Fingerprint image feature enhancement method, program product, and apparatus

Abstract

The application relates to a fingerprint image feature enhancement method, a program product and a device. The method comprises the following steps: dividing and analyzing a fingerprint image at different scales to obtain polar power spectrum of each block and determine angular marginal energy spectrum and ridge frequency estimation of each block; fusing angular marginal energy spectrum at different scales into a fused direction spectrum and determining direction estimation of each block; measuring the reliability of the direction estimation and counting a first number of blocks with reliable direction estimation; reconstructing a direction field according to the first number, the direction estimation and the reliability of the direction estimation; measuring the reliability of the ridge frequency estimation and counting a second number of reliable blocks; reconstructing a ridge frequency field according to the second number, the ridge frequency estimation and the reliability of the ridge frequency estimation; and continuously reconnecting and enhancing broken ridges in the fingerprint image based on the direction field and the ridge frequency field. The application can accurately enhance the ridge features in a small fingerprint image.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method, program product and apparatus for enhancing the features of fingerprint images. Background Technology

[0002] With the rapid development of smart terminals, wearable devices, and financial-grade identity authentication applications, small-area fingerprint sensors have become the mainstream solution. The effective acquisition area of ​​this type of sensor is usually only 8mm×8mm to 12mm×12mm, corresponding to a pixel size of approximately 100×100 to 200×200. The number of observable detail points is usually less than 15, and it often suffers severe degradation due to factors such as uneven pressure, dry or wet skin, and skin abrasion, resulting in broken ridges, adhered valleys, contrast degradation, and unstable orientation field estimation.

[0003] To reconstruct the ridge structure of small fingerprints, traditional methods rely on local orientation fields and local ridge frequency estimation. When there are large areas of low quality in the fingerprint image, traditional methods cause the ridge information in dry, wet, and worn areas to be almost completely lost, making it impossible to infer the orientation from local observations. At the same time, once the orientation field estimation of low quality areas is wrong, the method will strengthen along the wrong direction, which will generate false minutiae such as "false vortices" and "false bridging". This leads to the introduction of unreliable minutiae by arbitrarily generated ridge intersections under low signal-to-noise ratio, which poses a security risk to the recognition system.

[0004] In summary, finding an enhancement method for small-area fingerprint images that does not introduce false minutiae remains a key unsolved problem. Summary of the Invention

[0005] Therefore, it is necessary to provide a fingerprint image feature enhancement method, apparatus, computer device, computer-readable storage medium, and computer program product that can accurately enhance the ridge features in small fingerprint images, addressing the aforementioned technical problems.

[0006] In a first aspect, this application provides a fingerprint image feature enhancement method, including: The fingerprint image to be processed is sampled at different scales to obtain multiple blocks. The polar coordinates of each block are analyzed to obtain the polar power spectrum of each block. The angular marginal energy spectrum and ridge frequency estimate of each block are determined based on the polar coordinate power spectrum. The marginal energy spectra of the corner at different scales in each block are normalized and then fused to obtain the fused direction spectrum of each block, and the direction estimate of each block is determined based on the fused direction spectrum. The reliability of the direction estimate for each block is measured to obtain the reliability of the direction estimate for each block, and the first number of blocks with reliable direction estimates is counted. Based on the first quantity, the direction estimate of each block, and the reliability of the direction estimate, the direction field of each block included in the first quantity is reconstructed; The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation are counted. Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed. Based on the orientation field and the ridge frequency field, the broken ridges in the fingerprint image are continuously reconnected and enhanced. Under the condition of satisfying the preset stopping condition, the fingerprint image with enhanced features is obtained.

[0007] In one embodiment, reconstructing the orientation field of each block, which includes the first number, based on the orientation estimate of each block and the reliability of the orientation estimate, includes: The global orientation field of a block with reliable orientation estimation is derived by using a preset orientation model; Based on the orientation estimates of each pixel in the reliable block and the global orientation field, determine the observation residuals of the corresponding block; Based on the reliability of the direction estimation of the block with reliable direction estimation and the observation residual, the energy functional with the residual correction function as the variable is solved to obtain the residual correction function of the corresponding block; When the first quantity is greater than or equal to the first preset quantity, the orientation field of each block included in the first quantity is reconstructed based on the global orientation field and the residual correction function; If the first quantity is less than the first preset quantity, the preset global orientation constant is used as the orientation field of each pixel in the fingerprint image.

[0008] In one embodiment, the orientation model includes a zero-pole orientation model and a constant orientation model; the global orientation field of the block from which the orientation estimation is reliable is derived through the preset orientation model includes: If the first quantity is greater than or equal to the first preset quantity, determine whether there are singularities in the blocks where the direction estimation is reliable; In the presence of this singularity, a zero-pole orientation model is constructed based on the x-coordinate and y-coordinate of each pixel in the fingerprint image. The global orientation field of the block with reliable orientation estimation is derived through this zero-pole orientation model. In the absence of the singularity, the global orientation field of the block with reliable orientation estimation is derived based on the preset constant orientation model.

[0009] In one embodiment, determining whether a singularity exists in a block whose direction estimation is reliable includes: In a block where the orientation estimation is reliable, the detected center pixel is determined pixel by pixel, and a closed curve of a preset shape is generated with the center pixel as the center. Integrating the orientation field of the block to which the central pixel belongs along the closed curve yields the Poincaré index. If the Poincaré index is zero, it is determined that there are no singularities in the block to which the corresponding center pixel belongs; If the Poincaré index is not zero, it is determined that there is a singularity in the block to which the corresponding center pixel belongs.

[0010] In one embodiment, reconstructing the ridge frequency field of each block, which includes the second number, based on the ridge frequency estimate of each block and the reliability of the ridge frequency estimate, includes: When the second quantity is greater than or equal to the second preset quantity, based on the ridge frequency estimate of the block included in the second quantity and the reliability of the ridge frequency estimate, the energy functional with the ridge frequency field as the variable is solved to obtain the initial ridge frequency field of the corresponding block, and the initial ridge frequency field is clipped to a preset interval to obtain the ridge frequency field of the corresponding block. If the second quantity is less than the second preset quantity and is not zero, the average ridge frequency of the corresponding block is determined based on the ridge frequency estimate of the block included in the second quantity and the reliability of the ridge frequency estimate, and the average ridge frequency is determined as the ridge frequency field of the corresponding block. When the second quantity is zero, the preset default ridge frequency is determined as the ridge frequency field of the corresponding block.

[0011] In one embodiment, the different scales include a mesoscale, and determining the ridge frequency estimate based on the polar power spectrum includes: Spectral response analysis was performed on the polar coordinates of each block at the mesoscale to obtain the polar coordinate power spectrum of each block at the mesoscale, and the ridge frequency estimate of the mesoscale block was determined based on the polar coordinate power spectrum. The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation are counted. Specifically: The reliability of the ridge frequency estimation for the mesoscale block is measured to obtain the reliability of the ridge frequency estimation for each mesoscale block, and a second number of mesoscale blocks with reliable ridge frequency estimation and reliable orientation estimation are counted. Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed, specifically as follows: Based on the second quantity, the ridge frequency estimate of each mesoscale block, and the reliability of the ridge frequency estimate of the mesoscale block, the ridge frequency field of each block included in the second quantity is reconstructed.

[0012] Secondly, this application also provides a fingerprint image feature enhancement device, comprising: The block segmentation module is used to sample the fingerprint image to be processed at different scales to obtain multiple blocks, and to perform spectral response analysis on the polar coordinates of each block to obtain the polar coordinate power spectrum of each block. The frequency estimation module is used to determine the angular marginal energy spectrum and ridge frequency estimate of each block based on the polar coordinate power spectrum; The orientation estimation module is used to normalize and fuse the marginal energy spectra of the corner at different scales in each block to obtain the fused orientation spectrum of each block, and to determine the orientation estimate of each block based on the fused orientation spectrum. The directional reliability measurement module is used to measure the directional estimated reliability of each block, obtain the directional estimated reliability of each block, and count the first number of blocks with directional estimated reliability. The orientation field reconstruction module is used to reconstruct the orientation field of each block included in the first number based on the first number, the orientation estimate of each block, and the reliability of the orientation estimate; The frequency reliability measurement module is used to measure the reliability of the ridge frequency estimation of each block, obtain the ridge frequency estimation reliability of each block, and count the second number of blocks whose ridge frequency estimation is reliable and whose direction estimation is also reliable. The ridge frequency field reconstruction module is used to reconstruct the ridge frequency field of each block included in the second number based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate. The reconnection module is used to continuously reconnect and enhance the broken ridges in the fingerprint image based on the orientation field and the ridge frequency field, and obtain the feature-enhanced fingerprint image when the preset stopping condition is met.

[0013] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps: The fingerprint image to be processed is sampled at different scales to obtain multiple blocks. The polar coordinates of each block are analyzed to obtain the polar power spectrum of each block. The angular marginal energy spectrum and ridge frequency estimate of each block are determined based on the polar coordinate power spectrum. The marginal energy spectra of the corner at different scales in each block are normalized and then fused to obtain the fused direction spectrum of each block, and the direction estimate of each block is determined based on the fused direction spectrum. The reliability of the direction estimate for each block is measured to obtain the reliability of the direction estimate for each block, and the first number of blocks with reliable direction estimates is counted. Based on the first quantity, the direction estimate of each block, and the reliability of the direction estimate, the direction field of each block included in the first quantity is reconstructed; The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation are counted. Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed. Based on the orientation field and the ridge frequency field, the broken ridges in the fingerprint image are continuously reconnected and enhanced. Under the condition of satisfying the preset stopping condition, the fingerprint image with enhanced features is obtained.

[0014] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps: The fingerprint image to be processed is sampled at different scales to obtain multiple blocks. The polar coordinates of each block are analyzed to obtain the polar power spectrum of each block. The angular marginal energy spectrum and ridge frequency estimate of each block are determined based on the polar coordinate power spectrum. The marginal energy spectra of the corner at different scales in each block are normalized and then fused to obtain the fused direction spectrum of each block, and the direction estimate of each block is determined based on the fused direction spectrum. The reliability of the direction estimate for each block is measured to obtain the reliability of the direction estimate for each block, and the first number of blocks with reliable direction estimates is counted. Based on the first quantity, the direction estimate of each block, and the reliability of the direction estimate, the direction field of each block included in the first quantity is reconstructed; The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation are counted. Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed. Based on the orientation field and the ridge frequency field, the broken ridges in the fingerprint image are continuously reconnected and enhanced. Under the condition of satisfying the preset stopping condition, the fingerprint image with enhanced features is obtained.

