Iris biological characteristic comparison method based on three-dimensional fold embedding matching

Abstract

The invention discloses an iris biological feature comparison method based on three-dimensional fold embedding matching, and relates to the technical field of biological feature recognition, and the method comprises the following components: S1, iris image acquisition preparation, S2, iris segment acquisition, S3, three-dimensional structure extraction, S4, three-dimensional fold embedding matching, and S5, matching result output. According to the method, through the three-dimensional wrinkle embedding matching technology, the three-dimensional structure information, including the three-dimensional coordinates, the wave crest position, the wave trough position and the like of the wrinkles, of the iris wrinkles is deeply utilized for accurate matching, compared with a traditional two-dimensional iris recognition method, the three-dimensional structure information provides richer feature dimensions, and the recognition accuracy is improved. According to the method, the recognition accuracy and reliability are remarkably improved, particularly when tiny differences and complex textures in the iris image are processed, the features can be captured more accurately through three-dimensional wrinkle embedding matching, and therefore more accurate identity verification is achieved.

Description

Technical Field

[0001] This invention relates to the field of biometric recognition technology, specifically to an iris biometric comparison method based on three-dimensional fold embedding matching. Background Technology

[0002] With the rapid development of biometric identification technology, iris recognition, as a high-precision, contactless identity verification method, has been widely used in many fields such as security authentication, access control systems, and financial payments.

[0003] Traditional two-dimensional iris recognition technology has several drawbacks: First, two-dimensional images are easily affected by factors such as changes in lighting, occlusion, and image blurring, leading to a decrease in recognition accuracy. Second, two-dimensional images cannot fully reflect the three-dimensional structural information of the iris, such as the depth and curvature of folds, which can play a crucial role in distinguishing different individuals. In addition, traditional methods often fail to achieve ideal recognition results when dealing with subtle differences and complex textures in iris images, limiting their application in high-security scenarios.

[0004] In view of the problems of insufficient accuracy and reliability of traditional two-dimensional iris recognition technology, the iris biometric comparison method based on three-dimensional fold embedding matching proposed in this invention is particularly important. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an iris biometric matching method based on three-dimensional fold embedding matching. It can extract the three-dimensional structural information of the iris, including the three-dimensional coordinates of the folds and the position information of the peaks and troughs, and perform refined comparison analysis. Compared with traditional methods, this invention makes full use of the three-dimensional features of the iris, which significantly improves the accuracy and reliability of recognition. At the same time, through multi-scenario adaptive processing technology, this invention can adjust the matching strategy and output results according to different application scenarios, which enhances the practicality and flexibility of the method.

[0006] To solve the above-mentioned technical problems, this invention provides the following technical solution: an iris biometric comparison method based on three-dimensional fold embedding matching, the specific steps of which are as follows:

[0007] S1. Iris Image Acquisition Preparation: Before acquiring iris images, calibrate the acquisition device, prompt the user to cooperate with the acquisition operation, and select iris images as objects for subsequent processing.

[0008] S2. Iris fragment acquisition: Acquire iris images of the two users to be compared, and extract at least one iris fragment from each iris image. The iris fragment is a continuous region with a complete fold structure within the iris area.

[0009] S3. Three-dimensional structure extraction: Three-dimensional data acquisition is performed on each iris segment obtained in step S1, and the three-dimensional structure information of the iris segment is extracted. The three-dimensional structure information includes the three-dimensional coordinates of the iris folds, the position information of the fold peaks and the position information of the fold troughs.

[0010] S4. Three-dimensional fold embedding matching: Iris segments of two users are denoted as iris segment A and iris segment B, respectively. Iris segments A and B are aligned front to back, so that the peak of iris segment A extends into the trough of iris segment B, and at the same time, the peak of iris segment B extends into the trough of iris segment A. The embedding matching degree is calculated, which is the degree of fit between the peak and the corresponding trough. During the alignment process, one iris segment is controlled to translate left and right relative to the other iris segment. The embedding matching degree is recalculated after each translation, and the embedding matching degree values ​​at all translation positions are recorded.

[0011] S5. Matching result output: Select the maximum value from all the embedding matching degree values ​​recorded in step S3 as the final iris matching degree of the two users. When the final iris matching degree is 100%, it is determined that the iris segments of the two users have achieved perfect embedding matching.

[0012] Furthermore, in step S1, the iris image acquisition preparation employs a collaborative optimization technique combining device calibration and user guidance. Specifically, the lens focal length of the acquisition device is first calibrated in three stages by photographing a standard resolution plate, adjusting the near-focus, mid-focus, and far-focus positions respectively to ensure that the image resolution error at each focal length is ≤2%. Next, the depth measurement accuracy is calibrated by using a standard glass plate of known thickness and capturing 10 sets of data under different lighting conditions. Linear regression is used to correct the depth measurement values, ensuring that the depth error is ≤0.005mm. Regarding user guidance, the device's built-in infrared sensor detects the user's head position in real time; if the head deviates from the optimal acquisition position... When adjusting distance, the system uses voice prompts to guide the user. If an obstruction is detected in the user's eyes, an image recognition algorithm identifies it and displays a text prompt to remove the obstruction. During the acquisition process, real-time frame rate monitoring and blink detection are used. When a blink is detected, the frame is automatically skipped and recorded. After acquisition, 3-5 frames of images for each user are scored for sharpness and motion blur is detected. Images that meet both conditions are selected for processing. This technical step reduces acquisition errors from the source through multi-level device calibration and real-time user guidance. It supplements the calibration details and guidance mechanism not mentioned in the main claim and provides high-quality raw images for subsequent processing.

