Small area fingerprint matching method and device

Abstract

This application relates to the field of fingerprint matching technology, and discloses a method and device for small-area fingerprint matching. The method includes: acquiring a fingerprint image to be identified; determining corresponding neighborhood support domains for multiple local regions in the fingerprint image to be identified; extracting physical field features from the neighborhood support domains, calculating the information entropy corresponding to the extracted physical field features, and fusing them into a multi-physics fusion entropy; detecting feature points in the fingerprint image to be identified based on the multi-physics fusion entropy corresponding to the local regions and the local structural change features of the local regions; extracting image blocks of a preset size centered on the feature points in the fingerprint image to be identified, and performing pixel symmetry processing to generate polarity-independent compact descriptors; and performing fingerprint matching on the fingerprint image to be identified based on the feature points, compact descriptors, template feature points, and template compact descriptors to obtain the corresponding matching results. The embodiments of this application can effectively overcome the deficiency of insufficient feature information.

Description

Technical Field

[0001] This application relates to the field of fingerprint matching technology, and in particular to a method and device for small-area fingerprint matching. Background Technology

[0002] With the rapid proliferation of mobile devices and IoT terminals, small-area semiconductor fingerprint sensors have been widely used in smartphones, smartwatches, and access control systems due to their advantages such as small size and low power consumption. However, limited by the physical size of the sensor, the fingerprint image area acquired by such sensors is generally small, typically covering only 80×80 to 160×160 pixels, far smaller than the coverage area of ​​a traditional full fingerprint image. Due to the significantly limited image area, fingerprint ridges exhibit highly localized features within the acquisition area, resulting in a severe shortage of extractable reliable minutiae (such as endpoints and bifurcation points), often failing to reach the minimum threshold required for matching. Simultaneously, small-area images are susceptible to skin surface noise, uneven acquisition pressure, and changes in ambient lighting, causing ridge structures to become blurred, broken, or deformed, further exacerbating the difficulty of feature point extraction. Traditional minutiae-based matching methods rely on a sufficient number of stable feature points to construct matching relationships, but in small-area scenarios, feature points are scarce and their distribution is unstable, leading to an increased point-to-point matching error rate, a significant decrease in overall matching accuracy, and difficulty in ensuring system robustness. Summary of the Invention

[0003] The purpose of this application is to provide a method and device for matching small-area fingerprints, which can improve the matching accuracy of small-area fingerprint images and effectively overcome the deficiency of feature information.

[0004] This application provides a method for matching small-area fingerprints, including: Acquire the fingerprint image to be identified; For multiple local regions in the fingerprint image to be identified, a neighborhood support domain that is adaptive to the local ridge characteristics is determined respectively; Physical field features are extracted from the neighborhood support domain to obtain multiple physical field features. The information entropy corresponding to each physical field feature is calculated and fused to form a multi-physical field fusion entropy. Based on the multiphysics field fusion entropy corresponding to the local region and the local structural change characteristics of the local region, feature points are detected in the fingerprint image to be identified. In the fingerprint image to be identified, an image patch of a preset size centered on the feature point is extracted, and the image patch is subjected to pixel symmetry processing to generate a polarity-independent compact descriptor; Based on the feature points, the compact descriptor, and the preset template feature points and template compact descriptor, fingerprint matching is performed on the fingerprint image to be identified to obtain the corresponding matching results.

[0005] In some embodiments, prior to determining the neighborhood support domains that are adaptive to the local ridge characteristics, the method further includes: Based on the integral image of the fingerprint image to be identified, background estimation and illumination normalization are performed on the fingerprint image to be identified to obtain the illumination normalized image. The image after illumination normalization is locally normalized to adjust the local grayscale distribution to the target mean and standard deviation, thus obtaining the locally normalized image. The locally normalized image is then denoised and enhanced to obtain a preprocessed fingerprint image to be identified.

[0006] In some embodiments, determining the neighborhood support domains that are adaptive to the local ridge characteristics includes: The direction of the neighborhood support domain is determined based on the local ridge direction of the local region. The scale of the neighborhood support domain is determined based on the local ridge density of the local region. Based on the direction and scale of the neighborhood support domain, the neighborhood support domain is determined such that it is an elliptical region adapted to the local ridge characteristics.

[0007] In some embodiments, the physical field features include at least two of the following: a grayscale field characterizing image brightness variations, a direction field characterizing ridge flow direction, a frequency field characterizing ridge density, and a gradient field characterizing edge intensity.

[0008] In some embodiments, detecting feature points in the fingerprint image to be identified based on the multiphysics fusion entropy corresponding to the local region and the local structural change characteristics of the local region includes: Based on the distribution of the multiphysics field fusion entropy of each local region at multiple scales, the local optimal salient scale and cross-scale distribution variation degree corresponding to the local region are determined; Based on the multiphysics field fusion entropy at the local optimal salient scale, the cross-scale distribution change degree, and the local structural change characteristics, feature point response values ​​are generated. Based on the response values ​​of the feature points, the non-maximum suppression method is used to determine the feature points.

[0009] In some embodiments, the pixel symmetry processing of the image block includes: Perform pixel summation and pixel difference absolute value operation on the image blocks respectively to obtain symmetric summation feature map and absolute difference feature map; Flatten the symmetric summation feature map and the absolute difference feature map, and concatenate the two feature vectors obtained from the flattening to obtain a joint feature vector; The joint feature vector is subjected to dimensionality reduction and binary quantization to obtain the compact descriptor.

[0010] In some embodiments, the step of performing fingerprint matching on the fingerprint image to be identified based on the feature points, the compact descriptor, and preset template feature points and template compact descriptors includes: Based on the similarity between the compact descriptor and the template compact descriptor, the feature points and the template feature points that meet the similarity matching conditions are selected to form candidate matching point pairs; Based on the geometric relationship between the candidate matching point pairs, determine the geometric transformation parameters between the fingerprint image to be identified and the template fingerprint image; Based on the geometric transformation parameters, the feature points in the candidate matching point pair are transformed, and the geometric consistency between the transformed feature points and the template feature points is verified to determine the target matching point pair. Based on the target matching point pair, the degree of matching between the fingerprint image to be identified and the template fingerprint image is determined, and the matching result is generated based on the degree of matching.

[0011] In some embodiments, determining the geometric transformation parameters between the fingerprint image to be identified and the template fingerprint image based on the geometric relationship between the candidate matching point pairs includes: Based on the directional difference features between the candidate matching point pairs, a weighted voting method using a one-dimensional cyclic histogram is used to estimate the rotation parameters. Based on the positional difference features between the rotation parameters and the candidate matching point pairs, a weighted voting method using a two-dimensional voting array is used to estimate the translation parameters. The rotation parameters and translation parameters are iteratively optimized, and the optimized rotation parameters and translation parameters are used as the geometric transformation parameters.

[0012] In some embodiments, the geometric consistency verification of the transformed feature points and the template feature points includes: Candidate matching point pairs whose geometric difference features between the transformed feature points and the template feature points meet the first geometric difference condition are determined as the first matching point pairs, and otherwise determined as the second matching point pairs. For the second pair of matching points, within the geometric space that satisfies the geometric transformation parameters, search for potential matching points corresponding to the transformed feature points; When the potential matching point is found, and the geometric difference features between the potential matching point and the transformed feature point meet the second geometric difference condition, and the similarity between the potential compact descriptor of the potential matching point and the compact descriptor of the transformed feature point meets the similarity condition, a third matching point pair is constructed using the transformed feature point and the potential matching point. The first matching point pair and the third matching point pair are subjected to uniqueness constraints to obtain the target matching point pair.

[0013] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described small-area fingerprint matching method.

