based on the dynamic mask generation module, and determining a variance scaling map corresponding to the iris image data based on the convolution processing result;

[0062] Determining the dynamic feature mask for the iris image data based on a dynamic mask generation formula, the global grayscale mean, the global grayscale standard deviation, and the variance scaling map;

[0063] The dynamic mask generation formula is expressed as:

[0064]

[0065] Among them, M ij Indicates the value of the dynamic feature mask at the i-th and j-th pixel positions, which can be 0 or 1; I represents the grayscale value of the i-th and j-th pixel positions in the normalized iris image; μ Represents the global grayscale mean, I σ Represents the global grayscale standard deviation, A ij Represents the value of the i-th and j-th pixels in the variance scaling map.

[0066] In some possible embodiments, the preset iris template database includes multiple iris templates of known identities; and the identity recognition module is specifically configured to:

[0067] performing dynamic weighting processing on the iris feature template based on the dynamic feature mask, and using the iris feature template after the dynamic weighting processing as a query template;

[0068] Comparing the query template with the iris templates of each known identity in the preset iris template database in sequence, and calculating the similarity values ​​between the query template and the iris templates of each known identity in the preset iris template database respectively;

[0069] The identity information corresponding to the iris template with the largest similarity value that exceeds a preset threshold is determined as the target identity information corresponding to the iris image data.

[0070] An embodiment of the present disclosure provides a computer device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the computer device is running, the processor and the memory communicate via the bus. When the machine-readable instructions are executed by the processor, an iris recognition method based on uncertainty modeling as described in any possible embodiment described above is performed.

[0071] An embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the iris recognition method based on uncertainty modeling as described in any of the above possible implementations is implemented.

[0072] The iris recognition method, apparatus, medium, and device based on uncertainty modeling provided in the embodiments of the present disclosure extract stable features from iris images based on a pre-built iris recognition model, utilize dynamic feature masks to address environmental changes, and then match these features with a pre-set iris template database to ensure accurate identification of target identity information even under complex conditions. In this way, by combining the iris image feature template and dynamic feature mask, the accuracy and robustness of iris recognition can be effectively improved, especially maintaining high recognition accuracy under varying conditions such as varying lighting and angles.

[0073] In order to make the above-mentioned objectives, features and advantages of the present disclosure more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings that need to be cited in the embodiments. The drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present disclosure and, together with the specification, are used to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only illustrate certain embodiments of the present disclosure and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without inventive effort.

[0075] Figure 1 A flowchart of an iris recognition method based on uncertainty modeling provided by an embodiment of the present disclosure is shown;

[0076] Figure 2 A flowchart of an iris image preprocessing method provided by an embodiment of the present disclosure is shown;

[0077] Figure 3 A flow chart of a method for determining a dynamic feature mask provided by an embodiment of the present disclosure is shown;

[0078] Figure 4 A schematic structural diagram of an iris recognition device based on uncertainty modeling provided by an embodiment of the present disclosure is shown;

[0079] Figure 5 A schematic structural diagram of a computer device provided by an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0080] In order to make the purpose, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. The components of the embodiments of the present disclosure generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the drawings is not intended to limit the scope of the disclosure for which protection is sought, but merely represents selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present disclosure.

[0081] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0082] The term "and / or" herein simply describes an association relationship, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, and the existence of B alone. In addition, the term "at least one" herein refers to any combination of at least two of any one or more of a plurality of items. For example, "at least one of A, B, and C" can represent any one or more elements selected from the set consisting of A, B, and C.

[0083] Iris recognition has become a key technology in the field of biometric identification due to its high stability, uniqueness, and non-contact nature. As the human body's most unique biometric feature, virtually unaffected by the external environment, the iris' structure remains unchanged after birth, significantly improving the accuracy of identification. Consequently, iris recognition is widely used in fields such as identity authentication, financial payments, and security monitoring.

[0084] However, traditional iris recognition methods rely on the extraction of fixed features in images and are often affected by multiple factors such as lighting, equipment differences, eye movement, occlusions (such as eyelashes and eyelids), and shooting angles. These factors cause iris images to show large variations under different conditions, resulting in increased discreteness of iris images of the same identity in the feature space, thereby affecting recognition accuracy.

[0085] Research has found that most existing iris recognition technologies are based on traditional deterministic feature extraction methods, such as filtering, phase encoding, or deep learning models for feature extraction of iris images, followed by matching using binary templates. However, these methods are generally unable to effectively address the various uncertainties inherent in real-world applications. Most existing iris recognition methods rely on fixed thresholds for feature matching and mask generation, failing to dynamically adapt to changing acquisition conditions, resulting in limited recognition accuracy. Furthermore, while deep learning methods have made some progress in iris recognition, and the use of deep models such as convolutional neural networks for feature extraction can improve recognition performance, they still face the problem of feature ambiguity. Even with larger datasets or more complex network structures during training, further improving generalization is difficult. Furthermore, related technologies often overlook the diversity and uncertainty inherent in iris images, resulting in reduced model performance on unknown datasets and insufficient cross-domain performance across devices or databases.