[0013] Furthermore, the iris segment extraction process in step S2 employs multi-level optimization techniques. Specifically, the acquired iris image is first grayscaled, followed by Gaussian filtering, bilateral filtering, and morphological opening operations to achieve multi-level denoising. Gaussian filtering smooths high-frequency noise, bilateral filtering preserves the iris texture edges while removing low-frequency noise, and morphological opening operations eliminate minor dark spot interference. Subsequently, an iris region segmentation algorithm based on an active contour model separates the iris from the sclera and pupil regions. This algorithm dynamically adjusts the elastic coefficient and curvature weight of the contour curve to adaptively fit the iris edge, while precisely removing eyelid and eyelash-occluded areas using a combination of grayscale thresholding and edge gradient. Based on iris texture... For grayscale variance distribution, continuous regions with a grayscale variance ≥ 50 are selected as candidate segments. Candidate segments must meet the requirement that the continuity of folds in both the radial and circumferential directions of the iris is ≥ 85%. Then, candidate segments with an entropy value ≥ 7.5 are selected through image sharpness evaluation index. Next, the feature consistency of adjacent frames is verified to remove distorted segments caused by slight user shaking. The final determined iris segments must pass the fold integrity verification in at least 3 independent regions to ensure that the fold structure in the segment is unbroken and without missing parts. This technical step improves image purity through multi-level noise reduction and obtains high-quality iris segments through accurate segmentation and multi-dimensional screening, providing a reliable data foundation for subsequent 3D structure extraction, and making up for the deficiency of the sovereign item which only mentions segment truncation without specifying optimization techniques.

[0014] Furthermore, in step S3, the 3D structure extraction process incorporates a wrinkle feature saliency scoring algorithm to filter wrinkle regions with strong discriminative power. The formula is as follows: ,in The significance score for wrinkle features ranges from 0 to 10. Regions with a score ≥ 6.5 are considered to be highly discriminative regions. , , The weighting coefficients are 0.4, 0.3, and 0.3, respectively. The weights were determined using the analytic hierarchy process (AHP). Ten experts in biometrics were invited to conduct pairwise comparisons of the importance of the three indicators: wrinkle density, continuity, and depth variance. A judgment matrix was constructed, and the weight allocation was determined after passing a consistency test. The fold density of the target area. The average fold density of the entire iris region. The continuous length of the folds within the target area. The maximum diameter of the target area. For the first The depth value of each fold sampling point This represents the average depth value of all fold sampling points within the target area. The formula, which integrates the density advantage, structural continuity, and depth difference of folds to accurately select the fold regions that best reflect the uniqueness of the iris, solves the problem of unclear fold feature selection criteria in the sovereign term, improves the distinguishability and reliability of three-dimensional structural information, and the parameters in the formula are all defined based on the physiological characteristics of iris biometrics. The weight determination process is scientifically based and creative.

[0015] Furthermore, the 3D data optimization process in step S3 employs a multi-level calibration technique, specifically implemented as follows: First, statistical filtering is performed on the collected 3D point cloud data to remove isolated noise points. The number of neighboring points is set to 15. When the average distance between a point and its neighboring points exceeds three times the standard deviation, it is determined to be an isolated point and removed. Then, a point cloud registration algorithm based on plane fitting is used. Using the feature points of the iris and cornea edge as a reference, a local reference plane is constructed, and the point cloud data is projected onto this plane for preliminary calibration. Next, an improved version of the iterative nearest point algorithm is used for precise registration. The number of iterations is set to 150, and the registration error threshold is set to 0.003mm. After registration, coordinate system calibration is performed on the 3D coordinate data. With the pupil center as the origin and the normal vector of the iris surface as the Z-axis, the iris equatorial plane is fitted using the least squares method as the XY plane to ensure that the iris 3D data of different users are in a unified coordinate system. Finally, layered outlier detection is performed, first detecting depth values ​​exceeding the standard value. The point, among which The standard deviation of the depth value is used to detect curvature abrupt changes in the remaining points. After double verification, outliers are removed. This technical step significantly improves the accuracy and consistency of 3D structural information through multi-stage denoising, weighted registration and precise calibration. It also supplements the 3D data optimization details not mentioned in the sovereign term, and provides a high-quality feature foundation for subsequent embedding and matching.