[0014] The beneficial effects of this application are as follows: When matching fingerprint images to be identified, feature points are detected by introducing multi-physics fusion entropy and local structural change features. These feature points are not limited to traditional minutiae, but rather capture the local texture information of the fingerprint image more comprehensively. Even if the fingerprint image area is small, multiple discriminative feature points can be detected based on multi-physics fusion entropy and structural change features, rather than relying solely on sparse minutiae, effectively overcoming the deficiency of feature information in small-area fingerprint images. In addition, by performing pixel symmetry processing on image blocks, polarity-independent compact descriptors are generated. These descriptors not only have strong robustness but can also effectively cope with rotation and flipping during fingerprint acquisition. Moreover, their compactness reduces storage and computational overhead, providing a more stable and efficient feature representation in small-area fingerprint images, thereby improving the matching accuracy of small-area fingerprint images. Attached Figure Description

[0015] Figure 1 This is a flowchart of a small-area fingerprint matching method provided in an embodiment of this application.

[0016] Figure 2 This is a flowchart illustrating the process of determining neighborhood support domains that are adaptive to local ridge characteristics, as provided in the embodiments of this application.

[0017] Figure 3 This is a flowchart of detecting feature points in a fingerprint image to be identified, provided in an embodiment of this application.

[0018] Figure 4 This is a flowchart of pixel symmetry processing for image blocks provided in an embodiment of this application.

[0019] Figure 5 This is a flowchart of fingerprint matching of a fingerprint image to be identified, provided in an embodiment of this application.

[0020] Figure 6This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and drawings are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application. Furthermore, the information, data, and signals involved in the embodiments of this application are all authorized by relevant parties or have been fully authorized by all parties, and the collection, use, and processing of related data comply with the relevant laws, regulations, and standards of the relevant countries and regions.

[0024] In existing fingerprint matching technologies, the small size of fingerprint images limits the acquisition of a sufficient number of reliable feature points by traditional minutiae-based matching methods, resulting in reduced matching accuracy. Specifically, when the fingerprint image area is small, the number of extractable reliable minutiae is limited, making it difficult for minutiae-based matching algorithms to achieve high-precision matching. This problem directly affects the reliability and accuracy of the fingerprint recognition system. For example, in under-display fingerprint recognition applications on mobile devices, the image area captured by semiconductor fingerprint sensors is typically only 80×80 to 160×160 pixels. During user authentication, the limited acquisition area means that the image only contains a portion of the fingerprint ridges, limiting the number of reliable minutiae to be extracted. In this scenario, the matching process frequently fails due to insufficient feature points, leading to authentication delays or errors, impacting system reliability and user experience.

[0025] If the above problems are not solved, the performance of fingerprint recognition systems under small-area image conditions will not meet the needs of practical applications. The increase in false recognition rate and false rejection rate will make the system unreliable in security-critical scenarios and limit the application scope of small-area fingerprint sensors in mobile devices and IoT terminals.

[0026] Based on this, embodiments of this application provide a small-area fingerprint matching method and device. When matching fingerprint images to be identified, adaptive feature extraction and compact descriptor generation effectively overcome the deficiency of feature information in small-area fingerprint images and improve the matching accuracy of small-area fingerprint images.

[0027] See Figure 1 In one embodiment, a small-area fingerprint matching method is provided, wherein the execution subject of the method is an electronic device, including but not limited to steps S101 to S106.

[0028] Step S101: Obtain the fingerprint image to be identified.

[0029] The acquisition of a fingerprint image to be identified can be achieved in various ways, such as directly acquiring the fingerprint image using an optical sensor or reading an existing fingerprint image file from a storage medium. In one implementation, the raw grayscale image output by the sensor can be directly used as the fingerprint image to be identified.

[0030] Step S102: For multiple local regions in the fingerprint image to be identified, determine the neighborhood support domain that is adaptive to the local ridge characteristics.

[0031] A local region refers to a smaller image block that is divided within the fingerprint image to be identified. A local region can be the area corresponding to each pixel in the fingerprint image to be identified, or the area corresponding to sampling points obtained by densely sampling the fingerprint image to be identified according to a preset step size.

[0032] Local ridge characteristics refer to the geometric and topological properties of ridges, such as direction, density, and curvature, within a local region of a fingerprint image to be identified. Local ridge characteristics reflect the local structure of the fingerprint texture.

[0033] A neighborhood support domain refers to a specific region surrounding a local area in a fingerprint image, used for feature extraction and analysis. The shape and size of this support domain can be adjusted based on the characteristics of the local ridges to better capture local texture information.

[0034] Dividing the fingerprint image to be identified into multiple local regions can be done by centering on each pixel or by dividing the fingerprint image into multiple fixed-size grids, with each grid serving as a local region. For each local region, a fixed-size circular or rectangular region can be defined as its neighborhood support region. For example, a circular region with radius R can be specified as its neighborhood support region for each local region, and the size of this neighborhood support region does not change with the local ridge characteristics.

[0035] Step S103: Extract physical field features from the neighborhood support domain to obtain multiple physical field features, calculate the information entropy corresponding to each physical field feature, and fuse them to form a multi-physical field fusion entropy.

[0036] Information entropy is an indicator that measures the uncertainty or randomness of information. Calculating the information entropy of a physical field feature can quantify the amount of information contained in that feature; the higher the information entropy, the greater the discriminative power of that feature.

[0037] Multiphysics fusion entropy is a comprehensive index obtained by fusing the information entropy of features from multiple physical fields. By fusing the information entropy of different physical fields, the local characteristics of fingerprint images can be characterized more comprehensively and robustly, improving the expressive power of features.

[0038] Physical field features can be extracted from each neighboring support domain, such as grayscale distribution features. Then, the information entropy corresponding to each physical field feature is calculated. Subsequently, these information entropies are weighted and averaged to form a multi-physics fusion entropy.

[0039] Step S104: Based on the multi-physics field fusion entropy corresponding to the local region and the local structural change characteristics of the local region, feature points are detected in the fingerprint image to be identified.

[0040] Local structural variation features refer to the degree of variation in the ridge structure within a local area of ​​the fingerprint image to be identified. Examples include features such as ridge curvature, bifurcation, and endpoints.

[0041] Feature points are points in a fingerprint image that have salience, stability, and discriminative power. They are usually located in areas with dramatic changes in ridge structure or rich information content, and are key anchor points for fingerprint matching.

[0042] In detecting feature points in a fingerprint image to be identified, a fixed threshold can be set, and local regions where the multiphysics fusion entropy is higher than this threshold can be marked as potential feature point regions. Simultaneously, the gray-level variance of the local region can be calculated as a local structural change feature, and another threshold can be set. Only when both the fusion entropy and gray-level variance meet preset conditions is the center point of the local region determined as a feature point.

[0043] Step S105: In the fingerprint image to be identified, extract an image block of a preset size centered on the feature point, and perform pixel symmetry processing on the image block to generate a polarity-independent compact descriptor.

[0044] Pixel symmetry processing refers to performing symmetric operations on pixels in an image block to eliminate or reduce the sensitivity of the descriptor to image rotation or flipping, which can make the generated descriptor more robust.

[0045] A compact descriptor is a vector that encodes local features of a fingerprint. This descriptor is independent of the polarity of the fingerprint image (e.g., whether the fingerprint is forward or backward) and has low dimensionality.

[0046] Extract an image patch of a predetermined size centered on the feature point. Perform pixel symmetry processing on the image patch; for example, extract a 16x16 pixel image patch. To perform pixel symmetry processing on this image patch, simply flip the image patch horizontally along its center, then perform a pixel-level XOR operation between the original image patch and the flipped image patch to obtain a symmetric feature map. Subsequently, flatten the pixel values ​​of this feature map directly into a vector, which serves as a compact descriptor.