[0086] Based on the above research, the embodiments of the present disclosure provide an iris recognition method, apparatus, medium and equipment based on uncertainty modeling. The stable features of the iris image are extracted based on a pre-built iris recognition model, and a dynamic feature mask is used to cope with environmental changes. The model is then matched with a preset iris template database to ensure that the target identity information can still be accurately identified under complex conditions.

[0087] In the disclosed embodiment, by combining the feature template and dynamic feature mask of the iris image, the accuracy and robustness of iris recognition can be effectively improved, especially when facing different lighting, angle and other changing conditions, a high recognition accuracy can still be maintained.

[0088] To facilitate understanding of this embodiment, the execution subject of the iris recognition method based on uncertainty modeling provided by the embodiment of the present disclosure is first introduced in detail. The execution subject of the iris recognition method based on uncertainty modeling provided by the embodiment of the present disclosure is a computer device. The computer device can be a terminal device or a server. Among them, the terminal device can also be a mobile device, a user terminal, a terminal, a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. The server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud storage, big data and artificial intelligence platforms. Optionally, the method can also be applied to an implementation environment composed of computer devices and servers.

[0089] The following describes in detail the iris recognition method based on uncertainty modeling provided by the embodiment of the present application in conjunction with the accompanying drawings. Figure 1 FIG. 1 is a flow chart of an iris recognition method based on uncertainty modeling provided by an embodiment of the present disclosure. The method includes the following steps S101 to S105:

[0090] S101, acquiring iris image data.

[0091] It is understood that the iris is a disc-shaped membrane located between the cornea and the lens of the eye, with unique texture characteristics. Iris image data is an image containing iris texture information acquired by a specific image acquisition device. These images can be two-dimensional color or grayscale images, capable of reflecting detailed features such as the iris' texture and color distribution. An image acquisition device is a hardware device used to capture iris images, with iris cameras being a common example. Iris cameras offer high resolution, high contrast, and fast imaging speed, enabling accurate iris image capture under varying lighting conditions.

[0092] For example, referring to Figure 2 As shown, in order to improve the accuracy and stability of the iris recognition system and reduce the recognition error rate caused by poor iris image quality or inaccurate feature extraction, the present disclosure also proposes preprocessing the collected iris image, which can include the following steps S201 to S203:

[0093] S201: Acquire an initial iris image, and perform image segmentation on the initial iris image to obtain an initial iris binary mask.

[0094] Here, the initial iris image refers to the original iris image directly obtained by an iris image acquisition device (such as a professional iris camera). For example, in an airport security check scenario, a passenger stands in front of an iris recognition device, and the iris camera captures the passenger's iris image. Or, in an access control system, when a user approaches the access control device, the iris camera starts to obtain the user's iris image. The image may be affected by various factors, such as lighting conditions, eye movement, eyelash occlusion, etc., resulting in unclear edges of the iris area, noise, or interference from non-iris areas. After obtaining the initial iris image, an image segmentation algorithm can be used to segment the image to obtain an initial iris binary mask for separating the iris area from the non-iris area. Image segmentation is the process of separating the iris area in the initial iris image from the background and other non-iris areas.

[0095] In this disclosure, the initial iris image is segmented using existing image segmentation methods (e.g., the IrisParseNet algorithm). This algorithm, based on deep learning technology, automatically learns the characteristics of iris images and accurately locates the circular boundaries of the iris and pupil. Specifically, the algorithm analyzes the initial iris image, identifies the location and shape of the iris region, and distinguishes it from the surrounding area.

[0096] For example, after automatically obtaining the iris region by segmenting the image using the above method, manual review and adjustment of the machine segmentation results may be performed, as the machine segmentation results may contain mis-segments or incomplete boundaries. Based on their professional knowledge and experience, the manual reviewer will check the accuracy of the segmentation results, correct any mis-segmented areas, and supplement and improve any incomplete boundaries to prevent poor masking from affecting subsequent feature extraction and recognition quality.

[0097] For example, in a bank's iris recognition system, when a customer authenticates their identity, the iris camera captures an initial iris image. The system then segments this image using the IrisParseNet algorithm, generating a binary mask of the iris region. However, because the customer was wearing glasses, part of the iris region was obscured, resulting in an incorrect segmentation. A human reviewer discovered this issue and manually adjusted the segmentation boundaries to ensure the integrity and accuracy of the iris region.

[0098] S202 : performing edge fitting processing on the iris outer boundary and the pupil boundary in the initial iris binary mask based on an edge detection algorithm to obtain a circular iris region.

[0099] Specifically, an edge detection algorithm is used to detect the edges of objects in an image. In iris image preprocessing, edge detection algorithms can accurately identify the pixel locations of the outer iris and pupil boundaries. After detecting the pixel locations of the outer iris and pupil boundaries, a fitting algorithm (such as the least squares method) can be used to fit these discrete boundary points into a circular shape. Since the iris and pupil are normally approximately circular, boundary fitting can produce a more accurate and regular circular iris region, facilitating subsequent feature extraction and comparison.

[0100] Here, this disclosure uses the Hough circle detection algorithm to fit circles to edge points detected in the initial binary iris mask. The Hough circle detection algorithm is a voting-based circle detection method that determines the circle's parameters by voting on the circle center and radius in parameter space and finding the parameter combination with the highest number of votes. This method has high accuracy and robustness in detecting circular boundaries and can effectively handle noise and occlusion issues in images.