[0016] Furthermore, in step S4, the front-to-back alignment employs a multi-reference calibration technique using a three-dimensional coordinate system. Specifically, this involves: first, constructing an initial local coordinate system using the geometric center of iris segment A as the origin, and simultaneously combining three feature points—the pupil center and the corneal edge—to fit these four points using the least squares method, forming a stable coordinate system reference and avoiding coordinate system offset caused by a single origin; when mapping the three-dimensional data of iris segment B to this coordinate system, first, extracting the principal eigenvector of iris segment B through principal component analysis and aligning it with the principal eigenvector of iris segment A, ensuring the angle between their principal directions is ≤5°; then, adjusting the rotation angle of iris segment B and calculating the rotation matrix based on the extension direction vector of the folds. To ensure that the angle between the fold extension directions of the two is ≤8°, an incremental rotation method is used during rotation, and the angle is calculated in real time until the requirement is met. When adjusting the fitting distance, the distance in the Z-axis direction is first reduced from 1mm to 0.2mm at a speed of 1mm / s, and then slowly reduced at a speed of 0.2mm / s. At the same time, the distance between the peak and the trough is monitored in real time. When the distance is ≤0.05mm, the adjustment is stopped, and the fitting position at this time is recorded as the initial embedding position. This technical step improves the accuracy and stability of fitting alignment through the construction of multiple reference coordinate systems, alignment of principal feature vectors, and incremental rotation adjustment. It also supplements the alignment details not mentioned in the main claim and lays a precise positional foundation for embedding matching.

[0017] Furthermore, the calculation of the embedding matching degree in step S4 adopts a multi-dimensional fit formula, which is: ,in The embedding matching degree ranges from 0% to 100%. , , The weighting coefficients are 0.4, 0.3, and 0.3, respectively. These weights were determined through training with 1000 sets of real iris sample data. The least squares method was used to minimize the error between the calculated matching degree and the manually labeled fit, ultimately controlling the error to within 3%, ensuring that the weight allocation conforms to the actual matching scenario. The actual overlap depth between the wave crest and trough, i.e., the effective depth by which the wave crest extends into the trough, is obtained through distance calculation from the 3D point cloud. Iris segment The maximum height of the wave crest and the iris segment The larger of the maximum depths of the troughs Iris segment The average curvature of the wave crest surface, Iris segment The average curvature of the inner wall of the trough, The actual contact area between the wave crest and trough is calculated using the overlapping region of the 3D point cloud. The theoretical maximum contact area between the crest and trough is the smaller of the surface area of ​​the crest and the inner surface area of ​​the trough. This formula breaks through the limitations of traditional single-dimensional matching and integrates three core indicators: depth matching, curvature matching, and area coverage. Each parameter is defined based on the physical characteristics of the three-dimensional structure, and the weights are optimized through training with a large number of samples. It has significant creativity, solves the problem of the sovereign term not having a clear logic for calculating the degree of matching, and improves the accuracy and objectivity of the matching results.

[0018] Furthermore, the optimal step size for left and right translation in step S4 is calculated using a dynamic adjustment formula, which is: ,in This is the dynamic translation step size, ranging from 1 to 8 pixels. The calculated result is rounded to the nearest integer. The adjustment coefficient, set to 1.5, was determined through orthogonal experiments. Experimental factors included resolution, fold density, and initial matching degree, with three levels. The coefficient was determined through range analysis. The optimal value, This refers to the resolution of the iris segment, specifically the number of pixels per millimeter. The density of folds in the iris segment. The core innovation of this formula for the initial embedding matching degree lies in the dynamic adjustment of the step size based on resolution, wrinkle density, and initial matching degree: higher resolution and greater wrinkle density require a smaller step size to ensure matching accuracy; higher initial matching degree requires a smaller step size for fine-tuning the search for the optimal position, and vice versa, a slightly larger step size to improve search efficiency. During translation, the step size is recalculated after every 5 translations and dynamically adjusted according to the current trend of matching degree changes. If the matching degree improvement rate of 3 consecutive translations is ≥5%, the step size is reduced by 20%; if the improvement rate is ≤1%, the step size is increased by 10%, ensuring an optimal balance between accuracy and efficiency. This formula solves the contradiction between accuracy and efficiency caused by a fixed step size for the main term, demonstrating creativity, and also supplements the dynamic adjustment mechanism during the translation process, improving the level of intelligent matching.

[0019] Furthermore, in step S5, the matching result output adopts multi-scenario adaptive processing technology. Specifically, in the fun matching scenario, in addition to classifying matching levels, personalized matching tags are generated based on the matching degree value. A matching degree of 90-100% corresponds to the "soulmate" tag, 80-89% corresponds to the "high-dimensional compatibility" tag, 70-79% corresponds to the "tacit compatibility" tag, 60-69% corresponds to the "fate compatibility" tag, and below 60% corresponds to the "fun encounter" tag. At the same time, the dynamic demonstration animation adds highlighted annotations with embedded peaks and troughs and a matching degree change curve. Users can adjust the animation playback speed and perspective through their terminal devices. In the anti-fraud scenario, the matching threshold adopts a dynamic adjustment mechanism, set at 85%, 90%, and 95% respectively according to the fraud risk level of the application scenario. The risk level is determined using fraud case data from the local public security department over the past three months. When the matching degree is greater than or equal to the threshold, in addition to generating an early warning signal, key data from the comparison process is automatically extracted to generate a verification report. Encrypted storage uses the national cryptographic algorithm SM4, and the storage medium uses a distributed database, supporting hierarchical access control. In biometric rebuttal scenarios, the output results include the final matching degree, the confidence interval of the matching process, and the comparison details of key feature points. At the same time, a rebuttal report is generated, clearly stating the basis for verifying the uniqueness of the iris biometric feature. The report can be directly used in judicial or administrative verification scenarios. This technical step expands the applicability of the method through personalized processing in multiple scenarios, supplements the output details of the results not mentioned in the sovereign claims, and improves the practicality and flexibility of the method.