[0047] Step S106: Based on feature points, compact descriptors, and preset template feature points and template compact descriptors, perform fingerprint matching on the fingerprint image to be identified to obtain the corresponding matching results.

[0048] Template feature points refer to the feature points of template fingerprint images pre-stored in a fingerprint template library. These template feature points are compared with the feature points of the fingerprint image to be identified to determine the degree of fingerprint matching.

[0049] Template compact descriptors are compact descriptors of template fingerprint images pre-stored in a fingerprint template library. These template descriptors are compared with compact descriptors of the fingerprint image to be identified to evaluate the similarity between feature points.

[0050] Fingerprint matching of the fingerprint image to be identified can be performed by calculating the Euclidean distance between each compact descriptor of the fingerprint image to be identified and each template compact descriptor of the template fingerprint image. If the Euclidean distance is less than a preset threshold, the corresponding feature point is considered to be likely to match. Then, the number of matching points is counted, and the corresponding matching results are generated according to the ratio of the number of matching points to the total number of feature points.

[0051] The following example will provide a more detailed explanation of the above technical solution: Suppose user A needs to unlock a device. This device is equipped with a small fingerprint sensor. When user A places their finger on the sensor, an image of user A's fingerprint to be identified is first acquired. This image may only contain a portion of the fingerprint's texture information due to the sensor's size limitations.

[0052] The fingerprint image to be identified is then divided into multiple local regions. For each local region, a neighborhood support domain is determined. For example, a fixed-size square neighborhood can be assigned to each local region for subsequent feature analysis.

[0053] Next, physical field features are extracted for each neighborhood support domain. For example, grayscale features and simple ridge direction features can be extracted within that neighborhood. Then, the information entropy of each of these physical field features is calculated, and these information entropies are fused to form a multi-physics fusion entropy value. Simultaneously, local structural change features of each local region are calculated; for example, the activity level of the texture is characterized by calculating the standard deviation of local grayscale values.

[0054] Based on these fusion entropy values ​​and local structural change characteristics, feature points are detected in the fingerprint image to be identified. For example, regions with high fusion entropy and significant local structural changes can be identified, and the center points of these regions are marked as feature points. These feature points differ from traditional minutiae; they focus more on the richness and uniqueness of local texture information. Once feature points are detected, an image patch of a preset size is extracted centered on each feature point. For example, a 32x32 pixel image patch is extracted. To make the descriptor insensitive to fingerprint rotation and flipping, pixel symmetry processing is performed on these image patches. For example, radial symmetry analysis can be performed on the image patches, mapping the pixel values ​​of the image patches to a polar coordinate system and calculating their radial projection, thereby generating a polarity-independent compact descriptor.

[0055] Finally, these feature points and compact descriptors are compared with feature points and template compact descriptors of a pre-stored template fingerprint image. First, the similarity between each compact descriptor of the fingerprint image to be identified and the template compact descriptor is calculated to identify potential matching point pairs. Then, the geometric relationships between these potential matching point pairs are checked, such as whether their relative positions and orientations are consistent. Through these comparisons and verifications, the degree of matching between the fingerprint image to be identified and the template fingerprint image can be determined, and a matching result can be generated, for example, determining whether user A's fingerprint matches a registered fingerprint, thereby deciding whether to unlock the device.

[0056] Based on the above examples, the technical solution of this embodiment demonstrates a significant technical contribution to solving the problem of small-area fingerprint matching. In existing technologies, due to the limited area of ​​small-area fingerprint images, traditional minutiae-based methods struggle to obtain a sufficient number of reliable minutiae, severely impacting matching accuracy. This embodiment detects feature points by introducing multi-physics fusion entropy and local structural change features. These feature points are not limited to traditional minutiae but rather capture more comprehensively the local texture information of the fingerprint image. For example, in the above example, even with a small fingerprint image area, the system can detect multiple discriminative feature points based on fusion entropy and structural change features, rather than relying solely on sparse minutiae.

[0057] Furthermore, this embodiment generates a polarity-independent compact descriptor by performing pixel symmetry processing on image blocks. This descriptor not only possesses strong robustness, effectively handling rotation and flipping during fingerprint acquisition, but its compactness also reduces storage and computational overhead. Compared to existing descriptors that are sensitive to rotation or have high dimensionality, the descriptor in this embodiment provides a more stable and efficient feature representation in small-area fingerprint images. Through the combination of these techniques, the method in this embodiment effectively overcomes the challenge of insufficient minutiae in small-area fingerprint images, improving the reliability and accuracy of fingerprint matching.

[0058] In some embodiments, for multiple local regions in the fingerprint image to be identified, before determining the neighborhood support domains that are adaptive to the local ridge characteristics, the method further includes: performing background estimation and illumination normalization on the fingerprint image to be identified based on the integral image of the fingerprint image to be identified, to obtain an illumination-normalized image; performing local normalization on the illumination-normalized image to adjust the local gray-level distribution to the target mean and standard deviation, to obtain a locally normalized image; and performing denoising and enhancement on the locally normalized image to obtain a preprocessed fingerprint image to be identified.

[0059] This application's solution introduces a series of preprocessing steps to provide high-quality input images for subsequent fingerprint feature extraction and matching. First, background estimation and illumination normalization are performed based on the integral image of the fingerprint to be identified. This effectively eliminates the uneven illumination commonly found in the original image, making the grayscale difference between the fingerprint ridge region and the background region more stable and predictable. Subsequently, local normalization processing is applied to the illumination-normalized image to adjust the local grayscale distribution to a uniform target mean and standard deviation. This not only further enhances the contrast of the fingerprint ridges but also compensates for local grayscale variations caused by pressure applied or sensor differences, ensuring that the fingerprint ridges have similar visual characteristics in different regions. Finally, the locally normalized image is denoised and enhanced to filter out random noise introduced during image acquisition and sharpen the edges and orientation information of the fingerprint ridges, making the ridge structure clearer and more discernible. Through this series of orderly and complementary preprocessing steps, interference factors in the original fingerprint image to be identified are effectively suppressed, while the inherent ridge features of the fingerprint are significantly highlighted. This provides a cleaner, more stable, and more information-rich input for subsequent core matching steps such as neighborhood support domain determination, physical field feature extraction, feature point detection, and compact descriptor generation.

[0060] In one specific embodiment, the step of generating the preprocessed fingerprint image to be identified includes: Construct the integral image of the fingerprint to be identified: First-order integral graph (For fast local mean): , Second-order integral graph (For fast local variance): , in, The first fingerprint in the image to be identified Line number Column of pixels, and All are integers. , , The x-coordinate is the pixel index. The vertical coordinate is the pixel index; The integral image is constructed using a single traversal, which can be pipelined: , Similarly.

[0061] mean of any rectangular region and variance Quick calculations can be performed using integral graphs: , in, The distance from the center point (x, y) to the boundary of the rectangle.

[0062] Similarly, use You can get it immediately.

[0063] fingerprint image to be identified of Location, using an integral image, with a large window radius Calculate the local mean as its background estimate: , in, The first fingerprint in the image to be identified Line number Background estimation of the column's pixels, for, The preset threshold is generally selected based on experimental results; the preferred threshold can be chosen. , The number of columns in the fingerprint image to be identified. The number of rows in the fingerprint image to be identified. This is the floor function.

[0064] The fingerprint image to be identified is processed using a division model. To maintain the multiplicative relationship of its local gray levels: , in, The first image after illumination normalization Line number Column of pixels, The first fingerprint in the image to be identified Line number Column of pixels, The target is the global grayscale center.