[0101] In some other embodiments, the edge detection algorithm may also use a Sobel operator, a Canny operator, or the like. These algorithms can determine the position and direction of an edge by calculating the gradient change of pixels in an image, and are not specifically limited here. In actual applications, a suitable edge detection algorithm can be selected based on specific needs and image characteristics.

[0102] S203: Map the circular iris area into the iris image data of a uniform size.

[0103] It is understandable that due to the differences in iris size between individuals and the possible differences in shooting distance and angle during the acquisition process, the sizes of the segmented circular iris regions may be inconsistent. To facilitate subsequent feature extraction and comparison, the circular iris regions can be mapped to a uniform size. Typically, an interpolation algorithm (such as bilinear interpolation or bicubic interpolation) can be used to scale the circular iris regions to conform to a preset standard size.

[0104] For example, the present disclosure employs a rubber-sheet model to map a circular iris region into a normalized iris image of uniform size, eliminating scale and rotational deviations. The rubber-sheet model is a normalization method based on polar coordinate transformation. It converts the circular iris region from a rectangular coordinate system to a polar coordinate system, so that the iris texture appears as a regular texture pattern in the polar coordinate system. The iris texture in the polar coordinate system is then scaled and cropped according to a preset standard size to obtain a normalized iris image of uniform size. In this process, the polar coordinate transformation effectively eliminates rotational deviations in the iris image, ensuring that iris images captured at different angles have the same texture direction after normalization. Furthermore, the scaling and cropping operations eliminate scale differences, ensuring that iris images of different sizes have the same size after normalization. This normalization method can improve the accuracy and robustness of iris recognition and reduce recognition errors caused by differences in iris image size and angle.

[0105] Specifically, the uniformly sized circular iris region image obtained after the above preprocessing steps is the iris image data. This data contains the iris's primary texture features and structural information, which can be used in the subsequent iris recognition process. Thus, preprocessing the captured iris image through steps S201 to S203 can effectively improve the quality of the iris image and extract more accurate and stable iris features.

[0106] In some other embodiments, after obtaining the initial iris image, image preprocessing (e.g., denoising, contrast enhancement, etc.) may be performed to improve image quality. Accordingly, after obtaining the iris image data, enhancement processing may be performed to improve the quality of iris texture features and reduce noise interference.

[0107] S102: Input the iris image data into a pre-built iris recognition model.

[0108] It is understood that after obtaining iris image data, it can be input into a pre-built iris recognition model to achieve identity recognition based on the iris image data. The iris recognition model is a model constructed based on a deep learning algorithm in this disclosure, which is used to extract key features from iris images and perform identity recognition. This model is trained on a large amount of iris image data and is capable of learning the complex patterns and characteristics of iris texture. The iris recognition model includes a high-dimensional feature extraction module, an uncertainty modeling module, and a dynamic mask generation module.

[0109] Specifically, the high-dimensional feature extraction module can perform feature extraction and analysis on the image layer by layer to determine the high-dimensional iris features corresponding to the iris image data; the uncertainty extraction module is used to perform in-depth uncertainty extraction on the high-dimensional iris features to dig out the uncertain parts of the iris features (individual difference information); the dynamic mask generation module is used to generate a dynamic feature mask corresponding to the iris image data.

[0110] For example, after the iris image data is input into a pre-built iris recognition model, high-dimensional feature extraction can be performed on the iris image data based on the high-dimensional feature extraction module. Each layer of the network in the high-dimensional feature extraction model further abstracts and refines the features of the input iris image data, gradually extracting deeper and more essential feature information from the iris image. Through multiple layers of convolution, pooling and other operations, high-dimensional iris features can ultimately be obtained. These high-dimensional features contain rich information about the iris image, such as the direction of texture, the distribution of details, and the characteristics of local areas. It should be noted that the subsequent processing of the iris image data by the iris recognition model is based on the high-dimensional iris features of the iris image data.

[0111] Here, the implementation of the high-dimensional feature extraction module can be expressed as:

[0112] H=F θ (I iris )

[0113] Among them I iris is the iris image data, which is the input data of the model; F θ It is the backbone extraction network model used to extract image features, θ represents the set of trainable parameters of the network; H is the high-dimensional feature representation extracted from the input iris image. iris Input to the feature extraction network F θ In the above example, we obtain the deep image feature representation H.

[0114] S103, performing pixel-level modeling on the iris image data based on the uncertainty modeling module, extracting the mean feature and variance feature of each pixel in the iris image data, and generating an uncertainty feature representation corresponding to the iris image data based on the extraction result and a reparameter sampling strategy, and determining an iris feature template based on the uncertainty feature representation.

[0115] Specifically, the uncertainty modeling module extracts pixel-level statistical features from the iris' high-dimensional features, quantifying the deterministic components and noise interference in the iris texture. The uncertainty modeling module includes a mean extraction submodule and a variance extraction submodule. The mean extraction module extracts a mean feature map for each pixel from the iris' high-dimensional features, while the variance extraction submodule extracts a variance feature map for each pixel from the iris' high-dimensional features. The mean feature map reflects the overall trend of the iris features, while the variance feature map reveals the degree of variation in these features across individuals.