[0020] Compared with existing technologies, this iris biometric matching method based on three-dimensional fold embedding matching has the following advantages:

[0021] I. This method utilizes three-dimensional fold embedding matching technology to make in-depth use of the three-dimensional structural information of iris folds, including the three-dimensional coordinates, peak positions, and trough positions of the folds, for precise matching. Compared with traditional two-dimensional iris recognition methods, the three-dimensional structural information provides richer feature dimensions, significantly improving the accuracy and reliability of recognition. Especially when dealing with subtle differences and complex textures in iris images, three-dimensional fold embedding matching can more accurately capture these features, thereby achieving more accurate identity verification.

[0022] Second, this method employs multi-scenario adaptive processing technology, which can adjust the matching strategy and output results according to different application scenarios. For example, in a fun matching scenario, the system can generate personalized matching tags and dynamic demonstration animations to increase the user experience; in an anti-fraud scenario, the system can dynamically adjust the matching threshold and generate warning signals and verification reports when the matching degree exceeds the threshold, effectively preventing fraudulent behavior; in a biometric counter-evidence scenario, the system can provide detailed matching processes and key feature point comparison details, which can be directly used for judicial or administrative verification. This flexibility makes the method widely applicable to various security authentication and identity verification scenarios.

[0023] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0025] Figure 1 This is a flowchart of an iris biometrics comparison method based on 3D fold embedding matching.

[0026] Figure 2 A flowchart for determining the subdivision of index weights in an iris biometrics comparison method based on 3D fold embedding matching;

[0027] Figure 3 This is a flowchart of the TOPSIS health assessment process, a biometric matching method based on 3D fold embedding matching. Detailed Implementation

[0028] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0029] Example 1

[0030] In a fun and interactive scenario at a friend's gathering, a portable iris recognition device is used to capture images. Before acquisition, the device's lens focal length is calibrated in three stages by shooting a standard resolution board to adjust the near, medium, and far focal lengths, ensuring that the image resolution error at each focal length is ≤2%. Next, depth measurement accuracy is calibrated using a standard glass plate of known thickness. Ten sets of data are captured under different lighting conditions, including indoor lighting and natural light, and the depth measurement values ​​are corrected using linear regression. For user guidance, the device's built-in infrared sensor detects the head positions of the two participants in real time. When the head deviates from the optimal acquisition distance of 30-50cm, a voice prompt is issued through the device's built-in speaker: "Please adjust the distance to 30-50cm." If obstructions such as hair or reflections from glasses are detected, an image recognition algorithm identifies the obstruction and displays a text prompt on the device's screen asking the user to remove it. During the acquisition process, real-time frame rate monitoring and blink detection are used. When a user blinks, the frame is automatically skipped and recorded. After the acquisition is completed, 3-5 frames of images for each user are scored for sharpness and motion blur is detected. Images that meet both the sharpness standard and the motion blur are selected as the objects for subsequent processing.

[0031] After acquiring iris images from two friends, the images were first converted to grayscale. Then, Gaussian filtering, bilateral filtering, and morphological opening were sequentially performed to achieve multi-level denoising. Gaussian filtering smoothed high-frequency noise, bilateral filtering removed low-frequency noise while preserving the edges of the iris texture, and morphological opening eliminated small dark spot interference. Subsequently, an iris region segmentation algorithm based on an active contour model was used to accurately separate the iris from the sclera and pupil regions. Based on the grayscale variance distribution of the iris texture, continuous regions with a grayscale variance ≥ 50 were selected as candidate segments. Candidate segments needed to satisfy the requirement that the fold continuity in both the radial and circumferential directions of the iris be ≥ 85%. Then, candidate segments with an entropy value ≥ 7.5 were selected using image sharpness evaluation metrics. Next, feature consistency verification of adjacent frames was used to remove distorted segments caused by slight user shaking. The final determined iris segments needed to pass the fold integrity verification of at least 3 independent regions. One qualified iris segment was extracted from the selected images of each user.