[0065] The grayscale distribution of each local region is normalized to a uniform mean. and standard deviation For background areas (without texture), normalization should be avoided to amplify noise. , , in, For local gain coefficients, It is a limit threshold. and For the image after illumination normalization In the middle of the pixel Center, radius The local mean and standard deviation within the window. For the locally normalized image, the first... Line number Column of pixels, The average value is the unified value; the preferred value is... , , , .

[0066] use Neighborhood Denoising is achieved using truncated mean filtering: , in, For the image after noise removal, For the locally normalized image, the first... Line number Column of pixels, For indicator functions, It is a fixed threshold related to the standard deviation of semiconductor sensor noise; preferably, it can be taken as... .

[0067] After the preceding normalization and mild denoising, the gradient of the ridge-valley transition zone may be slightly attenuated. Finally, it is moderately enhanced through controllable desharpening to obtain the preprocessed fingerprint image to be identified. : , , , in, This is a normal enhancement value. This is the corrected enhancement value. For the preprocessed fingerprint image to be identified, the first... Line number Column of pixels, For denoising images Gaussian blur output, i.e. , The standard deviation is Gaussian filter kernel, For convolution operations, For gain factor, For overshoot protection threshold, the preferred value can be [value]. , , .

[0068] See Figure 2 In one embodiment, the method for determining the neighborhood support domains that are adaptive to the local ridge characteristics includes, but is not limited to, steps S201 to S203.

[0069] Step S201: Determine the direction of the neighborhood support domain based on the local ridge direction of the local region; Step S202: Determine the scale of the neighborhood support domain based on the local ridge density of the local region; Step S203: Based on the direction and scale of the neighborhood support domain, determine the neighborhood support domain so that the neighborhood support domain is an elliptical region that adapts to the local ridge characteristics.

[0070] Local ridge orientation refers to the direction or main orientation of ridges within a local region of a fingerprint image. It is one of the most fundamental directional features of a fingerprint image, reflecting the local arrangement trend of the ridges. This orientation can be obtained by calculating the statistical information of the pixel gray-level gradient direction within the local region, for example, by using a gradient orientation histogram or Gabor filter bank response. Alternatively, it can be determined by analyzing the Fourier spectrum of the local region to identify the direction of energy concentration.

[0071] Local ridge density refers to the number of ridges per unit length or area within a local region of a fingerprint image, reflecting the density of the ridges. This density can be obtained by calculating the reciprocal of the ridge period or frequency within the local region, for example, by analyzing the peak frequency of the local spectrum using Fourier transform. It can also be estimated by counting the number of intersections of ridges or valleys within the local region.

[0072] The direction of the neighborhood support domain refers to the principal axis direction of the neighborhood support domain in the image space. It is usually consistent with the direction of the local ridges. The direction of the local ridges can be directly used as the principal axis direction of the neighborhood support domain, or it can be finely adjusted based on the curvature of the ridges or the distribution of local feature points.

[0073] The scale of the neighborhood support region refers to the size or extent of the neighborhood support region, which is usually determined by the lengths of its major and minor axes. This scale can be dynamically adjusted according to the local ridge density. For example, when the ridge density is high, the scale can be appropriately reduced; when the ridge density is low, the scale can be appropriately increased. Alternatively, the scale can be set to an integer multiple of the ridge period, taking into account the periodicity of the local ridges, to better cover the complete ridge structure.

[0074] The proposed solution analyzes the local ridge direction and density in a local region of the fingerprint image to be identified, enabling precise determination of the direction and scale of the neighborhood support domain. Specifically, the local ridge direction is used as the principal axis of the neighborhood support domain, and the size of the neighborhood support domain is adjusted according to the local ridge density, thereby constructing an elliptical region that highly matches the direction and density of the local ridges as the neighborhood support domain. This adaptive elliptical support domain can more accurately cover and capture the ridge structure, allowing for more effective extraction of subsequent physical field features, resulting in more representative physical field features. This refined neighborhood support domain determination method ensures high-quality feature extraction in various local regions of the fingerprint image, especially in small fingerprints where there may be blurred, broken, or deformed areas. This lays a solid foundation for subsequent feature point detection and fingerprint matching, significantly improving the accuracy and robustness of the entire matching method.

[0075] In one specific embodiment, for candidate center points of the neighborhood support domain Its local ridge distance is : , in, Let p be the local frequency at point p.

[0076] use To cover the scale parameters of several ridge distance periods, i.e., the scale of the neighborhood support domain, in the scale parameters Below, the major and minor axes of the neighborhood support domain are defined as follows: , in, The semi-major axis along the ridge line, Let be the semi-minor axis along the ridge line normal. and All are weighted parameters. Preferred, acceptable or .

[0077] Candidate center point The local ridge tangential unit vector at that location is: , The normal unit vector is: , Then the orientation-adaptive neighborhood support domain Defined as: , in, To support domain The pixels within.

[0078] The above design makes the feature point extraction scale no longer directly dependent on the absolute size of pixels, but adapts to the local fingerprint ridge distance, making it more suitable for small-area fingerprints with different resolutions and different pressing states.

[0079] In some embodiments, the physical field features include at least two of the following: a grayscale field characterizing image brightness variations, a direction field characterizing ridge flow direction, a frequency field characterizing ridge density, and a gradient field characterizing edge intensity.

[0080] This application's scheme specifically defines multiple physical field features as at least two of the following: a grayscale field characterizing image brightness variations, a direction field characterizing ridge flow, a frequency field characterizing ridge density, and a gradient field characterizing edge intensity. This ensures that more comprehensive and discriminative local fingerprint information can be obtained when extracting physical field features from the neighborhood support domain. The grayscale field provides basic brightness distribution information, the direction field reveals the local direction of the ridges, the frequency field reflects the density of the ridges, and the gradient field emphasizes the edge intensity of the ridges. These physical fields characterize the local properties of the fingerprint image from different dimensions. When features are extracted from these specific physical fields and their information entropy is calculated, the information richness and structural complexity of the local region can be quantified more accurately. By fusing these multi-dimensional information entropies, the resulting multi-physical field fusion entropy can more robustly reflect the local structural features of the fingerprint image, providing a solid foundation for subsequent feature point detection based on multi-physical field fusion entropy and local structural change features, effectively improving the accuracy and stability of feature point detection in small-area fingerprint images.

[0081] In one specific embodiment, the physical field features include grayscale field, orientation field, frequency field, and gradient field.

[0082] Support domain for neighborhood Quantization is performed on local regions within the area to obtain the probability distribution of the grayscale histogram. That is, grayscale field.

[0083] Preprocessed fingerprint image to be identified Calculate the Scharr gradient and The calculation formula is as follows: , , Polarity is eliminated using the double-angle formula, and the direction at each point is calculated using the area averaging method. : , , in, It is a second-order statistic of the local gradient.

[0084] Directional field Defined as: , gradient field Defined as: , Within a local area, along the normal direction of the ridge line Projecting grayscale values, the ridge distance period is estimated. After rotating the neighborhood support domain to a local coordinate system, a one-dimensional projection sequence is calculated along the normal direction, and the average ridge distance is obtained from the interval between adjacent peaks or valleys. Then the frequency field Defined as: , For gradient field Mean filtering is performed to obtain the average gradient. Then for each pixel position Perform foreground / background judgment and calculation global mean and standard deviation .like Then Mark the area as background, otherwise mark it as foreground.

[0085] See Figure 3 In one embodiment, the method for detecting feature points in a fingerprint image to be identified includes, but is not limited to, steps S301 to S303.

[0086] Step S301: Based on the distribution of multiphysics field fusion entropy of each local region at multiple scales, determine the local optimal salient scale and cross-scale distribution variation degree corresponding to the local region.

[0087] Step S302: Based on the multi-physics field fusion entropy, cross-scale distribution change degree, and local structural change characteristics at the local optimal salient scale, generate the feature point response value.