[0116] Among them, the mean extraction submodule can extract local texture patterns, such as the fibrous, radial, or annular structures of the iris, through shallow convolutional layers (such as 3×3 convolution + ReLU activation). These patterns have high certainty and are key to identity recognition. The variance extraction submodule can learn the global noise distribution through global average pooling and 1×1 convolutional layers, such as pixel-level fluctuations caused by uneven lighting, eyelash occlusion, or imaging device noise. Through parallel computing, each pixel position can simultaneously obtain the mean vector (representing texture intensity) and variance vector (quantizing noise intensity), forming a statistical feature pair.

[0117] In addition, the uncertainty modeling module also includes a reparameterized sampling submodule, which is used to execute a reparameterized sampling strategy based on the mean feature map and variance feature map of each pixel to introduce perturbations, and generate a pixel-level uncertainty feature representation with distribution characteristics based on the execution results; and determine the iris feature template based on the uncertainty feature representation.

[0118] Specifically, after obtaining the mean feature map and the variance feature map, uncertainty features are added to them based on the reparameterized sampling submodule combined with the mean feature map and the variance feature map. In this disclosure, a method is adopted to combine standard Gaussian noise with the mean feature map and the variance feature map to generate an uncertainty feature representation with random perturbations that can characterize the uncertainty of iris features.

[0119] Here, for each pixel position, the uncertainty eigenvalue of the position can be calculated based on the mean eigenvalue and variance eigenvalue corresponding to the position, as well as a noise value sampled from a standard Gaussian distribution. The characteristic expression of the uncertainty feature can be expressed as:

[0120] S=P μ +ε×P σ ;

[0121] Among them, S represents the uncertainty feature representation; P μ Represented as mean feature map; P σ It is represented as a variance feature map; ε is represented as standard Gaussian noise.

[0122] It is understood that after obtaining the uncertain feature representation, it can be binarized by the binary feature template determination module, converting the uncertain feature representation into a concise and clear binary feature template to facilitate subsequent storage and matching operations. Here, the binarization process can use a fixed threshold method, including: by setting an appropriate threshold, mapping the pixel values ​​in the uncertain feature representation to 0 or 1, thereby obtaining a feature template composed of 0 and 1. This feature template retains the key information of the iris feature while simplifying the data representation.

[0123] In the disclosed embodiments, by constructing an uncertainty modeling module, the present disclosure is able to establish a Gaussian distribution feature representation for each pixel position. This feature representation not only includes the deterministic information of iris features (i.e., the information carried by the mean feature map), but also captures non-identity factors and local uncertainties that may be introduced during the acquisition process through the introduction of variance feature maps and standard Gaussian noise, enabling the model to perceive and adapt to uncertainties such as local noise, blur, and illumination changes in the input image. This design enables the model to more comprehensively consider various potential influencing factors when extracting iris features, thereby improving the robustness and accuracy of feature extraction.

[0124] In some other embodiments, a Bayesian convolutional neural network or an uncertainty estimation method based on Monte Carlo Dropout may also be used to realize the perception of uncertainty of input data in the feature extraction stage, which is not specifically limited here.

[0125] Here, the iris feature template is a representative feature set extracted from the iris image, which is used to uniquely identify a person's iris. Its expression can be a set of numerical vectors or a specific image feature description, which can reflect the main texture features and structural information of the iris.

[0126] S104 : generating a variance scaling map corresponding to the iris image data based on the dynamic mask generating module and the iris image data, and generating a dynamic feature mask corresponding to the iris image data based on the variance scaling map and a dynamic threshold determination strategy.

[0127] Here, to further improve the accuracy and robustness of iris image data processing, the iris image data can be normalized before generating the variance scaling map. The purpose of normalization is to map the pixel values ​​in the iris image to a uniform range, thereby eliminating the impact of external factors such as lighting and contrast on subsequent processing. Specifically, the iris image can be normalized using methods such as minimum-maximum normalization or Z-Score normalization, so that the grayscale value of each pixel is within a fixed range (e.g., between 0 and 1), thereby ensuring that iris image data collected from different sources or under different lighting conditions can be processed uniformly.

[0128] Furthermore, after the normalization process is completed, pixel-level grayscale statistical processing can be performed on the normalization results to obtain the global grayscale mean and global grayscale standard deviation of the iris image data. The global grayscale mean and global grayscale standard deviation reflect the overall grayscale distribution of the entire iris image, which can help distinguish stable iris features from dynamically changing areas affected by noise or environmental factors during the feature extraction process. For example, in an environment with unstable lighting conditions, the grayscale values ​​in the iris image may change significantly. At this time, by calculating the global grayscale mean and global grayscale standard deviation, it is possible to effectively identify which areas have undergone significant changes in features and appropriately adjust the weights of these areas to reduce their impact on the iris recognition results.

[0129] For example, referring to Figure 3 As shown in FIG. 1 , a dynamic feature mask determination method proposed in the present disclosure specifically includes the following S301 to S302:

[0130] S301 : performing convolution processing on the iris high-dimensional features based on the dynamic mask generation module, and determining a variance scaling map corresponding to the iris image data based on the convolution processing result.