[0032] Three-dimensional data were acquired from the iris fragments of two users, and the three-dimensional coordinates, peak positions, and trough positions of iris folds were extracted. During the extraction process, a fold feature saliency scoring algorithm was used to select fold regions with strong discriminative power, improving the fun and accuracy of the comparison. The formula is as follows: ,in The significance of the fold features is scored. , , These are the weighting coefficients. The fold density of the target area. The average fold density of the entire iris region. The continuous length of the folds within the target area. The maximum diameter of the target area. For the first The depth value of each fold sampling point This represents the average depth value of all fold sampling points within the target area. To optimize the 3D data, the collected 3D point cloud data is first subjected to statistical filtering to remove isolated noise points. The number of neighboring points is set to 15. When the average distance between a point and its neighboring points exceeds three times the standard deviation, it is identified as an isolated point and removed. Then, a point cloud registration algorithm based on plane fitting is used to construct a local reference plane based on the feature points of the iris and cornea edges. The point cloud data is projected onto this plane for preliminary calibration. Then, an improved version of the iterative nearest point algorithm is used for precise registration. The number of iterations is set to 150, and the registration error threshold is set to 0.003 mm. After registration, the iris equatorial plane is fitted as the XY plane using the least squares method with the pupil center as the origin and the normal vector of the iris surface as the Z-axis to complete the coordinate system calibration. Finally, layered outlier detection is used. Points with depth values ​​exceeding the average depth value ± two times the standard deviation are detected first, and then curvature abrupt change points are detected for the remaining points. After double verification, outliers are removed.

[0033] like Figure 3 As shown, iris segments from two friends are denoted as iris segment A and iris segment B, respectively. A multi-reference calibration technique using a three-dimensional coordinate system is employed for front-to-back alignment. First, an initial local coordinate system is constructed using the geometric center of iris segment A as the origin, combined with three feature points: the pupil center and the corneal edge. This initial local coordinate system is then fitted using the least squares method to form a stable coordinate reference. When mapping the three-dimensional data of iris segment B to this coordinate system, principal component analysis is used to extract the principal eigenvectors of iris segment B, aligning them with the principal eigenvectors of iris segment A, ensuring the angle between their principal directions is ≤5°. Then, the rotation angle of iris segment B is adjusted, and the rotation is calculated based on the extension direction vector of the folds. The matrix is ​​rotated to ensure the angle between the fold extension directions of the two segments is ≤8°. Incremental rotation is used during the rotation, and the angle is calculated in real time until the requirement is met. When adjusting the fitting distance, the distance in the Z-axis direction is first reduced from 1mm to 0.2mm at a speed of 1mm / s, and then slowly reduced at a speed of 0.2mm / s. At the same time, the distance between the peak and the trough is monitored in real time. When the distance is ≤0.05mm, the adjustment is stopped, and the fitting position at this time is recorded as the initial embedding position. Then, the peak of iris segment A is made to extend into the trough of iris segment B, and the peak of iris segment B is made to extend into the trough of iris segment A. The embedding matching degree is calculated using a multi-dimensional fit degree formula, which is: ,in For embedding matching degree, , , These are the weighting coefficients. The actual depth of the overlap between the crest and trough. Iris segment The maximum height of the wave crest and the iris segment The larger of the maximum depths of the troughs Iris segment The average curvature of the wave crest surface, Iris segment The average curvature of the inner wall of the trough, The actual contact area between the crest and trough. The theoretical maximum contact area between the crest and trough is represented by this matching degree, which reflects the degree of fit between the crest and the corresponding trough. During the fitting process, iris segment A is controlled to translate left and right relative to iris segment B. The optimal translation step size is calculated using a dynamic adjustment formula, which is: ,in For dynamic translation step size, For adjustment coefficients, For the resolution of the iris segment, The density of folds in the iris segment. The embedding matching degree is initially set at the initial fitting position. The embedding matching degree is recalculated after each translation, and the embedding matching degree values ​​at all translation positions are recorded. Figure 1 As shown.

[0034] The maximum value among all embedding matching scores is selected as the final iris matching score between the two friends, such as... Figure 2 As shown, based on the needs of interesting scenarios, in addition to classifying matching levels, personalized matching tags are generated based on the matching degree value: 90-100% corresponds to the soulmate tag, 80-89% corresponds to the high-dimensional matching tag, 70-79% corresponds to the tacit matching tag, 60-69% corresponds to the fate matching tag, and below 60% corresponds to the fun encounter tag. At the same time, a dynamic demonstration animation is generated, highlighting the embedding process of peaks and troughs in the animation, and synchronously displaying the matching degree change curve. Two friends can adjust the animation playback speed and viewing angle through the device touch screen to intuitively view the comparison process and the final result, increasing the interactive fun.

[0035] Example 2

[0036] In the user login process of the online financial platform, a dedicated iris recognition device is used to perform data collection preparation. First, the device's lens focal length is calibrated in three stages by adjusting the near, medium, and far focal lengths using a standard resolution chart, ensuring that the image resolution error at each focal length is ≤2%. Next, depth measurement accuracy is calibrated by using a standard glass slide of known thickness and capturing 10 sets of data under different lighting conditions. Linear regression is used to correct the depth measurement values, ensuring the accuracy of the depth data. During user guidance, the device's built-in infrared sensor detects the user's head position in real time. When the head deviates from the optimal collection distance, the platform's app provides a voice prompt asking the user to adjust their head within the collection range. If the user's eyes are detected to be obstructed by the edge of a mask, excessively long eyelashes, or dirty glasses, the image recognition algorithm identifies the obstruction and displays a red text prompt on the app interface asking the user to remove the obstruction and re-collect the data. The data collection process cannot continue until the obstruction is removed. During the acquisition process, real-time frame rate monitoring and blink detection are used to automatically skip and record the frame when the user blinks. After the acquisition is completed, the sharpness score and motion blur detection are performed on 3-5 frames of the user's image. Images that meet both conditions are selected as the objects for subsequent processing to ensure the quality of the comparison samples.