[0088] Step S303: Based on the feature point response values, the non-maximum suppression method is used to determine the feature points.

[0089] This application optimizes the feature point detection process based on the multiphysics fusion entropy and local structural change features corresponding to local regions by introducing multi-scale analysis and non-maximum suppression. First, for each local region in the fingerprint image to be identified, instead of relying solely on features at a single scale, the distribution of its multiphysics fusion entropy across multiple scales is analyzed to accurately determine the local optimal salient scale and the degree of cross-scale distribution variation of that local region. The local optimal salient scale captures the most essential and stable feature information of the region, while the degree of cross-scale distribution variation reflects the robustness of the feature under scale changes. Subsequently, this multi-scale information is effectively fused with the local structural change features to generate a comprehensive feature point response value. This response value not only considers the information richness and structural characteristics of the local region but also incorporates its stability and saliency in scale space, enabling the response value to more accurately indicate potential feature points. Finally, based on the generated feature point response value, a non-maximum suppression method is used to filter and simplify the response values. Non-maximum suppression can effectively remove redundant points with low response values ​​or those too close to stronger response points, ensuring that the finally determined feature points are locally most significant. This avoids false detections and false negatives caused by noise or local deformation in complex or low-quality fingerprint images, and significantly improves the accuracy and stability of feature point detection.

[0090] In one specific embodiment, at each candidate center point Each scale (of each foreground position) Corresponding neighborhood support domain Within the system, the statistical distributions of grayscale, direction, frequency, and gradient magnitude are calculated respectively. Based on this, the information entropy corresponding to each physical field feature is calculated, and a multi-physics fusion entropy is constructed accordingly.

[0091] Support domain for neighborhood Quantization is performed on local regions within the area to obtain the probability distribution of the grayscale histogram. , obtain grayscale entropy : , in, For the number of grayscale bins, the preferred value is... .

[0092] To handle the periodicity of direction, the neighborhood support domain is first calculated. Internal relative direction deviation : , in, for Inward-weighted mean; right Quantization yields the probability distribution of the radiation pattern. The directional deviation entropy is obtained. : , in, The number of bins after quantification of directional deviation; preferably, it can be taken as... In regions with regular parallel ridge lines, Smaller in size, in areas of local termination, bifurcation, and bending. It will increase.

[0093] definition Relative deviation of internal local frequency for: , in, for The average frequency within the range.

[0094] Quantification The frequency field probability distribution was then obtained. Frequency deviation entropy for: , in, The number of bins after frequency deviation quantization; preferably, it can be taken as... .

[0095] right gradient field inside Quantization is performed to obtain the probability distribution. Define gradient magnitude entropy for: , in, The number of bins after gradient bias quantization; preferably, it can be taken as... .

[0096] By weighting and fusing the above four types of information entropy, we obtain the corresponding multiphysics fusion entropy: , in, , Preferably, the weights of orientation deviation entropy and frequency deviation entropy can be slightly higher than those of grayscale entropy to enhance the sensitivity to fingerprint structural features. , , as well as .

[0097] In this embodiment of the application, a fixed scale is not used directly, but rather a set of scales is used. The search is conducted to find the locally optimal salient scale. Preferably, the scale can be... .

[0098] For candidate center points and scale parameters The histograms corresponding to grayscale, orientation deviation, frequency deviation, and gradient magnitude are concatenated into a joint statistical vector. : , definition Variation in cross-scale distribution for: , in, The chi-square distance, For any arbitrarily small scale deviation.

[0099] If a candidate center point has a high entropy value at a certain scale and its statistical distribution differs significantly from that of adjacent scales, it indicates that the scale is more likely to correspond to a real stable structure rather than random noise.

[0100] If scale satisfy ,and Then It is considered as a candidate feature scale for that point.

[0101] Further requirements , in, The threshold for variability can be determined by the validation set. This can eliminate spurious feature points that are "high in entropy but unstable across scales".

[0102] To avoid misdetecting noise, breakage artifacts, and background edges as feature points, this application's embodiments introduce consistency constraints on the orientation field, frequency field, and gradient field based on the entropy extremum. Directional consistency is defined as: , , , , in, , It is a second-order statistic of the local gradient. For traversal Index of all points, For The local neighborhood centered on, ; Frequency consistency is defined as: , in, This is the local frequency stability coefficient. The larger the value, the more stable the local ridge distance estimation. For sequence variance For sequence The mean, This is the sequence of adjacent peak-valley spacings in the local projection; Gradient consistency is: , , in , For pixels The polarity direction at that point Let be the gradient at point q. The direction of nonpolarity at point q is given.

[0103] Define the neighborhood support domain average directional consistency index Average frequency reliability index Orthogonality consistency index with average gradient They are respectively: , , ; Constructing multi-field mass constraint terms : , in, All are weighted indices; the preferred ones can be selected as... .

[0104] To further highlight the location of fingerprint structure changes, local curvature / rate of change terms for the orientation and frequency fields are introduced. : , in, To normalize to , For frequency gradient weights, The difference in polarity between point q and point p is the direction of polarity. Let q be the frequency difference between point q and point p.

[0105] By integrating multiphysics fusion entropy, cross-scale distribution variation, and local structural change characteristics, candidate points are defined at different scales. The final significant response for: , in, The curvature enhancement coefficient; The response function has the following characteristics: 1. Ensure that the local structure contains sufficient information; 2. Ensure that this information is identifiable at different scales; 3. Ensure that the structure conforms to the natural pattern of fingerprint texture, rather than noise; 4. This ensures that the extraction results are more biased towards regions with significant structural changes.

[0106] For all candidate center points in three-dimensional space Non-maximum suppression is applied, retaining only the candidate center point with the largest local significant response. That is, if the candidate center point... In its neighborhood Internal satisfaction: , Then retain the candidate center point.

[0107] Near the retained candidate center points, the significant response function Perform 3D quadratic surface fitting to obtain sub-pixel and sub-scale offsets, thereby improving positioning accuracy.

[0108] Let the response function be defined. The local Taylor expansion is: , The extreme value offset is then: ; like If the value is less than the preset threshold, the correction result is accepted.

[0109] Each feature point output is denoted as: , in, The coordinates of the feature point, The main direction.

[0110] See Figure 4 In one embodiment, the method for performing pixel symmetry processing on image blocks includes, but is not limited to, steps S401 to S403.

[0111] Step S401: Perform pixel summation and pixel difference absolute value operation on the image blocks respectively to obtain symmetric summation feature map and absolute difference feature map.

[0112] Step S402: Flatten the symmetric summation feature map and the absolute difference feature map, and concatenate the two feature vectors obtained from the flattening to obtain the joint feature vector.

[0113] Step S403: Dimensionality reduction and binary quantization are performed on the joint feature vector to obtain a compact descriptor.

[0114] The proposed scheme involves performing centrosymmetric pixel summation and pixel difference absolute value operations on the extracted image patches. Pixel summation captures the overall brightness information and low-frequency texture of the image patch, offering some noise suppression; while pixel difference absolute value operation highlights local contrast and edge information, exhibiting good invariance to illumination changes. These two symmetric operations extract the essential features of the image patch from different dimensions, generating symmetric summation feature maps and absolute difference feature maps respectively. These two feature maps are then flattened into one-dimensional feature vectors and concatenated to form a joint feature vector. This joint feature vector integrates multiple symmetric features of the image patch, containing richer and more comprehensive local texture information. To improve the efficiency and robustness of the descriptor, the joint feature vector undergoes dimensionality reduction, removing redundant information and reducing its dimensionality. Finally, binary quantization converts the dimensionality-reduced features into a compact binary descriptor. This processing flow enables the generated compact descriptors to not only effectively capture the local texture features of image patches, but also exhibit strong robustness to potential rotation, uneven illumination, and noise interference in fingerprint images. Furthermore, its compact binary form significantly improves the efficiency of subsequent fingerprint matching. In this way, even with limited information in small fingerprint images, high-quality descriptors can be generated, laying the foundation for accurate fingerprint matching.