[0015] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps: The fingerprint image to be processed is sampled at different scales to obtain multiple blocks. The polar coordinates of each block are analyzed to obtain the polar power spectrum of each block. The angular marginal energy spectrum and ridge frequency estimate of each block are determined based on the polar coordinate power spectrum. The marginal energy spectra of the corner at different scales in each block are normalized and then fused to obtain the fused direction spectrum of each block, and the direction estimate of each block is determined based on the fused direction spectrum. The reliability of the direction estimate for each block is measured to obtain the reliability of the direction estimate for each block, and the first number of blocks with reliable direction estimates is counted. Based on the first quantity, the direction estimate of each block, and the reliability of the direction estimate, the direction field of each block included in the first quantity is reconstructed; The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation are counted. Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed. Based on the orientation field and the ridge frequency field, the broken ridges in the fingerprint image are continuously reconnected and enhanced. Under the condition of satisfying the preset stopping condition, the fingerprint image with enhanced features is obtained.

[0016] The aforementioned fingerprint image feature enhancement methods, apparatus, computer equipment, computer-readable storage media, and computer program products sample the fingerprint image to be processed at different scales to obtain multiple blocks. Then, the marginal energy spectra of the angular direction at different scales in each block are normalized and fused to obtain the fused direction spectrum of each block. The direction estimate of each block is determined based on this fused direction spectrum, and the reliability of the direction estimate for each block is measured to obtain the reliability of the direction estimate for each block. This facilitates the identification of blocks with reliable direction estimates in the global image as the basis for reconstructing the direction field of each block, resulting in a reconstructed direction field with globality and high accuracy. On the other hand… This application measures the reliability of the ridge frequency estimation for each block, so that when reconstructing the ridge frequency field, it relies on the ridge frequency estimation of blocks with reliable ridge frequency estimation and reliable orientation estimation, and the reliability of these ridge frequency estimations. This makes the reconstructed ridge frequency field global and highly accurate, thus enabling more accurate inference of orientation during the continuous reconnection and enhancement of broken ridges in fingerprint images based on the orientation field and the ridge frequency field. The feature restoration of reconnected and enhanced ridges is more accurate, effectively solving common problems such as orientation estimation failure in low-quality areas of small fingerprints, difficulty in reconnecting broken ridges, and easy generation of false minutiae. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating a fingerprint image feature enhancement method in one embodiment; Figure 2 This is a schematic diagram illustrating the sampling of fingerprint images at different scales in one embodiment; Figure 3 This is a schematic diagram of the Cartesian power spectrum in one embodiment; Figure 4 This is a schematic diagram of the polar coordinate power spectrum in one embodiment; Figure 5 This is a schematic diagram of the direction estimation determined based on the fused direction spectrum in one embodiment; Figure 6 This is a schematic diagram illustrating the reliability weighting of fusing angular marginal energy spectra at different scales into a fused directional spectrum in one embodiment. Figure 7 This is a schematic diagram of the global orientation field of a block for which orientation estimation is reliable, derived through a zero-pole orientation model in one embodiment. Figure 8 A heatmap for the reliability of block orientation estimation in one embodiment; Figure 9 This is a schematic diagram illustrating the process of reconnecting and enhancing broken ridges in a fingerprint image in one embodiment. Figure 10 This is a schematic diagram illustrating the implementation process of a fingerprint image feature enhancement method in one embodiment; Figure 11 This is a schematic diagram comparing the effects of applying the fingerprint image feature enhancement method of the present invention in one embodiment; Figure 12 This is a structural block diagram of a fingerprint image feature enhancement device in one embodiment; Figure 13 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0019] 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.

[0020] It should be noted that the terms "first," "second," etc., used in this application may be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish the first element from the second element. The terms "comprising" and "having," and any variations thereof, used in this application, are intended to cover non-exclusive inclusion. The term "multiple" used in this application refers to two or more.

[0021] In one exemplary embodiment, such as Figure 1 As shown, a fingerprint image feature enhancement method is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes steps 101 to 108, wherein: Step 101: Sample the fingerprint image to be processed at different scales to obtain multiple blocks, and perform spectral response analysis on the polar coordinates of each block to obtain the polar coordinate power spectrum of each block.

[0022] According to one example of this embodiment, the input fingerprint image I to be processed can be sampled at three scales k∈{S, M, L}. In other embodiments, the input fingerprint image I to be processed can also be sampled at five scales, but the number of different scales should be odd to ensure that there is a unique medium scale, which is convenient for subsequent reconstruction of the global ridge frequency field.

[0023] Figure 2 This is a schematic diagram illustrating fingerprint image sampling at three different scales in one embodiment, as shown below. Figure 2 As shown, the green box indicates a small scale. The blue box indicates the mesoscale. Orange boxes indicate large scale. The purpose of multi-scale sampling and segmentation of fingerprint images is to facilitate subsequent steps at a smaller scale (…). ), mesoscale ( ), large scale ( Windowed FFT analysis is performed on the same central location at different sizes to complementaryly capture fine frequency, dominant directional components and global continuity.

[0024] According to one example of this embodiment, small scale The squares used below have a side length of 12 pixels and are medium scale. The squares used below have a side length of 16 pixels; Large scale. The square used below has a side length of 24 pixels.

[0025] Furthermore, sampling is performed at the center of blocks at different scales using a uniform grid spacing to obtain the spacing (in pixels) between the centers of adjacent blocks at the corresponding scale. This invention uniformly sets s=8 to ensure sufficient overlap between adjacent blocks and to obtain a dense orientation field sampling grid. With an 8-pixel spacing between the centers of adjacent blocks, for... For small-scale blocks, such as Figure 2 As shown, the overlap width between adjacent blocks ( Figure 2 The bold green portion is 4px. For mesoscale blocks, the overlap width between adjacent blocks ( Figure 2 The blue bold portion is 8px. For large-scale blocks, the overlap width between adjacent blocks ( Figure 2 The bold orange portion is 16px.

[0026] The fingerprint image can be a small-area grayscale image, and can be represented as follows:

[0027] in, This represents a small-area grayscale image of the fingerprint input (whose pixel values ​​have been normalized to...). (interval); Represents the set of real numbers (bold, hollow), indicating Each pixel in the array is a real number; Dimensions are OK The two-dimensional real matrix space of the column; Indicates the number of pixels (rows) in the height direction of the image; : Number of pixels (columns) in the width direction of the image; This invention is primarily aimed at small-area fingerprint scenarios, therefore a typical upper bound of up to 200 pixels is given for the image height and width.

[0028] According to an example of this embodiment, the first Scale of the first Each block is defined as:

[0029] Pixels falling outside the image are filled with zeros. The symbols in the formula have the following meanings: Indicated in scale Below, with An image patch extracted from the center; Indicates scale index. ; Represents the discrete index (integer pair) of the block in the sampling grid; Represents the local coordinates (horizontal and vertical) within the block, with a range of values. arrive ; Indicates the first The side length of the block of the scale (and) correspond); This represents the center-to-center spacing of the blocks, synonymous with the formula above. In this embodiment... The preferred value is 8; This represents the grayscale value of the original fingerprint image at a specified pixel location.

[0030] in, Indicates all that satisfy and A set of integer pixel positions.

[0031] According to an example of this embodiment, spectral response analysis is performed on the polar coordinates of each block to obtain the polar coordinate power spectrum of each block, including the following steps 4.2.2 to 4.2.4, wherein: Step 4.2.2: Locally remove the mean value from the pixel grayscale of each block, and then apply a Gaussian window to the blocks after removing the mean value.

[0032] The effect of local grayscale de-averaging in each block is to eliminate DC components and suppress window boundary leakage. The de-averaging process is performed on each block sequentially using the following formula:

[0033] in, This represents the image patch after the mean has been removed; Indicated in scale Below, with An image patch extracted from the center; This represents an operator that calculates the arithmetic mean of the gray levels of all pixels within a block. Representation block The grayscale mean (scalar) is used to eliminate zero-frequency components in the spectrum.

[0034] In this embodiment, applying a Gaussian window to the mean-removed block can be represented as:

[0035] in, This indicates a block that has been mean-trimmed and then windowed with a Gaussian window, used for subsequent FFT (fast Fourier transform) calculations; This represents the mean-removed block obtained in the previous step; This represents the pixel-by-pixel (Hadamard) product operator; The scale is represented as A two-dimensional isotropic Gaussian window function, at this scale Same size as the block.

[0036] Furthermore, the analytical form of the Gaussian window is:

[0037] in, Indicates the position of the Gaussian window within the block. The weight value at the location; Indicates local coordinates within the block; This indicates the center coordinates of the block, ensuring that the peak of the Gaussian window is located at the center of the block; Indicates the first The bandwidth (standard deviation) of the Gaussian window of the scale is taken in this embodiment. This ensures that the effective working range of the window is coordinated with the block size; This represents the natural exponential function.

[0038] Step 4.2.3: Perform a two-dimensional discrete Fourier transform on each windowed block, and convert each block after the discrete Fourier transform into a centered Cartesian power spectrum.

[0039] The purpose of performing a two-dimensional discrete Fourier transform on the windowed blocks in this embodiment is to ensure that the output describes the periodic frequency characteristics of the coordinate sequence changes, such as the repeating texture of the contour and the frequency of shape fluctuations. Specifically, a two-dimensional discrete Fourier transform can be performed on each windowed block according to the following formula to obtain the spectrum. :

[0040] in, The spectrum obtained after applying a windowed two-dimensional discrete Fourier transform is a complex number defined on a frequency domain grid. ; This represents discrete coordinates in the frequency domain (horizontal and vertical frequency indices), where... ; This represents the windowed block obtained in step 4.2.2; Indicates spatial coordinates within the block; Indicates block size; Represents the imaginary unit, where, It belongs to engineering symbols (and) distinguish); Represents the standard two-dimensional DFT complex exponential kernel; This indicates that the summation is performed on all pixel positions within the block.

[0041] According to one example of this embodiment, zero frequency can be transmitted via... Translate to the center The Cartesian power spectrum is obtained, and its formula is expressed as:

[0042] in, Let represent the centered two-dimensional Cartesian power spectrum, which is a non-negative real number; Represents the center coordinates in the frequency domain. ; The square of the modulus of a complex number, i.e., the sum of the squares of the real and imaginary parts, is used to convert a complex spectrum into a real power spectrum; other notation is the same as above. The visualization of this Cartesian power spectrum is as follows: Figure 3 As shown.