[0037] After acquiring real-time iris images of logged-in users, multi-level optimization techniques are used to extract iris segments. First, the images are converted to grayscale, then Gaussian filtering, bilateral filtering, and morphological opening are performed sequentially to achieve multi-level noise reduction. Gaussian filtering smooths high-frequency noise, bilateral filtering preserves iris texture edges while removing low-frequency noise, and morphological opening eliminates minor dark spot interference. Subsequently, an iris region segmentation algorithm based on an active contour model is used to separate the iris from the sclera and pupil regions. Based on the grayscale variance distribution of the iris texture, continuous regions with a grayscale variance ≥50 are selected as candidate segments. These candidate segments must satisfy a fold continuity of ≥85% in both the radial and circumferential directions. Candidate segments with an entropy value ≥7.5 are selected using image sharpness evaluation metrics. Then, feature consistency verification between adjacent frames is used to eliminate distorted segments caused by slight shaking of the user's handheld device. The final determined iris segments must pass fold integrity verification in at least three independent regions. One compliant iris segment is extracted from the selected images as a real-time comparison sample. At the same time, the iris template image pre-stored by the user is retrieved from the platform's encrypted database, and the corresponding iris segment is extracted as a comparison benchmark using the same process.

[0038] Three-dimensional data was acquired from real-time iris segments and iris template segments, and the three-dimensional coordinates, peak positions, and trough positions of iris folds were extracted. A fold feature saliency scoring algorithm was used during the extraction process to select fold regions with strong discriminative power, improving the uniqueness and reliability of anti-fraud verification. During 3D data optimization, statistical filtering is first performed on the acquired 3D point cloud data to remove isolated noise points. The number of neighboring points is set to 15. A point is considered an isolated point and is removed when the average distance between it and its neighbors exceeds three times the standard deviation. Then, a point cloud registration algorithm based on plane fitting is used to construct a local reference plane based on the feature points of the iris and cornea edges. The point cloud data is projected onto this plane for preliminary calibration. Next, an improved version of the iterative nearest point algorithm is used for precise registration. The number of iterations is set to 150, and the registration error threshold is set to 0.003 mm. After registration, the coordinate system is calibrated by fitting the iris equatorial plane as the XY plane using the least squares method, with the pupil center as the origin and the normal vector of the iris surface as the Z-axis. Finally, layered outlier detection is performed. Points with depth values ​​exceeding the average depth value ± two times the standard deviation are detected first, and then curvature abrupt change points are detected for the remaining points. After double verification, outliers are removed to ensure the accuracy of the 3D structural information.

[0039] like Figure 3 As shown, the real-time acquired iris segment is denoted as iris segment A, and the pre-stored iris template segment is denoted as iris segment B. The front and back segments are aligned using a multi-reference calibration technique in a three-dimensional coordinate system. First, an initial local coordinate system is constructed using the geometric center of iris segment A as the origin, combined with three feature points: the pupil center and the corneal edge. A stable coordinate system reference is formed by fitting these four points using the least squares method. When mapping the three-dimensional data of iris segment B to this coordinate system, the principal eigenvectors of iris segment B are extracted using principal component analysis and aligned with the principal eigenvectors of iris segment A, ensuring the angle between their principal directions is ≤5°. Then, the rotation angle of iris segment B is adjusted, and the rotation matrix is ​​calculated based on the extension direction vector of the folds, ensuring the angle between the fold extension directions is ≤8°. Incremental rotation is used during the rotation process, and the angle is calculated in real-time until the requirement is met. When adjusting the bonding distance, the Z-axis distance is first reduced from 1mm to 0.2mm at a speed of 1mm / s, and then slowly reduced at a speed of 0.2mm / s, while simultaneously monitoring the distance between the peaks and troughs in real-time. Adjustment is stopped when the distance is ≤0.05mm, and the bonding position at this point is recorded as the initial embedding position. Then, the crest of iris segment A is aligned with the trough of iris segment B, and vice versa. The embedding matching degree is calculated using a multi-dimensional fit formula. During the bonding process, iris segment A is translated left and right relative to iris segment B. The optimal translation step size is calculated using a dynamic adjustment formula. The embedding matching degree is recalculated after each translation, and the embedding matching degree values ​​at all translation positions are recorded. Figure 1 As shown.