[0131] It can be understood that the dynamic mask generation module will perform convolution processing on the high-dimensional features of the iris to extract its local feature information, and determine the variance scaling map based on the convolution processing results. The variance scaling map calculates the change in each local area, and then reflects the flexibility coefficient of different areas in the iris image when generating masks. The larger the value corresponding to each pixel in the variance scaling map, the higher the tolerance of the pixel position, and the smaller the value, the more strict constraints are required.

[0132] S302 : Determine the dynamic feature mask for the iris image data based on a dynamic mask generation formula, the global grayscale mean, the global grayscale standard deviation, and the variance scaling map.

[0133] Specifically, the dynamic feature mask dynamically adjusts iris features during the matching process to improve matching accuracy and robustness. By comprehensively considering the overall distribution of the iris' high-dimensional features (i.e., the global grayscale mean and global grayscale standard deviation) and local information (the variance scaling map), a dynamic feature mask is generated that adaptively adjusts the weights of iris features. During the matching process, the dynamic feature mask is dynamically weighted by element-by-element multiplication with the iris feature template, thereby more accurately locating and matching iris features.

[0134] Finally, based on the dynamic mask generation formula, the global grayscale mean, the global grayscale standard deviation, and the variance scaling map, the dynamic feature mask of the iris image data can be determined. The dynamic mask generation formula comprehensively considers the global statistical characteristics and local feature changes of the iris image. By setting a reasonable threshold and weight distribution strategy, a mask matrix that can adaptively adjust the weight of the iris features is generated. Here, the dynamic mask generation formula can be expressed as:

[0135]

[0136] Among them, M ij Indicates the value of the dynamic feature mask at the i-th and j-th pixel positions, which can be 0 or 1; I represents the grayscale value of the i-th and j-th pixel positions in the normalized iris image; μ Represents the global grayscale mean, I σ Represents the global grayscale standard deviation, A ij Represents the value of the i-th and j-th pixels in the variance scaling map.

[0137] Since iris images can be affected by factors such as lighting variations and eye movement during acquisition, causing certain features to dynamically change, dynamic feature masks can be used to identify and mark these dynamically changing feature regions. This effectively suppresses noise interference, highlights key information within iris features, and thus improves iris recognition performance. For example, in environments with significant variations in lighting intensity, dynamic feature masks can identify regions where features change due to these changes, preventing these variations from interfering with identification results.

[0138] In some other embodiments, when determining a dynamic feature mask, a feature-aware mask generation method based on an attention mechanism (e.g., channel attention or spatial attention) is also used to dynamically adjust pixel weights. Furthermore, the traditional fixed threshold can be replaced with an adaptive learning-based multi-layer perceptron (MLP) output, or a small Transformer structure can be used to perceive contextual information and generate a mask, without specific limitation here.

[0139] In the disclosed embodiment, by introducing a pixel-level uncertainty feature modeling mechanism and a dynamic threshold mask generation method, the model's adaptability to complex conditions such as noise, occlusion, and blur is improved, thereby enhancing the accuracy and robustness of iris recognition.

[0140] It should be noted that the iris recognition model proposed in the present disclosure is highly versatile and applicable to a variety of deep learning network frameworks. In addition to verifying the effect in a single model, the present invention also verifies its good versatility and adaptability by integrating the high-dimensional feature extraction module and the feature template determination module into different types of backbone networks. Through application practice in different architectural models (such as traditional residual network structures, dilated convolutional networks, etc.), the module proposed in the present disclosure does not need to modify the underlying network structure, can be flexibly integrated in a plug-in manner, and brings stable performance improvements. This design makes the module not only limited to a specific network structure, but also applicable to a variety of existing and future deep iris recognition frameworks. At the same time, experimental results show that the module integration process does not require complex adjustments or manual parameter adjustment, the model can converge in a short time, and the inference speed is almost unaffected. This good modular and pluggable design greatly simplifies the application deployment process, facilitating subsequent rapid application and large-scale promotion in a variety of device terminals and application scenarios.

[0141] S105 , obtaining a preset iris template database; and determining target identity information corresponding to the iris image data based on the preset iris template database, the iris feature template, and the dynamic feature mask to complete iris identity recognition.

[0142] It is understood that the preset iris template database is a database that stores iris feature templates for a large number of registered users and may include multiple iris templates. Each user's iris template is unique and is used to compare with the iris image data to be identified. The target identity information may be the user's name, ID number, permission level, etc. For example, in a company's access control system, the preset iris template database stores the iris feature templates of all employees. When an employee approaches the access control device, the system will obtain the employee's iris image data and extract the iris feature template and dynamic feature mask.

[0143] Here, the dynamic feature mask can dynamically weight the feature template according to the actual situation of the iris image data. It can adjust the feature template in real time according to the quality of the iris image and the actual situation to improve the accuracy of recognition. Furthermore, a comparison algorithm is used to compare the extracted feature template with the templates in the preset iris template database one by one. The comparison algorithm mainly determines whether they match based on the similarity between the iris feature templates. Generally, the similarity value between each iris template in the preset database and the extracted feature template can be calculated, and finally, by selecting the iris template with the highest similarity value, the target identity information corresponding to the iris image data is determined. For example, if the comparison result shows that the iris feature template of an employee has the highest similarity with the iris template pre-stored in the database, and its similarity value exceeds the set threshold, the identity information of the employee can be confirmed, and further, based on the identity information of the employee, it can be determined whether to allow him to enter a specific area.