[0043] Step 4.2.4: Obtain the polar coordinate power spectrum of each block based on the Cartesian power spectrum.

[0044] First, define polar coordinates. ,in , By performing bilinear interpolation sampling on the Cartesian power spectrum, the polar coordinate power spectrum is obtained, which can be expressed by the formula:

[0045] in, Representing scale ,piece The polar power spectrum (real number); Represents the radial coordinate, corresponding to the frequency magnitude (related to the ridge frequency). ; Represents angular coordinates, corresponding to the spatial frequency direction (i.e., the ridge normal; the ridge orientation is orthogonal to it), and the domain is... This finite domain derives from the conjugate symmetry of real signals. ; Indicates the first The radial maximum value of the polar coordinate sampling on the scale is usually taken as ; This represents the coordinates of the frequency domain center (zero frequency) of the power spectrum after FFT-shift centering in step 4.2.3 above. , which serves as the pole (origin) for polar coordinate sampling; , Represented by frequency domain center Using the origin as the coordinate system, the polar coordinates are... Convert to Cartesian frequency domain sampling coordinates (sampling point relative to the center along the direction) Offset ); This indicates that due to the even symmetry of the power spectrum of the real signal about the center, the true values ​​of this pair of centrally symmetric frequency points are theoretically equal; in the discrete implementation, the two values ​​fall into different sampling grid neighborhoods through bilinear interpolation and have slightly different values. Explicit addition (equivalent to their mean) is used to expressly add them. (Times) can average interpolation errors and suppress single-point interpolation deviations; This represents the complex conjugate operator.

[0046] To focus on the normal frequency range of fingerprint ridges, the radial direction is calculated according to the following formula. Restricted valid range:

[0047] in, This represents the lower radial bound (low), corresponding to the maximum permissible period of the ridge line. In this embodiment, it is taken as... ; This represents the upper radial limit (high), corresponding to the minimum allowable period of the ridge line. In this embodiment, it is taken as... ; Indicates the first Scale block side length. This interval corresponds to approximately... Normalized frequency per week / pixel (i.e., ridge period approximately) (pixels) to cover the typical fingerprint ridge frequency range and achieve global measurement of ridge frequency.

[0048] The visualization of this polar coordinate power spectrum is as follows: Figure 4 As shown.

[0049] Step 102: Determine the angular marginal energy spectrum and ridge frequency estimate of each block based on the polar coordinate power spectrum.

[0050] According to an example of this embodiment, the determination of the angular marginal energy spectrum of each block based on the polar coordinate power spectrum in this step can be achieved using the following formula:

[0051] in, This represents the polar coordinate power spectrum. Indicated in scale ,piece The angular marginal energy spectrum obtained above is about direction A one-dimensional function; Represents the direction angle variable, with values... ; The radial frequency variable is used as the integrand. This indicates the lower and upper limits of the radial integration. The effective frequency range of this ridge line is described in step 4.2.4 above. This represents a radial weighting term, which physically means a volume element in polar coordinates. Provide the actual energy integral; Indicates radial Integrating over a one-dimensional variable. Based on the angular marginal energy spectrum in this embodiment. exist Precisely locate the peak within the ridge normal, such as Figure 5 As shown, the ridge normal represents the dominant frequency direction, and the ridge direction is its orthogonal direction, such as... Figure 5 It can be seen that when the true direction angle θ0, represented by the red dashed line, is 60°, the angular boundary determined based on the energy spectrum of that angular boundary ( Figure 5 The peak value of the blue line is slightly smaller, indicating that the energy spectrum at the edge of the angle cannot accurately reflect the direction angle in the global diagram.

[0052] Step 103: Normalize the marginal energy spectra of the corner at different scales in each block and then fuse them to obtain the fused direction spectrum of each block, and determine the direction estimate of each block based on the fused direction spectrum.

[0053] According to an example of this embodiment, this step involves normalizing and fusing the angular marginal energy spectra under small-scale blocks, medium-scale blocks, and large-scale blocks respectively to obtain the fused directional spectrum.

[0054] The orientation estimates for each block at each scale can be expressed as:

[0055] in, Representing scale ,piece The initial estimate of the single-scale ridge normal (in radians) is obtained from the direction where the angular value of the marginal spectrum reaches its maximum value—because the power spectrum energy is concentrated in the dominant frequency direction perpendicular to the ridge, this peak direction is the ridge normal (dominant frequency direction), and the ridge direction is... Direction estimation in this embodiment With the reconstruction direction field All refer to the direction of the ridge line normal (dominant frequency); Indicates in The independent variable that makes the function being calculated reach its maximum value within the range of its values.

[0056] According to an example of this embodiment, the angular marginal energy spectrum at different scales in each block is normalized using the following formula to obtain the normalized spectrum:

[0057] in, This represents the normalized single-scale angular spectrum, with values ​​ranging from 1 to 2. ; Indicates single-scale spectrum in The maximum value on the denominator is used as the normalized denominator. This indicates that it is used only as a temporary variable for finding the maximum value, and is related to the external... Irrelevant; This represents the aforementioned angular marginal energy spectrum.

[0058] Furthermore, the normalized spectrum is fused using the following formula to obtain the fused directional spectrum. :

[0059] in, Indicates in block The fused direction spectrum obtained after three-scale fusion; Indicates a set of indices at three scales. The product operator on; Represents a normalized single-scale angular spectrum; Representing scale In block The fusion weight at the location.

[0060] According to an example of this embodiment, the step of determining the orientation estimate of each block based on the fused orientation spectrum further comprises determining the final orientation estimate of each block after fusion using the following formula:

[0061] in, Representation block The final orientation estimate (in radians) after multi-scale fusion; the meanings of the remaining symbols are the same as above, except that the single-scale quantities are replaced. Replace with fusion spectrum .

[0062] This embodiment normalizes and fuses the marginal energy spectra at different scales in each block to obtain the fused direction spectrum of each block. Based on this fused direction spectrum, the direction estimate of each block is determined. The fusion is performed using a power product (weighted geometric mean) method. Since the product is suppressed when the spectral value at any scale is low, this fusion method can significantly suppress noise spurious peaks that only appear at a few scales, thereby sharpening the true main peak supported by consistent values ​​across all three scales. Furthermore, the normalized spectrum can be further refined by taking… To avoid degradation at zero, where, It is represented as a very small positive number.

[0063] According to an example of this embodiment, the fusion weight can be given by a reliability softmax function:

[0064] in, , Representing scale The reliability weight, k takes the value of The sum of the weights of the three scales is ; This represents the softmax sharpness parameter, which is dimensionless. A larger value indicates a stronger bias towards high reliability metrics; this embodiment recommends... interval ; Representing scale ,piece Reliability; Represents the temporary scale index traversed during summation; This represents the natural exponential function.

[0065] The reliability of each block at each scale can be determined using the following formula:

[0066] in, Representing scale ,piece Reliability of direction estimation (dimensionless, non-negative real number); molecule The denominator represents the energy value of the angular marginal energy spectrum at the direction of the main peak; Indicates the angular marginal energy spectrum in The mean of the above, its physical meaning is: when This indicates that the spectrum is almost uniformly distributed in the angular direction (direction is unreliable). This indicates that the energy is concentrated at the main peak (the direction is reliable).

[0067] Step 104: Measure the reliability of the direction estimate for each block, obtain the reliability of the direction estimate for each block, and count the first number of blocks with reliable direction estimates.

[0068] According to an example of this embodiment, the reliability of the orientation estimation for a certain block can be determined based on a preset orientation estimation threshold. To determine this, the preferred direction estimation threshold is 1.8. When the reliability of the direction estimation for a certain block is greater than this direction estimation threshold... At that time, determine the estimated direction of the block. Reliable, and this block will be included in the first quantity.

[0069] According to an example of this embodiment, the reliability of the directional estimation for each block can be measured by the following formula:

[0070] in, This indicates the reliability of the orientation estimation for each block after fusion (dimensionless). Figure 6 This is a schematic diagram illustrating the reliability weighting of fusing angular marginal energy spectra at different scales into a fused directional spectrum in one embodiment. The reliability of the directional estimation at each scale before and after fusion is shown in the figure. Figure 6As shown in the figure, red indicates the reliability of the direction estimation of the fused direction spectrum, green, blue, and orange indicate the reliability of the direction estimation of the direction spectrum at small, medium, and large scales, respectively, and the vertical gray dashed line indicates that the true direction angle is 75°. Figure 6 It can be seen that single-scale peaks exhibit obvious spurious peaks due to noise or local defects, and the fused directional spectrum... The main peak is sharpened, and spurious peaks are suppressed, significantly improving reliability. The reliability thermogram estimated based on this direction is shown below. Figure 8 As shown. The red area represents the low reliability zone (…). ), which requires physically consistent directional field extrapolation.

[0071] It should be noted that the reliability of the orientation estimation of the fused block is... Reliability of orientation estimation for the block at a single scale Both are spectral concentrations of "peak value / interval mean", and they are constant. Furthermore, the dimensions and physical meanings are consistent, differing only in the underlying spectrum; the direction estimation threshold... It is aimed at the fusion spectrum of The calibrated empirical values. The above formula calculates the fusion weights using softmax. In practice, it is customary to first subtract [the amount of something]. Take the index again to ensure No numerical overflow occurs.

[0072] The set of reliable blocks estimated according to the direction defined in this embodiment. It can be represented as:

[0073] in, This represents the set of indices of blocks whose direction estimation is reliable, where R stands for Reliable; The direction estimation threshold (scalar, typically taken as...) ), used to delineate reliable and unreliable regions; flat / background blocks are nearly isotropic in their angular spectrum, Automatically excluded In addition (if the sensor background contains structural noise, a foreground mask based on block variance / gradient consistency can be added before block segmentation to further remove background blocks); This indicates that the condition within the parentheses is satisfied. "All" A set of index pairs.

[0074] Step 105: Based on the first quantity, the direction estimate of each block, and the reliability of the direction estimate, reconstruct the direction field of each block included in the first quantity.

[0075] In one embodiment, reconstructing the orientation field of each block, which includes the first number, based on the orientation estimate of each block and the reliability of the orientation estimate, includes the following steps 01 to 05: Step 01: Derive the global orientation field of the block with reliable orientation estimation using a preset orientation model.