[0115] Since the orientation of feature points is not polar, it is necessary to design compact descriptors that are independent of polarity.

[0116] In one specific embodiment, in the preprocessed fingerprint image to be identified, feature points are... , with its coordinates With the origin as the starting point, and the direction as the starting point. Establish a rectangular coordinate system along the positive x-axis. Within this coordinate system, construct a coordinate system centered at the origin with a dimension of... square area In the embodiments of this application, A square region is interpolated from the preprocessed fingerprint image to be identified using bilinear interpolation. Treat it as an image patch, feature points Location as an image block The center The Middle Line number The grayscale value in the column is used express, The row and column indices start from 0. For image patches... Perform bidirectional folding to generate two sets of quarter-size feature maps ( ): , The two sets of feature maps remain strictly invariant under polarity reversal; Flatten the two feature maps into one-dimensional vectors in row-major order. : , , It is a 256-dimensional real number space; Concatenate them into a joint feature vector: , It is a 512-dimensional real number space; PCA projection is performed on the joint feature vectors, and fingerprints are collected from the training fingerprint database. Feature points (suggested) Extract the joint feature vector set according to the above process. .

[0117] Mean vector: , Calculate the covariance matrix: , Eigenvalue decomposition: ,in, , The corresponding orthogonal eigenvector matrix; Before selection Principal components: ; Joint vector for each feature point Projection yields dimensionality-reduced features: ; use Hadamard matrix Acting on the feature vector get : , Among them, the 128th order Hadamard matrix The constructor is as follows: , , in, For Kronecker product; Finally, symbolic quantization was used to... Hash to a 128-dimensional Hamming space to obtain a compact descriptor in the final hash form. : , Each feature point in the final output is denoted as .

[0118] See Figure 5 In one embodiment, the method for fingerprint matching of the fingerprint image to be identified includes, but is not limited to, steps S501 to S504.

[0119] Step S501: Based on the similarity between the compact descriptor and the template compact descriptor, select feature points and template feature points that meet the similarity matching conditions to form candidate matching point pairs.

[0120] Step S502: Based on the geometric relationship between candidate matching point pairs, determine the geometric transformation parameters between the fingerprint image to be identified and the template fingerprint image.

[0121] Step S503: Based on the geometric transformation parameters, the feature points in the candidate matching point pair are transformed, and the geometric consistency between the transformed feature points and the template feature points is verified to determine the target matching point pair.

[0122] Step S504: Based on the target matching point pair, determine the degree of matching between the fingerprint image to be identified and the template fingerprint image, and generate a matching result based on the degree of matching.

[0123] The proposed scheme utilizes compact descriptors for initial feature point screening. Compact descriptors, as a compact representation of local texture information, effectively distinguish different feature points. By calculating the similarity between the compact descriptors of feature points in the fingerprint image to be identified and the template compact descriptors in the template fingerprint image, feature point pairs with potential correspondences in local texture can be quickly identified, forming candidate matching point pairs. This effectively reduces the computational load of subsequent geometric verification and eliminates a large number of obviously mismatched point pairs. To overcome the geometric differences that may exist in small-area fingerprints, such as deformation, rotation, and translation, this scheme estimates the overall geometric transformation parameters between the fingerprint image to be identified and the template fingerprint image based on the geometric relationships between these candidate matching point pairs. This process, by analyzing the relative position and orientation information of the point pairs, accurately captures the global alignment relationship between the two fingerprint images. Estimating accurate geometric transformation parameters provides a foundation for subsequent geometric consistency verification, enabling effective matching of fingerprint images obtained under different acquisition conditions. Next, using the determined geometric transformation parameters, the feature points of the fingerprint image to be identified in the candidate matching point pairs are transformed to make them geometrically aligned with the template fingerprint image. Subsequently, a rigorous geometric consistency verification is performed on the transformed feature points and their corresponding template feature points. This verification process effectively eliminates erroneous matching point pairs that, although similar in descriptors, have inconsistent geometric positions, thus ensuring that the final target matching point pairs are highly consistent in both local texture and global geometry. This geometric consistency verification mechanism is crucial for improving matching accuracy, especially when the amount of fingerprint information in a small area is limited, effectively suppressing the effects of noise and deformation. Finally, based on the rigorously screened and verified target matching point pairs, this scheme can accurately quantify the degree of matching between the fingerprint image to be identified and the template fingerprint image. By statistically analyzing the number of target matching point pairs or their weighted sum, a reliable matching score can be obtained. Comparing this matching score with a preset threshold generates the final matching result. This multi-verification-based matching strategy enables high-precision and robust fingerprint matching even in cases of incomplete or deformed fingerprint information in a small area, significantly improving overall matching performance.

[0124] In some embodiments, determining the geometric transformation parameters between the fingerprint image to be identified and the template fingerprint image based on the geometric relationship between candidate matching point pairs includes: estimating rotation parameters by performing weighted voting using a one-dimensional cyclic histogram based on the directional difference features between candidate matching point pairs; estimating translation parameters by performing weighted voting using a two-dimensional voting array based on the rotation parameters and the positional difference features between candidate matching point pairs; iteratively optimizing the rotation parameters and translation parameters, and using the iteratively optimized rotation parameters and translation parameters as geometric transformation parameters.

[0125] This application employs a step-by-step and iterative optimization strategy to accurately determine the geometric transformation parameters between the fingerprint image to be identified and the template fingerprint image. First, utilizing the directional difference features between candidate matching point pairs, a weighted voting process using a one-dimensional cyclic histogram effectively handles the periodicity of orientation. This weighted voting mechanism ensures that matching point pairs with high descriptor similarity and reliability play a greater role in estimating the rotation parameters, thereby suppressing interference from erroneous matches and improving the robustness of rotation estimation. After obtaining the initial rotation parameters, the method further performs rotation correction on the feature points in the fingerprint image to be identified based on these parameters, and then calculates the positional difference features between the corrected feature points and the template feature points. These positional difference features are then weighted using a two-dimensional voting array to estimate the translation parameters. By prioritizing rotation before translation, the complex two-dimensional rigid body transformation is decomposed into two relatively independent steps, simplifying the parameter estimation. The two-dimensional voting array, also using a weighted mechanism, ensures that reliable matching point pairs dominate the estimation of translation parameters. To further improve the accuracy and stability of the geometric transformation parameters, the scheme in this application iteratively optimizes the initially estimated rotation and translation parameters. This iterative process allows the system to re-evaluate the contribution of matching point pairs and refine the parameters in each iteration based on the current parameter estimation results. For example, in one iteration, all candidate matching point pairs can be transformed using the currently estimated rotation and translation parameters, then their orientation and position differences can be recalculated, and weighted voting can be performed again, thereby gradually converging to more accurate geometric transformation parameters. This iterative optimization mechanism can effectively cope with initial estimation errors and enhance the algorithm's adaptability in the presence of noise or local deformation, ultimately obtaining more accurate geometric transformation parameters and laying a solid foundation for subsequent geometric consistency verification.

[0126] In one specific embodiment, the fingerprint image to be identified is marked as Mark the template fingerprint image as Fingerprint image to be identified The feature points and their corresponding compact descriptors are represented as follows: Template fingerprint image The template feature points and their corresponding template compact descriptors are represented as follows: .