[0144] In some possible embodiments, an iris feature template that has been dynamically weighted according to a dynamic feature mask may be used as a query template. The query template is then sequentially compared with the iris templates of each known identity in a preset iris template database for similarity, and similarity values ​​between the query template and the iris templates of each known identity are calculated. Finally, the target identity information corresponding to the iris image data is determined based on the calculation result (i.e., the iris template with the largest similarity value that exceeds a preset threshold).

[0145] Here, when calculating the similarity value between the query template and the iris templates of each known identity in the preset iris template database, the present disclosure uses the fractional Hamming distance matching method to calculate the minimum offset distance between the iris template and the feature template to achieve the calculation of the similarity value, which is expressed as follows:

[0146]

[0147] Among them, S1 represents the iris template, that is, the extracted iris feature matrix; S2 represents the feature template; FD represents the template resulting from horizontally shifting iris template S1 by b pixels. FD represents the fractional distance, which measures the difference between templates S1 and S2. S1 is shifted within a range, and the difference between it and S2 is calculated (using the fractional distance). The smallest difference is taken as the final matching score. The smaller the difference between the templates, the higher the match. This difference is calculated only within the valid pixel area (i.e., where the mask of both templates is 1), avoiding calculations of invalid areas (such as noise or occlusions).

[0148] In some other embodiments, cosine distance, log loss distance or KL divergence can also be used to construct a matching metric formula, or the translation matching process can be changed to a spatial convolution kernel sliding calculation to replace the traditional bit-shifting; no specific limitation is made here.

[0149] The iris recognition method, apparatus, medium, and equipment based on uncertainty modeling provided in the embodiments of the present disclosure can effectively improve the accuracy and robustness of iris recognition by combining the feature template and dynamic feature mask of the iris image, especially in the face of different lighting, angle and other changing conditions, while still maintaining a high recognition accuracy.

[0150] Those skilled in the art will understand that in the above-mentioned method of the specific implementation method, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process. The specific execution order of each step should be determined by its function and possible internal logic.

[0151] Based on the same inventive concept, the embodiments of the present disclosure also provide an iris recognition device based on uncertainty modeling corresponding to the iris recognition method based on uncertainty modeling. Since the principle of solving the problem by the device in the embodiments of the present disclosure is similar to the above-mentioned iris recognition method based on uncertainty modeling in the embodiments of the present disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be repeated.

[0152] Reference Figure 4 FIG. 4 is a schematic diagram of an iris recognition device 400 based on uncertainty modeling according to an embodiment of the present disclosure, wherein the device includes:

[0153] Data acquisition module 401, used to acquire iris image data;

[0154] A data input module 402 is configured to input the iris image data into a pre-built iris recognition model; wherein the iris recognition model includes an uncertainty modeling module and a dynamic mask generation module;

[0155] a template determination module 403 configured to perform pixel-level modeling on the iris image data based on the uncertainty modeling module, extract the mean feature and variance feature of each pixel in the iris image data, generate an uncertainty feature representation corresponding to the iris image data based on the extraction result and a reparameter sampling strategy, and determine an iris feature template based on the uncertainty feature representation;

[0156] a mask determination module 404 configured to generate a variance scaling map corresponding to the iris image data based on the dynamic mask generation module and the iris image data, and to generate a dynamic feature mask corresponding to the iris image data based on the variance scaling map and a dynamic threshold determination strategy;

[0157] The identity recognition module 405 is configured to obtain a preset iris template database and determine target identity information corresponding to the iris image data based on the preset iris template database, the iris feature template, and the dynamic feature mask to complete iris identity recognition.

[0158] In some possible embodiments, the data acquisition module 401 is specifically configured to:

[0159] Acquire an initial iris image, and perform image segmentation on the initial iris image to obtain an initial iris binary mask for separating the iris area from the non-iris area;

[0160] Performing edge fitting processing on the iris outer boundary and pupil boundary in the initial iris binary mask based on an edge detection algorithm to obtain a circular iris region;

[0161] The circular iris area is mapped into the iris image data of uniform size.

[0162] In some possible embodiments, the iris recognition model further includes a high-dimensional feature extraction module; and the data input module 402 is further configured to:

[0163] Performing high-dimensional feature extraction on the iris image data based on the high-dimensional feature extraction module to obtain iris high-dimensional features;

[0164] Accordingly, the template determination module 403 is specifically configured to:

[0165] Performing pixel-level modeling on the iris high-dimensional features based on the uncertainty modeling module;

[0166] Accordingly, the mask determination module 404 is specifically configured to:

[0167] A variance scaling map corresponding to the iris image data is generated based on the dynamic mask generation module and the iris high-dimensional features.