[0076] According to one example of this embodiment, the orientation model includes, but is not limited to, a zero-pole orientation model and a constant orientation model. The global orientation field can be represented as... .

[0077] Step 02: Determine the observation residual of the corresponding block based on the orientation estimates of each pixel in the reliable block and the global orientation field.

[0078] According to an example of this embodiment, the block can be calculated using the following formula. Observation residuals:

[0079] in, Indicates the first The directional residual of the block is used to obtain the model using the above formula. Then fell into (and direction field) (consistent cycle) This represents the final direction estimate after multi-scale fusion; Indicates the first The global orientation field of the block is obtained through the zero-pole orientation model or the constant orientation model.

[0080] Step 03: Based on the reliability of the direction estimation of the block with reliable direction estimation and the observation residual, solve the energy functional with the residual correction function as the variable to obtain the residual correction function of the corresponding block.

[0081] According to an example of this embodiment, the residual correction function can be obtained by minimizing the bending energy functional, and the energy functional for solving the residual correction function can be expressed as:

[0082] in, Indicated by The energy functional (scalar) is a variable. Describes a functional; This represents the two-dimensional residual function to be solved; Represents a reliable set of blocks; Indicates a reliable block index; Indicates the reliability weight after fusion; This indicates the value of the residual function at the center of the block; Represents the observation residual; This represents the TPS regularization weight (which controls the balance between the data fidelity term and the bending energy term). This embodiment recommends taking ; express Second-order partial derivatives with respect to the horizontal and vertical axes; Represents the two-dimensional bending energy density (the physical bending energy integral kernel of a thin plate); This represents the integration over a two-dimensional area element across the entire image plane.

[0083] Understandably, the TPS energy functional is in the form of a reliability-weighted data term for a standard thin-plate spline, and its minimum solution is closed-form: it can be expressed as a radial basis function. A linear combination with a first-order affine term, the coefficients of which are determined by a single... The system of linear equations (including reliability-weighted diagonal matrices and affine orthogonal constraints, taking a full-plane natural boundary) is uniquely determined, thus in The residual is approximated internally, and then smoothly extrapolated in the low-quality region using the minimum bending energy.

[0084] Step 04: If the first quantity is greater than or equal to the first preset quantity, the orientation field of each block included in the first quantity is reconstructed based on the global orientation field and the residual correction function.

[0085] According to an example of this embodiment, the first preset number is, for example, five. The reconstruction of the orientation field of each block included in the first number, based on the global orientation field and the residual correction function, can be achieved using the following formula:

[0086] in, This represents the final reconstructed global continuous direction field; Represents the global direction field; Indicates TPS at position The direction residual correction value is given at the location; Indicated by For the modulo operation, the direction field is guaranteed. Periodicity, making ; It represents the position of any pixel in the image.

[0087] Figure 7 This is a schematic diagram illustrating the global orientation field of a block with reliable orientation estimation derived through a zero-pole orientation model in one embodiment. The orientation field reconstructed according to this embodiment is as follows: Figure 7 As shown, Figure 7 The red circle in the diagram represents the core point, and the green triangle within it represents the triangular point.

[0088] Step 05: If the first quantity is less than the first preset quantity, the preset global orientation constant is used as the orientation field of each pixel in the fingerprint image.

[0089] According to an example of this embodiment, when the first quantity is less than the first preset quantity, preferably, the following two cases are discussed: This indicates insufficient control points for the thin-plate spline: skip TPS residual correction and directly take... ; This indicates that there are no reliable blocks in the entire graph, and the extreme degradation is to take a single constant direction given by the principal direction of the whole graph structure tensor as a catch-all (this quantity is a global statistic, which is different from the block-by-block local gradient method abandoned in the background technology of this invention. At this time, the enhancement target degenerates into conservative smoothing along this direction), and the diffusion iteration number in step four is increased accordingly. (e.g., take the upper end of the interval) ).

[0090] In one embodiment, the orientation model includes a zero-pole orientation model and a constant orientation model; deriving the global orientation field of a block with reliable orientation estimation through a preset orientation model includes steps 21 to 23, wherein: Step 21: If the first quantity is greater than or equal to the first preset quantity, determine whether there are singularities in the blocks where the direction estimation is reliable.

[0091] The first preset quantity is, for example, 5, that is This singularity includes, but is not limited to, the core and the delta.

[0092] In one embodiment, determining whether a singularity exists in a block with reliable direction estimation includes steps 31 to 34, wherein: Step 31: Determine the detected center pixel pixel by pixel in the block with reliable orientation estimation, and generate a closed curve of a preset shape centered on the center pixel. .

[0093] According to one example of this embodiment, the closed curve is a small square or a small circle. The center pixel is, for example, a point. .

[0094] Step 32: Integrate the orientation field of the block to which the center pixel belongs along the closed curve to obtain the Poincaré index.

[0095] Among them, the fingerprint orientation field has a period of Directional field (i.e.) Its global topology is determined by a finite number of singularities. It is understood that the estimated and reconstructed topology in this embodiment differs from that of the original. This is the ridge normal direction field, which differs from the ridge direction field only by a constant value. Since both have the same singularity location and Poincaré exponent, the following topological constraints and zero-pole modeling are applicable. The same applies. In this embodiment, along the closed curve The Poincaré integral of the direction field can be expressed by the formula:

[0096] in, Indicated by point Centered closed curve The Poincaré index around that point; Indicates the coordinates of the center pixel being detected; Indicates A small closed curve centered on; Indicates along a closed curve Perform a loop integral over the direction field; Represents the normalization coefficient, which makes the exponent value fall within... Three discrete values.

[0097] Step 33: If the Poincaré index is zero, determine that there are no singular points in the block to which the corresponding center pixel belongs.

[0098] Corresponding to the above formula middle, express This is a normal (non-singular) point.

[0099] Step 34: If the Poincaré index is not zero, determine that there is a singularity in the block to which the corresponding center pixel belongs.

[0100] Corresponding to the above formula middle," "express As the core point, "express It is a triangle point (delta).

[0101] It is understandable that, in a discrete directional field, along a closed curve The loop integral is based on the directional difference between adjacent pixels (taking the period). The minimum angle difference is approximated by accumulation, and the index is adjusted during detection. Set tolerance (typical ),Right now Judged as the core point The point identified as a triangulation point is considered a triangle point; the rest are considered normal points. In the formula... This is the singularity detection tolerance (dimensionless); increase this when it is necessary to relax the singularity detection threshold. This allows for the accommodation of greater directional field noise in strongly degraded scenarios.

[0102] Step 22: In the presence of the singularity, construct a zero-pole orientation model based on the horizontal and vertical coordinates of each pixel in the fingerprint image, and derive the global orientation field of the block with reliable orientation estimation through the zero-pole orientation model.

[0103] In this embodiment, the key to constructing the zero-pole orientation model based on the horizontal and vertical coordinates of each pixel in the fingerprint image lies in modeling the orientation field as the argument of a zero-pole rational function, which can be achieved through the following formula:

[0104] in, , is a complex variable introduced based on the x and y coordinates of the pixel. Represents the position variable on the complex plane; : The horizontal and vertical coordinates of pixels on the real plane; Represents the imaginary unit ( ); Indicates zero point and extreme points Describing rational functions (complex values); Indicates the number of zero points (core points); Indicates the number of poles (triangulation points); This indicates that it is used only as a product index in the formula, traversing the zero or pole indices (note the difference from the scale index in other steps such as step 4.2). The meanings are different, this formula within (It is a local variable); Indicates the first The position of the core point in the complex plane ( ); Indicates the first The order of a core point (typical) ); This indicates the position of the k-th triangulation point in the complex plane; Indicates the first The order of a triangulation point (typical) ); This represents the direction field value (in radians) derived from the zero-pole model. Representing complex numbers Argument (i.e., the angle between the real axis and the positive real axis); The period of the direction field is Therefore, half of the argument is taken.

[0105] In this embodiment, each singular point is taken as a first-order ( Therefore, each core point and triangulation point contributes to the Poincaré index. , This ensures the model strictly satisfies the topological constraints of the fingerprint orientation field; for small-area local fingerprints, the number of zeros... and the number of poles Unaffected by "core points" Triangle Points "This is constrained by the topological laws of the entire fingerprint (the field of view may contain only some singularities or even none). Furthermore, at the discrete pixels where the zero / extreme point exactly falls..." Undefined, the pixel orientation value is given by the limiting interpolation of its neighborhood orientation (or excluded by the singularity mask), and does not affect subsequent thin-plate spline correction and diffusion enhancement.

[0106] According to an example of this embodiment, the construction of the zero-pole orientation model needs to be performed under reliable region constraints. The parameter identification of the reliable region during the construction of the zero-pole orientation model can be achieved through steps 41 and 42, wherein: Step 41: Implement the following formula in the reliable region. Minimize the direction matching error and use a period of Orientation difference metric:

[0107] in, Indicates two directions , The period between Distance (minimum angular difference); and Indicates two direction angles (in radians) to be compared; Represents the absolute value of the difference in normal angles; This indicates that the period of the directional field is considered. The corresponding "bypass" distance; This indicates that the smaller value is taken.

[0108] Step 42: Determine the parameters of the zero-pole direction model using the following formula:

[0109] in, The set of core points and triangle points obtained through optimization; The value that minimizes the objective function among all combinations of zero and pole locations; A set of blocks with reliable orientation estimation; Only for reliable blocks Summation of blocks within; Reliability weighting makes more reliable blocks contribute more to parameter identification; :cycle The square of the direction difference (guaranteed to be non-negative and differentiable almost everywhere, except when...) The fact that it is not differentiable on the zero test set does not affect Levenberg's algorithm. Marquardt numerical optimization); Zero-pole model at the center of the block The predicted direction given at the location; : Final direction estimation after fusion; The physical pixel coordinates of the block center, i.e. .

[0110] The number of zeros and poles is determined by the Poincaré test within tolerance. The number of core points and triangulation points obtained below (i.e. and ), each take one order ( The location was solved using Levenberg-Marquardt numerical optimization, with initial values ​​given from the detected singularity candidates.

[0111] Step 23: In the absence of the singularity, derive the global orientation field of the block with reliable orientation estimation based on the preset constant orientation model.

[0112] According to an example of this embodiment, the preset constant direction model can be represented as follows: ,in, This represents the estimated single global direction constant. However, if no singularities are detected (common in small areas), it degenerates into a constant direction model. Only by reliable areas Reliability weighted least squares estimation of single global direction constant for upward direction estimation Then, the residuals are corrected by TPS (Thin Plate Spline).