[0040] The maximum value among all embedding matching scores is selected as the final iris matching score, such as... Figure 2 As shown, in the anti-fraud scenario, the matching threshold adopts a dynamic adjustment mechanism. This online financial login scenario sets the fraud risk level to medium-high risk, corresponding to a matching threshold of 90%. When the final matching degree is ≥90%, the verification is deemed successful, and the user is allowed to complete the login process. When the matching degree is <90%, a warning signal is generated, and the APP interface pops up a prompt that the iris verification failed and asks the user to verify their identity and try again. The device is also limited to a maximum of 3 verification opportunities per hour. At the same time, regardless of whether the verification is successful or not, key data in the comparison process is automatically extracted to generate a verification report. The report is encrypted using the national cryptographic algorithm SM4 and stored in a distributed database. It supports hierarchical access control and can only be viewed by the platform's security management department and authorized personnel, providing complete data evidence for tracing potential fraud cases in the future.

[0041] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for iris biometric matching based on three-dimensional fold embedding matching, characterized in that, The specific steps of this method are as follows: S1. Iris Image Acquisition Preparation: Before acquiring iris images, calibrate the acquisition device, prompt the user to cooperate with the acquisition operation, and select iris images as objects for subsequent processing. S2. Iris fragment acquisition: Acquire iris images of the two users to be compared, and extract at least one iris fragment from each iris image. The iris fragment is a continuous region with a complete fold structure within the iris area. S3. Three-dimensional structure extraction: Three-dimensional data acquisition is performed on each iris segment obtained in step S1, and the three-dimensional structure information of the iris segment is extracted. The three-dimensional structure information includes the three-dimensional coordinates of the iris folds, the position information of the fold peaks and the position information of the fold troughs. S4. Three-dimensional fold embedding matching: Iris segments of two users are denoted as iris segment A and iris segment B, respectively. Iris segments A and B are aligned front to back, so that the peak of iris segment A extends into the trough of iris segment B, and at the same time, the peak of iris segment B extends into the trough of iris segment A. The embedding matching degree is calculated, which is the degree of fit between the peak and the corresponding trough. During the alignment process, one iris segment is controlled to translate left and right relative to the other iris segment. The embedding matching degree is recalculated after each translation, and the embedding matching degree values ​​at all translation positions are recorded. S5. Matching result output: Select the maximum value from all the embedding matching degree values ​​recorded in step S3 as the final iris matching degree of the two users. When the final iris matching degree is 100%, it is determined that the iris segments of the two users have achieved perfect embedding matching.

2. The iris biometrics comparison method based on three-dimensional fold embedding matching according to claim 1, characterized in that, In step S1, the iris image acquisition preparation adopts a collaborative optimization technology of device calibration and user guidance. Specifically, the lens focal length of the acquisition device is first calibrated in three stages by shooting a standard resolution board and adjusting the near-focus, mid-focus, and far-focus positions respectively, so that the image resolution error at each focal length is ≤2%. Then, the depth measurement accuracy is calibrated by using a standard glass plate of known thickness and taking 10 sets of data under different lighting conditions, and correcting the depth measurement value through linear regression. In terms of user guidance, the user's head position is detected in real time by the built-in infrared sensor of the device. When the head deviates from the optimal acquisition distance, the distance is adjusted through voice prompts. When an obstruction is detected in the user's eyes, the image recognition algorithm identifies it and pops up a text prompt to remove the obstruction. During the acquisition process, real-time frame rate monitoring and blink detection are used. When a user blinks, the frame is automatically skipped and recorded. After the acquisition is completed, sharpness scoring and motion blur detection are performed on 3-5 frames of images for each user. Images that meet both conditions are selected as processing objects.

3. The iris biometric comparison method based on three-dimensional fold embedding matching according to claim 1, characterized in that, The iris segment extraction process in step S2 adopts a multi-level optimization technique. Specifically, the acquired iris image is first grayscale processed, and then Gaussian filtering, bilateral filtering and morphological opening operation are performed in sequence to achieve multi-level noise reduction. Among them, Gaussian filtering is used to smooth high-frequency noise, bilateral filtering is used to remove low-frequency noise while preserving the edge of iris texture, and morphological opening operation is used to eliminate small dark spot interference. Subsequently, the iris, sclera, and pupil regions are separated using an iris region segmentation algorithm based on an active contour model. Based on the gray-level variance distribution of the iris texture, continuous regions with a gray-level variance ≥ 50 are selected as candidate segments. Candidate segments must satisfy that the fold continuity in both the radial and circumferential directions of the iris is ≥ 85%. Then, candidate segments with an entropy value ≥ 7.5 are selected through image sharpness evaluation index. Finally, through feature consistency verification of adjacent frame images, distorted segments caused by slight user shaking are eliminated. The final determined iris segments must pass the fold integrity verification of at least 3 independent regions.

4. The iris biometric comparison method based on three-dimensional fold embedding matching according to claim 1, characterized in that, In step S3, the 3D structure extraction process incorporates a wrinkle feature saliency scoring algorithm to filter wrinkle regions with strong discriminative power. The formula is as follows: ,in The significance of the fold features is scored. , , These are the weighting coefficients. The fold density of the target area. The average fold density of the entire iris region. The continuous length of the folds within the target area. The maximum diameter of the target area. For the first The depth value of each fold sampling point This represents the average depth value of all fold sampling points within the target area. This represents the number of sampling points.