[0168] In some possible embodiments, the uncertainty modeling module includes a mean extraction submodule, a variance extraction submodule, and a reparameterization sampling submodule;

[0169] The mean extraction submodule is used to extract the mean feature map of each pixel from the iris high-dimensional features;

[0170] The variance extraction submodule is used to extract the variance feature map of each pixel from the iris high-dimensional features;

[0171] The reparameterized sampling submodule is used to execute the reparameterized sampling strategy according to the mean feature map and the variance feature map of each pixel to introduce disturbance, and generate the pixel-level uncertainty feature representation with distribution characteristics based on the execution result; and determine the iris feature template based on the uncertainty feature representation.

[0172] In some possible embodiments, the mask determination module 404 is further configured to:

[0173] Normalization processing is performed on the iris image data, and pixel-level grayscale statistical processing is performed on the normalization processing result to obtain a global grayscale mean and a global grayscale standard deviation of the iris image data.

[0174] In some possible embodiments, the mask determination module 404 is specifically configured to:

[0175] performing convolution processing on the iris high-dimensional features based on the dynamic mask generation module, and determining a variance scaling map corresponding to the iris image data based on the convolution processing result;

[0176] Determining the dynamic feature mask for the iris image data based on a dynamic mask generation formula, the global grayscale mean, the global grayscale standard deviation, and the variance scaling map;

[0177] The dynamic mask generation formula is expressed as:

[0178]

[0179] Among them, M ij Indicates the value of the dynamic feature mask at the i-th and j-th pixel positions, which can be 0 or 1; I represents the grayscale value of the i-th and j-th pixel positions in the normalized iris image; μ Represents the global grayscale mean, I σ Represents the global grayscale standard deviation, A ij Represents the value of the i-th and j-th pixels in the variance scaling map.

[0180] In some possible embodiments, the preset iris template database includes multiple iris templates of known identities; the identity recognition module 405 is specifically configured to:

[0181] performing dynamic weighting processing on the iris feature template based on the dynamic feature mask, and using the iris feature template after the dynamic weighting processing as a query template;

[0182] Comparing the query template with the iris templates of each known identity in the preset iris template database in sequence, and calculating the similarity values ​​between the query template and the iris templates of each known identity in the preset iris template database respectively;

[0183] The identity information corresponding to the iris template with the largest similarity value that exceeds a preset threshold is determined as the target identity information corresponding to the iris image data.

[0184] Based on the same technical concept, the embodiment of the present disclosure also provides a computer device. Figure 5 , which is a schematic diagram of the structure of a computer device 500 provided in an embodiment of the present disclosure, includes a processor 501, a memory 502, and a bus 503. The memory 502 is used to store execution instructions and includes a memory 5021 and an external memory 5022. The memory 5021 is also referred to as internal memory and is used to temporarily store operation data in the processor 501 and data exchanged with an external memory 5022 such as a hard disk. The processor 501 exchanges data with the external memory 5022 through the memory 5021.

[0185] In the embodiment of the present application, the memory 502 is specifically used to store application code for executing the solution of the present application, and the execution is controlled by the processor 501. That is, when the computer device 500 is running, the processor 501 communicates with the memory 502 via the bus 503, so that the processor 501 executes the application code stored in the memory 502, thereby performing the method described in any of the aforementioned embodiments.

[0186] Among them, the memory 502 can be, but is not limited to, random access memory (RAM), read only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0187] The processor 501 may be an integrated circuit chip with signal processing capabilities. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The various methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0188] It should be understood that the structure illustrated in the embodiments of the present application does not constitute a specific limitation on the computer device 500. In other embodiments of the present application, the computer device 500 may include more or fewer components than shown, or may combine or separate certain components, or arrange the components differently. The components shown in the illustrations may be implemented in hardware, software, or a combination of software and hardware.

[0189] The present disclosure also provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program executes the steps of the iris recognition method based on uncertainty modeling described in the above method embodiment. The storage medium may be a volatile or non-volatile computer-readable storage medium.

[0190] The embodiments of the present disclosure also provide a computer program product that carries program code. The instructions included in the program code can be used to execute the steps of the iris recognition method based on uncertainty modeling described in the above method embodiment. For details, please refer to the above method embodiment and will not be repeated here.

[0191] The computer program product may be implemented in hardware, software, or a combination thereof. In one embodiment, the computer program product is implemented as a computer storage medium. In another embodiment, the computer program product is implemented as a software product, such as a software development kit (SDK).

[0192] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system and device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the several embodiments provided in the present disclosure, it should be understood that the disclosed system and method can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0193] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0194] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0195] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present disclosure, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0196] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present disclosure, which are used to illustrate the technical solutions of the present disclosure, rather than to limit them. The scope of protection of the present disclosure is not limited thereto. Although the present disclosure has been described in detail with reference to the above-described embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-described embodiments within the technical scope disclosed in the present disclosure, or replace some of the technical features therein with equivalents. Such modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be included in the scope of protection of the present disclosure. Therefore, the scope of protection of the present disclosure shall be subject to the scope of protection of the claims.