[0127] For each feature point Calculate its compact descriptor with each template feature point Similarity of template compact descriptors : , in, Represents the XOR operation; The closer to 1, the better. and The higher the similarity, the higher the probability of a match. The closer to 0, the more likely it is to be zero. and The lower the similarity, the lower the probability of a match. right Sort from largest to smallest, and take the top three as... The candidate pairing points are denoted as , respectively with This forms a candidate matching point pair.

[0128] like and For the true corresponding points, under the similarity transformation, we have: , Therefore, for Each feature point and its 3 candidate matching points ( ), using the direction difference to estimate the rotation angle (modulus 180°): , Constructing a one-dimensional cyclic histogram (Total 180 bins, each bin corresponds to 1°), initialization , Weighted voting: , Take the peak position of the histogram : , Due to the 180° periodicity of the direction, the true rotation angle (scope There are two candidate values: and That is, the rotation parameter.

[0129] For two candidate rotation angles Perform the following procedures respectively: a. Initialize the two-dimensional voting array ,scope , ; b. Each feature point and its 3 candidate matching points ( First, check for consistency in direction: , If satisfied, then based on the transformation equation Should be with Overlapping constraints, accurate calculation of candidate translation parameters and : , , in, Let x be the x-coordinate of the i-th feature point. Let be the ordinate of the i-th feature point. Let x be the x-coordinate of the t-th candidate matching point of the i-th feature point. Let be the ordinate of the t-th candidate matching point of the i-th feature point; In a two-dimensional array Weighted voting is conducted in the following ways: , c. Records The maximum value in and their corresponding positions .

[0130] d. Choose to make Larger As the final rotation angle Corresponding As the initial optimal translation parameter .

[0131] Collect all candidate matching sets that match the optimal parameters. : , The translation parameters for weighted refinement, with similarity as the weight. and : , , in, For feature points Compact descriptor and template feature points The similarity of template compact descriptors for and The corresponding horizontal translation parameters, for and The corresponding vertical translation parameters.

[0132] Optionally, the refined translation parameters can be used. and Redefining The weighted average is then applied again, and the iteration is repeated for 2 to 3 rounds or until the parameter change is less than 0.1, which further improves the accuracy. The rotation and translation parameters optimized by the iteration are then used as the geometric transformation parameters.

[0133] In some embodiments, geometric consistency verification of the transformed feature points and template feature points includes: determining candidate matching point pairs whose geometric difference features between the transformed feature points and template feature points meet the first geometric difference condition as first matching point pairs, and otherwise as second matching point pairs; for the second matching point pairs, searching for potential matching points corresponding to the transformed feature points within the geometric space that satisfies the geometric transformation parameters; when a potential matching point is found, and the geometric difference features between the potential matching point and the transformed feature points meet the second geometric difference condition, and the similarity between the potential compact descriptor of the potential matching point and the compact descriptor of the transformed feature points meets the similarity condition, constructing a third matching point pair using the transformed feature points and the potential matching point; and applying uniqueness constraints to the first matching point pair and the third matching point pair to obtain the target matching point pair.

[0134] The first geometric difference condition is used to initially determine whether the geometric difference between the transformed feature points and the template feature points is within an acceptable range. It can be a threshold set for the distance, angle, or relative position deviation between the two.

[0135] The second geometric difference condition is used to make a more refined judgment on the geometric consistency between the searched potential matching points and the transformed feature points. It may be more lenient than the first geometric difference condition, or optimized for specific geometric deformation patterns. For example, it can be set to require that the Euclidean distance between the potential matching point and the transformed feature point is less than another preset threshold, or it can consider the distance under a local deformation model. The potential compact descriptor refers to the compact descriptor corresponding to the potential matching point.

[0136] This application employs a hierarchical and refined verification strategy. In the geometric consistency verification stage of fingerprint matching, it first classifies the initially formed candidate matching point pairs. By setting a first geometric difference condition, candidate matching point pairs with small geometric differences are directly identified as first matching point pairs, considered highly reliable matches. For second matching point pairs that do not meet the first geometric difference condition (i.e., have slightly larger geometric differences), this application does not discard them directly. Instead, it actively searches for potential matching points within the geometric space defined by the geometric transformation parameters, centered on the transformed feature points. This search mechanism aims to compensate for local biases caused by fingerprint deformation, noise, or inaccurate initial geometric transformation parameter estimation, avoiding the erroneous removal of true matching point pairs. When a potential matching point is found, this application simultaneously evaluates the second geometric difference condition between the potential matching point and the transformed feature points, as well as the similarity condition between their respective compact descriptors. Only when both conditions are met is the transformed feature point and the potential matching point constructed as a third matching point pair. This dual verification mechanism ensures that even with some geometric discrepancies, highly similar feature descriptor pairs can still be effectively identified. Finally, to ensure the accuracy and uniqueness of the matching results, the proposed solution applies uniqueness constraints to the first and third matching pairs, eliminating potential redundant or conflicting matches to obtain the final target matching pair. Through this layered and refined verification strategy, the proposed solution can more comprehensively capture the true matching relationships in fingerprint images, effectively addressing situations with poor fingerprint image quality or deformation.

[0137] In one specific embodiment, when transforming the feature points in the candidate matching point pair based on geometric transformation parameters, the following is used: , and Generate from arrive rigid transformation : , calculate All feature points exist The image below is used to obtain the transformed feature points: , in, The x-coordinates of the transformed feature points are: The ordinates are the ordinates of the transformed feature points.

[0138] Then, geometric consistency verification is performed on the transformed feature points and the template feature points. Candidate matching point pairs whose geometric differences between the transformed and template feature points meet the first geometric difference condition are determined as the first matching point pair; otherwise, they are determined as the second matching point pair. Specifically: First round of matching: If Candidate pairing points , , There is a certain point satisfy: , , Then mark match If multiple candidate pairings meet the conditions, the pair with the smallest distance is selected as the first pair; otherwise, it is selected as the second pair.

[0139] Second round of matching: For each second matching pair, iterate through... All potential matching points that have not yet been matched If the following conditions are met simultaneously: , , , Then mark match ,Increase Similarity constraints are applied to avoid mismatches with completely dissimilar descriptors. When the similarity constraints are satisfied, the transformed feature points are... with potential matching points Construct a third matching point pair.

[0140] Finally, a uniqueness constraint is applied to the first and third matching point pairs. If multiple transformed feature points match the same template feature point, only the first and third matching point pairs are retained. The highest-ranking matching pair is selected, and the remaining matching pairs are discarded. The above steps will yield the desired result. and If the number of target matching point pairs that ultimately match exceeds a certain threshold, it can be considered as a valid match. Matched Otherwise, it will be judged. and Mismatch.

[0141] This application also provides an electronic device. (See appendix.) Figure 6 , Figure 6This is a schematic diagram of the main structure of an electronic device according to an embodiment of this application. Figure 6 As shown, the electronic device in this embodiment mainly includes a processor 601 and a memory 602. The memory 602 can be configured to store a program for executing the small-area fingerprint matching method of the above-described method embodiments. The processor 601 can be configured to execute the program in the memory 602, which includes, but is not limited to, a program for executing the small-area fingerprint matching method of the above-described method embodiments. For ease of explanation, only the parts related to the embodiments of this application are shown. For specific technical details not disclosed, please refer to the method section of the embodiments of this application.

[0142] In some possible embodiments of this application, the electronic device may include multiple processors 601 and multiple memories 602. The program executing the small-area fingerprint matching method of the above-described method embodiments can be divided into multiple subroutines. Each subroutine can be loaded and run by a processor 601 to execute different steps of the small-area fingerprint matching method of the above-described method embodiments. Specifically, each subroutine can be stored in a different memory 602, and each processor 601 can be configured to execute programs in one or more memories 602 to jointly implement the small-area fingerprint matching method of the above-described method embodiments. That is, each processor 601 executes different steps of the small-area fingerprint matching method of the above-described method embodiments to jointly implement the small-area fingerprint matching method of the above-described method embodiments.