5. The iris biometric comparison method based on three-dimensional fold embedding matching according to claim 1, characterized in that, The three-dimensional data optimization process in step S3 adopts a multi-level calibration technique. Specifically, the collected three-dimensional point cloud data is first subjected to statistical filtering to remove isolated noise points. The number of neighboring points is set to 15. When the average distance between a point and its neighboring points exceeds 3 times the standard deviation, it is determined to be an isolated point and is removed. Subsequently, a point cloud registration algorithm based on plane fitting was employed. Using feature points at the edges of the iris and cornea as a reference, a local reference plane was constructed, and the point cloud data was projected onto this plane for initial calibration. Then, an improved version of the iterative nearest-point algorithm was used for precise registration, with 150 iterations and a registration error threshold of 0.003 mm. After registration, coordinate system calibration was performed on the 3D coordinate data, using the pupil center as the origin and the normal vector of the iris surface as the Z-axis. The iris equatorial plane was fitted using the least squares method as the XY plane. Finally, layered outlier detection was performed, first detecting depth values ​​exceeding... The point, among which The standard deviation of the depth value is used to detect curvature abrupt changes in the remaining points. After double verification, outliers are removed.

6. The iris biometric comparison method based on three-dimensional fold embedding matching according to claim 1, characterized in that, In step S4, the front-to-back alignment employs a multi-reference calibration technique using a three-dimensional coordinate system. Specifically, the following steps are taken: First, an initial local coordinate system is constructed using the geometric center of iris segment A as the origin, combined with three feature points: the pupil center and the corneal edge. A stable coordinate system reference is formed by fitting these four points using the least squares method. When mapping the three-dimensional data of iris segment B to this coordinate system, the principal eigenvector of iris segment B is extracted using principal component analysis and aligned with the principal eigenvector of iris segment A, ensuring the angle between their principal directions is ≤5°. Then, the rotation angle of iris segment B is adjusted, and a rotation matrix is ​​calculated based on the extension direction vector of the folds, ensuring the angle between their fold extension directions is ≤8°. Incremental rotation is used during rotation, and the angle is calculated in real-time until the requirement is met. When adjusting the bonding distance, the Z-axis distance is first reduced from 1mm to 0.2mm at a speed of 1mm / s, and then slowly reduced at a speed of 0.2mm / s, while simultaneously monitoring the distance between the peak and trough. Adjustment is stopped when the distance is ≤0.05mm, and the bonding position at this point is recorded as the initial embedding position.

7. The iris biometric comparison method based on three-dimensional fold embedding matching according to claim 1, characterized in that, The embedding matching degree calculation in step S4 adopts a multi-dimensional fit degree formula, which is: ,in For embedding matching degree, , , These are the weighting coefficients. The actual depth of the overlap between the crest and trough. Iris segment The maximum height of the wave crest and the iris segment The larger of the maximum depths of the troughs Iris segment The average curvature of the crest surface, Iris segment The average curvature of the inner wall of the trough, The actual contact area between the crest and trough. This represents the theoretical maximum contact area between the crest and trough.

8. The iris biometric comparison method based on three-dimensional fold embedding matching according to claim 1, characterized in that, The optimal step size for left and right translation in step S4 is calculated using a dynamic adjustment formula, which is: ,in For dynamic translation step size, For adjustment coefficients, For the resolution of the iris segment, The density of folds in the iris segment. The embedding matching degree of the initial fitting position.

9. The iris biometric comparison method based on three-dimensional fold embedding matching according to claim 1, characterized in that, In step S5, the matching result output adopts multi-scenario adaptive processing technology. Specifically, in the fun matching scenario, in addition to classifying the matching level, personalized matching tags are generated based on the matching degree value. A matching degree of 90-100% corresponds to the soulmate tag, 80-89% corresponds to the high-dimensional matching tag, 70-79% corresponds to the tacit matching tag, 60-69% corresponds to the fate matching tag, and below 60% corresponds to the fun encounter tag. At the same time, the dynamic demonstration animation adds high-brightness markings embedded with peaks and troughs and a matching degree change curve. Users can adjust the animation playback speed and perspective through their terminal devices. In the anti-fraud scenario, the matching threshold adopts a dynamic adjustment mechanism, which is set to 85%, 90%, and 95% respectively according to the fraud risk level of the application scenario. When the matching degree is ≥ the threshold, in addition to generating a warning signal, key data in the comparison process is automatically extracted to generate a verification report. Encrypted storage adopts the national cryptographic algorithm SM4, and the storage medium adopts a distributed database, supporting hierarchical permission management. In biometric counter-evidence scenarios, the output includes the final matching degree, the confidence interval of the matching process, and the comparison details of key feature points. At the same time, a counter-evidence report is generated, which clearly explains the basis for verifying the uniqueness of the iris biometric feature. The report can be directly used in judicial or administrative verification scenarios.

Images