[0113] in, Indicates modulus Argument is plural; It represents the estimated global direction constant (scalar), which can also be preset manually.

[0114] Step 106: Measure the reliability of the ridge frequency estimation for each block, obtain the reliability of the ridge frequency estimation for each block, and count the second number of blocks where the ridge frequency estimation is reliable and the direction estimation is also reliable.

[0115] According to one example of this embodiment, the reliability of the ridge frequency estimation can be compared with a preset frequency reliability threshold to determine whether the ridge frequency estimation is reliable. The frequency reliability threshold is, for example, 1.5. When the reliability of the ridge frequency estimation is greater than or equal to this frequency reliability threshold, the ridge frequency estimation of that block is determined to be reliable. For reliability.

[0116] According to an example of this embodiment, in order to obtain the ridge frequency estimate for each block, this embodiment needs to define the radial marginal energy spectrum based on the polar coordinate power spectrum:

[0117] in, Represents the power spectrum in polar coordinates. The radial marginal energy spectrum is related to the frequency magnitude. A one-dimensional function used to estimate ridge frequencies; This represents the direction variable, which is used as the integrand. Indicates direction Integrate the direction angle above; other symbols are the same as above.

[0118] Note: Angular marginal spectrum Multiply the polar coordinate volume element by the radial weighting factor To give the true energy integral; while the radial marginal spectrum Deliberately not adding Weighting, thereby avoiding high-frequency components due to The factor is over-amplified, which is beneficial for the accurate localization of the main peak of the ridge frequency. In the numerical implementation, exist Uniform sampling One point (typical) ,Right now Angular resolution) exist The samples were taken at equal intervals, and the integrals described above were approximated by discrete summation.

[0119] Furthermore, in this embodiment, the ridge frequency estimate of each block is determined using the following formula:

[0120]

[0121]

[0122] in, Representation block Ridge frequency estimation, Indicates in The independent variable that maximizes the function within its range of values; Representing scale ,piece The initial single-scale radial estimate (corresponding to the ridge frequency multiplied by the block size) is obtained by taking the maximum value of the radial marginal spectrum; Representing mesoscale blocks Radial initial estimate; Indicates the side length of the mesoscale block; This indicates that the units are "intra-block frequency indexes". Converted to physical frequency in "weeks per pixel" (i.e.) ).

[0123] In this embodiment, the frequency estimation uses the mesoscale radial spectrum to balance resolution (high noise at small scales) and stability (insufficient resolution at large scales); this block frequency Valid only on reliable frequency blocks, the frequency field for Gabor enhancement is ultimately obtained by extrapolating to low-quality regions and performing pixel-level continuation. .

[0124] Due to the above ridge frequency estimation The given information is a ridge frequency estimate for the block, but subsequent directional Gabor kernels require accurate ridge frequencies at each pixel. and cosine phase Unlike the directional field, the ridge frequency has no... A periodic scalar field without topological singularities represents a continuous variation in the density of ridges in adjacent regions. Since there is no directional winding around the core / triangle point, zero-pole topology modeling is unnecessary. The aforementioned reliability-weighted thin-plate spline (TPS) framework can be directly used to transform the block frequency into a globally consistent, pixel-level continuous frequency field. This fills the gap in "block frequency". The data stream breakpoints at the "pixel-level frequency" also provide physically consistent frequency extrapolations for low-quality (to be repaired) areas.

[0125] According to an example of this embodiment, the first [number] is calculated according to the following formula. Reliability of the ridge frequency estimation of the block (ratio of the main peak value to the interval mean):

[0126] in, Indicates the first Block ridge frequency reliability metric (dimensionless, non-negative real number), superscript This indicates that it is for frequency (as opposed to directional reliability). ); Representing mesoscale blocks The radial marginal energy spectrum, ;molecular This represents the energy value of the radial spectrum at the main peak frequency; Denotes the endpoints of the effective radial interval, where, ; denominator This represents the mean of the radial spectrum over the effective interval, achieving the addition of a minimum positive number. To avoid division by zero; This indicates that the radial spectrum is nearly flat within the frequency band (frequency is unreliable). This indicates that energy is concentrated in a single ridge cycle (frequency reliable).

[0127] Furthermore, the set of blocks for which both the ridge frequency estimation and the direction estimation are reliable can be represented as:

[0128] in, This represents the index of the block set whose ridge frequency estimation is reliable and whose direction estimation is also reliable; "F" stands for Frequency. Additional constraints. Used to exclude pseudo-reliable blocks that are "instable in direction and whose radial spectrum main peaks are mostly caused by noise" in order to avoid introducing false features.

[0129] Step 107: Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, reconstruct the ridge frequency field of each block included in the second quantity.

[0130] According to one example of this embodiment, the reconstruction of the ridge frequency field of each block depends on the number of blocks for which the ridge frequency estimation is reliable and the direction estimation is also reliable. Different second numbers determine how the ridge frequency field is reconstructed.

[0131] In one embodiment, reconstructing the ridge frequency field of each block, which includes the second number, based on the ridge frequency estimate of each block and the reliability of the ridge frequency estimate, includes steps 51 to 53, wherein: Step 51: When the second quantity is greater than or equal to the second preset quantity, based on the ridge frequency estimate of the block included in the second quantity and the reliability of the ridge frequency estimate, the energy functional with the ridge frequency field as the variable is solved to obtain the initial ridge frequency field of the corresponding block, and the initial ridge frequency field is clipped to a preset interval to obtain the ridge frequency field of the corresponding block.

[0132] According to one example of this embodiment, the second preset quantity is, for example, 5, and the preset interval is, for example... When the second quantity is greater than or equal to the second preset quantity, the energy functional used to solve the ridge frequency field can be expressed as:

[0133] The meanings of the symbols in the formula are as follows: : with frequency field The energy functional (scalar) of the variable, subscript This indicates that it pertains to frequency; : The two-dimensional initial ridge frequency field to be solved, in cycles per pixel; A set of blocks whose ridge frequency estimation is reliable and whose direction estimation is also reliable; Frequency reliability weighting makes more reliable blocks contribute more to the fit; The frequency field is at the center of the block. The value at; : Ridge frequency estimation of the block, in cycles per pixel; Frequency field TPS regularization weight (controls the balance between data fidelity and bending energy terms). This embodiment recommends taking Because the data items for frequency and direction have different dimensions (cycle / pixel vs. radians). With direction field They cannot be directly interchanged and each needs to be calibrated separately. : Second-order partial derivatives with respect to the horizontal and vertical axes; Two-dimensional bending energy density, i.e., the physical bending energy integral kernel of a thin plate; Integrate the two-dimensional area element over the entire image plane.

[0134] According to an example of this embodiment, the initial ridge frequency field is clipped to a preset interval to obtain the ridge frequency field of the corresponding block, which can be achieved by the following formula:

[0135] in, This represents the final reconstructed global continuous, pixel-level ridge frequency field, measured in cycles per pixel, which is used for pixel-by-pixel directional Gabor enhancement in subsequent steps. This represents the frequency field TPS functional. The frequency field obtained before clipping; This indicates a pixel-by-pixel cropping operator that limits the frequency to [a certain value]. interval; Represents the lower and upper bounds (cycles / pixel) of the physical frequency band of the ridge line, taken as... , The frequency band indicated by this preset interval is strictly consistent with the radial effective interval in step 4.2.4 above—due to the physical frequency. Therefore This eliminates non-physiological sparse / dense ridge frequencies.

[0136] This embodiment uses the initial ridge frequency field Pixel-by-pixel cropping to the physical frequency band yields the final pixel-level frequency field for use by 108.

[0137] Step 52: If the second quantity is less than the second preset quantity and is not zero, determine the average ridge frequency of the corresponding block based on the ridge frequency estimate of the blocks included in the second quantity and the reliability of the ridge frequency estimate. The average ridge frequency The ridge frequency field of the corresponding block is determined.

[0138] When there are too few reliable frequency blocks (e.g.) When there is a large area of ​​defect or extremely low signal-to-noise ratio, the TPS fitting is no longer robust and degenerates into a reliability-weighted global average frequency. This serves as a constant frequency field to ensure that no spurious features are introduced during the feature enhancement process of the fingerprint image. According to an example of this embodiment, the average ridge frequency of the corresponding block can be calculated using the following formula:

[0139] in, Represents frequency reliability on reliable blocks Weighted average ridge frequency (weeks / pixel); This indicates the number of blocks whose ridge frequency estimation is reliable and whose direction estimation is also reliable.

[0140] Step 53: When the second quantity is zero, the preset default ridge frequency is determined as the ridge frequency field of the corresponding block.

[0141] According to an example of this embodiment, the preset default ridge frequency The unit is cycles per pixel, and the preferred value for the default ridge frequency is 8, corresponding to approximately The pixel ridge period is the typical width of a small-area fingerprint ridge, and .

[0142] In one embodiment, the different scales include a mesoscale. The ridge frequency estimate is determined based on the polar coordinate power spectrum. Specifically, spectral response analysis is performed on the polar coordinates of each block at the mesoscale to obtain the polar coordinate power spectrum of each block at the mesoscale, and the ridge frequency estimate of the mesoscale block is determined based on the polar coordinate power spectrum. .

[0143] The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block. A second number of blocks with reliable ridge frequency estimation and reliable orientation estimation is then counted. Specifically, the reliability of the ridge frequency estimation for each mesoscale block is measured to obtain the reliability of the ridge frequency estimation for each mesoscale block, and a second number of mesoscale blocks with reliable ridge frequency estimation and reliable orientation estimation are then counted. .

[0144] Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed. Specifically, based on the second quantity, the ridge frequency estimate of each mesoscale block, and the reliability of the ridge frequency estimate of the mesoscale block, the ridge frequency field of each block included in the second quantity is reconstructed.

[0145] Step 108: Based on the orientation field and the ridge frequency field, the broken ridges in the fingerprint image are continuously reconnected and enhanced. Under the condition of satisfying the preset stopping condition, the fingerprint image with enhanced features is obtained.

[0146] According to one example of this embodiment, the preset stopping condition is, for example, the number of iterations reaches a preset number T, or the result obtained from two adjacent iterations. , This is the preset error.