[0143] The aforementioned multiple processors 601 can be processors deployed on the same device. For example, the aforementioned electronic device can be a high-performance device composed of multiple processors, and the aforementioned multiple processors 601 can be processors configured on that high-performance device. Alternatively, the aforementioned multiple processors 601 can also be processors deployed on different devices. For example, the aforementioned electronic device can be a server cluster, and the aforementioned multiple processors 601 can be processors on different servers within the server cluster.

[0144] This application also provides a computer-readable storage medium. In one embodiment of the computer-readable storage medium according to this application, the computer-readable storage medium can be configured to store a program that performs the small-area fingerprint matching method of the above-described method embodiments. This program can be loaded and run by a processor to implement the above-described small-area fingerprint matching method. For ease of explanation, only the parts related to the embodiments of this application are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of this application. The computer-readable storage medium can be a memory formed by various electronic devices. Optionally, in the embodiments of this application, the computer-readable storage medium is a non-transitory computer-readable storage medium.

[0145] The small-area fingerprint matching method and device provided in this application detect feature points by introducing multi-physics field fusion entropy and local structural change features when matching fingerprint images to be identified. These feature points are not limited to traditional minutiae, but rather capture the local texture information of the fingerprint image more comprehensively. Even if the fingerprint image area is small, multiple discriminative feature points can be detected based on multi-physics field fusion entropy and structural change features, rather than relying solely on sparse minutiae, effectively overcoming the deficiency of feature information in small-area fingerprint images. In addition, by performing pixel symmetry processing on image blocks, a polarity-independent compact descriptor is generated. This descriptor not only has strong robustness, but can also effectively cope with rotation and flipping during fingerprint acquisition. Moreover, its compactness reduces storage and computational overhead, providing a more stable and efficient feature representation in small-area fingerprint images, thereby improving the matching accuracy of small-area fingerprint images.

[0146] Exemplary embodiments of this disclosure have been specifically shown and described above. It should be understood that this disclosure is not limited to the detailed structures, arrangements, or implementations described herein; rather, this disclosure is intended to cover various modifications and equivalent arrangements contained within the spirit and scope of the appended claims.

Claims

1. A small area fingerprint matching method, characterized by, include: Acquire the fingerprint image to be identified; For multiple local regions in the fingerprint image to be identified, a neighborhood support domain that is adaptive to the local ridge characteristics is determined respectively; Physical field features are extracted from the neighborhood support domain to obtain multiple physical field features. The information entropy corresponding to each physical field feature is calculated and fused to form a multi-physical field fusion entropy. Based on the multiphysics field fusion entropy corresponding to the local region and the local structural change characteristics of the local region, feature points are detected in the fingerprint image to be identified. In the fingerprint image to be identified, an image patch of a preset size centered on the feature point is extracted, and the image patch is subjected to pixel symmetry processing to generate a polarity-independent compact descriptor; Based on the feature points, the compact descriptor, and the preset template feature points and template compact descriptor, fingerprint matching is performed on the fingerprint image to be identified to obtain the corresponding matching results; The step of detecting feature points in the fingerprint image to be identified based on the multiphysics fusion entropy corresponding to the local region and the local structural change features of the local region includes: Based on the distribution of the multiphysics field fusion entropy of each local region at multiple scales, the local optimal salient scale and cross-scale distribution variation degree corresponding to the local region are determined; Based on the multiphysics field fusion entropy at the local optimal salient scale, the cross-scale distribution change degree, and the local structural change characteristics, feature point response values ​​are generated. Based on the response values ​​of the feature points, the non-maximum suppression method is used to determine the feature points.

2. The small area fingerprint matching method according to claim 1, wherein, Before determining the neighborhood support domains that are adaptive to the local ridge characteristics, the method further includes: Based on the integral image of the fingerprint image to be identified, background estimation and illumination normalization are performed on the fingerprint image to be identified to obtain the illumination normalized image. The image after illumination normalization is locally normalized to adjust the local grayscale distribution to the target mean and standard deviation, thus obtaining the locally normalized image. The locally normalized image is then denoised and enhanced to obtain a preprocessed fingerprint image to be identified.

3. The small area fingerprint matching method of claim 1, wherein, The determination of neighborhood support domains that are adaptive to local ridge characteristics includes: The direction of the neighborhood support domain is determined based on the local ridge direction of the local region. The scale of the neighborhood support domain is determined based on the local ridge density of the local region. Based on the direction and scale of the neighborhood support domain, the neighborhood support domain is determined such that it is an elliptical region adapted to the local ridge characteristics.

4. The small area fingerprint matching method of claim 1, wherein, The physical field features include at least two of the following: a grayscale field characterizing image brightness variations, a direction field characterizing ridge flow direction, a frequency field characterizing ridge density, and a gradient field characterizing edge intensity.

5. The small area fingerprint matching method of claim 1, wherein, The pixel symmetry processing of the image block includes: Perform pixel summation and pixel difference absolute value operation on the image blocks respectively to obtain symmetric summation feature map and absolute difference feature map; Flatten the symmetric summation feature map and the absolute difference feature map, and concatenate the two feature vectors obtained from the flattening to obtain a joint feature vector; The joint feature vector is subjected to dimensionality reduction and binary quantization to obtain the compact descriptor.

6. The small area fingerprint matching method of claim 1, wherein, The step of performing fingerprint matching on the fingerprint image to be identified based on the feature points, the compact descriptor, and preset template feature points and template compact descriptors includes: Based on the similarity between the compact descriptor and the template compact descriptor, the feature points and the template feature points that meet the similarity matching conditions are selected to form candidate matching point pairs; Based on the geometric relationship between the candidate matching point pairs, determine the geometric transformation parameters between the fingerprint image to be identified and the template fingerprint image; Based on the geometric transformation parameters, the feature points in the candidate matching point pair are transformed, and the geometric consistency between the transformed feature points and the template feature points is verified to determine the target matching point pair. Based on the target matching point pair, the degree of matching between the fingerprint image to be identified and the template fingerprint image is determined, and the matching result is generated based on the degree of matching.

7. The small area fingerprint matching method of claim 6, wherein, The step of determining the geometric transformation parameters between the fingerprint image to be identified and the template fingerprint image based on the geometric relationship between the candidate matching point pairs includes: Based on the directional difference features between the candidate matching point pairs, a weighted voting method using a one-dimensional cyclic histogram is used to estimate the rotation parameters. Based on the positional difference features between the rotation parameters and the candidate matching point pairs, a weighted voting method using a two-dimensional voting array is used to estimate the translation parameters. The rotation parameters and translation parameters are iteratively optimized, and the optimized rotation parameters and translation parameters are used as the geometric transformation parameters.

8. The small area fingerprint matching method of claim 6, wherein, The geometric consistency verification of the transformed feature points and the template feature points includes: Candidate matching point pairs whose geometric difference features between the transformed feature points and the template feature points meet the first geometric difference condition are determined as the first matching point pairs, and otherwise determined as the second matching point pairs. For the second pair of matching points, within the geometric space range that satisfies the geometric transformation parameters, search for potential matching points corresponding to the transformed feature points; When the potential matching point is found, and the geometric difference features between the potential matching point and the transformed feature point meet the second geometric difference condition, and the similarity between the potential compact descriptor of the potential matching point and the compact descriptor of the transformed feature point meets the similarity condition, a third matching point pair is constructed using the transformed feature point and the potential matching point. The first matching point pair and the third matching point pair are subjected to uniqueness constraints to obtain the target matching point pair.

9. An electronic device, comprising: The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the small-area fingerprint matching method according to any one of claims 1 to 8.

Images