Claims

1. An iris recognition method based on uncertainty modeling, characterized in that: include: Obtain iris image data; Inputting the iris image data into a pre-built iris recognition model; wherein the iris recognition model includes an uncertainty modeling module and a dynamic mask generation module; Performing pixel-level modeling on the iris image data based on the uncertainty modeling module, extracting mean features and variance features of each pixel in the iris image data, generating an uncertainty feature representation corresponding to the iris image data based on the extraction results and a re-parameter sampling strategy, and determining an iris feature template based on the uncertainty feature representation; generating a variance scaling map corresponding to the iris image data based on the dynamic mask generation module and the iris image data, and generating a dynamic feature mask corresponding to the iris image data based on the variance scaling map and a dynamic threshold determination strategy; Acquire a preset iris template database; and determine target identity information corresponding to the iris image data based on the preset iris template database, the iris feature template, and the dynamic feature mask to complete iris identity recognition.

2. The method according to claim 1, characterized in that The obtaining of iris image data comprises: Acquire an initial iris image, and perform image segmentation on the initial iris image to obtain an initial iris binary mask for separating the iris area from the non-iris area; Performing edge fitting processing on the iris outer boundary and pupil boundary in the initial iris binary mask based on an edge detection algorithm to obtain a circular iris region; The circular iris area is mapped into the iris image data of uniform size.

3. The method according to claim 1, characterized in that The iris recognition model further includes a high-dimensional feature extraction module; after the iris image data is input into the pre-built iris recognition model, the method further includes: Performing high-dimensional feature extraction on the iris image data based on the high-dimensional feature extraction module to obtain iris high-dimensional features; Accordingly, performing pixel-level modeling on the iris image data based on the uncertainty modeling module includes: Performing pixel-level modeling on the iris high-dimensional features based on the uncertainty modeling module; Accordingly, the generating of a variance scaling map corresponding to the iris image data based on the dynamic mask generating module and the iris image data includes: A variance scaling map corresponding to the iris image data is generated based on the dynamic mask generation module and the iris high-dimensional features.

4. The method according to claim 3, characterized in that The uncertainty modeling module includes a mean extraction submodule, a variance extraction submodule and a reparameterization sampling submodule; The mean extraction submodule is used to extract the mean feature map of each pixel from the iris high-dimensional features; The variance extraction submodule is used to extract the variance feature map of each pixel from the iris high-dimensional features; The reparameterized sampling submodule is configured to execute the reparameterized sampling strategy according to the mean feature map and the variance feature map of each pixel to introduce disturbances, and generate the pixel-level uncertainty feature representation with distribution characteristics based on the execution result; And, determining the iris feature template based on the uncertain feature representation.

5. The method according to claim 3, characterized in that Before generating a variance scaling map corresponding to the iris image data based on the dynamic mask generation module and the iris high-dimensional features, the method includes: Normalization processing is performed on the iris image data, and pixel-level grayscale statistical processing is performed on the normalization processing result to obtain a global grayscale mean and a global grayscale standard deviation of the iris image data.

6. The method according to claim 5, characterized in that The generating of a variance scaling map corresponding to the iris image data based on the dynamic mask generating module and the iris image data, and generating a dynamic feature mask corresponding to the iris image data based on the variance scaling map and a dynamic threshold determination strategy, includes: performing convolution processing on the iris high-dimensional features based on the dynamic mask generation module, and determining a variance scaling map corresponding to the iris image data based on the convolution processing result; Determining the dynamic feature mask for the iris image data based on a dynamic mask generation formula, the global grayscale mean, the global grayscale standard deviation, and the variance scaling map; The dynamic mask generation formula is expressed as: Among them, M ij Indicates the value of the dynamic feature mask at the i-th and j-th pixel positions, which can be 0 or 1; I represents the grayscale value of the i-th and j-th pixel positions in the normalized iris image; μ Represents the global grayscale mean, I σ Represents the global grayscale standard deviation, A ij Represents the value of the i-th and j-th pixels in the variance scaling map.

7. The method according to claim 6, characterized in that The preset iris template database includes a plurality of iris templates of known identities; and determining the target identity information corresponding to the iris image data based on the preset iris template database, the iris feature template, and the dynamic feature mask includes: performing dynamic weighting processing on the iris feature template based on the dynamic feature mask, and using the iris feature template after the dynamic weighting processing as a query template; Comparing the query template with the iris templates of each known identity in the preset iris template database in sequence, and calculating the similarity values ​​between the query template and the iris templates of each known identity in the preset iris template database respectively; The identity information corresponding to the iris template with the largest similarity value exceeding a preset threshold is determined as the target identity information corresponding to the iris image data.

8. An iris recognition device based on uncertainty modeling, characterized in that: include: A data acquisition module, used for acquiring iris image data; A data input module, configured to input the iris image data into a pre-built iris recognition model; wherein the iris recognition model includes an uncertainty modeling module and a dynamic mask generation module; a template determination module, configured to perform pixel-level modeling on the iris image data based on the uncertainty modeling module, extract mean features and variance features of each pixel in the iris image data, generate an uncertainty feature representation corresponding to the iris image data based on the extraction results and a reparameter sampling strategy, and determine an iris feature template based on the uncertainty feature representation; a mask determination module, configured to generate a variance scaling map corresponding to the iris image data based on the dynamic mask generation module and the iris image data, and to generate a dynamic feature mask corresponding to the iris image data based on the variance scaling map and a dynamic threshold determination strategy; The identity recognition module is used to obtain a preset iris template database; and determine the target identity information corresponding to the iris image data based on the preset iris template database, the iris feature template and the dynamic feature mask to complete iris identity recognition.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images