[0147] According to one example of this embodiment, the continuous reconnection and enhancement of broken ridges in the fingerprint image can be achieved through orientation-aware anisotropic diffusion—Gabor iterative collaborative enhancement (DADAGE, Dual Attention Geometry Estimation), which uses a pre-reconstructed orientation field. With ridge frequency field As a priori, an iterative operator is constructed that first performs anisotropic diffusion smoothing and then directional Gabor enhancement to avoid the pseudo-structure generated by single-step Gabor filtering at low signal-to-noise ratios. Specifically, this includes steps 4.5.1 to 4.5.4, wherein: Step 4.5.1, Anisotropic diffusion tensor.

[0148] Define the directional unit vector and its orthogonal vectors:

[0149] in, : in pixels Along the ridge line ( A two-dimensional unit vector (column vector); : in pixels The location is perpendicular to the ridge line (along the ridge line normal). A two-dimensional unit vector (column vector); The reconstructed global orientation field (ridge normal / dominant frequency direction; ridge orientation is...) ); : Vector transpose (converts horizontal notation to column vector); and Standard trigonometric functions.

[0150] Define the anisotropic diffusion tensor:

[0151] in: : in pixels place Anisotropic diffusion tensor (symmetric positive definite matrix); The diffusion coefficient (positive real number) along the ridge direction is taken in this invention. ; The diffusion coefficient (positive real number) in the direction perpendicular to the ridge line is taken in this invention. ; , The outer product of unit vectors yields the rank. of Matrix; satisfying This results in strong conduction along the ridge line and weak conduction in the vertical direction, thus connecting the break along the ridge line while maintaining edge sharpness in the vertical direction.

[0152] Step 4.5.2, Partial Differential Equation for Anisotropic Diffusion.

[0153] The diffusion process satisfies:

[0154] in, : in the Images at each time step (time is a continuous variable, corresponding to the diffusion iteration process); : Diffusion time variable (dimensionless, corresponding to iteration progress); : The partial derivative of the image with respect to diffusion time; Spatial gradient vector of an image ; The anisotropic diffusion tensor defined in the above steps; : Matrix-vector product of diffusion tensor and gradient (resulting in a two-dimensional flux vector). : Divergence operator (divergence of a vector field).

[0155] Furthermore, this embodiment employs an explicit difference scheme for discretization (step size). ):

[0156] in: Indicates the first Images from the next diffusion iteration; This represents the image for the next iteration.

[0157] Indicates the time step (real number) in the explicit format; in the unit grid ( Under the standard central difference dispersion discretization, the stability (CFL) condition is taken as follows: (Based on the maximum eigenvalue of the diffusion tensor) (Decision); including non-zero cross terms The strict stability bound is slightly tighter than this value, therefore the present invention takes... To allow for a safety margin; the image boundaries are subject to the Neumann zero-throughput condition (…). To avoid boundary artifacts; : These represent the partial derivatives with respect to the horizontal and vertical axes (implemented using difference); , and The elements of the diffusion tensor matrix, a symmetric matrix, satisfy... .

[0158] Furthermore, the analytical expression for the elements of the square matrix is:

[0159]

[0160]

[0161] in This indicates that the orientation field value at each pixel location is taken.

[0162] Iteration The image is obtained after smoothing along the direction. ,in It is the number of inner iteration steps (a dimensionless integer, usually taken as...). When there is a large area of ​​damage, it can increase to ).

[0163] Step 4.5.3, directional Gabor core enhancement.

[0164] In each pixel Construction direction Frequency-adaptive Gabor core:

[0165] The meanings of the symbols in the formula are as follows: : Directional Gabor kernel function (depends on parameter direction) Frequency field ); Local coordinates (horizontal and vertical) inside the kernel function, range of values ; Gabor kernel sampling window side length (odd number); to avoid the Gaussian envelope and cosine carrier being truncated by a fixed small window at low frequencies (sparse ridges), Adaptive to ridge frequency and ridge period Proportional, take (Approximately covered) One ridge cycle, in The above is approximately ); The meanings of the other symbols in the formula are as follows: : The orientation field value of this pixel; The ridge frequency of this pixel is taken from the pixel-level continuous frequency field reconstructed in 4.4.5. (as a parameter); : Crossing the ridge line direction (along the ridge line normal, i.e., rotating coordinates) The standard deviation of the Gaussian envelope (where the cosine carrier oscillates across the ridge along this direction) is taken as... ; Along the ridgeline direction (rotation coordinates) The standard deviation of the Gaussian envelope (direction) is taken as... , (A longer envelope along the ridge line facilitates aggregation and connection of broken parts along the ridge). In this formula, the aspect ratio of the Gabor envelope is represented (as shown in 4.2 Mesoscale Index). Unlike the previous example, this one uses local parameters. :Will Along direction Local coordinates obtained after rotation; : Natural exponential function; : Cosine function; Cosine carrier phase.

[0166] To ensure that the real-valued kernel has no response to uniform grayscale (i.e., zero DC gain), zero-mean normalization is applied to the kernel during implementation. ( Within the sampling window (mean); the Gabor filter output is then normalized by local contrast to make it consistent with the anisotropic diffusion smoothed image. Being at the same grayscale ensures that the collaborative iteration in step 4.5.4 is performed correctly. The weighted fusion has a consistent numerical meaning between the two.

[0167] This can be obtained through coordinate rotation relationships:

[0168]

[0169] in, Crossing the ridge line (along the ridge line normal direction) The rotated coordinates of the cosine carrier wave oscillate across the ridge line along this direction; : Along the ridgeline ( The rotated coordinates of ); the other symbols are the same as above.

[0170] Convolve the Gabor kernel with the diffuse-smoothed image:

[0171] The meanings of the symbols in the formula are as follows: Gabor-enhanced images at the pixel level The grayscale value at that location; The image obtained after the anisotropic diffusion iteration in step 4.5.2.

[0172] in, :exist The Gabor core constructed at location (parameters vary with location); Represents a two-dimensional convolution operator; due to kernel parameters The filtering is done pixel-by-pixel; this is actually a position-dependent filtering process—applying to each pixel... The neighborhood constructed from the kernel's local parameters is obtained by weighted summation using the following formula (rather than global translation-invariant convolution):

[0173] Step 4.5.4, Diffusion Gabor Cooperative Iteration.

[0174] Define the cooperative iterative operator:

[0175] The meanings of the symbols in the formula are as follows: : Diffusion Gabor cooperative operator, Represents operator mapping; Images before and after iteration; First, perform anisotropic diffusion on the current image (from step 4.5.2, iteration). The smoothed image obtained in the first step; it reflects the sequential order of "diffusion first, then Gabor"—the Gabor operator acts on the diffused image. (Same as step 4.5.3); Mixed weights (real numbers, values) This embodiment recommends ), in diffusion smoothing results The results are weighted and fused with the Gabor enhancement results to avoid pseudo-structures caused by over-enhancement of single-step Gabor while reconnecting ridges; : Parameters are The directional Gabor convolution operator, from step 4.5.3; Among them, the number of external collaborative iterations (Dimensionless integer, conventionally taken) When there is a large area of ​​damage, it can increase to ), final output Diffusion tensor With Gabor parameters The direction / frequency field remains unchanged throughout all outer iterations and is determined by the direction / frequency field reconstructed in a single iteration; therefore, the total number of diffusion steps is... ; each round Weighted fusion reconnects fractured ridges while continuously resetting the ridge-valley structure with Gabor enhancement terms, avoiding repeated diffusion and degradation into trivial (overly smooth) solutions. Iteration is performed by default with a fixed execution method. Step, also available Stop in advance.

[0176] Figure 9 This is a schematic diagram illustrating the process of reconnecting and enhancing broken ridges in a fingerprint image in one embodiment. (DA) The process of DAGE processing is as follows: Figure 9 As shown, Figure 9 (a) represents the input fracture ridge line. (b) indicates that the anisotropic diffusion tensor extends into an ellipse along the ridge direction; (c) indicates that the diffusion is smooth along the direction and remains perpendicular to the direction; (d) indicates that Gabor synergistic enhancement causes the ridge to break and be reconnected.

[0177] To avoid excessive smoothing during the above steps, which could lead to a loss of edge contrast between ridges and valleys, a dual-tree complex wavelet transform (DTCWT) is introduced in the direction perpendicular to the ridges to compensate for high-frequency details. This is applied to the original image. With collaborative iterative output Each work Layer DTCWT decomposition:

[0178] The meanings of the symbols in the formula are as follows: Dual-Tree Complex Wavelet Transform operator; Original image In all The set of wavelet subband coefficients on; Output The set of wavelet subband coefficients; Decomposition level index, ; Total number of DTCWT decomposition levels (typical) ); Directed Subband Index (DTCWT) is generated at each level. Each directional sub-band corresponds to approximately Six directions, therefore ); : No. Layer, First The complex wavelet coefficients (complex matrix) of the subband.

[0179] According to one example of this embodiment, the original image details are compensated only on the sub-band closest to the direction perpendicular to the ridge line:

[0180] The meanings of the symbols in the formula are as follows. : The synthesized target wavelet coefficients; : Wavelet coefficients output by collaborative iteration; Wavelet coefficients of the original image; Other sub-bands; High-frequency detail compensation intensity (real number, This invention recommends ); : Selects the subband index mapping and returns the DTCWT orientation subband index that is closest to the given direction, i.e. ,in For the first The center direction of each directional sub-band (corresponding to) ), The period defined in the above steps Directional difference; here, the direction is taken as the direction perpendicular to the ridge line, i.e., the ridge line normal. It itself (first take the mold) Remapping); The direction field (ridge line normal) reconstructed by the above steps; it is itself perpendicular to the ridge line, crosses the ridge line, and bears the ridge line. The direction of the high-frequency edge of the valley, therefore the high-frequency compensation should be selected to be closest to it. directional sub-bands; : Piecewise function notation, which gives different rules for taking values ​​under two conditions.

[0181] Two points need to be clarified: First, the selection of the directional sub-band. Compensation is performed point-by-point according to the spatial location of the coefficients—for each wavelet coefficient, compensation is performed based on its spatial location. (Spine normal) Selects the corresponding sub-band, so compensation is spatially localized within each sub-band, rather than being uniformly switched on and off for the entire sub-band; secondly, to avoid treating original image noise as "detail" in low-quality areas, the compensation intensity is gated according to reliability: only in reliable areas (this location) Injecting high-frequency and low-quality regions from the original image. Maintain the smooth extrapolation result from step three.

[0182] The final image is obtained by reversing DTCWT: .

[0183] in, This represents the dual-tree complex wavelet inverse transform operator; This represents the set of subband coefficients after synthesis; This represents the intermediate image (before cropping) obtained after wavelet detail compensation, and the final output is the cropped image. .

[0184] Figure 10 This is a schematic diagram illustrating the implementation flow of a fingerprint image feature enhancement method in one embodiment, as shown below. Figure 10 As shown in the example of this embodiment, the final output is a feature-enhanced fingerprint image. It can be represented as:

[0185] in, This represents the wavelet-compensated intermediate image obtained before cropping, as described in the above steps. This represents the inverse complex wavelet transform of a dual-tree system. This represents the set of synthetic subband coefficients obtained in the above steps; This indicates a pixel-by-pixel cropping operator that limits the pixel grayscale value to a certain range. interval (less than) Pick ,big Pick ).

[0186] Figure 11 This is a schematic diagram comparing the effects of applying the fingerprint image feature enhancement method of the present invention in one embodiment. Figure 11 In the image, (a) represents the truth map of the fingerprint image. Figure 11In this context, (e) represents the true orientation field of the fingerprint image. Figure 11 (b) in the figure indicates that the truth graph is degraded. Figure 11 In the diagram, (c), (f), and (g) represent the effect of using traditional methods to enhance the features of degraded fingerprint images. Figure 11 In the diagram, (d) represents the effect of using the present invention to enhance the features of a degraded fingerprint image. Figure 11 In the diagram, (h) represents a schematic diagram of the orientation field for reconstructing a degraded fingerprint image using the present invention.

[0187] According to one application scenario of this embodiment, for example: targeting large-area defects (defect rate) ), singularity detection tolerance Relax (increase its value) and allow the directional model to be a degenerate constant directional model; at the same time, cooperative iteration will be implemented. Diffusion internal iteration Increase to the upper end of the recommended range respectively , This enhances the ability to extrapolate and repair large-area defects; it can achieve a Rank 1 hit rate. Its recognition rate is superior to traditional methods such as FingerGAN and it significantly reduces false details.

[0188] The beneficial effects of this embodiment include: The orientation estimation is more robust. The reliability-weighted fusion of multi-scale polar spectral responses significantly improves the robustness of orientation estimation over small areas, and the orientation error is reduced compared to the single-scale STFT short-time Fourier transform. ; Large-area low-quality regions can be physically and consistently extrapolated. The zero-pole topology model ensures that the orientation field satisfies the Poincaré constraint; thin-plate spline hybrid correction provides accurate approximation in reliable regions and smooth extrapolation in unreliable regions, avoiding erroneous enhancement. Ridge frequencies are also physically and consistently extrapolated to pixel-level frequency fields using the same reliability-weighted thin-plate splines. This avoids the directional Gabor from amplifying along the wrong period due to frequency inaccuracy in the low-quality region, thus preventing enhancement failure. No false details are introduced. Direction awareness diffusion. Gabor iteration strictly reinforces ridges in the direction of topological constraints. This pure operator design effectively suppresses the "created ridges" problem that is prone to occur in GAN-like methods. With annotation and further enabling optional differentiable minutiae fidelity loss, the false minutiae rate is lower than that of FingerGAN. ; It features low computational complexity and can be deployed in real-time on edge devices. The core operators include only FFT, TPS closed-form solution, explicit differential diffusion, and small Gabor convolution; no large neural networks are involved. Typical computation time... A fingerprint security island that can be embedded in a mobile SoC; It is highly interpretable and easy to tune. Each step has a clear mathematical and physical meaning, and can be controllably adjusted for different sensors and skin conditions. Its cross-domain generalization is significantly better than that of black-box deep learning models. Seamlessly compatible with existing matchers. Output is a standard grayscale enhanced fingerprint image, which can be directly integrated with commercial and open-source matchers such as VeriFinger, NBIS, and MCC without requiring modification to downstream processes.

[0189] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages in other steps. It is understood that the steps in different embodiments can be freely combined as needed, and all non-contradictory solutions formed by such combinations are within the scope of protection of this application.

[0190] Based on the same inventive concept, this application also provides a fingerprint image feature enhancement apparatus for implementing the fingerprint image feature enhancement method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more fingerprint image feature enhancement apparatus embodiments provided below can be found in the limitations of the fingerprint image feature enhancement method described above, and will not be repeated here.

[0191] In one exemplary embodiment, such as Figure 12 As shown, a fingerprint image feature enhancement device 100 is provided, including: a block segmentation module 11, a frequency estimation module 12, a direction estimation module 13, a direction reliability measurement module 14, a direction field reconstruction module 15, a frequency reliability measurement module 16, a ridge frequency field reconstruction module 17, and a reconnection module 18, wherein: Block module 11 is used to sample the fingerprint image to be processed at different scales to obtain multiple blocks, and to perform spectral response analysis on the polar coordinates of each block to obtain the polar coordinate power spectrum of each block. Frequency estimation module 12 is used to determine the angular marginal energy spectrum and ridge frequency estimation of each block based on the polar coordinate power spectrum; The direction estimation module 13 is used to normalize and fuse the marginal energy spectra of the corner at different scales in each block to obtain the fused direction spectrum of each block, and to determine the direction estimate of each block based on the fused direction spectrum. The directional reliability measurement module 14 is used to measure the directional estimated reliability of each block, obtain the directional estimated reliability of each block, and count the first number of blocks with reliable directional estimates. The orientation field reconstruction module 15 is used to reconstruct the orientation field of each block included in the first number based on the first number, the orientation estimate of each block, and the reliability of the orientation estimate. The frequency reliability measurement module 16 is used to measure the reliability of the ridge frequency estimation of each block, obtain the ridge frequency estimation reliability of each block, and count the second number of blocks whose ridge frequency estimation is reliable and whose direction estimation is also reliable. Ridge frequency field reconstruction module 17 is used to reconstruct the ridge frequency field of each block included in the second number based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate. The reconnection module 18 is used to continuously reconnect and enhance the broken ridges in the fingerprint image based on the orientation field and the ridge frequency field, and obtain a feature-enhanced fingerprint image when the preset stopping conditions are met.

[0192] In one embodiment, the orientation field reconstruction module 15 specifically includes: The global orientation field export unit is used to export the global orientation field of a block with reliable orientation estimation through a preset orientation model. The observation residual determination unit is used to determine the observation residual of the corresponding block based on the orientation estimates of each pixel in the block with reliable orientation estimates and the global orientation field. The residual correction function determination unit is used to solve the energy functional with the residual correction function as a variable based on the direction estimation reliability of the block with reliable direction estimation and the observation residual, so as to obtain the residual correction function of the corresponding block. The orientation field reconstruction unit is used to reconstruct the orientation field of each block included in the first number based on the global orientation field and the residual correction function when the first number is greater than or equal to the first preset number. The orientation field determination unit is used to use a preset global orientation constant as the orientation field of each pixel in the fingerprint image when the first quantity is less than the first preset quantity.

[0193] In one embodiment, the orientation model includes a zero-pole orientation model and a constant orientation model; the global orientation field derivation unit further includes: The singularity determination unit is used to determine whether there are singularities in the blocks with reliable direction estimation when the first number is greater than or equal to the first preset number. The first global orientation field derivation unit is used to construct a zero-pole orientation model based on the horizontal and vertical coordinates of each pixel in the fingerprint image when the singular point exists, and derive the global orientation field of the block with reliable orientation estimation through the zero-pole orientation model. The second global orientation field derivation unit is used to derive the global orientation field of a block with reliable orientation estimation based on a preset constant orientation model when the singular point does not exist.

[0194] In one embodiment, the singularity determination unit further includes: The curve generation unit is used to determine the detected center pixel pixel pixel by pixel in the block with reliable orientation estimation, and generate a closed curve of a preset shape with the center pixel as the center. The integration unit is used to integrate the orientation field of the block to which the center pixel belongs along the closed curve to obtain the Poincaré index. The first judgment unit is used to determine that there are no singular points in the block to which the corresponding center pixel belongs when the Poincaré index is zero. The second judgment unit is used to determine whether there is a singular point in the block to which the corresponding center pixel belongs when the Poincaré index is not zero.

[0195] In one embodiment, the ridge frequency field reconstruction module 17 further includes: The first ridge frequency field determining unit is used to solve the energy functional with the ridge frequency field as a variable based on the ridge frequency estimate of the blocks included in the second number and the reliability of the ridge frequency estimate when the second number is greater than or equal to the second preset number, to obtain the initial ridge frequency field of the corresponding block, and to cut the initial ridge frequency field to a preset interval to obtain the ridge frequency field of the corresponding block. The second ridge frequency field determining unit is used to determine the average ridge frequency of the corresponding block based on the ridge frequency estimate of the block included in the second number and the reliability of the ridge frequency estimate when the second number is less than the second preset number and is not zero, and to determine the average ridge frequency as the ridge frequency field of the corresponding block. The third ridge frequency field determination unit is used to determine the preset default ridge frequency as the ridge frequency field of the corresponding block when the second quantity is zero.

[0196] In one embodiment, the different scales include the mesoscale, and the frequency estimation module is specifically used to: perform spectral response analysis on the polar coordinates of each block at the mesoscale to obtain the polar coordinate power spectrum of each block at the mesoscale, and determine the ridge frequency estimate of the mesoscale block based on the polar coordinate power spectrum. The frequency reliability measurement module 16 is specifically used to: measure the reliability of the ridge frequency estimation of the mesoscale block, obtain the ridge frequency estimation reliability of each mesoscale block, and count the second number of mesoscale blocks where the ridge frequency estimation is reliable and the direction estimation is also reliable. The ridge frequency field reconstruction module 17 is specifically used to: reconstruct the ridge frequency field of each block included in the second number based on the second quantity, the ridge frequency estimate of each mesoscale block, and the reliability of the ridge frequency estimate of the mesoscale block.

[0197] Each module in the aforementioned fingerprint image feature enhancement device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0198] In one exemplary embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 13As shown, the computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When executed by the processor, the computer program implements a fingerprint image feature enhancement method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of the computer device can be a touch layer covering the display screen, or buttons, trackballs, or touchpads set on the casing of the computer device, or external keyboards, touchpads, or mice, etc.

[0199] Those skilled in the art will understand that Figure 13 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0200] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps: The fingerprint image to be processed is sampled at different scales to obtain multiple blocks. The polar coordinates of each block are analyzed to obtain the polar power spectrum of each block. The angular marginal energy spectrum and ridge frequency estimate of each block are determined based on the polar coordinate power spectrum. The marginal energy spectra of the corner at different scales in each block are normalized and then fused to obtain the fused direction spectrum of each block, and the direction estimate of each block is determined based on the fused direction spectrum. The reliability of the direction estimate for each block is measured to obtain the reliability of the direction estimate for each block, and the first number of blocks with reliable direction estimates is counted. Based on the first quantity, the direction estimate of each block, and the reliability of the direction estimate, the direction field of each block included in the first quantity is reconstructed; The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation are counted. Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed. Based on the orientation field and the ridge frequency field, the broken ridges in the fingerprint image are continuously reconnected and enhanced. Under the condition of satisfying the preset stopping condition, the fingerprint image with enhanced features is obtained.

[0201] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: The fingerprint image to be processed is sampled at different scales to obtain multiple blocks. The polar coordinates of each block are analyzed to obtain the polar power spectrum of each block. The angular marginal energy spectrum and ridge frequency estimate of each block are determined based on the polar coordinate power spectrum. The marginal energy spectra of the corner at different scales in each block are normalized and then fused to obtain the fused direction spectrum of each block, and the direction estimate of each block is determined based on the fused direction spectrum. The reliability of the direction estimate for each block is measured to obtain the reliability of the direction estimate for each block, and the first number of blocks with reliable direction estimates is counted. Based on the first quantity, the direction estimate of each block, and the reliability of the direction estimate, the direction field of each block included in the first quantity is reconstructed; The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation are counted. Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed. Based on the orientation field and the ridge frequency field, the broken ridges in the fingerprint image are continuously reconnected and enhanced. Under the condition of satisfying the preset stopping condition, the fingerprint image with enhanced features is obtained.

[0202] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps: The fingerprint image to be processed is sampled at different scales to obtain multiple blocks. The polar coordinates of each block are analyzed to obtain the polar power spectrum of each block. The angular marginal energy spectrum and ridge frequency estimate of each block are determined based on the polar coordinate power spectrum. The marginal energy spectra of the corner at different scales in each block are normalized and then fused to obtain the fused direction spectrum of each block, and the direction estimate of each block is determined based on the fused direction spectrum. The reliability of the direction estimate for each block is measured to obtain the reliability of the direction estimate for each block, and the first number of blocks with reliable direction estimates is counted. Based on the first quantity, the direction estimate of each block, and the reliability of the direction estimate, the direction field of each block included in the first quantity is reconstructed; The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation are counted. Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed. Based on the orientation field and the ridge frequency field, the broken ridges in the fingerprint image are continuously reconnected and enhanced. Under the condition of satisfying the preset stopping condition, the fingerprint image with enhanced features is obtained.

[0203] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0204] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0205] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0206] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for feature enhancement of fingerprint images, characterized in that, The method includes: The fingerprint image to be processed is sampled at different scales to obtain multiple blocks. The polar coordinates of each block are analyzed to obtain the polar power spectrum of each block. The angular marginal energy spectrum and ridge frequency estimate of each block are determined based on the polar coordinate power spectrum. The angular marginal energy spectra at different scales in each block are normalized and then fused to obtain the fused direction spectrum of each block, and the direction estimate of each block is determined based on the fused direction spectrum. The reliability of the orientation estimation for each block is measured to obtain the reliability of the orientation estimation for each block, and a first number of blocks with reliable orientation estimation is counted. Based on the first quantity, the direction estimate of each block, and the reliability of the direction estimate, the direction field of each block included in the first quantity is reconstructed; The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation are counted. Based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, the ridge frequency field of each block included in the second quantity is reconstructed. Based on the orientation field and the ridge frequency field, the broken ridges in the fingerprint image are continuously reconnected and enhanced. Under the condition of satisfying the preset stopping condition, a fingerprint image with enhanced features is obtained. The step of reconstructing the orientation field of each block, which includes the first number, based on the orientation estimate of each block and the reliability of the orientation estimate, includes: The global orientation field of a block with reliable orientation estimation is derived by using a preset orientation model; Based on the orientation estimates of each pixel in the block with reliable orientation estimates and the global orientation field, the observation residuals of the corresponding blocks are determined; Based on the direction estimation reliability of the block with reliable direction estimation and the observation residual, the energy functional with the residual correction function as the variable is solved to obtain the residual correction function of the corresponding block; When the first quantity is greater than or equal to the first preset quantity, the orientation field of each block included in the first quantity is reconstructed according to the global orientation field and the residual correction function; When the first quantity is less than the first preset quantity, the preset global orientation constant is used as the orientation field of each pixel in the fingerprint image.

2. The fingerprint image feature enhancement method according to claim 1, characterized in that, The orientation model includes a zero-pole orientation model and a constant orientation model; deriving the global orientation field of a block with reliable orientation estimation through a preset orientation model includes: If the first quantity is greater than or equal to the first preset quantity, determine whether there are singularities in the blocks where the direction estimation is reliable; In the presence of the singularity, a zero-pole orientation model is constructed based on the x-coordinate and y-coordinate of each pixel in the fingerprint image, and the global orientation field of the block with reliable orientation estimation is derived through the zero-pole orientation model. In the absence of the singularity, the global orientation field of the block with reliable orientation estimation is derived based on a preset constant orientation model.

3. The fingerprint image feature enhancement method according to claim 2, characterized in that, The determination of whether there are singularities in the blocks with reliable direction estimation includes: In a block where the orientation estimation is reliable, the detected center pixel is determined pixel by pixel, and a closed curve of a preset shape is generated with the center pixel as the center. The Poincaré index is obtained by integrating the orientation field of the block to which the center pixel belongs along the closed curve. When the Poincaré index is zero, it is determined that there are no singular points in the block to which the corresponding center pixel belongs; If the Poincaré index is not zero, it is determined that there is a singularity in the block to which the corresponding center pixel belongs.

4. The fingerprint image feature enhancement method according to claim 1, characterized in that, The step of reconstructing the ridge frequency field of each block included in the second number, based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, includes: When the second quantity is greater than or equal to the second preset quantity, based on the ridge frequency estimation of the blocks included in the second quantity and the reliability of the ridge frequency estimation, the energy functional with the ridge frequency field as the variable is solved to obtain the initial ridge frequency field of the corresponding block, and the initial ridge frequency field is clipped to a preset interval to obtain the ridge frequency field of the corresponding block. When the second quantity is less than the second preset quantity and is not zero, the average ridge frequency of the corresponding block is determined based on the ridge frequency estimation of the block included in the second quantity and the reliability of the ridge frequency estimation, and the average ridge frequency is determined as the ridge frequency field of the corresponding block. When the second quantity is zero, the preset default ridge frequency is determined as the ridge frequency field of the corresponding block.

5. The fingerprint image feature enhancement method according to claim 1, characterized in that, The different scales include mesoscale, and the ridge frequency estimate is determined based on the polar coordinate power spectrum, including: Spectral response analysis is performed on the polar coordinates of each block at the mesoscale to obtain the polar coordinate power spectrum of each block at the mesoscale, and the ridge frequency estimate of the mesoscale block is determined based on the polar coordinate power spectrum. The reliability of the ridge frequency estimation for each block is measured to obtain the reliability of the ridge frequency estimation for each block, and a second number of blocks with reliable ridge frequency estimation and reliable direction estimation is counted. Specifically: The reliability of the ridge frequency estimation of the mesoscale block is measured to obtain the reliability of the ridge frequency estimation of each mesoscale block, and a second number of mesoscale blocks with reliable ridge frequency estimation and reliable orientation estimation are counted. The step of reconstructing the ridge frequency field of each block included in the second quantity, based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate, specifically involves: Based on the second quantity, the ridge frequency estimates of each mesoscale block, and the reliability of the ridge frequency estimates of the mesoscale blocks, the ridge frequency field of each block included in the second quantity is reconstructed.

6. A fingerprint image feature enhancement device, characterized in that, The device includes: The block segmentation module is used to sample the fingerprint image to be processed at different scales to obtain multiple blocks, and to perform spectral response analysis on the polar coordinates of each block to obtain the polar coordinate power spectrum of each block. The frequency estimation module is used to determine the angular edge energy spectrum and ridge frequency estimate of each block based on the polar coordinate power spectrum. The orientation estimation module is used to normalize and fuse the angular marginal energy spectra at different scales in each block to obtain the fused orientation spectrum of each block, and to determine the orientation estimate of each block based on the fused orientation spectrum. The orientation reliability measurement module is used to measure the orientation estimation reliability of each block, obtain the orientation estimation reliability of each block, and count the first number of blocks with reliable orientation estimation. The orientation field reconstruction module is used to reconstruct the orientation field of each block included in the first number based on the first quantity, the orientation estimate of each block, and the reliability of the orientation estimate; The frequency reliability measurement module is used to measure the reliability of the ridge frequency estimation of each block, obtain the ridge frequency estimation reliability of each block, and count the second number of blocks where the ridge frequency estimation is reliable and the direction estimation is also reliable. The ridge frequency field reconstruction module is used to reconstruct the ridge frequency field of each block included in the second number based on the second quantity, the ridge frequency estimate of each block, and the reliability of the ridge frequency estimate. The reconnection module is used to continuously reconnect and enhance the broken ridges in the fingerprint image based on the orientation field and the ridge frequency field, and obtain the feature-enhanced fingerprint image when a preset stopping condition is met. The orientation field reconstruction module includes: The global orientation field export unit is used to export the global orientation field of a block with reliable orientation estimation through a preset orientation model. The observation residual determination unit is used to determine the observation residual of the corresponding block based on the orientation estimation of each pixel in the block with reliable orientation estimation and the global orientation field. The residual correction function determination unit is used to solve the energy functional with the residual correction function as a variable based on the direction estimation reliability of the block with reliable direction estimation and the observation residual, so as to obtain the residual correction function of the corresponding block. The orientation field reconstruction unit is used to reconstruct the orientation field of each block included in the first number based on the global orientation field and the residual correction function when the first number is greater than or equal to the first preset number. The orientation field determination unit is used to use a preset global orientation constant as the orientation field of each pixel in the fingerprint image when the first quantity is less than the first preset quantity.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the fingerprint image feature enhancement method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the fingerprint image feature enhancement method according to any one of claims 1 to 5.

9. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the fingerprint image feature enhancement method according to any one of claims 1 to 5.

Images